Copyright Undertaking



By reading and using the thesis, the reader understands and agrees to the following terms:

1. The reader will abide by the rules and legal ordinances governing copyright regarding the

use of the thesis.

2. The reader will use the thesis for the purpose of research or private study only and not for

distribution or further reproduction or any other purpose.

3. The reader agrees to indemnify and hold the University harmless from and against any loss,

damage, cost, liability or expenses arising from copyright infringement or unauthorized

usage.

IMPORTANT

If you have reasons to believe that any materials in this thesis are deemed not suitable to be

distributed in this form, or a copyright owner having difficulty with the material being included in

our database, please contact [email protected] providing details. The Library will look into

your claim and consider taking remedial action upon receipt of the written requests.

Pao Yue-kong Library, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong

http://www.lib.polyu.edu.hk

SAFETY OF PROFESSIONAL DRIVERS IN HONG KONG

TIANTIAN CHEN

PhD

The Hong Kong Polytechnic University

2021

The Hong Kong Polytechnic University

Department of Civil and Environmental Engineering

Safety of Professional Drivers in

Hong Kong

Tiantian Chen

A thesis submitted in partial fulfilment of the requirements for the degree

of Doctor of Philosophy

July 2020

CERTIFICATE OF ORIGINALITY

I hereby declare that this thesis is my own work and that, to the best of my knowledge

and belief, it reproduces no material previously published or written, nor material that has

been accepted for the award of any other degree or diploma, except where due

acknowledgement has been made in the text.

(Signed)

CHEN Tiantian (Name of student)

Dedicated with love and gratitude to the memory of my grandfathers,

Zhenxiong Chen and Winglam Chong.

Abstract

Safety of professional drivers is of great concern around the world, especially in the public

transit-oriented cities like Hong Kong. The higher rates of traffic violation rates, fatigue

driving, aggressive driving, as well as the increasing proportion of older drivers are the

main contributory factors to the crashes involving commercial vehicles. In this study,

safety of professional drivers in Hong Kong is assessed from the behavioural,

psychological, and empirical perspectives. Therefore, effective traffic control and driver

management measures can be implemented to improve the safety of professional drivers.

First, effects of driving time, age, traffic condition and road environment on the driving

performance of professional drivers are evaluated using the driving simulator approach,

with which the difference in the driving performance between professional and non-

professional drivers are considered. Results indicate that age-related impairments on

driving performance could be reduced by the driving experience and task familiarity of

professional drivers. Furthermore, two modified traffic conflict measures were used to

investigate the compensatory strategy and strategic adaptation of professional drivers. It

is found that, despite the longer brake reaction time of older drivers, the likelihood of

more severe traffic conflict of older drivers was lower, especially for professional drivers.

Second, attitudes of professional drivers towards the enforcement and penalties against

speeding violations are evaluated using a perception survey. A stated preference survey

approach is adopted to gauge the trade-off between enforcement strategies, penalty levels

and speed choice of the professional drivers. A panel mixed logit regression model is

adopted to account for the effects of unobserved heterogeneity. Results indicated that the

professional drivers are more sensitive to the increase in driving-offence points as

compared to monetary fines. Also, presence of a warning sign is effective in enhancing

speed compliance. Several demographic and employment characteristics, driving history

and perception variables also influence drivers’ choices of speed compliance.

Third, safety effects of the composition of commercial vehicles including taxis, public

buses, light goods vehicles, and heavy goods vehicles are assessed based on the integrated

traffic and crash data. A Bayesian random-parameter Tobit approach is adopted to

measure the relationship between explanatory factors and the overall crash rates by injury

severity. Results reveal significant increasing effects of the proportions of taxi, buses,

light goods vehicles on the overall crash risk. Additionally, a Bayesian multivariate Tobit

model is applied to identify possible risk factors to the crash rates across different vehicle

types. Results indicate that crash rates of private car and light commercial vehicle would

increase with the increase in average lane width and presence of on-street parking, while

such finding is not valid for heavy commercial vehicle. Moreover, intersection density

has significant increasing effect only for the crash rate of light commercial vehicle.

Overall, findings of driving simulator study, stated preference survey and crash risk

analysis of professional drivers should be indicative to the driver licensing, training and

education, enforcement, and driver management strategies of the authorities and transport

operators that can enhance the safety performance of professional drivers in the long run.

(498 words)

iii

Publications arising from the thesis

Referred Journal:

Chen, T., Sze, N. N., & Bai, L. (2019). Safety of professional drivers in an ageing

society–A driving simulator study. Transportation Research Part F: Traffic Psychology

and Behaviour, 67, 101-112.

Chen, T., Sze, N. N., Saxena, S., Pinjari, A. R., Bhat, C. R., & Bai, L. (2020). Evaluation

of penalty and enforcement strategies to combat speeding offences among professional

drivers: a Hong Kong stated preference experiment. Accident Analysis & Prevention, 135,

105366.

Chen, T., Sze, N.N., Chen Sikai, Samuel Labi (2020). Urban road space allocation

incorporating the safety and construction cost impacts of lane and footpath widths.

Journal of Safety Research, 75, 222-232.

Chen, T., Sze, N.N., Sharon Newnam, Bai, L. (2021). Effectiveness of the Compensatory

Strategy Adopted by Older Drivers: Difference between Professional and Non-

professional drivers. Transportation Research Part F: Traffic Psychology and Behaviour,

accepted.

Chen, T., Sze, N.N., Chen Sikai, Samuel Labi, Zeng, Qiang (2021). Analysing the main

and interaction effects of commercial vehicle mix and roadway attributes on crash rates

using a Bayesian random-parameter Tobit model. Accident Analysis & Prevention, under

review.

Chen, T., Sze, N.N., Zeng, Qiang (2020). Multivariate Tobit modeling of crashes

involving private cars, light commercial vehicles, and heavy commercial vehicles.

Accident Analysis & Prevention, to be submitted

Conference paper

Sze, N.N. & Chen, T. (2018) Factors affecting the driving performance of elderly

professional drivers: A driving simulator study. Paper presented at the 23

International

Conference of Hong Kong Society for Transportation Studies, 8-10 December, Hong

Kong.

Sze, N.N. & Chen, T. (2018) Temporal change in the driving performance of professional

driver: a driving simulator study. Proceedings of the 6

International Symposium on

Reliability Engineering and Risk Management, 31 May-1 June, Singapore.

Chen, T., Bai, L., & Sze, N. N. (2019, July). Factors affecting the severity of rear-end

conflicts: a driving simulator study. In 2019 5th International Conference on

Transportation Information and Safety (ICTIS) (pp. 1182-1187). IEEE. (Best paper

award)

Chen, T., Sze, N.N. & Bai, L. (2019) Effective enforcement and penalty strategies to

combat speeding offence of professional drivers. Paper presented at the 24

International

Conference of Hong Kong Society for Transportation Studies, 14-16 December, Hong

Kong.

Chen, T., Sze, N. N., Saxena, S., Pinjari, A. R., Bhat, C. R., & Bai, L. (2020). Evaluation

of penalty and enforcement strategies to combat speeding offences among professional

drivers: a Hong Kong stated preference experiment. Paper presented at the 99

Annual

Meeting of the Transportation Research Board (TRB 2020), 12-16 January, Washington,

DC.

Acknowledgements

First and foremost, I would like to express my most sincere gratitude to my chief

supervisor, Dr. Nang-ngai Sze, for his unreserved guidance and supervision. The priceless

time he spent with me on this research is gratefully appreciated. I have gained the once-

in-a-lifetime chance of PhD study and a lot of knowledge from him. He shows much

patience and kindness throughout my postgraduate study. He cares about the health and

well-being of the students and encourages us a lot when we face difficulties either in our

research or daily life. He has also provided me with valuable opportunities for improving

my research abilities and interpersonal skills. He is an excellent supervisor, a friend, a

family, and a life mentor to me. I would not be here without him.

I would like to deliver my heartfelt thanks to my co-supervisor, Prof. Anthony Chen, for

always sharing interesting ideas with me, encouraging me to expand my research area,

and providing professional advice and support to me. He is very meticulous and careful

in scientific research and he always cares about the student's future development.

Next, I must sincerely thank the members of my Thesis Examining Committee, including

Professor Yuhong Wang, who kindly chaired the committee during the COVID-19

pandemic, as well as the two examiners, Professor Becky Loo, and Dr. Shimul (Md.

Mazharul) Haque, who kindly provided invaluable suggestions and guidance before and

during the oral examination. Without their help, I can hardly improve my thesis to this

finalized version.

Also, I would like to thank Prof. Chandra R. Bhat (from the University of Texas at Austin)

for his valuable guidance and critical comments on some of my works. I have learned a

lot from him during our individual meetings. Many thanks to my collaborators, Dr. Abdul

Rawoof Pinjari and Mr. Shobhit Saxena (from the Indian Institute of Science), Dr. Sharon

Newnam (from Monash University), Dr. Lu Bai (currently at the University of Hong

Kong), Dr. Qiang Zeng (from the South China University of Technology), Dr. Sikai Chen

and Prof. Samuel Labi (from Purdue University), for their wisdom provided to enhance

the quality of the research works. My sincere thanks are also sent to my groupmates in

the transportation team of the Hong Kong Polytechnic University, for the fruitful

discussions over the past few years. I also wish to thank our laboratory technician Emily

Fung for her efficient assistance during my experimental works.

And of course, to my close friends, Man Lin and Jiayi Su, thanks for bringing me laughter

during my tough times and for our friendship since junior school.

Throughout my life, my mom and dad provide me with unwavering love and encourage

me to fight for my pursuits. Thanks, my dearest parents, Sihong Chen and Minmin

Zhuang, this journey would not have been possible without your support. After a long

day with the research works, there is nothing better to go home and chat with my brother

and grandmother. Indeed, I am greatly indebted to my beloved brother Wongwai Chan

and grandmother Honin Lam, for their patience, understanding and care.

Finally, I know I rarely mention two of my grandfathers. In fact, I do want to show my

sincere love to them at this juncture. My grandfather Mr. Chen passed away before I was

born. My father told me that he dreamed about grandpa before mom gave birth to me, and

in that dream, grandpa was taking a little girl home. Since I was a little girl, I have heard

a lot of stories about grandpa, such as how he loved to study, and how he faced life

positively. Since I came to Hong Kong for postgraduate study, my grandfather Mr. Chong

took very good care of me. However, he passed away before I received my PhD offer.

When I was applying for admission, my family meanwhile held a funeral for my grandpa

just beside PolyU. During these years of study, it is hard for me to walk through that street

as it evokes grieved memories. This thesis is dedicated with love and gratitude to the

memory of my grandfathers, and I strongly believe they would watch over me from the

stars.

Tiantian Chen

Jan 2021

 
vii 
Table of Contents 
Abstract ............................................................................................................................. i 
Acknowledgements ......................................................................................................... v 
List of Figures ................................................................................................................. xi 
List of Tables ................................................................................................................. xii 
Chapter 1 Introduction .................................................................................................. 1 
1  Background of this study ............................................................................. 1 
1.1  Driving performance of professional drivers ........................................ 2 
1.2  Attitudes and behaviours of professional drivers ................................. 3 
1.3  Crash risk of commercial vehicles ........................................................ 4 
2  Research aims and objectives ...................................................................... 5 
3  Thesis organization ...................................................................................... 6 
Chapter 2 Literature review .......................................................................................... 8 
1  Driving performance .................................................................................... 8 
1.1  Driving performance of professional drivers ........................................ 8 
1.2  Risk compensation and driving performance ..................................... 10 
1.3  Driving performance indicators .......................................................... 11 
2  Perceptions, attitudes, and behaviours ....................................................... 13 
2.1  Attitudes and behaviours of professional drivers ............................... 13 
2.2  Effect of driver age on attitudes and behaviours ................................ 14 
2.3  Driver’s perception towards traffic legislation ................................... 15 
3  Crash risk analysis ..................................................................................... 19 
3.1  Effects of road environment on crash risk .......................................... 19 
3.2  Crash risks of commercial vehicles .................................................... 20 
3.3  Effect of commercial vehicle mix ....................................................... 21 
3.4  Modelling crash counts and crash rates .............................................. 25 
4  Concluding remarks ................................................................................... 27 

 
viii 
Chapter 3 Evaluation of driving performance using driving simulator .................. 29 
1  Introduction ................................................................................................ 29 
2  Simulator experiment design ..................................................................... 30 
2.1  Participants .......................................................................................... 30 
2.2  Apparatus, driving scenario and test procedures ................................ 31 
3  Driving performance indicators ................................................................. 34 
4  Statistical method ....................................................................................... 37 
5  Results ........................................................................................................ 38 
5.1  Lateral stability ................................................................................... 45 
5.2  Stability of steering control ................................................................. 45 
5.3  Steering error ...................................................................................... 46 
5.4  Speed stability ..................................................................................... 46 
6  Discussion .................................................................................................. 47 
6.1  Effect of age on driving performance ................................................. 47 
6.2  Performance of professional drivers ................................................... 48 
6.3  Effect of other factors on driving performance .................................. 49 
7  Concluding remarks ................................................................................... 50 
Chapter 4 Evaluation of conflict risk using driving simulator ................................. 53 
1  Introduction ................................................................................................ 53 
2  Simulator experiment design ..................................................................... 54 
2.1  Participants .......................................................................................... 54 
2.2  Driving scenario and test procedures .................................................. 55 
3  Surrogate safety measures ......................................................................... 57 
4  Statistical method ....................................................................................... 58 
5  Results ........................................................................................................ 60 
5.1  Driving capability ............................................................................... 61 
5.2  Compensatory behaviour .................................................................... 61 
5.3  Safety effectiveness of the compensatory strategy ............................. 62 
6  Discussion .................................................................................................. 67 
6.1  Effective compensatory strategy of professional drivers .................... 68 
6.2  Strategic adaptation of professional driver ......................................... 69 

 
ix 
6.3  Study limitations ................................................................................. 70 
7  Concluding remarks ................................................................................... 71 
Chapter 5 Perceptions and attitudes of professional drivers .................................... 73 
1  Introduction ................................................................................................ 73 
2  Stated preference survey design ................................................................ 75 
3  Data collection and sample used ................................................................ 79 
3.1  Driver demographics and employment characteristics ....................... 82 
3.2  Driver history and safety perceptions ................................................. 84 
4  Statistical method ....................................................................................... 85 
5  Results and discussion ............................................................................... 88 
5.1  Effects of penalty level and enforcement strategy .............................. 93 
5.2  Effects of demographic characteristics of professional drivers .......... 95 
5.3  Effects of operational characteristics of professional drivers ............. 96 
5.4  Driver history and safety perceptions ................................................. 97 
5.5  Marginal effects due to changes in SP attributes ................................ 98 
6  Concluding remarks ................................................................................. 101 
Chapter 6 Effects of commercial vehicle mix and multivariate analysis of crash 
rates by vehicle type .................................................................................. 104 
1  Introduction .............................................................................................. 104 
2  Data collection ......................................................................................... 106 
2.1  Overall crash rates by injury severity ............................................... 106 
2.2  Crash rates by vehicle type ............................................................... 110 
3  Model specification and goodness-of-fit ................................................. 114 
3.1  Random parameter Tobit model ....................................................... 114 
3.2  Multivariate Tobit model .................................................................. 117 
3.3  Goodness-of-fit ................................................................................. 118 
4  Results and discussion on the effects of commercial vehicle mix ........... 120 
4.1  Effects of geometric factors .............................................................. 126 
4.2  Effects of Traffic flow and traffic control ........................................ 127 
4.3  Temporal effect ................................................................................. 128 
4.4  Main effects of commercial vehicle proportions (CVPs) ................. 128 

 
x 
4.5  Interaction effects of CVPs and roadway attributes ......................... 130 
5  Results and discussion on crash rates by vehicle type ............................. 132 
5.1  Effects of geometric factors .............................................................. 135 
5.2  Effects of traffic condition and time ................................................. 136 
6  Concluding remarks ................................................................................. 137 
Chapter 7 Conclusions and recommendations ......................................................... 140 
1 General conclusions ...................................................................................... 140 
2 Main findings and contributions .................................................................... 144 
3 Limitations ..................................................................................................... 147 
4 Recommendations for future research ........................................................... 148 
4.1  Conventional driving performance measured by simulator .............. 148 
4.2  Traffic conflict analysis using driving simulator .............................. 148 
4.3  Perception survey .............................................................................. 149 
4.4  Crash risk analysis for commercial vehicles ..................................... 149 
References .................................................................................................................... 151 
Appendix ...................................................................................................................... 180 
 
   

 
xi 
List of Figures 
Figure 3. 1 OKTAL CDS-650 driving simulator .................................................... 31 
Figure 3. 2 Typical scenarios for the simulated driving experiment ...................... 32 
Figure 3. 3 Illustration of the heading error θ ......................................................... 35 
Figure 3. 4 Driving performance of professional and non-professional drivers over     
time ....................................................................................................... 44 
Figure 4. 1 Typical simulated driving scenarios ..................................................... 56 
Figure 4. 2 Illustration of proposed driving performance indicators ...................... 58 
Figure 5. 1 Illustration of the location type and a hypothetical scenario for the 
stated preference game ......................................................................... 78 
Figure 6. 1 Study area (Hong Kong) showing the road segments studied and safety 
trends ................................................................................................... 108 
Figure 6. 2 Variation of commercial vehicle percentage by time of day .............. 108 
Figure 6. 3 Normalized overall crash rates by injury severity under different 
commercial vehicle mix ...................................................................... 110 
Figure 6. 4 Histograms of private car, light commercial vehicle, and heavy 
commercial vehicle crashes ................................................................ 113 
 
 
   

 
xii 
List of Tables 
Table 2. 1 Summary of past related work ............................................................... 24 
Table 3. 1 Summary of the participants of driving simulator study ....................... 31 
Table 3. 2 Distribution of the completed driving simulator tests ........................... 33 
Table 3. 3 Summary statistics for the simulated driving tests (N=1128) ............... 36 
Table 3. 4 Results of parameter estimates of the (overall) random intercept models 
 .............................................................................................................. 39 
Table 3. 5 Results of market segmentation analysis ............................................... 41 
Table 3. 6 Results of parameter estimates for professional and non-professional       
drivers ................................................................................................... 42 
Table 4. 1 Summary of participants of driving simulator study ............................. 55 
Table 4. 2 Summary statistics for the simulated driving tests ................................ 60 
Table 4. 3 Estimation results of random intercept models for driving capability .. 64 
Table 4. 4 Estimation results of random intercept models for compensatory 
behavior ................................................................................................ 65 
Table 4. 5 Estimation results of random parameter Tobit models for safety risk ... 66 
Table 5. 1 Distribution of speed choices by location type ...................................... 80 
Table 5. 2 Crosstabulation of SP attributes with speed choices at plain and warning 
sections .................................................................................................. 81 
Table 5. 3 Distribution of the sample ..................................................................... 82 
Table 5. 4 Parameter estimates of a panel MMNL model for the speed choice of 
professional drivers* ............................................................................. 90 
Table 5. 5 Marginal effects due to changes in the SP attributes ........................... 100 
Table 6. 1 Descriptive statistics of the variables .................................................. 109 
Table 6. 2 Descriptive statistics of the variables .................................................. 114 
Table 6. 3 Results of the goodness-of-fit tests ...................................................... 121 
Table 6. 4 Results of random-parameter Tobit model .......................................... 124 
Table 6. 5 Random parameter Tobit model for KSI crash rate ............................. 125 
Table 6. 6 Marginal effects results for the refined models ................................... 126 
Table 6. 7 Comparison of model performance ..................................................... 132 
Table 6. 8 Multivariate Tobit model error parameters and goodness-of-fit ......... 133 
Table 6. 9 Results of multivariate Tobit model .................................................... 134 

Chapter 1 Introduction

1.1 Background of this study

Economic losses attributed to road injuries cost the society around 3% of the national

income. The share of these losses contributed by road crashes involving commercial

vehicles are considerable (Health and Safety Commission, 2001; WHO, 2018). It is

because the crashes involving passengers (e.g. light bus, bus and taxi) and goods vehicles

do not only result in severe injury and death of the convicted drivers, but also those of

other innocent road users including passengers on the vehicles and pedestrians on roads

(Barua and Tay, 2010; Mooren et al., 2014; Meng et al., 2019). In Hong Kong, although

commercial vehicles merely constitute about 20% of total vehicle fleets, they are involved

in over 70% of road crashes. One possible reason may be the high mileages of commercial

vehicles (Transport Department, 2018a; Pei et al., 2012). However, safety of professional

drivers is also of great concern since the rates of professional drivers involved in crashes

and traffic violation are higher, compared to non-professional drivers (Chen et al., 2020;

Öz et al., 2010a; Wong et al., 2008).

The road-based public transportation modes in Hong Kong primarily include a regular

bus mode (operated either publicly or privately), a light bus mode (or mini-bus mode that

typically carries up to 19 passengers, again operated publicly or privately), and taxis

(while the taxi mode may not be traditionally viewed as a public transportation mode, it

is not uncommon in Hong Kong for the use of taxis to access bus stations and rail stations,

making it an integral component of public transportation use in the country). The

substantial dependence on the road-based public transportation modes contribute to, on a

per capita basis, a low vehicle miles of travel (VMT) in Hong Kong. This low exposure,

along with low speeds (due to high vehicle densities) and the protective cushion offered

by large buses, has resulted in a relatively low number of crashes in Hong Kong,

especially those resulting in serious injuries/death. In particular, there were 108 fatalities

and 2,214 individuals seriously injured in road traffic crashes in 2017 (Transport

Department, 2017). Based on a population estimate of about 7.4 million in Hong Kong in

2017, this translates to a per capita fatality rate of 14.6 deaths per million population

(relative to, for example, 28 road traffic fatalities per million population in the UK and

107 deaths per million population in the US).

Clearly, Hong Kong’s traffic safety record, at least on a per capita basis, is superior

relative to many other western nations. However, an issue of concern in Hong Kong is

that, unlike many western countries, a vast majority of the vehicles being driven on the

roads are by professional drivers (interestingly, ride-hailing services have yet to be

legalized in Hong Kong, and, as indicated earlier, taxi rides are a common way to access

the road-based public transportation modes, in addition to walking; and taxi drivers are

carefully regulated in terms of licensing requirements). Thus, it is of concern in Hong

Kong that the crash involvement rate of public transport vehicles is seven times higher

than that of the private car (Transport Department, 2017). It certainly brings into spotlight

the safety performance of professional drivers and the licensing regulations in place for

such drivers. While professional driver-related crashes and the organization/travel culture

has been examined at some length in the west and the middle-east (for example, see

Mallia et al., 2015; Newnam et al., 2018; Öz et al., 2010a, 2010b; Rosenbloom and Shahar,

2007), there has been relatively little research into the causes and considerations

associated with professional driver-related crashes in the far-east. This is particularly

surprising, given that professional drivers make up more of the pool of overall drivers in

Hong Kong relative to the west and the middle east. Therefore, such an obvious research

gap existing for the professional drivers motivates us to assess their safety performance

from the perspectives of driving performance, attitudes, and crash risks.

1.1.1 Driving performance of professional drivers

The proportion of older drivers in the transport sector has been increasing because of the

ageing population, shortage of labor and economic incentives (Duke et al., 2010). In Hong

Kong, the percentage of the population aged 60 or above increased from 16.8% in 2008

to 23.6% in 2017 (Census and Statistics Department of HKSAR, 2017). Accordingly, the

percentage of the full driving license holders aged above 60 in Hong Kong increased from

8% in 2008 to 16% in 2017 (Transport Department of HKSAR, 2017). The effect of age

on driving performance is thus of increasing concern. Driving performance is recognized

to be deteriorated with age (Islam and Mannering, 2006; Shanmugaratnam et al., 2010).

It in turn increases the associated crash and injury risks (Hole, 2007). Increase in crash

risk of older driver is found to be associated with the degradation in physical, mental and

cognitive conditions of human being (Lundberg et al., 1998). These associations have

been examined using the driving simulator experiments. For example, older drivers

generally show a degraded neuropsychological performance, which is in turn associated

with the degraded lateral control performance (Andrews and Westerman, 2012;

Shanmugaratnam et al., 2010). Also, older drivers perceive greater mental workload from

driving (Cantin et al., 2009), and perform worse at controlling the vehicle simultaneously

than younger drivers (Bélanger et al., 2010). Despite that there is negative correlation

between age and driving performance, professional drivers tend to have better driving

skills since they have more on-road experience. It is controversial that whether the age-

related impairments on driving performance could be offset by the driving experience and

task familiarity of professional drivers (Andrews and Westerman, 2012). On the other

hand, older (non-professional) drivers tend to drive less and the reduction in exposure

could be a more significant factor to crash risk, compared to driving experience and task

familiarity (Tay, 2006, 2008). To our knowledge, the interaction between driver type (i.e.

professional driver or not) and age on the driving performance is still unclear, and

therefore it is of our great interest.

1.1.2 Attitudes and behaviours of professional drivers

Driver aggressiveness and prevalence of traffic offences are identified to be the factors

contributing to crashes involving professional drivers (Rosenbloom et al., 2007; Öz et al.,

2010a). It could be attributed to the fact that professional drivers have to withstand high

work and time pressure due to their job nature (Öz et al., 2013). More importantly,

professional drivers deliberate over the trade-off between traffic offence-penalties and

potential income (Rosenbloom et al., 2007). Yet, the attitude of professional driver could

be controversial. Studies also suggest that professional drivers show a high intention to

avoid risky behaviours due to the perceived social responsibility (Rohani et al., 2013;

Shams et al., 2011). In this case, research is urgently needed to better understand the

perceptions and attitudes of professional drivers towards traffic offences and therefore to

tackle the problem of their impaired safety performance.

1.1.3 Crash risk of commercial vehicles

Commercial vehicles in Hong Kong presented higher crash involvement rates over the

years, as compared with private cars (Transport Department, 2018a). Previous studies

reveal that increase in the overall proportion of commercial vehicle is associated with the

increases in severe crashes on motorways and at signalized intersections (Wong et al.,

2007; Xu et al., 2014). Despite that the significance of mixed traffic situation on road

safety performance has been established by earlier research (Dinu and Veeraragavan,

2011; Srinivas et al., 2007), study limitation comes from the availability of elaborated

traffic data for the proportions of various commercial vehicle types. Nevertheless, a recent

study found that increases in the proportion of private cars, as well as that of the medium

trucks, are associated with the increase in injury crashes on highway segments (Wen et

al., 2018). As such, taking the advantage of more detailed breakdown of the proportions

of various commercial vehicle types, this study aims to shed light on the safety effects of

the composition of commercial vehicles. On the other hand, numerous studies have

proposed a need to model crashes by type (e.g. injury severity, collision type,

transportation mode, road users), as the effects of risk factors vary across crash types

(Guo et al., 2019; Ulak et al., 2018; Wang et al., 2017; Lee et al., 2015). However, studies

considering the crashes by vehicle type have been so far scanty. The only study

simultaneously estimating crashes by vehicle type (larger trucks and passenger cars) was

by Dong et al. (2014), suggesting that understanding the effects of geometric factors on

crash frequencies by vehicle type is of significance. In order to develop effective safety

measures for different commercial vehicles, identifying possible risk factors to crashes of

different vehicle types is therefore a must.

This study attempts to assess safety of professional drivers from the behavioural,

psychological, and empirical perspectives. Five research questions are therefore proposed

here: 1) whether age-related impairments on driving performance can be reduced by the

driving experience and task familiarity of professional drivers, 2) whether the

compensatory strategies of older drivers are different between professional and non-

professional drivers, 3) how the penalty and enforcement strategies deter professional

drivers from traffic violations, 4) whether the relationship between commercial vehicle

proportions and crashes can be moderated by roadway attributes, and 5) whether the

effects of risk factors vary across crashes categorized by vehicle type.

1.2 Research aims and objectives

This study aims to assess safety of professional drivers in Hong Kong from the

behavioural, psychological, and empirical perspectives. It is of great importance to assess

the driving performance of professional drivers, to evaluate their attitudes towards

legislation and law enforcement, to investigate the safety effects of commercial vehicle

mix, and to identify risk factors across crashes by vehicle type. Driving simulator

experiments, perception survey, and crash risk analysis are employed to achieve the aim.

The specific objectives of this study are as follows:

1) Driving performance

⚫ To examine the effects of driving time, age, traffic condition and road environment

on the driving performance of professional drivers, with which the difference in the

driving performance between professional and non-professional drivers is considered.

⚫ To evaluate effects of driving time, age and traffic condition on driver’s conflict risk

using surrogate safety measures, with the concern of difference between professional

and non-professional drivers.

⚫ To explore the relationship between traffic conflict risk and compensatory behaviour

of professional drivers

2) Perception evaluation

⚫ To investigate the perceptions and attitudes of professional drivers towards the

enforcement and penalties using a stated preference survey approach.

3) Crash risk analysis

⚫ To measure the association between the commercial vehicle proportion and crash rate

and to examine the mediating (moderating or magnifying) effects of roadway

attributes on this association.

⚫ To measure the relationships between possible risk factors and the crash rates of

different vehicle types using multivariate approach, with which correlations between

crash rates across vehicle types are considered.

It is expected that findings from this study would support the decision making of transport

operators regarding the driver recruitment and management, enhance the current

understanding and effectiveness of penalties and speed-enforcement strategies, and

provide useful insights into relevant countermeasures (that can enhance the safety culture

and awareness of professional drivers and crashworthiness of commercial vehicles).

Therefore, safety performance of professional drivers can be improved in the long run.

1.3 Thesis organization

Chapter 2 reviews the literature on various aspects of driver safety studies, including

driving performance and behaviours, driver perception and attitudes, as well as the crash

risk of commercial vehicles.

Chapter 3 assesses the driving performance of professional drivers using a driving

simulator study. Effects of driving time, age, traffic condition and road environment on

the driver’s speed, lateral and steering control performances are revealed. Difference in

contributory factors to driving performance between professional and non-professional

drivers is discussed. Interaction between age-related impairments on driving performance

and task familiarity of professional drivers is also explored.

Chapter 4 focuses on the professional drivers’ traffic conflict risk. Two surrogate safety

measures: time exposed time-to-collision (TET) and time integrated time-to-collision

(TIT) are adopted to indicate the risk of more severe rear-end traffic conflict in the car-

following tasks. Other performance indicators include the brake reaction time (BRT),

lateral control, average driving speed, and time headway. A driving simulator study is

used to evaluate the effects of driving time, age, and traffic condition on the risk of rear-

end conflict. Furthermore, relationship between traffic conflict risk and compensatory

behaviour of professional drivers is discussed.

Chapter 5 investigates the perceptions and attitudes of professional drivers towards the

enforcement and penalties in Hong Kong to deter speeding. A stated preference survey is

used to gauge the trade-offs among enforcement strategies, penalty levels and speed

choice of the professional drivers. Also, effects of factors including driver demographic,

socioeconomics, driving experience and crash record are considered.

Chapter 6 measures the relationship between the proportions of different commercial

vehicle types and the overall crash rates by injury severity, with which the confounding

factors including road geometry, traffic control and time period are considered. Also, risk

factors affecting the crash risks of different vehicle types are identified using multivariate

analysis, which accommodates possible correlations between crash rates across vehicle

types.

Chapter 7 concludes the study with a summary of the findings, implications, limitations,

and future research directions.

Chapter 2 Literature review

This chapter reviews the literature on driver safety from the behavioural, psychological,

and empirical perspectives. Section 2.1 discusses the driving performance of professional

drivers, contributory factors to driving performance, compensatory driving, and the

indicators that commonly used to measure the performance. Section 2.2 reviews the

literature with respect to drivers’ perceptions, attitudes, and behaviours. Previous findings

regarding the attitudes and behaviours of professional drivers and elderly drivers are

presented first. Then, drivers’ perceptions towards traffic legislation including penalty

and enforcement are discussed. Lastly, Section 2.3 reviews the crash risks of commercial

vehicles, safety effects of the percentages of various commercial vehicle types, and

crashes modelled by types.

2.1 Driving performance

2.1.1 Driving performance of professional drivers

Hong Kong is a city with high population density and limited road space. The ability of

public transport to serve high density cities well, as well as the relatively high costs of

private vehicle ownership and high operational costs (especially parking costs) resulting

from the limited road space, has resulted, in Hong Kong, in the dominance of public

transport as the primary mode for work-related as well as non-work travel. Of particular

note is the relatively seamless integration of road-based and rail-based metro public

transportation services in Hong Kong, with transfers between the two broad modes of

public transportation commonplace. Overall, over 90% of commute trips as well as over

46 % of non-commute trips in the territory are undertaken by road-based and/or rail-based

public transport (Transport and Housing Bureau, 2017; Transport Department, 2014).

However, crash involvement rates (per million vehicle-km) of commercial vehicles,

particularly taxi, light bus and bus, are higher than that of the private car in Hong Kong

(Transport Department of HKSAR, 2017). It certainly brings into spotlight the safety

performance of professional drivers.

Indeed, driving under the influence of fatigue has been a significant safety issue (Bunn et

al., 2005; Duke et al., 2010; Rosenbloom and Shahar, 2007). Professional drivers are more

vulnerable to the fatigue as they have to drive for longer time, as compared to the general

drivers. Also, aggressive driving behavior (attributed to desire for higher revenue and

expectation from the customers/employers) can contribute to higher crash rates

(Matthews et al., 1999; Sullman et al., 2002; Kontogiannis, 2006, Öz et al., 2010a). In

addition, increase in the exposure of professional drivers is also associated with the

increase in crash involvement rate. Regarding the difference in the behaviors between

professional and non-professional drivers, one possible factor is the vehicle ownership.

Professional driver who does not own the vehicle may have a higher propensity of

committing convicted driving behavior. Hence, moral hazard may occur since the driver

(who is not the owner) has less incentive to avoid any risky event (Tay and Choi, 2016).

However, professional drivers are believed to have better driving skills (Andrews and

Westerman, 2012; Borowsky and Oron-Gilad, 2013). It is therefore crucial to assess the

driving performance of professional drivers (i.e. whether it is better than that of non-

professional drivers), and the possible contributory factors.

Moreover, ageing population is now a common problem faced by many countries or

regions because of the reduction in fertility rates and increased life expectancy. By 2035,

proportion of population of age above 65 in Hong Kong would reach 25% (Sze and

Christensen, 2017). In recent years, of the peoples who hold the valid driving licenses,

percentage of elderly has been increasingly rapidly in other ageing societies (Newman et

al., 2018, 2019). In Hong Kong, proportion of drivers who held valid public transport

vehicle (e.g. taxi, light bus and bus) driving license of age above 60 was 37-46% in 2017

(Lee, 2018). Prevalence of older drivers in the transport industry can be attributed to the

issues including labour shortage, lack of social welfare and seeking of social engagement

(Duke et al., 2010; Navarro et al., 2007). However, it was recognized that cognitive

performance could be deteriorated when driver age increased, and the potential crash risk

of elderly driver could be higher than that of the younger counterpart (Hole, 2007; Islam

et al, 2006; Shanmugaratnam et al., 2010). Performance deterioration of older drivers can

be the results of audio and visual loss and extended perception-reaction time (Yan et al.,

2005; Yan and Radwan, 2006). However, the impacts on road safety because of the ageing

population and prevalence of older drivers in the transport sector have not been

thoroughly investigated.

2.1.2 Risk compensation and driving performance

Despite of the age-related deterioration, some older drivers, especially the professional

drivers who spend more time on roads, can still demonstrate satisfactory driving

performance. Then, no evidence can be established for the elevated crash risk of older

drivers in some studies (Braitman et al., 2007; Langford et al., 2006). Additionally,

satisfactory driving performance of older drivers could be attributed to self-regulation.

Older drivers might drive more cautiously and avoid driving under the adverse conditions,

such as traffic congestion, peak hours, high speed roads, bad weather and poor visibility

conditions. Modification of driving behavior that offsets the perceived risk attributed to

deteriorated physiological, cognitive and driving performances is known as compensatory

strategy (Lyman et al., 2001). Compensatory strategies are prevalent for older drivers who

had known cognitive impairment, traffic violations and crash involvement records

(Charlton et al., 2006; Molnar et al., 2008). For instances, older drivers may drive at a

lower speed, maintain a longer headway with leading vehicle, and avoid complicated

roads and maneuvers (Andrews and Westerman, 2012; Merat et al., 2005). Hence, the

potential crash risk of older drivers could be reduced by the compensatory strategies (De

Raedt et al., 2000; Molnar et al., 2008).

Moreover, professional drivers have higher capability to identify road hazards, and a brief

perception training intervention can improve the hazard perception skills of such

experienced drivers (Borowsky and Oron-Gilad, 2013; Li et al., 2015). Also, professional

drivers can have better risk anticipation and quicker response to potential road hazards

(Damm et al., 2011; De Craen et al., 2008). Consider the above, it is necessary to evaluate

the difference in the compensatory strategies between older professional and non-

professional drivers.

On the other hand, fatigue is one of the leading causes of crashes involving professional

drivers (Duke et al., 2010; Meng et al., 2015). Professional drivers often need to drive for

long distance and extended period per trip and/or per day (Öz et al., 2010a; Williamson

and Boufous, 2007). Additionally, fatigue can also increase the risk of fatality and severe

injury of professional drivers (Bunn et al., 2005). However, some researchers argued that

drivers could self-detect the occurrence of fatigue, and accommodate the impairment

while driving (Filtness et al., 2012; Williamson et al., 2014; Meng et al., 2015). For

example, drivers would slow down when they feel tired to mitigate the potential crash

risk attributed to driving under the influence of fatigue (Williamson et al., 2002). Strategic

adaptation refers to the intentional modification of driving behavior to adapt for the

impairment or driving under the influence of distraction, therefore driving and safety

performance can be optimized. Considered the experience in prolonged driving and high-

demand situations, it is expected that strategic adaptation of professional drivers is more

prevalent, and the elevated crash risk of professional drivers after prolonged driving can

be marginal.

2.1.3 Driving performance indicators

Driving simulator experiment is a safe and cost-effective approach to evaluate the driving

performance. In particular, the effects of road design and traffic condition on the driving

performance could be assessed in a controlled manner (Boyle et al., 2010; Lee et al.,

2003). Using a driving simulator, Otmani et al. (2005) found that sleepiness of male

professional drivers increases over time during the 90-min simulated driving task.

Consistently, Oron and Ronen (2007) indicated that fatigue of truck drivers was detected

through their degraded steering performance over time. Indeed, fatigue impaired driver’s

capability in terms of vehicle control and collision avoidance through withdrawing

driver's attention to the road and traffic condition progressively (Brown, 1994).

Measures of driving performance are commonly used to detect driver’s fatigue or

alertness (Brookhuis and De Waard, 1993; Brown, 1997). For example, the impaired

lateral control of the vehicle was revealed strongly associated with the driver fatigue

resulted from prolonged driving (Van der Hulst et al., 2001; Boyle et al., 2008). Standard

deviation of vehicle speed (SD-SPEED), standard deviation of the lateral position

(SDLP), standard deviation of steering wheel angle (SDSWA) reflect driver’s

longitudinal, lateral and steering control of the vehicle, which have been frequently

adopted in simulator studies (Boyle et al., 2008; Cantin et al., 2009; Shanmugaratnam et

al., 2010; Li et al., 2016).

In addition to conventional driving performance indicators, traffic conflict, as a road

safety surrogate measure, provides rich information for road safety assessment. A traffic

conflict was defined as “… an observable situation in which two or more road users

approach each other in space and time to such an extent that there is a risk of collision if

their movements remain unchanged” (Hydén, 1987). It has been revealed that driving

simulator could be an efficient alternative for traffic conflict analysis (Chang et al., 2019;

Yan et al., 2008). Using a driving simulator, near-departure events of drivers could be

identified based on the corresponding thresholds of the selected indicators (e.g. lateral

distance to departure, time to departure). Then, the expected number of lane departures

could be estimated (Tarko, 2012). On the other hand, critical events (e.g. pedestrian

crossing; car incursion; sudden brake by the leading vehicle) can be incorporated into the

driving simulator experiment to assess the driver performance (Andrews Westerman,

2012; Bélanger et al., 2010). Meanwhile, the effects of driver characteristics on the driving

performance during the critical situations can be evaluated. For example, elderly drivers

tend to show impairments on driving performance when facing a challenging event that

required synchronous reactions (Bélanger et al., 2010).

Time to collision (TTC) is one of the most widely used conflict indicators. It is defined as

the time left before two vehicles collide when there is no evasive maneuver to avoid a

collision (Hayward, 1972). The minimum TTC (TTC

min

) value is the shortest time span

required for a driver’s perception and reaction to prevent a collision from happening.

TTC

min

can be used to distinguish the conflict events once it is less than the predetermined

threshold (i.e. 1.0 s to 5.0 s) (Autey et al., 2012; Sayed et al., 2013; Zheng et al., 2014).

Time exposed time-to-collision (TET) and time integrated time-to-collision (TIT) are two

modified indicators based on TTC. When using TET and TIT for safety analysis, a TTC

threshold should be selected to differentiate the critical conditions from the safe

conditions. TET is defined as “… a summation of all moments (over the considered time

period) that a driver approaches a front vehicle with a TTC value below the threshold value

TTC” (Minderhoud and Bovy, 2001). In other words, TET reflects the duration of safety-

critical situation, given that the TTC values are lower than the threshold. TIT is defined as

“… uses the integral of the TTC profile of drivers to express the level of safety (in s

)”

(Minderhoud and Bovy, 2001). TIT considers both the duration of safety-critical situation

and the absolute difference between TTC value and the threshold. Other common conflict

indicators include post encroachment time (Cooper 1984), time to accident (Hydén 1987),

headway (Vogel 2003), braking time (Lu et al. 2012), etc.

2.2 Perceptions, attitudes, and behaviours

2.2.1 Attitudes and behaviours of professional drivers

A driver’s crash risk was revealed to be correlated with his/her perceived stress caused by

the road traffic (Matthews et al., 1999). Professional drivers are more likely to experience

driver stress, which contributes to the prevalence of driving aggression and traffic violation

(Öz et al., 2010a; Kontogiannis, 2006; Westerman and Haigney, 2000). Wong et al. (2008)

revealed that minibus drivers in Hong Kong tend to drive aggressively and violate traffic

rules due to the keen market competition. Machin and De Souza (2004) also found that

taxi drivers with aggressive personality reported more frequent risky behaviors. Sullman

et al (2002) examined the relationship between aberrant driving behaviors (including three

categories, i.e. error, lapse and violation) and crash involvement of truck drivers in New

Zealand. Results suggest that violation behavior of truck drivers was the only significant

predictor to their crash involvements.

Indeed, compared with non-professional drivers, risky driving behaviors are more

prevalent among professional drivers. One possible reason is the difference in risk

perception between professional and non-professional drivers. Due to the higher exposure

on roads, professional drivers are more familiar with the traffic hazards and therefore

perceive certain situations less risky (Öz et al., 2010a). Rosenbloom and Shahar (2007)

reported that professional drivers may sometimes regard committing minor offences as a

possible way to increase their income. For example, professional drivers may earn more

money by driving faster or stopping in restricted area to pick up more passengers. As a

result, professional drivers are often caught by minor traffic offences. However,

professional drivers are less likely to commit egregious risky behaviors (e.g. drink driving,

drug driving) than the non-professional drivers (Knipling et al., 2004), which could be

attributed to their greater sense of social responsibility (Paleti et al., 2010).

Referring to the theory of planned behavior (TPB) (Ajzen, 1991), attitude, subjective

norms, and perceived control are the three factors that determine the behavioral intentions.

While the behavioral intention is the best predictor of a person’s behavior. TPB has been

widely used to explain and predict human behaviors including driver and pedestrian

behaviors (Poulter et al., 2008). Applying the TPB model, Newnam et al. (2004) found

that drivers show lower speeding intention when driving a work vehicle compared with

his/her personal vehicle. This could be explained by the difference in subjective norms

(that is, attitudes of important others towards his/her behavior) between driving the work

vehicle and personal vehicle. It is suggested that subjective norms could have significant

effects on driver behavior particularly in a work environment. In addition, factor of

perceived control is revealed to affect the law-abiding behaviors of drivers. For example,

truck drivers are found more likely to comply with traffic regulations when they perceive

greater controllability over the work.

Moreover, professional drivers are usually regular employees of an organization, such as

logistic firms or bus companies. It is found that safety culture of the organization

significantly affects the attitudes of professional drivers towards risky driving (Öz et al.,

2010b, 2014). This could be attributed to the company’s driver management system. For

example, penalties would be issued to the drivers in accordance with the safety driving

policy because of their risky behaviors. In Hong Kong, the public bus companies

encourage the bus drivers to meet the driving safety requirements with monetary bonus.

While others who have the record of risky driving may need to attend additional training

and counseling courses. As a result, the organizational safety climate would help lower the

drivers’ intention to commit risky driving (Newnam et al., 2004).

2.2.2 Effect of driver age on attitudes and behaviours

Driving behavior can be predicted by attitudes (Ajzen, 1991), while risk perception

directly affect road safety attitude (Ram and Chand, 2016). Rosenbloom et al. (2008)

indicated that older drivers perceive higher risk while driving, which contributes to their

cautious and careful behaviors (Ram and Chand, 2016). Consistently, Tseng (2013) found

that elderly drivers tend to drive in a cautious and law-abiding manner. For example, they

are less likely to commit speeding violations compared with younger drivers. In addition,

involvements of aggressive driving were reported to decrease as the driver age increases

(Sullman et al., 2002).

Moreover, avoidance behaviour is found prevalent among elderly drivers (De Raedt and

Ponjaert-Kristoffersen, 2000; Molnar et al., 2008). For example, previous studies

revealed that elderly drivers tend to avoid driving under heavy-traffic condition, in peak

hours, on expressways, under the adverse weather and at night (Abdel-aty et al., 1998;

Ball et al., 1998; Charlton et al., 2006; Molnar et al., 2008). This is because they try to

adjust their exposure for a satisfactory driving performance or a lower crash risk

(Langford et al., 2006; Lyman et al., 2001). Besides, elderly drivers who are aware of

their functional and cognitive impairments often accommodate the driving task by

modifying driving behaviour (Lyman et al., 2001). For example, they tend to drive at a

lower speed or keep a longer headway with the vehicle in front to ameliorate their crash

risk (Ni et al., 2010; Shinar et al., 2005).

2.2.3 Driver’s perception towards traffic legislation

Earlier studies in other regions of the world, such as those referenced earlier, suggest that

driver aggressiveness, caused by high work and time pressure and resulting in a trade-off

deliberation between traffic offence-penalties and potential income gains from saved time

in the face of congested travel conditions, contribute to the high crash risk of professional

drivers (Öz et al., 2010a; Rosenbloom and Shahar, 2007). In particular, speeding has been

identified as a common aggressive driving behavior exhibited by professional drivers,

and speeding has also been identified in many earlier studies as being the single most

important factor impacting the occurrence and severity of roadway traffic crashes

(Fitzpatrick et al., 2017; Watson et al., 2015; WHO, 2018). In this context, in some OECD

countries, the proportion of drivers who self-report being guilty of excessive speeding is

as high as 80% (WHO, 2018). The same situation manifests itself in Hong Kong, with

speeding being one of the most common recorded traffic offences among professional

drivers and drivers at large. According to the number of prosecutions against traffic

offences in 2017, speeding accounted for over 42%, while red light running and drunk

driving accounted for 13% and 0.17% of the total number of prosecutions in Hong Kong,

respectively (Hong Kong Police Force, 2018). Admittedly, these statistics from Hong

Kong do not necessarily reflect the relative prevalence of speeding compared to other

illegal driving behaviors, because the statistics may simply be an indication of the type

and intensity of resources dedicated to enforcing speed limits relative to other illegal

driving behaviors. Even so, the very fact that more investment is made in preventing

speeding relative to other behaviors is in and of itself an acknowledgment that

countermeasures aimed at speed reduction are considered one of the most cost-effective

ways to enhance traffic safety.

Monetary fine, driving disqualification and imprisonment are the common penalties to

address and reduce speeding offence occurrences (as well as other driving offences; see

Hössinger and Berger, 2012; Li et al., 2014). In Hong Kong, the Driving-offence Points

(DOPs) system was introduced in 1984. Over 50 items of traffic offences carry DOPs in

addition to a monetary penalty. As would be logical, more DOPs and higher monetary

fines are issued as the level of speeding increases. Thus, a severe speeding offence (excess

of speed limit by more than 30 km/h but less than or equal to 45 km/h) incurs five DOPs

and HK$ 600 penalty (Transport Department, 2018b). Under this DOP system, persons

who have incurred 15 points or more within two years are disqualified from driving.

Some previous studies have revealed a significant negative correlation between the

monetary fine level imposed and penalty points, and the occurrence of traffic offences

(Hössinger and Berger, 2012; Li et al., 2014; Wong et al., 2008). For example, an increase

of fine by 10 Euros is associated with the reduction in speeding frequency by 5% among

Austrian drivers (Hössinger and Berger, 2012). However, there are studies suggesting that

monetary fine levels and penalty points alone have only a relatively minor deterrent effect

on the speeding offence (Elvik and Christensen, 2007; Fleiter et al., 2010; Langlais, 2008;

Ritchey and Nicholson-Crotty, 2011; Sagberg and Ingebrigtsen, 2018). Specifically, these

studies raise the issue of not only the level of the penalty on speeding deterrence, but the

risk of being subjected to that penalty (Kergoat et al., 2017; Li et al., 2014; Tay, 2009).

That is, the propensity for speeding depends on both the level of penalty as well as the

prevalence of speed enforcement operations, with some studies finding that the latter is

much more effective in curbing speeding offences than the former (see, for example,

Gargoum and El-Basyouny, 2018; Lawpoolsri et al., 2007; Ryeng, 2012; Truelove et al.,

2017). In other words, fines and DOPs penalty, according to these earlier studies, do not

function very well when the level of speed enforcement is not adequate (and thus the risk

of being subjected to the penalties is low). This finding also has backing in criminal

justice-based deterrence theory (Gibbs, 1985), which stems from the notion that

individuals effectively undertake a cost-benefit analysis of pursuing a “crime”, and the

effectiveness of a dissuasive mechanism originates from the costs being perceived as

higher than the benefits. The cost-benefit analysis itself is conducted within a frame of

three criteria: the certainty, celerity (swiftness or rapidity of imposition), and the severity

of a sanction. While the relative contributions of these three criteria may vary based on

the crime under question, lower “crime” activities (at least as viewed traditionally by

society, such as illegal driving behaviors) are typically dominated by the “certainty of

being apprehended” criterion in the cost-benefit evaluation of individuals (Høye, 2014;

Watson et al., 2015). In the context of speeding, this “certainty” criterion is directly

related to the level of enforcement of speed limits.

The automated speed enforcement camera (ASEC) system is generally considered as a

promising and cost-effective enforcement technique that increases the certainty of being

apprehended if speeding (Carnis and Blais, 2013; De Pauw et al., 2014a; Tay, 2009).

Once the cameras are installed, such systems obviate the need for more costly human

police patrols along roadways. Of course, some studies suggest that human police patrols

are still effective, when combined with ASEC systems, because many drivers feel

embarrassed when confronted by a fellow human (that is, a police person) who is

perceived as passing a judgment on one’s societal conduct. In addition, the fear of a verbal

reprimand by the police also can add to the embarrassment factor, elevating the

cumulative cost of being detained by a human police to be even higher than the fear of

risking one’s life or that of others through speeding (Kergoat et al., 2017; Silcock et al.,

2000). But drivers also understand that human agents, even if equipped with hand-held

radar/laser speed guns that provide accurate and reliable readings, can get fatigued over

long periods of time in terms of holding and directing the speed guns in appropriate

directions, and cannot have a consistent level of vigilance over extended periods of time,

leading to speeding event “misses” (see Kergoat et al., 2017). On the other hand, properly

functioning ASEC systems are more reliable in detecting speeding violations over

extended stretches of time. Even so, there is the issue of driver ability to dodge the dangers

posed by spatially fixed ASEC systems (that is, an ASEC with overtly announced camera

locations, as opposed to covert or unpublicized camera locations). In particular, according

to the integrative social-cognitive protection-motivation theory (PMT) (see Rogers,

1983), the effectiveness of a “threat” (that is, a speed enforcement mechanism in the

context of roadway speeding) is based both on threat appraisal (by way of the certainty,

celerity, and severity, as proposed by deterrence theory) as well as coping appraisal (that

is, the ability to cope with and dodge the danger). As an individual’s self-efficacy (the

ability to perform an action needed to dodge a threat) and the response efficacy (the

efficacy of the response to actually dodge the danger) increase, there will be less incentive

to not commit an offence based on a positive coping appraisal.

In the context of a spatially fixed ASEC systems, drivers typically perceive more

controllability and a positive coping appraisal (that is, a higher belief that they have the

capability to effectively dodge the speeding enforcement threat) by simply reducing

speeds in the immediate vicinity of the camera locations. This so-called “kangaroo effect”

(abrupt reductions close to camera locations and abrupt speed jumps upstream and

downstream of locations relatively removed from the camera range) has been well-

identified in earlier studies (De Pauw et al., 2014a, 2014b; Elvik, 1997; Marciano et al.,

2015). On the other hand, previous studies (see, for example, Cameron et al., 2003;

Dowling and Holloman, 2008) have shown the higher effectiveness of covert (or

unmarked and unpublicized) ASEC systems relative to fixed ASEC systems because of a

lower coping appraisal and higher uncontrollability to dodge a threat on the part of

drivers. However, such covert ASEC systems are not legally allowed in Hong Kong and

many other countries, both due to privacy regulations as well as the notion that ASEC

systems should be fundamentally aimed at preventing speeding rather than apprehending

offenders (Høye, 2014).

2.3 Crash risk analysis

2.3.1 Effects of road environment on crash risk

Environmental factors including road type and traffic flow condition are revealed to affect

the association between crash and possible risk factors. For examples, sleep-related road

crashes are more prevalent on the motorways, as compared to the urban roads (Horne and

Reyner, 1999; Maycock, 1996). Crash risks on the rural roads are also higher than that on

the urban roads because of the monotonous road environments and limited stimuli on

roads (Blower et al., 1993). For the traffic flow condition, increase in traffic volume and

presence of moderate traffic congestion are associated with the reduction in crash risk and

crash severity (Martin, 2002; Yau, 2004). The association between road environment,

traffic flow condition and crash risk could be attributed to the variation in driving

performance across different environments. For instances, variation in the steering and

lateral stability are associated with the complexity of driving task (e.g. reduced horizontal

and vertical curvatures, traffic interactions and roadside stimuli) (Thiffault and Bergeron,

2003; Jamson and Merat, 2005; Arnedt et al., 2005; Boyle et al. 2008; Teh et al., 2014).

Several studies have discussed the relationships between road lane width and crash

frequency and severity (Chen et al., 2017a; Wu et al., 2015; Pei et al., 2012; Gross and

Jovanis, 2007). Park et al. (2012) and Chen et al. (2017b) determined the extent to which

the traffic lane width, within a certain standard range, correlates negatively with the

frequency of different crash patterns or severity levels. An increase in the lane width by

1 ft could result in a 2% decrease in crash frequency (Abdel-Rahim and Sonnen, 2012).

Therefore, wider lanes generally promote vehicle safety. A few studies suggested that

beyond a certain limit, an increase in total traffic road width could possibly increase the

crash risk (Mohamed et al., 2013). Tulu et al. (2015, 2013) explained that this could be

due to the effects of lane width on drivers’ perception and driving speed.

With regard to the effect of road alignment, an increase in horizontal curve density is

revealed to be correlated to fewer casualties in previous research (Lamptey, 2004; Labi,

2011). A few past studies for major roads in Hong Kong had found little or no evidence

of relationships between road curvature and crash frequency (Zeng et al., 2016; Pei et al.,

2016, 2012), while Elvik (2019) indicated that increased number of curves for the road

segment could lower the crash rate.

2.3.2 Crash risks of commercial vehicles

In Hong Kong, over 46 % of non-commute trips and 90% of commute trips are made by

public transport. In particular, over 60% of public transport are road-based. They include

buses, light buses, and taxis (Transport and Housing Bureau, 2017; Transport

Department, 2014). This can be attributed to limited parking infrastructure, high density

development, high burden of private car ownership, and more importantly, availability of

economical, efficient and reliable public transport services (Chen et al., 2020). On the

other hand, as an important entrepot in Asia, freight logistics industry constituted to 3.2%

of Hong Kong’s GDP, provided more than 180,000 job opportunities, and accounted for

29% of service export in 2017 (Hong Kong Trade Development Council, 2019). The long-

term economic development, to a certain extent, depends upon the efficient and safety

movements of passengers and goods on Hong Kong roads.

Commercial vehicles in Hong Kong (including buses, light buses, taxis, goods vehicles)

merely constitute about 20% of total vehicle fleets. However, they are involved in more

than 70% of road crashes. This could be attributed to the higher exposure of commercial

vehicles on roads (Pei et al., 2012). On the other hand, safety of professional drivers

should be of great concern. Previous studies indicate that professional drivers tend to

involve in more road crashes and traffic violations compared with non-professional

drivers (Chen et al., 2020; Öz et al., 2010; Wong et al., 2008). In particular, crash

involvement of taxi drivers was positively associated with the increase in driver

workload, driving hours, and more aggressive driving behavior (Wang et al., 2019a,

2019b). Additionally, elevated crash risk of taxi was associated with the ageing of taxi

drivers, which their performances were more likely impaired by the deteriorating health,

fatigue and distraction (Chen et al., 2019a; Meng et al., 2017).

In Hong Kong, franchised bus and public light bus constituted to 49% and 24% of overall

road-based public transport patronage (Transport Department, 2014). Bus drivers are

usually more risk averse since they usually have stronger sense of social responsibility.

Also, they have lower tendency to commit traffic offense because of the better safety

management of operators (Paleti et al., 2010; Chen et al., 2020; Öz et al., 2010, 2013).

For example, all franchised bus companies in Hong Kong have comprehensive driver

surveillance, reward and penalty system to deter against speeding and other driving

offences and increase the safety awareness of bus drivers. However, safety of public light

bus has been of major concern in Hong Kong. Public light bus drivers are usually self-

employed, and their salaries are trip-based. They are more aggressive and are likely to

commit various driving offenses to compete for the business (Wong et al., 2008). Hence,

fatal and severe crash risk of public light bus-related crash is higher than that of franchised

bus (Transport Department, 2017a).

Goods vehicles in Hong Kong are classified into three categories - light, medium and

heavy, depending on the weight and dimensions. Light goods vehicle (also known as

‘light van’) drivers are self-employed and their salaries are trip-based, and medium and

heavy goods vehicle drivers are usually employed by the logistic firms and their salaries

are monthly or daily-based. Indeed, employment type can also affect the drivers’ attitudes

and behaviors. Light van drivers are usually more aggressive (Chen et al., 2020). Studies

also indicated that number of light trucks is positively associated with overall fatal crash

and pedestrian fatalities and severe injuries (Ballesteros et al. 2004; Desapriya et al.,

2010). Moreover, road environments and working conditions also affect the driving

performance of goods vehicle drivers. Driving under the influence of fatigue is prevalent

when a light goods vehicle driver is exposed to congested urban traffic environment for

extended period. In contrast, driving fatigue is prevalent when a heavy goods vehicle

driver is driving on monotonous rural highway (Friswell and Williamson, 2013). It is

necessary to estimate the effects of driver characteristics, road environment, traffic

condition and time period on the commercial vehicle crash risk. Therefore, effective

driver education and training programs can be introduced to enhance the safety awareness

and thus reduce the crash risk of commercial vehicle drivers.

2.3.3 Effect of commercial vehicle mix

The commercial vehicle proportion (CVP) refers to the ratio of commercial vehicles to

all vehicles in the traffic stream. Previous studies indicate that the relationship between

commercial vehicle proportion and crashes is influenced by the class of commercial

vehicle in question (Tay, 2003; Ballesteros et al., 2004; Desapriya et al., 2010).

Furthermore, even for a given class of commercial vehicles, it has been determined that

the direction of the effect of vehicle proportion on crash rate can vary with the roadway

feature type and crash severity level. For example, with regard to buses, Dinu and

Veeraragavan (2011) found that an increase in their proportion is associated with an

increase in night-time highway crashes, while Xu et al. (2014) determined that an increase

in the bus proportion is associated with a decrease in slight-injury crashes at intersections.

With regard to trucks, Wen et al. (2018) indicated that an increase in truck proportion is

associated with an increase in injury crashes at road segments, while Dinu and

Veeraragavan (2011) found that an increase in truck proportion is associated with a

decrease in night-time crashes. Dong et al. (2014) revealed that an increase in the

percentage of heavy trucks in traffic stream is associated with the increase in intersection

crashes.

It is interesting to note in the literature that a number of studies have used commercial

vehicle data to develop indicators of potential crash risk, and have discussed the policy

implications of doing this (Bao et al., 2019; Zhou and Zhang, 2019). Bao et al. (2019)

found that the spatial distribution of taxi trips exhibits a similar pattern with that of

crashes, and that locations with higher density of taxi trips are positively correlated with

those with high daytime crashes. However, the authors emphasized that the relationship

between taxi trips and crashes is non-monotonic, meaning that it can be moderated by

other environmental factors such as weather and land-use variables.

Indeed, the argument for the need to investigate the moderating effects of road attributes

on the CVP-crash relationship is rooted in the longstanding realization that the roadway

environment (physical and operational) profoundly influences the crash experience

(Zegeer at al., 1988; Hauer, 1988; Dumbaugh, 2006; Gross and Jovanis, 2007). Recently,

there has been research efforts that have thrown more light on the safety effects of road

environment features including geometric characteristics, ambient natural factors, the

nature of traffic flow, and types of traffic control facilities (Chen S. et al., 2019a; Zeng et

al., 2016, 2017b; Sze et al., 2019; Álvarez et al., 2020).

Interaction is present where two or more objects have an effect upon one another. In a

statistical model, an interaction is a term in which the effect of two (or more) variables is

not additive. In other words, the effect of factor A plus the effect of factor B is different

than the effect of factors A and B combined. Therefore, interaction effects refer to the

modification of the effect of one independent variable on the dependent variable due to

the presence of a second independent variable. Such modification may be a diminished

or mitigating effect (moderation) or an exacerbated effect (magnified). Failure to account

for such interaction effects could lead to poor model performance.

A large number of past research studies have addressed interactions between various road

crash factors without explicitly identifying or explaining any moderating or magnifying

effects of the interacting variables. Ahmed et al. (2012) indicated that the positive

association between steep grades and mountainous freeway crashes is magnified

significantly in the snow season, suggesting that the interaction of road geometry and

weather condition has a significant effect on crashes. Wen et al. (2019) found that an

increase in roadway vertical gradient generally contributes to reduced highway crashes;

however, the interactions between vertical gradient and weather variables (such as wind

speed and visibility) was found to have positive coefficients, suggesting that increase in

crash propensity with increased interaction (of vertical gradient, wind speed, and poor

visibility) which is intuitive. It seems clear that introducing interaction terms in a crash

prediction model indeed has several benefits including improvement of the model’s

goodness-of-fit and intuitiveness, identification of potential sources of heterogeneity

(Azimi et al., 2020), and quantification of the moderating effects of the roadway

environment on the relationships between crash frequency and any specific crash factor.

The conclusions of the Bao (2019) and similar studies lend credence to the notion that the

interaction effect (which can be considered as a third variable) potentially influences the

relationship between commercial vehicle proportion and crash risk.

However, there is relatively limited research on the safety effects of the interaction of

roadway attributes and commercial vehicle proportion. This can be considered a research

gap in the literature because (a) commercial vehicles are a critical aspect of urban

transportation and therefore, urban social and economic development, (b) commercial

vehicles have been found to contribute significantly to crashes, (c) roadway design and

operations attributes significantly influence crashes (d) unlike most crash factors,

roadway design and operations attributes are within the control of the city authorities.

Therefore, we feel that the stated research gap is very significant, from a practical

perspective.

Table 2. 1 Summary of past related work

Study

scope

Reference

Study

region/

period

Independent variables

Outcome

variable

Only

examined

main

effects

Tay, 2003

Queensland/

1997-2001

Commercial vehicle factors: Proportion of truck (+), Proportion of bus (-),

Proportion of van (+)

Other factors: season

Nr. of fatal

crashes

Wong et al.,

2007

Hong Kong/

2002-2003

Commercial vehicle factors: Proportion of overall commercial vehicles (+)

Geometric factors: Curvature (+), Average lane width (-), Presence of tram stops

(+)

Traffic flow & traffic control factors: AADT (IS), pedestrian flow (+)

Nr. of fatal

& severe

intersection

crashes

Dinu and

Veeraragavan,

2011

Chennai,

India/

2001-2003

Commercial vehicle factors: Proportion of bus (+), Proportion of trucks (-)

Geometric factors: Curvature (+), Segment length (+), Driveway density (+)

Traffic flow & traffic control factors: Hourly traffic volume (+)

Nr. of night-

time

highway

crashes

Xu et al.,

2014

Hong Kong/

2002-2003

Commercial vehicle factors: Proportion of commercial vehicles (-)

Geometric factors: Curvature (+), Four or more approaches (+), Presence of a

turning pocket (+), Nr of pedestrian crossings (+)

Traffic flow & traffic control factors: AADT (+)

Nr. of slight

intersection

crashes

Dong et al.,

2014

Tennessee/

2001-2005

Commercial vehicle factors: Proportion of trucks (+)

Geometric factors: Angle of intersection (-), Shoulder width (+), Nr. of left-turn

lanes (+), Roughness index (+), Rutting depth (+)

Traffic flow & traffic control factors: AADT (+), Speed limit (+)

Nr. of

intersection

crashes

Zeng et al.,

2017a

Hong Kong/

2002-2006

Geometric factors: Curvature (-), Average lane width (+), Nr. of lanes (+), Nr. of

merging ramps (-), Presence of median barrier (-), Presence of bus stop (+)

Traffic flow & traffic control factors: AADT (-), speed limit (-), Nr. of

intersections (-)

Other factors: Rainfall (IS)

Crash rates

of road

segments

Wen et al.,

2018

Guangdong,

China/ 2014

Commercial vehicle factors: Proportion of medium bus & medium truck (+),

Proportion of large bus & large truck (IS)

Geometric factors: Curvature (IS), Vertical gradient (IS), Part of a bridge (IS),

presence of ramps (IS)

Traffic flow & traffic control factors: daily vehicle-km traveled (+)

Highway

injury crash

frequency

Considered

interaction

effects

Wen et al.,

2019

Guangdong,

China/ 2014

Commercial vehicle factors : proportion of medium bus & medium truck (IS),

proportion of large bus & large truck (-)

Geometric factors: Vertical gradient (-), Curved road (IS)

Traffic flow & traffic control factors: Monthly vehicle-km travelled (+)

Interaction terms: Gradient × wind speed (+), Gradient × precipitation (-),

Gradient × visibility (+), Curve × precipitation (+)

Highway

injury crash

frequency

Zhao et al.,

2019

State of

Connecticut/

2011-2015

Geometric factors: Nr. of through lanes (-), Rural setting (+), Outside shoulder

width (+), Segment length (+), Inside shoulder width (-)

Traffic flow & traffic control factors: Monthly traffic volume (+)

Other factors: temperature (-), precipitation (+), wind speed (+)

Interaction terms: Nr. of through lanes × rural setting (-), Nr. of through lanes ×

monthly traffic (+), Nr. of through lanes × outside shoulder width (-)

Highway

injury crash

frequency

Zhai et al.,

2019

Hong Kong/

2015

Commercial vehicle factors: Goods vehicle (+), taxi (+), bus (+)

Geometric factors: One-way road (-)

Interaction terms: Raining × pedestrian jaywalking (+), Raining × Careless

driving (+), Raining × footpath overcrowded (-), Above 30 ℃ × driver inattention

(+), Above 30℃ × pedestrian run onto the road (+)

Severity of

pedestrian

crashes

Azimi et al.,

2020

State of

Florida/

2007-2016

Geometric factors: Dry and sand road surface (+), Unpaved shoulder (+),

Downhill grade (+), Curve right alignment (+)

Traffic flow & traffic control factors: Vehicle speed of 20 to 49 (mph) (+),

Vehicle speed of 50 to 75 (mph) (+)

Severity of

truck

rollover

crashes

Interaction terms: Vehicle speed of 20 to 49 (mph) × clear vision (-), Vehicle

speed of 20 to 49 (mph) × Driver careless driving (+), Dark condition × driver

speeding (+), Dark condition × fog weather (+)

Note: (direction of the parameter: (+)positive; (-)negative; (IS)examined but not statistically significant

2.3.4 Modelling crash counts and crash rates

To address the issue of excessive zeros in crash count observations, earlier studies applied

the zero-inflated Poisson and negative binomial models (ZIP and ZINB), assuming a

dual-state generating process of crash data (Shankar et al., 2003; Qin et al., 2004).

Mannering and Bhat (2014) reviewed that zero-inflated models have been the most

popular model to handle the excessive zeros, and they are still frequently used in recent

research (Raihan et al., 2019; Gu et al., 2020). However, this approach has also been

criticized by Lord et al. (2005), Lord et al. (2007) and Lord and Mannering (2010),

suggesting that zero-inflated crash count models can create theoretical inconsistencies

regarding the source of the predominance of zeros. For example, instead of being

generated from inherently safe sites, excessive zeros can be cause by the roadway entities

with low vehicle exposure and high heterogeneity, short data collection time, small spatial

scale, underreported crashes, etc. Though many studies tried to avoid excessive zeros

using the aggregated analysis, it creates another problem of ignoring the time-varying

factors (Behnood and Mannering, 2019; Mannering, 2018). In this context, Pei et al.

(2016) developed a highly disaggregated model that incorporated the variables of year,

day of the week, and time of day. To accommodate the predominance of zero-crash

observations, the authors adopted a bootstrap resampling approach while still applying a

traditional negative binomial model. Nonetheless, the computation time would increase

remarkably when generating large bootstrap samples to reduce the effects of random

sampling errors (Efron and Tibshirani, 1994; Efron, 2014).

On the other hand, modelling the crash rate (e.g. crash count per million vehicle

kilometres travelled) is now advocated as an alternative to conventional crash count

modelling. Crash rate analysis has its advantages such as providing a standardized

measure of the safety performance of roadway entities (Xu et al., 2014; Zeng et al., 2017a,

2017b). Tobit regression approach is commonly applied to address the left-censoring

problem at zero of crash rates since many roadway entities can have zero crash record.

Using a multivariate random parameter Tobit model, Ulak et al. (2018) estimated the

crashes involving elderly (aged above 65) drivers, passengers, bicyclists, and pedestrians.

This study adopted the same set of explanatory variables for different crash types to gain

comparable results. It is revealed that there are positive correlations between crashes by

types of elderly road users, except between pedestrians and bicyclists. Results also

indicate that the effects of risk factors on crashes vary by road user type. For examples,

increase in density for aging population would increase the crash rates of elderly drivers

and passengers, but decrease the crash rates of elderly bicyclists and pedestrians.

Additionally, increase in traffic flow would decrease the crash rates of elderly drivers,

passengers, and bicyclists, while it has no significant effect on elderly pedestrians.

Similarly, Guo et al. (2019) looked into crash rates by collision types such as rear-end,

sideswipe, and angle crashes at freeway diverge areas. Results show that effects of

contributing factors on crash rates vary across collision types.

In the context of substantial differences in vehicle feature and driver behaviour across

vehicle types, the effects of explanatory variables on crashes can vary across vehicle

types. Indeed, rather than simply predicting the overall crash counts or crash rates in either

a microscopic (roadway entities, e.g. road segments, intersections) or a macroscopic level

(spatial units, e.g. traffic analysis zones), recent studies have investigated the crash

counts/rates by crash types using multivariate analysis (Lee et al., 2015; Li et al., 2015;

Wang et al., 2017; Ulak et al., 2018; Alarifi et al., 2018; Guo et al., 2019). It has been

proven by previous research that there could be strong correlations between crashes by

types within each study unit. Lee et. al (2015) applied a multivariate Poisson log-normal

regression approach to simultaneously investigate the motor vehicle, bicycle, and

pedestrian crashes. Shared unobserved factors across crashes by different transportation

modes were captured by the multivariate model, in which positive correlations were

found. Moreover, significant explanatory factors to motor vehicle, bicycle, and pedestrian

crashes are different. For examples, ‘proportion of household without vehicles’ is

significant only for pedestrian crashes, while ‘proportion of roadway with speed limit less

than or equal to 20 mph’ is significant only for bicycle crashes. Similarly, Wang et al.

(2017) investigated crash counts by severity level (i.e. property damage only, non-

incapacitating injury, and fatal and incapacitating injury) and collision type (i.e. same-

direction, intersecting-direction, opposite-direction, and single-vehicle crashes).

Multivariate Poisson log-normal model was used to accommodate the unobserved shared

effects across severity levels and collision types. More importantly, results of the

parameter estimates indicate that the coefficient for a risk factor can have different signs

(positive or negative) across crash types. As such, looking into crashes by type would

provide more useful insights since it identifies different contributory factors across crash

types.

To summarize, these studies justify a need to consider variations in crash counts or crash

rates by crash type (e.g., severity, transportation mode, collision type). It is suggested that

modelling crashes by type could be more favourable when identifying crash contributing

factors. However, estimating crashes by type using separated prediction models has its

limitation due to the assumption of independence in crash counts/rate across crash types.

Univariate models ignoring possible correlation resulted from common unobserved

factors across crash types within each site would have degraded modelling performance

and biased parameter estimation, compared with multivariate models. As such, due to the

differences in driver behaviour and crashworthiness among vehicle types, there is a need

to simultaneously model the crashes by vehicle type to identify risk factors while

accommodating possible correlation between crash types. In this way, effective

countermeasures can be tailored to enhance the safety performance of a specific vehicle

type.

2.4 Concluding remarks

This chapter demonstrates the results of the literature survey on driver safety assessment

studies. In particular, previous findings in terms of the professional drivers’ driving

performance, perception, attitude, behaviour and the corresponding crash risk are

presented. Conventional driving performance indicators and surrogate safety measures

are introduced. The phenomenon of increasing elderly drivers in the transport sector and

its possible negative effect on the safety performance of professional drivers are also

discussed. For the perceptions and attitudes of professional drivers, contributory factors

revealed in previous studies and their corresponding effects are explored. Deterrence

theory to measure driver’s perception towards traffic legislation, protection-motivation

theory, as well as the theory of planned behaviour are reviewed. This chapter also

discussed crash risks of professional drivers by commercial vehicle types and safety

effects of commercial vehicle mix.

There are several research gaps identified in the literature because (a) interaction between

better driving skill of professional driver and impaired driving performance of elderly

drivers is still unclear; (b) for the effectiveness of penalty and enforcement strategies,

there has been relatively little research focusing on the perception and attitude of

professional drivers; (c) while the compensatory strategy adopted by older general drivers

has been studied in some depth, there is little work that researches into the compensatory

behavior and its safety implications of older professional drivers; (d) interactions between

road safety factors, including road user behaviour, weather, and road geometric factors,

have been investigated in past research. However, this has rarely been done for the

proportion of commercial vehicles and roadway features. While consideration of the

interactions between these two specific factors may be a contribution that seems to be

only incremental, it is important to address this gap in the literature because both roadway

features and commercial vehicle proportion have been found to be significant factors of

urban road crash propensity; (e) last but not least, research is urgently needed to estimate

crashes by vehicle type simultaneously to better understand different effects of

contributing factors. Therefore, effective countermeasures can be tailored to tackle the

higher crash risk of commercial vehicles. These deficiencies are of great concern, given

that professional drivers make up more of the pool of overall drivers in Hong Kong. It

provides us with the motivation to fill the research gap.

Chapter 3 Evaluation of driving performance using driving

simulator

3.1 Introduction

Professional drivers, due to their work nature, spend much longer time on roads than the

non-professional drivers, which may contribute to their better driving skill. However,

fatigue and risk-taking behaviours are also prevalent among professional drivers, which

could increase the associated crash injury risk (Bunn et al., 2005; Duke et al., 2010;

Rosenbloom and Shahar, 2007). On the other hand, the proportion of elderly drivers in

transport sector has been increasing because of the ageing population, shortage of labor

and economic incentives (Duke et al., 2010). In Hong Kong, the percentage of the

population aged 60 or above has increased from 16.8% in 2008 to in 23.6% 2017 (Census

and Statistics Department, 2017). Accordingly, the percentage of the full driving license

holders who aged above 60 in Hong Kong increased from 8% in 2008 to 16% in 2017

(Transport Department, 2017a). The effect of age on driving performance is of increasing

concern since driving performance was recognized to be deteriorated with age (Islam and

Mannering, 2006; Shanmugaratnam et al., 2010).

This chapter aims to explore the interaction between age-related impairments on driving

performance and task familiarity of professional drivers using a driving simulator

approach. Two hypotheses are thus proposed, (1) age-related impairments on driving

performance can be reduced by the driving experience and task familiarity of professional

drivers; and (2) contributory factors to driving performance of professional drivers are

different from that of non-professional drivers. Section 3.2 provides the details of

experimental design and procedures of driving simulator test. Section 3.3 and 3.4 describe

the driving performance indicators and method of analysis used in this study. The results

and implications will be discussed in Section 3.5 Eventually, Section 3.6 provides the

concluding remarks. It is anticipated that the results will be indicative to the dispatch

policies and driver management strategies of the transport operators. This is particularly

important to a compact and ageing society like Hong Kong, where the public transport

usage is very high.

3.2 Simulator experiment design

3.2.1 Participants

A total of 50 male drivers were recruited for the driving simulator study. The selection

criteria were: (1) holding a valid full driving license; (2) driving for at least 5 hours a

week; and (3) having good health condition. Of the 50 participants, 26 were professional

drivers and 24 were non-professional drivers respectively. For the professional drivers,

one must be a full-time driver of taxi, public light bus, bus, or goods vehicle. In the

subsequent analysis, the professional drivers will be stratified into two groups: (i)

passenger vehicle (i.e. taxi, light bus and bus) drivers and (ii) goods vehicle drivers. It is

because the difference in the experience and driving skills between vehicle types may

moderate the association between driving performance and possible factors. Age of the

participants ranged from 40 to 69 years. The participants were classified into two

categories by age: (i) “mid-aged” referred to the drivers of age from 40 to 55 years; and

(ii) “older” referred to the drivers of age from 56 to 69 years. Such classification was

consistent to that of previous study (Li et al., 2016). Informed consent of the participation

was obtained, and monetary reward was provided. US$50 and US$25 were paid to the

professional drivers and non-professional drivers respectively for the participation. All

participants were required to have a good rest and abstain from the consumption of

alcohol and caffeinated beverages on the day before the experiment. Table 3.1 provides

the summary of the participants. Overall, the mean age was 53.2 years and the mean

driving experience (year holding driving license) was 29.0 years respectively.

A number of driving simulator studies have been carried out to examine the effects of

fatigue induced by prolonged driving time on driver performance. The duration of driving

tests varied from 30 minutes to 2 hours (Thiffault and Bergeron, 2003; Filtness et al.,

2012; Ahlström et al., 2018). In this study, duration of 60 minutes is adopted for each

simulated driving test. This is consistent to the time duration of a typical bus or goods

vehicle trip in Hong Kong. Also, according to the results of pilot tests, 60-minute driving

is long enough to reveal the driving performance under the influence of fatigue, while

avoiding the simulator sickness.

Table 3. 2 Summary of the participants of driving simulator study

Driver Group

Number of

participants

Age

Year holding full

driving license

Annual driving

distance (10

km)

Mean (s.d.)

Older professional

63.7 (2.9)

41.3 (4.3)

35.4 (10.1)

Mid-aged professional

43.9 (2.8)

22.1 (4.7)

51.6 (13.2)

Older non-professional

59.4 (3.3)

31.9 (6.7)

7.3 (2.1)

Mid-aged non-professional

47.2 (4.8)

21.7 (8.5)

10.7 (4.5)

Overall

53.2

29.0

27.3

3.2.2 Apparatus, driving scenario and test procedures

The apparatus is a fix-based driving simulator - OKTAL CDS-650. The apparatus is

depicted in Figure 3.1. As shown in Figure 3.1, three 32’’ full HD LED displays were set

up to provide 100

horizontal field of view. The simulator is equipped with clutch, brake

and throttle pedals, steering wheel (real Peugeot wheel with OKTAL force-feedback

system), signaler, dashboard, and a sound system. They are expected to provide realistic

feedback to the participants.

Figure 3. 1 OKTAL CDS-650 driving simulator

The simulated driving scenario is generated by the SCANeR

studio software. The

scenarios are depicted in Figure 3.2. The typical road environments in Hong Kong are

simulated. In particular, two distinct road environments are set out: (1) Inner city road

with numerous roadside activities including but not limited to off-street parking, cyclists,

and pedestrians (walking or standing); and (2) Dual carriageway three-lane motorway

with no roadside activity. Speed limits of inner city road and motorway are 50km/h and

80km/h respectively, which are consistent to the actual driving environment in Hong

Kong.

Figure 3. 2 Typical scenarios for the simulated driving experiment

There are four types of simulated driving scenarios, with respect to road environment and

traffic flow condition, namely (i) Motorway – high traffic flow condition (Figure 3.2a);

(ii) Motorway – low traffic flow condition (Figure 3.2b); (iii) Inner city road – high

traffic flow condition (Figure 3.2c); and (iv) Inner city road – low traffic flow condition

(Figure 3.2d). Each participant was asked to complete two driving simulator tests. In the

experiment, a private car was simulated. Drivers were asked to drive on the middle lane

and were not allowed to make any overtake. In the low traffic flow condition, two vehicles

travelling around the subject vehicle were simulated. In the high traffic flow condition,

ten vehicles travelling around the subject vehicle were simulated. Also, a car would be

following the subject vehicle, while a safe following distance should be maintained. For

the geometric design, average lane width of the motorway is 3.5 m, and that of the inner

city road is 3.3 m respectively. To simulate the environment of urban area, a grid street

network in Shum Shui Po district, with traffic signals, buildings and shops, were

presented in the inner-city scenario. Also, walking and standing pedestrians (number of

pedestrians varied with traffic volume) on the footpaths would be simulated. To simulate

the environment of motorway, a highway section with bridges, interchanges, and roadside

features like plantation and slopes were presented. Also, the horizontal curvature and

vertical grade could vary. It is expected that the road environment (motorway versus

inner-city road) and traffic condition (low versus high traffic conditions) could moderate

the association between driver type, age and driving performance.

Table 3.2 provides the distribution of the simulated driving tests by driver group, road

environment and traffic flow condition. Because of the simulator sickness and

unavailability, some participants only completed one simulated driving test. Hence, a total

of 94 tests (instead of 100) were completed. Also, the distribution of the tests by road

environment, traffic flow condition and driver group, were not perfectly balanced (as

shown in Table 3.2).

Table 3. 3 Distribution of the completed driving simulator tests

Driver Group

Inner city road

Motorway

Total

High

traffic flow

Low traffic

flow

High

traffic flow

Low traffic

flow

Number of tests

Older professional

Mid-aged professional

Older non-professional

Mid-aged non-professional

Overall

Prior to the experiment, a 15-minute practice was provided to help the participants

familiarize with the control of simulator and detect the possible syndromes of simulator

sickness. After the practice, each participant was asked to complete the two 60-minute

driving simulator tests: one was on the motorway and the other was on the inner city road

respectively. The driving tasks were similar for the two tests. Between the tests, a 30-

minute break was given. In addition, the orders of the tests were randomized and

counterbalanced. Furthermore, the experiment would be stopped immediately when a

participant felt unwell.

3.3 Driving performance indicators

Four indicators are used to assess the driving performance. They are standard deviation

of lateral position (SDLP), standard deviation of heading error (SDHE), mean heading

error (MeanHE), and standard deviation of speed (SDspeed). In particular, lateral

position is defined as the perpendicular distance between the centerlines of a traffic lane

and the vehicle. On the other hand, as shown in Figure 3.3, the heading error is defined

as the angular deviation of the vehicle centerline from the tangent to the (curved) road

centerline (Mollenhauer et al., 1994; Comte et al., 2000). Increases in SDLP, SDHE,

MeanHE and SDspeed imply the prevalence of lateral instability, steering control

instability, steering error and inability of speed control respectively (Shanmugaratnam et

al., 2010; Li et al., 2016; Meng et al., 2019). Additionally, these indicators also imply the

existence of driver fatigue and sleepiness, especially after the prolonged driving (Boyle

et al., 2008; Meng et al., 2019). Furthermore, they are sensitive to the interactions between

road environment, driver fatigue and driving performance (Thiffault and Bergeron, 2003;

Ahlström et al., 2018). In this study, all the driving performance indicators were recorded

at a very high frequency (100Hz) throughout the test. In addition, the data was aggregated

into twelve time periods, i.e. [0-5) minute, [5-10) minute, [10-15) minute, [15-20) minute,

[20-25) minute, [25-30) minute, [30-35) minute, [35-40) minute, [40-45) minute, [45-50)

minute, [50-55) minute, [50-60) minute respectively. Therefore, there were 1128

observations (94 tests x 12 time periods) in total for each driving performance indicator.

Figure 3. 3 Illustration of the heading error θ

Table 3.3 summarizes the data collected in the simulated driving tests. As shown in Table

3.3, the professional drivers may have better driving performance than the non-

professional drivers, as the average SDLP, MeanHE and SDspeed of professional drivers

are lower than that of the counterpart. On the other hand, performances of the mid-aged

drivers are better than that of the older drivers. For the road environment, driving

performances on inner city road are better than that on motorway, given that the average

SDLP, SDHE, MeanHE and SDspeed are lower on inner city road, as compared to that

on motorway. Furthermore, the driving performance tends to degrade over time in general,

as the average SDLP, SDHE, MeanHE and SDspeed in the last two time periods are all

higher than that in the first two time periods, and so on and so forth.

Table 3. 4 Summary statistics for the simulated driving tests (N=1128)

Variable

Number of

observations

SDLP

SDHE

MeanHE

SDspeed

Mean (s.d.)

Professional driver

612

0.20 (0.08)

0.33 (0.21)

0.33 (0.14)

3.49 (1.59)

Passenger vehicle driver

408

0.21 (0.07)

0.32 (0.20)

0.34 (0.13)

3.59 (1.45)

Goods vehicle driver

204

0.19 (0.08)

0.33 (0.22)

0.33 (0.15)

3.29 (1.61)

Non-professional driver

516

0.21 (0.07)

0.32 (0.21)

0.36 (0.13)

3.52 (1.49)

Older driver

540

0.22 (0.08)

0.34 (0.21)

0.36 (0.14)

3.76 (1.69)

Mid-aged driver

588

0.20 (0.06)

0.32 (0.20)

0.33 (0.13)

3.26 (1.34)

High traffic flow

564

0.21 (0.08)

0.32 (0.20)

0.35 (0.11)

3.56 (1.54)

Low traffic flow

564

0.21 (0.08)

0.34 (0.21)

0.34 (0.16)

3.44 (1.56)

Motorway

588

0.26 (0.05)

0.51 (0.12)

0.44 (0.11)

4.43 (1.38)

Inner city road

540

0.15 (0.05)

0.13 (0.02)

0.24 (0.05)

2.49 (0.95)

Time period 1 (0-5 min)

0.20 (0.07)

0.28 (0.16)

0.31 (0.10)

3.13 (1.19)

Time period 2 (6-10 min)

0.20 (0.07)

0.28 (0.17)

0.32 (0.11)

3.26 (1.37)

Time period 3 (11-15 min)

0.18 (0.06)

0.29 (0.16)

0.32 (0.11)

3.36 (1.45)

Time period 4 (16-20 min)

0.19 (0.07)

0.28 (0.16)

0.32 (0.11)

3.63 (1.60)

Time period 5 (21-25 min)

0.21 (0.08)

0.35 (0.23)

0.35 (0.14)

3.35 (1.49)

Time period 6 (26-30 min)

0.22 (0.08)

0.34 (0.22)

0.35 (0.14)

3.54 (1.48)

Time period 7 (31-35 min)

0.20 (0.07)

0.33 (0.20)

0.35 (0.12)

3.52 (1.57)

Time period 8 (36-40 min)

0.20 (0.06)

0.31 (0.19)

0.33 (0.12)

3.62 (1.68)

Time period 9 (41-45 min)

0.21 (0.08)

0.36 (0.23)

0.36 (0.14)

3.55 (1.53)

Time period 10 (46-50 min)

0.22 (0.08)

0.35 (0.23)

0.37 (0.17)

3.50 (1.55)

Time period 11 (51-55 min)

0.22 (0.08)

0.38 (0.24)

0.38 (0.15)

3.65 (1.57)

Time period 12 (56-60 min)

0.23 (0.08)

0.37 (0.24)

0.38 (0.15)

3.88 (1.76)

3.4 Statistical method

Multiple regression approach is applied to measure the effects of driver type, age, road

environment, traffic flow condition and driving time on driving performance. Also, the

interaction effect between driver type and age is considered. Then, disaggregated models

by driver type (i.e. professional and non-professional drivers) are developed based on the

results of market segmentation analysis (Wong et al., 2008; Szeto et al., 2013).

In this study, each participant was asked to complete two simulated driving tests (each

for 60 minutes). On the other hand, there were twelve observations (by twelve time

periods) for each test. The observations within the same test and of the same participant

would be correlated because they shared the common (unobserved) random effect. To

allow for the correlation between observations, the panel random intercept regression

approach was applied to measure the association between driving performance and

possible factors, including road environment, traffic flow condition, driver occupation,

age and driving time. The random intercept models (θ

for SDLP, θ

for SDHE, θ

for

MeanHE and θ

for SDspeed respectively) are specified as follows,









 



 







 



(1)









 



 







 



(2)









 



 







 



(3)









 



 







 



(4)

where i refers to the test (i = 1, 2, 3, …, and 94); t refers to the time period (t = 1, 2, 3 …

12); μ

refers the independent residual (between tests); ε

refers to the independent residual

(between observations); x is the value of explanatory variable (including road

environment, traffic flow condition, driver type, age, and time period) and β is the

corresponding coefficient respectively.

The coefficients were estimated using the maximum likelihood approach. To assess the

goodness-of-fit of the proposed model, the likelihood ratio test statistics are given by,

LR = -2 [LL(β

) – LL(β

)]

(5)

where LL(β

) is the restricted log likelihood function and LL(β

) is the unrestricted log

likelihood function respectively. Under the null hypothesis, LR is χ

distributed with q

degree of freedom (q is the difference in the number of parameters between the restricted

and unrestricted models). A good fit was indicated by a statistically significant LR. In this

study, the statistical package NLOGIT 5.0 was used to establish the proposed random

intercept models.

3.5 Results

In this study, the random intercept approach was used to measure the effects of factors

including driving time, age, driver type, road type and traffic flow condition on the driving

performance, with which the unobserved effect of correlation between observations of

the same participant and in the same test was controlled for. There were 1128 observations

for each model (SDLP, SDHE, MeanHE and SDspeed). Table 3.4 presents the results of

parameter estimation of the overall models. Since driver type, age, traffic flow condition

and driving time are the variables of interest in this study, they are all considered in the

proposed models, even no evidence can be established for significant correlation with the

driving performance indicators. Additionally, two types of professional drivers (i.e.

passenger vehicle and goods vehicle drivers) are considered. This is to control for the

effects of differences in experience and skills between vehicle types. For the interaction

effect, focus was paid on the two interested variables (i.e. older and professional driver).

However, no significant evidence related to the driving performance of older professional

drivers could be established.

Table 3. 5 Results of parameter estimates of the (overall) random intercept models

Variable

SDLP

SDHE

MeanHE

SDspeed

Coeff.

(z-stat.)

Coeff.

(z-stat.)

Coeff.

(z-stat.)

Coeff.

(z-stat.)

Constant

0.140

(14.26)**

0.098

(7.16)**

0.206

(12.61)**

1.875

(8.00)**

High traffic flow

-0.0005

(-0.06)

-0.031

(-3.00)**

0.004

(0.32)

0.115

(0.59)

Motorway

0.109

(13.24)**

0.370

(35.70)**

0.204

(14.98)**

1.946

(9.95)**

Goods vehicle driver

-0.027

(-2.37)*

-0.011

(-0.79)

-0.031

(-1.63)

-0.264

(-0.97)

Passenger vehicle driver

-0.005

(-0.58)

-0.010

(-0.90)

-0.023

(-1.52)

0.085

(0.39)

Older driver

0.023

(2.70)**

0.017

(1.66)

0.021

(1.52)

0.472

(2.40)*

Time period 2 (5-10 min)

0.001

(0.14)

0.006

(0.65)

0.005

(0.71)

0.105

(1.07)

Time period 3 (10-15 min)

-0.020

(-4.92)**

0.016

(1.66)

0.006

(0.86)

0.201

(2.04)*

Time period 4 (15-20 min)

-0.007

(-1.65)

0.002

(0.21)

0.010

(1.40)

0.471

(4.79)**

Time period 5 (20-25 min)

0.015

(3.59)**

0.076

(7.67)**

0.041

(5.84)**

0.194

(1.97)*

Time period 6 (25-30 min)

0.015

(3.75)**

0.062

(6.31)**

0.040

(5.72)**

0.384

(3.91)**

Time period 7 (30-35 min)

0.004

(0.92)

0.054

(5.47) **

0.039

(5.47)**

0.362

(3.68)**

Time period 8 (35-40 min)

-0.0002

(-0.04)

0.037

(3.76)**

0.019

(2.74)**

0.469

(4.77)**

Time period 9 (40-45 min)

0.015

(3.60)**

0.079

(7.96)**

0.050

(6.95)**

0.399

(4.05)**

Time period 10 (45-50 min)

0.015

(3.73)**

0.077

(7.84)**

0.065

(9.19)**

0.344

(3.50)**

Time period 11 (50-55 min)

0.022

(5.33)**

0.098

(9.95) **

0.066

(9.41)**

0.491

(4.99)**

Time period 12 (55-60 min)

0.028

(6.85)**

0.097

(9.87)**

0.070

(9.98)**

0.761

(7.72)**

Goodness-of-fit

Unrestricted loglikelihood

2281.89

1348.18

1671.35

-1299.14

Restricted loglikelihood

1320.83

178.47

671.08

-2084.54

Likelihood ratio test

1922.12**

2339.42**

2000.54**

1570.80**

** Statistical significance at the 1% level

* Statistical significance at the 5% level

In addition, the market segmentation analysis using the Watson and Westin pooling

approach (Wong et al., 2008; Szeto et al., 2013) was conducted to examine the possible

intervention effect by driver type (i.e. professional versus non-professional drivers) on

the relationship between driving performance and possible factors. Disaggregated models

for professional and non-professional drivers were then established based on the results

of market segmentation analysis (shown in Table 3.5). Therefore, differences in possible

factors between professional and non-professional drivers could be assessed. Table 3.6

presents the results of the disaggregated analyses. Overall, the proposed models fit well

with the observations, all at the 1% level of significance. Results of parameter estimation

for each of the four driving performance indicators (i.e. SDLP, SDHE, MeanHE and

SDspeed) are described one by one in the following Section 3.5.1, 3.5.2, 3.5.3 and 3.5.4.

Table 3. 6 Results of market segmentation analysis

Indicator

LL(β

)

LL(β

)

Degrees of

freedom

Likelihood

Ratio

SDLP

2280.24

2294.37

28.26*

SDHE

1348.18

1379.61

62.86**

MeanHE

1671.27

1684.44

26.34*

SDspeed

-1299.90

-1280.12

41.38**

** Significant at the 1% level

* Significant at the 5% level

Table 3. 7 Results of parameter estimates for professional and non-professional drivers

Variable

SDLP

SDHE

MeanHE

SDspeed

Pro

Non-Pro

Pro

Non-Pro

Pro

Non-Pro

Pro

Non-Pro

Constant

0.121**

0.148**

0.095**

0.101**

0.180**

0.226**

2.026**

1.668**

High traffic flow

0.001

0.000

-0.040*

-0.051**

-0.002

-0.026

-0.142

0.443

Motorway

0.117**

0.098**

0.365**

0.379**

0.205**

0.206**

1.803**

2.094**

Older driver

0.021*

0.021

0.006

0.001

0.023

-0.021

0.690**

0.212

Older x High traffic flow

0.060*

0.076*

Time period 2

0.006

-0.005

0.018

-0.008

0.017

-0.009

0.032

0.193

Time period 3

-0.016**

-0.025**

0.026

0.005

0.011

-0.001

0.143

0.271*

Time period 4

-0.003

-0.011

0.001

0.003

0.012

0.008

0.391**

0.567**

Time period 5

0.020**

0.009

0.097**

0.050**

0.043**

0.038**

0.071

0.342**

Time period 6

0.015*

0.016**

0.068**

0.055**

0.039**

0.042**

0.326*

0.455**

Time period 7

0.001

0.007

0.056**

0.051**

0.042**

0.035**

0.345*

0.385**

Time period 8

-0.005

0.005

0.030*

0.046**

0.019

0.019*

0.425**

0.523**

Time period 9

0.015**

0.014*

0.074**

0.084**

0.043**

0.056**

0.214

0.619**

Time period 10

0.021**

0.008

0.076**

0.079**

0.071**

0.057**

0.253

0.454**

Time period 11

0.026**

0.017**

0.106**

0.088**

0.072**

0.060**

0.314*

0.671**

Time period 12

0.031**

0.024**

0.104**

0.090**

0.070**

0.689**

0.849**

Variable

SDLP

SDHE

MeanHE

SDspeed

Pro

Non-Pro

Pro

Non-Pro

Pro

Non-Pro

Pro

Non-Pro

Goodness-of-fit

Unrestricted Loglikelihood

1228.76

1065.61

670.85

711.28

868.62

818.49

-752.79

-527.33

Restricted Loglikelihood

711.38

613.19

93.37

85.22

341.32

336.59

-1121.04

-963.07

Likelihood ratio test

1034.58**

904.84**

1154.96**

1252.12**

1054.60**

963.80**

736.50**

871.48**

Notes:

** Statistical significance at the 1% level

* Statistical significance at the 5% level

“Pro” refers to professional driver and “non-Pro” refers to non-professional driver respectively

(a) Lateral control stability

(b) Steering stability

(d) Speed Control

Figure 3. 4 Driving performance of professional and non-professional drivers over time

Note:

1. The box plots indicate the minimum, 1st quartile, median, 3rd quartile and maximum

of the performance indicators (i.e. SDLP, SDHE, MeanHE, SDspeed)

2. “X”s indicate the average value

3.5.1 Lateral stability

Standard deviation of lateral position is widely used to reflect the driver’s ability to

maintain the lateral stability (Boyle et al., 2008). Increase in SDLP implies the

degradation in lateral control performance. As shown in Table 3.4, for the main effect,

driver type, driver age, road type and driving time are found associated with SDLP. In

particular, SDLP of older driver is higher than that of the mid-aged driver. SDLP of goods

vehicle driver is lower than that of the non-professional driver. Also, SDLP tends to

increase when driving on the motorway. In addition, SDLP increases remarkably when

the driving time is greater than 25 minutes (Time period 5).

As shown in Table 3.6, driver age, road type and driving time significantly affected the

lateral stability of professional driver. SDLP of older professional driver tends to be

higher than that of the mid-aged professional driver. Also, remarkable increase in SDLP

could be observed after driving for 25 minutes (Time period 5). On the other hand, no

evidence could be established for the association between driver age and SDLP of non-

professional driver. Yet, obvious increase in SDLP of non-professional driver could be

observed after 30 minutes (Time period 6).

3.5.2 Stability of steering control

Standard deviation of heading error reflects the driver’s stability of steering wheel control.

Increase in SDHE indicates the degradation in steering performance. As shown in Table

3.4, for the main effects, traffic flow condition, road type and driving time are found

associated with SDHE, all at the 1% level of significance. In particular, SDHE increases

when driving on the motorway and decreases when driving under the high traffic flow

condition. In addition, SDHE increases remarkably when the driving time is greater than

25 minutes (Time period 5).

For the disaggregated analyses, as shown in Table 3.6, traffic flow condition, road type

and driving time significantly affected the SDHE, both of professional and non-

professional drivers. However, the interaction effect between driver age and traffic flow

condition was significant only among non-professional drivers. In particular, steering

stability of older non-professional driver tends to be worse than that of the mid-aged non-

professional drivers, when driving under the high traffic flow condition.

3.5.3 Steering error

Mean heading error refers to the mean angular deviation of the vehicle trajectory from

that of the road centerline. Increase in MeanHE indicates the degradation in steering

accuracy. As shown in Table 3.4, for the main effects, road type and driving time are

found associated with MeanHE at the 1% level of significance. In particular, MeanHE

increases when driving on the motorway. Also, MeanHE increases remarkably when the

driving time is greater than 25 minutes (Time period 5).

For the disaggregated analyses, as shown in Table 3.6, main effects of road type and

driving time on MeanHE were found significant both for the professional and non-

professional drivers. However, the interaction effect between driver age and traffic flow

condition on MeanHE is significant only among the non-professional drivers. In

particular, steering error of older non-professional driver was greater than that of the mid-

aged non-professional driver, when driving under the high traffic flow condition.

3.5.4 Speed stability

Standard deviation of speed is widely used to reflect the driver’s ability to maintain the

stability of driving speed. Increase in SDspeed implies the degradation in speed control

performance. As shown in Table 3.4, for the main effect, driver age, road type and driving

time are found associated with SDspeed. In particular, SDspeed of older driver is higher

than that of the mid-aged driver. Also, SDspeed tends to increase when driving on the

motorway. In addition, SDspeed increases remarkably when the driving time is greater

than 15 minutes (Time period 3).

For the disaggregated analyses, as shown in Table 3.6, driver age, road type and driving

time all affect the speed control performance of professional drivers. SDspeed of older

professional driver tends to be higher than that of the mid-aged professional driver. Also,

significant increase in the SDspeed of professional driver could be observed after 20

minutes (Time period 4). On the other hand, no evidence could be established for the

association between driver age and SDspeed of non-professional driver. Yet, remarkable

increase in the SDspeed could be observed for non-professional driver after 15 minutes

(Time period 3).

3.6 Discussion

In this study, we hypothesized that (1) the age-related impairments on driving

performance could be reduced by the driving experience and task familiarity of

professional drivers; and (2) contributory factors to the driving performance of

professional drivers were different from that of non-professional drivers. Results of

overall model indicate that increase in age and driving on the motorway are associated

with the degradation of lateral and speed control stability. In addition, road type and

traffic flow condition are associated with the steering performance. Furthermore, results

of disaggregated models indicate that effects of possible environmental factors on the

driving performances of professional drivers are similar to that of non-professional

drivers. Nevertheless, interaction effect by driver age on the association between traffic

flow condition and driving performance is remarkable exclusively for the steering

performance of non-professional drivers. In particular, the older non-professional drivers

tend to have higher MeanHE and SDHE when driving under the high traffic flow

condition. On the other hand, effect of age is exclusive for the lateral and speed control

performances of professional drivers.

3.6.1 Effect of age on driving performance

Overall, older drivers show poorer lateral and speed control performance than the mid-

aged drivers in this study. It is consistent to the findings of previous studies that driving

performance deteriorates with age (Islam and Mannering, 2006; Shanmugaratnam et al.,

2010). This could be attributed to the degraded physical, mental and cognitive capabilities

(Lundberg et al., 1998; Andrews and Westerman, 2012). In addition, older people have

higher perceptual sensitivity to the mental workload induced by driving, as compared to

the younger counterpart (Cantin et al., 2009).

Estimation results for non-professional driver indicate that interaction between age and

traffic condition is statistically significant. For non-professional driver, degradation in

steering performance by age is found remarkable under the high traffic flow condition.

Such findings in line with that of previous studies (Cantin et al., 2009; Trick et al., 2010).

In particular, older drivers often consider driving under the high traffic flow condition as

a challenging task (with higher mental workload). Since older drivers tend to be risk

averse, they would avoid driving under the unfavorable road environment and traffic

condition (Molnar and Eby, 2008; Cantin et al., 2009; Trick et al., 2010; Teh et al., 2014).

In contrast, no evidence could be established for the degraded driving performance of

older professional drivers under the high traffic flow condition. This could be attributed

to the task familiarity of older professional drivers (Andrews and Westerman, 2012). Yet,

it is recommended that more rigorous medical assessment should be implemented for the

renewal of driving license of older drivers, considering their high crash involvement rates.

Also, the moderating effects by other environmental conditions, such as lighting and

weather, on the association between age and driving performance must be explored in the

extended study.

3.6.2 Performance of professional drivers

Results of overall model indicate that goods vehicle drivers have better lateral control

performance than the non-professional drivers. This can be attributed to the higher driving

experience and better driving skill of goods vehicle drivers (Borowsky and Oron-Gilad,

2013). Seemingly, goods vehicle drivers in Hong Kong demonstrate better driving skills

and attitudes. Indeed, the crash involvement rate (per million vehicle-km) of goods

vehicle was lower than that of other commercial vehicles (i.e. taxi, light bus and bus) in

Hong Kong (Transport Department of HKSAR, 2017). Yet, no evidence can be

established for significant difference in driving performance between passenger vehicle

drivers and non-professional drivers. Seemingly, superior driving skill of passenger

vehicle drivers related to driving experience on different types of the roads, vehicle size

relative to road width, and work-related trips could be offset by the aggressive driving

behaviors (Kontogiannis, 2006; Öz et al., 2010a; Li et al., 2019). Examples of aggressive

driving behaviors include but are not limited to speeding, red light running, and improper

lane changing. They are indeed more prevalent for passenger vehicle drivers because of

the desire for higher revenues and expectations of the employers/customers (Wong et al.,

2008). In this study, effects of operation characteristics and attitudes of the drivers on the

driving performance are however not considered. It is worth exploring the effects of

driver perceptions and characteristics (e.g. perceived or observed aggressive behaviours,

risk-taking traits, traffic offenses, and crash involvement) on the driving performance in

the extended study. Furthermore, it is of essence to measure the association between the

road geometry, driving performance (especially on steering performance) and crash risk,

when comprehensive vehicle trajectory and crash data of professional drivers are

available (Ahlström et al., 2018).

For the difference in effects of possible factors between professional and non-professional

drivers, age effect is found significant only among the professional drivers. In particular,

lateral and speed control performances of mid-aged professional drivers are better than

that of the older professional drivers. Seemingly, the reduction in exposure by age could

be a significant contributory factor to the degraded performance of professional drivers.

Older professional drivers have lower annual driving distance that their mid-aged

counterparts. It could be because the working and/or driving hours of older professional

drivers tend to be lower. It is likely that the older professional drivers proactively reduce

their exposure on road to mitigate the elevated crash risk due to age-related impairments.

On the other hand, the driving hours of older professional drivers can be limited by the

safety management policy of the transport operators. To this end, it is proposed that

additional driver training, particularly on lateral stability and speed control, could be

provided to the older professional drivers to mitigate the age-related impairments.

3.6.3 Effect of other factors on driving performance

As revealed in this study, the lateral performance of driver degrades over the 60-minute

drive in general. It is consistent to the findings of previous studies (Oron-Gilad and

Ronen, 2007; Ting et al., 2008; Farahmand and Boroujerdian, 2018). This could reflect

the increase in fatigue level resulted from the prolonged driving (Du et al., 2015;

Ahlström et al., 2018). Degradation of speed control, lateral stability and steering

performance are associated with the increase in driver sleepiness and fatigue, particularly

after prolonged driving (Boyle et al., 2008; Meng et al., 2019). In addition, the results of

overall and disaggregated models show that degraded lateral, steering and speed control

performances tend to occur when driving on the motorways. This could be attributed to

the drowsiness due to the monotonous driving environment, and limited roadside

activities and interactions with other road users (Oron-Gilad and Ronen, 2007;

Williamson et al., 2014; Du et al., 2015; Ahlström et al., 2018). Yet, such findings could

be verified when information on both subjective and physiological indicators are

available in the extended study (Oron-Gilad et al., 2008). On the other hand, the degraded

driving performance on the motorways could also be attributed to the geometric and

operational characteristics of the roadway. Indeed, degradation of lateral and speed

stabilities could be profound when speed limit and road width increase (Ahlström et al.,

2018; Meng et al., 2019). It would be worth exploring the effect of road curvature (in

term of the number of curves, interval between curves and radius of curvature), road width

and speed limit on the association between driving performance and driving time in the

extended study.

It was expected that the effect of driving time on driving performance should be different

between professional and non-professional drivers, since the professional drivers are

more skillful in general. Results of disaggregated model indicated that onset of the

significant degradations in lateral and speed control performance are different between

professional and non-professional drivers. Moreover, the lateral and steering performance

of professional drivers started to degrade after driving for 25 minutes. This should be

indicative to the safety management strategies of the transport operators, especially the

design and development of in-vehicle driver monitoring and assistance system on the

commercial vehicle fleets (Davidse et al., 2009). For example, eye tracking unit for the

detection of driver fatigue, and electronic stability control system could be installed on

the passenger vehicles (for franchised buses in Hong Kong, a subsidization scheme has

been introduced to retrofit smart safety devices including electronic stability control on

the existing bus fleets).

3.7 Concluding remarks

Professional drivers are considered more skillful and experienced. However, the overall

crash involvement rate of professional drivers is higher than their counterparts in Hong

Kong. Also, the population of older professional driver is increasing because of the ageing

population. In this chapter, we examined two hypotheses 1) the impairment of driving

performance by age could be reduced by the driving experience and task familiarity of

professional drivers; 2) the contributory factors to the driving performance of professional

drivers should be different from that of non-professional drivers. The driving performance

indicators considered are standard deviation of lateral position, standard deviation of

heading error, mean heading error and standard deviation of driving speed.

Results of overall model indicate that goods vehicle drivers tend to have better lateral

stability than the non-professional drivers. Driving performances of mid-aged drivers

tend to be better than that of the older drivers. Based on the results of disaggregated

analysis, the impairments on driving performance by age (i.e. older) are more prevalent

when driving under the high traffic flow condition among non-professional drivers. No

evidence could be established for the degraded driving performance of older professional

drivers under the high traffic flow condition. Although older drivers are often risk averse

and would avoid driving under the high traffic flow condition, age-related impairments

could be reduced by the driving experience and task familiarity of professional drivers.

Therefore, for the driver recruitment and management, decision making of transport

operators should not be solely based on driver age. Instead, rigorous assessment of driving

skills and enhanced training could be provided for the older drivers. Furthermore, results

of the disaggregated models indicate that the effect of age is found prevalent only among

the professional drivers. Lateral and speed control performance of mid-aged professional

drivers were superior than that of older professional drivers. Seemingly, reduction in

exposure could be a contributory factor to the impaired driving performance of older

professional drivers. It is recommended that driver training could be provided to the older

drivers, particular on vehicle control. This is to mitigate the increase in collision risk

attributed to reduced exposure. As for the second hypothesis, results of disaggregated

analyses indicate that effects of possible environmental factors (i.e. motorway and high

traffic flow condition) of professional drivers are similar to that of non-professional

drivers. However, the interaction between age and traffic flow condition are exclusive to

the non-professional drivers only.

In this study, the interaction effects by the driver perception and attitude on the association

between driving performance and driver characteristics are not considered. Additionally,

the driving performances of male drivers only are assessed in the driving simulator

experiment, given the relatively small sample size. Effect of gender on driving

performance is therefore not attempted. It is worth exploring the effects of driver

characteristics (in term of crash involvement, traffic offense and risk perception) on the

driving performance when the comprehensive information is available from the

attitudinal survey (Wong et al., 2008; Li et al., 2014). Furthermore, effects of road design

and environmental condition (e.g. lighting and weather) on the crash risk of professional

driver can be revealed based on comprehensive vehicle trajectory and crash data in

extended study.

Chapter 4 Evaluation of conflict risk using driving simulator

4.1 Introduction

The percentage of elderly who hold a valid driving license has been increasing rapidly in

those ageing societies (Newnam et al., 2020, 2018). Indeed, Hong Kong is facing the

problem of ageing population because of the reduction in fertility rates and increased life

expectancy. By 2035, proportion of population of age above 65 in Hong Kong would

reach 25% (Sze and Christensen, 2017). The proportion of drivers aged above 60 with a

valid public transport vehicle (e.g. taxi, light bus, and bus, etc.) driving license was 37-

46% in 2017 (Lee, 2018).

Safety of professional drivers is of great concern since they have much higher exposure

on roads. In Hong Kong, 50% of work trips are made by taxi, public light bus, and bus

(Hong Kong Transport Department, 2014). More importantly, the proportion of older

drivers in the transport sector increases dramatically because of the shortage of labour.

Given the age-related declines in driving performance, the strategy adopted by older

drivers to compensate for their elevated crash risk has drawn increasing attention in recent

years. This issue is of importance to employers given their multiple responsibilities to

keep the drivers, passengers, and cargo safe, as well as to support their older employees

who want to stay in the industry. This information can be used to review and revise control

measures, as well as develop new intervention, designed to promote the safety, health and

wellbeing of older professional drivers.

While the compensatory strategy adopted by older general drivers has been studied in

some depth, there is little work that researches into the compensatory behavior and its

safety implications of older professional drivers. This is surprising given that the

proportion of older drivers in the transport sector has been increasing dramatically

because of the shortage of labor. Therefore, we are motivated to study the driving

performance of professional drivers from the behavioral perspective. The aim of this

study is to address the research question that whether the older professional drivers reduce

the crash risk more effectively by capitalizing on their rich experience, and to provide

suggestions on driver training and management policy for transport authority and

operators.

In this chapter, two hypotheses are proposed, (i) older professional drivers have lower

likelihood and severity of rear-end conflict, as compared with older non-professional

drivers; and (ii) likelihood and severity of rear-end conflict after prolonged driving would

be higher for older non-professional drivers. The hypotheses are proposed considering

the possible differences in driving skill, exposure, and experience (especially driving long

hours) between professional and non-professional drivers. A car-following scenario with

sudden brakes of the leading vehicle is used to test the hypotheses. Time exposed time-

to-collision (TET) and time integrated time-to-collision (TIT) are adopted as performance

measure indicating the severity of traffic conflict. Also, drivers’ brake reaction times for

the sudden events are recorded. A random-parameter Tobit regression approach is applied

to investigate the associations between performance measures and possible factors

including driver occupation, driver age and traffic flow condition.

The remainder of this chapter is structured as follows. Section 4.2 describes the

experimental design and data collection. Section 4.3 and 4.4 illustrate the surrogate safety

measures and analysis method, respectively. Results of the hypotheses are presented in

Section 4.5 and implications of the results are discussed in Section 4.6. Finally, Section

4.7 provides the concluding remarks.

4.2 Simulator experiment design

4.2.1 Participants

Forty-four male drivers were recruited for the driving simulator experiment. The

inclusion criteria are having a full driving license, minimum driving time of 5 hours per

week and (self-declared) good health condition. The exclusion criteria are feeling unwell

and having any syndrome of simulator sickness (e.g. headache, nausea, blurred vision and

dizziness, etc.). Participants were asked to have enough rest and abstain from alcohol and

caffeinated beverages 24 hours prior the simulator test. Prior to the experiment, a 15-

minute training session was provided to each participant to help familiarize the

participants with the driving simulator controls. Informed consent in accordance with the

requirements of university research ethics committee was obtained, and monetary

compensation (US$25-50) was provided for the participation.

Table 4. 1 Summary of participants of driving simulator study

Professional driver

Non-professional driver

Overall

Mid-aged

Older

Mid-aged

Older

Number of participants

Mean (S.D.)

Age

43.5

(2.5)

63.4

(3.1)

46.5

(4.5)

60.1

(3.2)

53.3

(9.1)

Year holding full

driving license

21.5

(4.5)

41.4

(4.7)

21.4

(8.9)

32.1

(7.0)

29.1

(10.4)

Annual driving distance

(10

km)

51.6

(13.8)

34.4

(9.8)

10.8

(4.7)

7.0

(2.0)

26.0

(2.1)

All participants need to complete a short questionnaire survey to provide information on

driver age, annual driving distance, occupation, record of traffic convictions and accident

involvement. Table 4.1 presents the summary of participants of driving simulator study.

Of the 44 participants, three (i.e. 1 mid-aged professional and 2 older professional drivers)

were excluded since they had driving simulator sickness. Data collected from 19

professional drivers and 22 non-professional drivers were used for the analysis. In this

study, professional drivers refer to the full-time taxi, public light bus, public bus and

goods vehicle drivers. Furthermore, the participants were classified into two categories:

(i) mid-aged and (ii) older drivers, in accordance with the classification in some recent

studies in Hong Kong (Li et al., 2016; Chen et al., 2019). Specifically, “mid-aged” drivers

refer to those who are aged from 40 to 55 years, and “older” drivers refer to those who

are aged from 56 to 69 years.

4.2.2 Driving scenario and test procedures

In this study, the experiments were carried out using the OKTAL CDS-650 compact fix-

based simulator and the simulated driving scenarios were developed using the

SCANeRTM studio package. For instance, there are three 32’’ full HD LED monitors

providing a 100o horizontal field of view. Particularly, the simulator is equipped with

force-feedback pedals, steering wheel and indicator of real vehicle (e.g. Peugeot). It is to

provide the participants realistic control experience, especially the road texture and kerb

side. Driving and vehicle movement attributes including acceleration, speed, lateral

position, pedal force and steering angle are recorded at a frequency of 100 Hz.

As depicted in Figure 4.1, typical Hong Kong road environment is simulated in the

driving experiment. In particular, buildings, roads, intersections and road furniture in

Sham Shui Po District (a densely populated urban district in Hong Kong) are simulated.

The roads simulated are of three lanes (with on-street parking space on the left hand side)

and single direction. They form a grid network and the speed limit is 50km/h. Also, two

traffic conditions: (i) high traffic flow, more pedestrians on the footpaths and more

surrounding vehicles; and (ii) low traffic flow, less pedestrians and less surrounding

vehicles, are simulated. In high traffic flow condition, there are 10 vehicles moving

around the subject vehicle and 10 pedestrians per 100 meter long footpath. In low traffic

flow condition, there are 2 vehicles moving around the subject vehicle and 0.5 pedestrians

per 100 meter long footpath.

(a) High traffic flow

(b) Low traffic flow

Figure 4. 1 Typical simulated driving scenarios

Each participant was asked to complete one simulated driving session, either under the

low or high traffic condition. Duration of every session was 60 minutes. The scenarios

(high versus low traffic conditions) presented were randomized and counterbalanced

across the participants. In the experiment, participants were asked to drive as if driving a

small passenger car. They were instructed to drive on the middle lane and follow a leading

vehicle, with a comfortable and constant following distance. They were also required not

to overtake during the simulated driving. To assess the drivers’ response, ‘events’ (sudden

brake of leading vehicle as indicated by the ‘brake light’) were induced after 5 minutes

and 55 minutes of driving respectively. In particular, the leading vehicle would decelerate

from 50 km/h to complete stop within 3 seconds, stop for 2 seconds, and then accelerate

gradually to 50 km/h again.

4.3 Surrogate safety measures

In 1970s, researcher first attempted to evaluate safety using traffic conflict technique as

defined by the shortest time-to-collision (TTC) (Hayward, 1972). TTC refers to the

remaining time before two vehicles would collide, if there was no evasive maneuver to

avoid a collision. For car-following scenario, there is always a definite TTC when the

speed of leading vehicle is lower than that of following vehicle. When the separation

reduces or the speed difference increases, value of TTC decline. Value of TTC should be

sufficiently small to define a conflict. Threshold of minimum TTC can be defined based

on the driver perception-reaction time plus the time required for evasive maneuver to

avoid collision. In conventional studies, threshold of minimum TTC ranges from 1 to 5

second (Autey et al., 2012; Sayed et al., 2013; Zheng et al., 2014). To indicate the severity

of traffic conflict, two modified TTC-based measures - time exposed time-to-collision

(TET) and time integrated time-to-collision (TIT) were developed (Minderhoud and

Bovy, 2001). Values of TET and TIT are sensitive to the threshold of minimum TTC as

which a traffic conflict is defined. Threshold of 3 second is adopted for this study (Sayed

et al., 2013)

In this study, surrogate safety measures – TET and TIT are depicted in Figure 4.2. In

particular, TET refers to the duration when a safety-critical situation (i.e. TTC is lower

than the threshold) persists, and TIT refers to the integral that gives the area bounded by

the TTC curve and TTC threshold (during which TTC is lower than the threshold)

respectively. Increases in TET and TIT both indicate the increase in the severity level of

traffic conflict. It should be noted that the TET and TIT could become zero when the

minimum TTC is higher than the threshold (i.e. 3 second). It implies the absence of traffic

Different thresholds of minimum TTC (from 1 second to 5 second) were considered in preliminary

analysis. However, influences on the TET and TIT estimates and modeling results were marginal when

reducing the threshold further below 3 second. Hence, threshold of 3 second is considered appropriate.

conflict. Additionally, as shown in Figure 2, brake reaction time refers to time lag for the

onset of evasive action (i.e. when the driver of following vehicle presses the brake pedal,

tbr) in response to an event (i.e. sudden deceleration of a leading vehicle, tse). Also,

standard deviation of lateral position (SDLP) and standard deviation of driving speed

(SD_Speed) are employed to assess driver’s lateral and longitudinal controls (Chen et al.,

2019; Li et al., 2016; Shanmugaratnam et al., 2010). Moreover, average speed and time

headway are measured to examine the possible compensatory behaviors (Andrews and

Westerman, 2012; Martchouk et al., 2011; Ni et al., 2010).

To sum up, the performance indicators can be classified into three categories: (1) driving

capability, i.e. SDLP, SD_Speed, and brake reaction time (BRT), etc.; (2) compensatory

behavior, i.e. average speed and time headway, etc.; and (3) safety risk , i.e. TET and TIT,

etc. In particular, SDLP, SD_Speed, average speed, and time headway during the five-

minute period prior to the onsets of sudden brake, i.e. [0-5) minute and [50-55) minute

were measured.

Figure 4. 2 Illustration of proposed driving performance indicators

4.4 Statistical method

To accommodate the censoring nature (either left-censored or right-censored) of

dependent variable, Tobit regression was proposed (Tobin, 1958). In road safety research,

Tobit regression approach is commonly used to model the crash rate, which is left-

censored (Zeng et al., 2017, 2018; Anastasopoulos et al., 2008, 2012). In this study, the

surrogate safety measures - TET and TIT, are non-negative, continuous, and left censored

at zero. To address the problem of unobserved heterogeneity attributed to repeated

observations (at different driving time), random parameter Tobit regression should be

applied to measure the association between conflict risk and possible factors including

driver age, driving time, and traffic flow condition (Anastasopoulos et al., 2012).

Separated prediction models were established for professional and non-professional

drivers since the effects of possible factors could be different. For instance, the proposed

Tobit model can be specified as,











 









 

































(1)

where 



denotes the performance indicator (i.e. TET and TIT), x denotes the

explanatory variable, β denotes the corresponding coefficient, and ε

denotes the

independent residual (ε

~ N(0, σ

)), of ith participant (i = 1, 2, 3, …, 41) and tth event (t

= 1, 2) respectively.

The parameters are estimated using maximum likelihood approach. To evaluate the effect

of possible factor on the likelihood of traffic conflict, zero sensitivity is estimated using

the formulation specified as (Anastasopoulos et al., 2008),











   









































(2)

where E denotes the expectation of occurrence of traffic conflict (i.e. TET and TIT

being greater than zero), z denotes the normalized variable, 







denotes the probability

distribution function, and 







denotes the probability density function respectively.

To model the brake reaction time, the random parameter linear model can be specified as,









 









 



(3)

where 



denotes the brake reaction time, x denotes the explanatory variable, β denotes

the corresponding coefficient, and ε

denotes the independent residual (ε

~ N(0, σ

)), of

ith participant (i = 1, 2, 3, …, 41) and tth event (t = 1, 2) respectively.

To access the goodness-of-fit of proposed models, Maddala R

and likelihood ratio test

statistics would be estimated (Maddala, 1986; Anastasopoulos et al., 2008). In this study,

parameter estimations of proposed models are carried out using the software package

NLOGIT 5.0.

4.5 Results

Table 4.2 summarizes the performances (i.e. driving capability, compensatory behavior,

and safety risk) of simulated driving tests, with respect to driver type and age group. As

shown in Table 4.2, TIT and TET of older drivers are lower than that of the counterpart.

Also, average BRT of mid-aged drivers is lower than that of older drivers.

Table 4. 2 Summary statistics for the simulated driving tests

Scope of work

Driving

performance

indicator

Professional driver

Non-professional driver

Mid-aged

Older

Mid-aged

Older

Mean

S.D.

Mean

S.D.

Mean

S.D.

Mean

S.D.

Driving capability

SDLP (m)

0.14

0.03

0.15

0.05

0.16

0.04

0.17

0.07

SD_Speed (km/h)

2.64

1.13

2.61

1.01

2.64

0.84

2.32

0.62

BRT (s)

1.18

0.32

1.35

0.42

1.26

0.32

1.46

0.47

Compensatory

behavior

Average speed

(km/h)

51.20

1.25

51.38

1.60

51.91

2.16

50.61

1.26

Time headway (s)

2.25

0.56

2.50

0.59

2.20

0.44

2.42

0.59

Collision risk

TET (s)

1.32

0.43

0.70

0.63

1.17

0.53

0.78

0.60

TIT (s

)

1.10

0.71

0.51

0.68

1.01

0.79

0.63

0.70

4.5.1 Driving capability

In this study, driving capability measures considered are SDLP, SD_Speed, and BRT.

SDLP reflects the ability of a driver to maintain lateral stability. Increase in SDLP

indicates the degradation of lateral control performance. As shown in Table 4.3, traffic

flow condition, event time, driver type, and driver age are found significantly associated

with the lateral stability all at the 5% level. For instance, SDLP of professional drivers

are lower than that of the non-professional drivers. Also, SDLP of older drivers are higher

than that of the younger drivers. In addition, SDLP after prolonged driving and under the

high traffic flow condition are found to be higher.

On the other hand, SD_Speed reflects the driver’s capability of longitudinal control.

Increase in SD_Speed implies the incapability to maintain good longitudinal control. As

also shown in Table 4.3, SD_Speed when driving under the high traffic flow condition

are higher than that when driving under the low traffic flow condition. However, no

evidence can be established for the association between longitudinal control, driver type,

and driving time. Lastly, BRT of older drivers are significantly longer than that of mid-

aged drivers. Also, BRT increase when driving under the high traffic flow condition and

after driving for 55-minute, all at the 1% significant level.

4.5.2 Compensatory behaviour

In this study, compensatory behaviour is indicated by the average driving speed and time

headway in the car-following task. As shown in Table 4.4, separated prediction models

are established for professional drivers and non-professional drivers. As shown in Table

4.4(a), there exists difference in the contributing factors to average driving speed between

professional and non-professional drivers. For non-professional drivers, driver age and

event time significantly affect the average speed, at the 1% level. Results indicate that

older non-professional drivers tend to drive at a lower speed in the car-following task, as

compared with the mid-aged non-professional drivers. In addition, non-professional

drivers would reduce the driving speed after prolonged driving. In contrast, no evidence

can be established for the association between the average driving speed of professional

drivers and the factors including driver age, driving time, and traffic flow condition. As

shown in Table 4.4(b), driver age and traffic flow condition are found associated with the

time headway of professional drivers. Older professional and older non-professional

drivers tend to keep a longer time headway when following a leading vehicle. In addition,

professional drivers tend to keep a longer time headway when driving under the high

traffic flow condition.

4.5.3 Safety effectiveness of the compensatory strategy

To indicate the effectiveness of compensatory driving behavior in enhancing driving

safety, two safety surrogate measures - TET and TIT are used. Table 4.5 illustrates the

results of random parameter models for the association between safety risk and possible

contributory factors. Zero sensitivity indicates to the changes in the likelihood of the

prevalence of traffic conflict (i.e. TET and TIT being greater than zero) given the per unit

change of possible attribute. As shown in Table 4.5(a), TET of older drivers is

significantly lower than that of mid-aged drivers, both for the professional drivers (at the

1% level) and non-professional (at the 5% level) drivers. This implies the lower likelihood

of severe traffic conflict of older drivers. In particular, the zero sensitivity of older driver

is -3.8% for professional drivers and -2.6% for non-professional drivers respectively. In

other words, the compensatory driving behaviors of older drivers are effective in reducing

the likelihood of traffic conflict, especially for the professional drivers. Similar findings

could be revealed for TIT. Again, as shown in Table 4.5(b), TIT of older drivers is

significantly lower than that of mid-aged drivers, both for the professional (at the 1%

level) and non-professional (at the 5% level) drivers. Also, reduction in the likelihood of

traffic conflict of older professional drivers (-4.0%) is more remarkable than that of older

non-professional drivers (-2.8%).

For the effect of time, as shown in Table 4.5(a), TET after driving for 55 minutes is

significantly lower than that after driving for 5 minutes among the professional drivers.

This implies the reduction in possible collision risk after prolonged driving. Specifically,

the zero sensitivity of event time for TET of professional driver is -2.1%. In contrast, no

evidence can be established for the relationship between TET and event time among the

non-professional drivers. Similar findings are revealed for TIT. Again, as shown in Table

4.5(b), TIT after driving for 55 minutes is significantly lower than that after driving for 5

minutes among the professional drivers, at the 1% level. In particular, the zero sensitivity

of event time for TIT of professional driver is -3.3%.

For the effect of traffic flow condition, as shown in Table 4.5(a) and Table 4.5(b), except

for the TET of professional driver, the likelihood of severe traffic conflict under the high

traffic flow condition is higher than that under the low traffic flow condition, both for the

professional and non-professional drivers. In particular, increase in the likelihood of

traffic conflict among non-professional drivers (4.2%) is apparently higher than that

among professional drivers (2.5%).

Table 4. 3 Estimation results of random intercept models for driving capability

Factor

Attribute

SDLP

SD_Speed

BRT

Coefficient

z-statistic

Coefficient

z-statistic

Coefficient

z-statistic

Constant

Mean

0.145**

24.08

2.321**

19.00

0.933**

13.67

S.D.

0.046**

19.29

0.731**

12.96

0.154**

4.65

Traffic flow

condition

High traffic flow

0.011*

2.08

0.305**

2.85

0.269**

4.41

(Control: Low traffic)

Event time

55 minutes

0.016**

3.11

0.192

1.76

0.360**

5.69

(Control: 5 minutes)

Driver type

Professional

-0.023**

-4.50

0.146

1.36

-0.075

-1.20

(Control: Non-professional)

Driver age

Older

0.014**

2.63

-0.192

-1.80

0.199**

3.20

(Control: Mid-aged)

Unrestricted loglikelihood

147.29

-88.93

-18.90

Restricted loglikelihood

126.21

-104.99

-40.66

Likelihood ratio test statistics

42.16**

32.12**

43.52**

* Statistical significance at the 5% level

** Statistical significance at the 1% level

Table 4. 4 Estimation results of random intercept models for compensatory behavior

(a) Average speed

Factor

Attribute

Professional driver

Non-professional driver

Coefficient

z-statistic

Coefficient

z-statistic

Constant

Mean

51.338**

143.74

52.134**

129.03

S.D.

0.792**

3.37

1.165**

6.65

Traffic flow

condition

High traffic

-0.011

-0.03

0.582

1.50

(Control: Low traffic)

Event time

55 min

-0.261

-0.69

-1.084**

-2.63

(Control: 5 min)

Driver age

Older

0.178

0.46

-1.299**

-3.41

(Control: Mid-aged)

Unrestricted loglikelihood

-65.09

-81.27

Restricted loglikelihood

-66.45

-90.08

Likelihood ratio test statistics

2.72

17.62**

* Statistical significance at the 5% level

** Statistical significance at the 1% level

(b) Time headway

Factor

Attribute

Professional driver

Non-professional driver

Coefficient

z-statistic

Coefficient

z-statistic

Constant

Mean

2.038**

14.79

2.264**

22.15

S.D.

0.371**

5.75

0.376**

6.96

Traffic flow

condition

High traffic

0.302*

2.42

-0.129

-1.25

(Control: Low traffic)

Event time

55 minutes

0.099

0.78

0.010

0.09

(Control: 5 minutes)

Driver age

Older

0.276*

2.18

0.216*

2.05

(Control: Mid-aged)

Unrestricted loglikelihood

-25.38

-28.36

Restricted loglikelihood

-32.76

-34.33

Likelihood ratio test statistics

14.76*

11.94*

* Statistical significance at the 5% level

** Statistical significance at the 1% level

Table 4. 5 Estimation results of random parameter Tobit models for safety risk

(a) TET

Factor

Attribute

Professional driver

Non-professional driver

Coefficient

Zero

sensitivity

Coefficient

Zero

sensitivity

Constant

1.38**

1.05**

Traffic flow

condition

High traffic

0.29

0.44*

2.31%

(Control; Low traffic)

Event time

55 minutes

-0.40**

-2.14%

-0.28

(Control: 5 minutes)

Driver age

Older

(Control: Mid-aged)

Mean

-0.72**

-3.84%

-0.50*

-2.63%

S.D.

0.58**

0.52**

Unrestricted loglikelihood

-30.53

-39.64

Restricted loglikelihood

-41.42

-47.71

Likelihood ratio test statistics

21.78**

16.14**

Maddala R

0.44

0.31

* Statistical significance at the 5% level

** Statistical significance at the 1% level

(b) TIT

Factor

Attribute

Professional driver

Non-professional

driver

Coefficient

Zero

sensitivity

Coefficient

Zero

sensitivity

Constant

1.18**

0.75*

Traffic

flow

condition

High traffic

0.47**

2.48%

0.78**

4.15%

(Control: Low

traffic)

Event

time

55 minutes

-0.63**

-3.34%

-0.40

(Control:5

minutes)

Driver age

Older

(Control:

Mid-aged)

Mean

-0.76**

-4.00%

-0.53*

-2.79%

S.D.

0.64**

0.52**

Unrestricted loglikelihood

-36.84

-46.57

Restricted loglikelihood

-42.26

-54.74

Likelihood ratio test statistics

10.84**

16.34**

Maddala R

0.25

0.31

* Statistical significance at the 5% level

** Statistical significance at the 1% level

4.6 Discussion

This study attempts to investigate the compensatory behavior of professional drivers,

coping with the elevated safety risk attributed to ageing, using the driving simulator study.

Many modern societies are facing the problem of ageing population. The proportion of

older peoples who hold driving licensing has been increasing rapidly. There is great

concern for the prevalence of older drivers in the transport sector, since the drivers’

physiological and cognitive performances may deteriorate with the increase in age.

However, as evidenced in this study, older drivers could reduce the anticipated risk by

compensatory behaviors, such as intentionally reducing the speed and increasing the time

headway when following a car. Furthermore, professional drivers usually have better

driving skills, such as detecting the road hazards and adapting for the demanding driving

task. This research provides a significant contribution to furthering our understanding of

the safety of older professional drivers by filling the knowledge gap on the difference in

compensatory driving behavior between professional and non-professional drivers.

This study found that degraded lateral control performance and longer brake reaction time

among older drivers, as compared with mid-aged drivers. This aligns with the previous

findings that driving performance deteriorates with age because of the impairments on

physical and cognitive performance (Biernacki and Lewkowicz, 2020; Chen et al., 2019;

Andrews and Westerman, 2012; Shanmugaratnam et al., 2010). Indeed, older drivers tend

to have longer perception-reaction time. It could then result in the delay and even absence

of evasive maneuver (Islam and Mannering, 2006; Yan et al., 2005).

However, this study also found that older drivers tend keep a longer time headway and

lower driving speed in the car-following tasks. Such findings justify that compensatory

behavior is prevalent among the older drivers (Dykstra et al., 2020; Molnar et al., 2008).

Indeed, previous studies revealed that older drivers tend to compensate for the elevated

crash risk resulted from cognitive impairment by reducing the driving speed and

increasing the time headway in car-following process (Bao et al., 2020; Martchouk et al.,

2011; Ni et al., 2010; Shinar et al., 2005). Furthermore, it is revealed that compensatory

behaviors (e.g. to keep a longer time headway) are more effective in reducing the safety

risk of older professional drivers, as compared to the non-professional drivers.

Above finding indicates that age should not be the only consideration for human resource

management of transport operators. For example, some bus drivers in Hong Kong are

compelled to work part-time or retire when they reach their 60s, without undergoing any

health and driving capability assessment. This study also provides useful insights into the

driver licensing policy for the transport authority. Licensing requirements for older

drivers vary among jurisdictions. Policy strategies including shortened time intervals

between license renewals, mandatory health assessments, visual acuity checks and

driving tests, and defensive driving courses for older drivers are implemented (Transport

Department of HKSAR, 2020; Transport for New South Wales, 2015; Thomas et al.,

2013). It would be of essence to assess the capability of compensatory strategy of older

drivers in the driving tests for license renewals. In addition, older professional drivers can

shed some light on the driving skills for the younger cohorts.

4.6.1 Effective compensatory strategy of professional drivers

The findings of this study showed that older professional drivers showed a greater

reduction in the likelihood of traffic conflict. In other words, the compensatory strategy

adopted by the older professional drivers is more effective as compared with the older

non-professional drivers. This is consistent to the findings of previous studies that

existence of compensatory strategy is closely related to driving experience (Farrow and

Reynolds, 2012; Andrews and Westerman, 2012). Professional drivers are good at

identifying hazards since they have higher exposure on roads. Therefore, their driving

performances are better than that of non-professional drivers (Borowsky and Oron-Gilad,

2013). As experience accumulated over times, older professional drivers can maintain the

satisfactory driving performance (Newnam et al., 2018; Chen et al., 2019). While a recent

study reported that the driving experience of older professional drivers can not

compensate for their reaction slowed down by aging (Medic-Pericevic et al., 2020), our

results provide evidence of effective compensatory strategy reducing their safety risk.

Findings of this research contribute further to the literature on the safety of older

professional drivers from the behavioral perspectives. Transport operators can develop

tailored management strategies for older drivers to keep the drivers in the workforce for

as long as safely possible (Newnam et al., 2020; Newnam and Watson, 2011). For

example, regular assessments of cognitive performance and driving skills (including

effective compensatory behaviors) for the older drivers can be implemented. In addition,

training courses and driver enhancement programs on hazard identification and defensive

driving skills can be provided. Yet, it is worth exploring the relationship between driver

performance, safety perception and hazard identification skills based on cognitive

assessment and perception survey in the extended study (Chen et al., 2020).

4.6.2 Strategic adaptation of professional driver

Strategic adaptation refers to the intentional modification of driving behavior to adapt for

the prolonged driving or hazardous conditions. It is expected that strategic adaptation of

professional drivers is more prevalent than the non-professional drivers, and the elevated

crash risk of professional drivers after prolonged driving can be marginal. For the effect

of driving time, our results showed the impaired lateral control and longer brake reaction

time after driving for 55 minutes, as compared with those after 5 minutes. This could be

attributed to the existence of possible psychological fatigue. As also indicated in previous

driving simulator studies, greater variations in lateral position, longitudinal speed and

steering angle were observed after driving for 30 to 60 minutes (Chen et al., 2019; Ting

et al., 2008; Otmani et al., 2005). Interestingly, despite the degraded driving performance

and slower response over time, results indicate that the likelihood of traffic conflict of

professional drivers remarkably reduced after driving for 55 minutes, while there was no

such finding for the non-professional drivers. One possible explanation is that

professional drivers adopted strategic adaptation – that is, adjusting their behaviors to

accommodate the driving task. For example, professional drivers may adapt to the

situation by reducing the driving speed where appropriate (Smiley and Rudin-Brown,

2020; Williamson et al., 2002; Cnossen et al., 2004). In particular, detection of possible

fatigue and potential road hazards could trigger the strategic adaptation of professional

drivers (Filtness et al., 2012; Williamson et al., 2014). Meng et al.’s (2015) study suggests

that professional drivers are usually more confident in coping with fatigue given the rich

experience in long driving and working time. Moreover, Iseland et al.’s (2018) study

affirms that long-haul truck drivers usually engage in various secondary tasks

intentionally to get rid of the tedious driving task and maintain the level of alertness. Just,

no evidence could be established for the association between prolonged driving time and

presence of adaptation behaviors (e.g. reduction in driving speed or increase in time

headway) among the professional drivers in this study. As such, it is worth exploring the

adaption behavior of professional driver using alternate behavioral and psychological

metrics in the extended study.

Moreover, current study also considers the effects of traffic flow condition on the driving

capability, compensatory behavior, and safety risk. Results indicate the increase in brake

reaction time, and the degradations in lateral and longitudinal controls when driving under

high traffic condition. It could be attributed to the increase in visual stimuli and mental

workload, given the increase in surrounding vehicular traffic and pedestrian (Cantin et

al., 2009). However, though professional drivers tend to adopt the longer time headway

under the high traffic condition, current results indicate that their safety risk still increases.

It is worth exploring the relationship between traffic volume, strategic adaptation and

potential crash risk based on empirical observation survey in the extended study.

4.6.3 Study limitations

The findings from this research should be interpreted in the context of the limitations.

First, the ability of simulator studies to reflect realistic driving is often questioned.

However, many previous studies have demonstrated the absolute and relative validity of

the simulator experiment (Wynne et al., 2020; Meuleners and Fraser, 2015). Moreover,

the real-world driving data have been successfully explained by the findings from the

simulator research (Saifuzzaman et al., 2015). In this study, a high-fidelity driving

simulator was used. The driving scenario replicating the local environment in Hong Kong

was created with high-resolution images trying to simulate the real-life scenery as much

as possible. Nevertheless, naturalistic driving studies could aid in understanding the

interaction between the compensatory strategy of older drivers and the increased driving

experience of professional drivers.

Second, as compared with the random parameters linear model, hazard-based duration

models can be applied to analyse drivers’ reaction times to improve the estimation

accuracy in future study (Ali et al., 2019; Choudhary and Velaga, 2017). For example,

Haque and Washington (2014, 2015) used a parametric accelerated failure time (AFT)

duration model with a Weibull distribution to identify the factors affecting drivers’

reaction times. Results confirmed that the Weibull-AFT model with gamma heterogeneity

showed the best statistical fit.

Third, the cognitive ability of older drivers is not examined in current study. As

compensatory strategy is prevalent for the drivers who have known cognitive impairment,

traffic violation and crash involvement records (Wong et al., 2012; Molnar et al., 2008;

Charlton et al., 2006), it is worth collecting the data of safety perception, hazard

identification skills, and cognitive ability. Therefore, the association between these

human factors and compensatory strategy of older professional drivers can be measured.

Forth, the driving simulator experiment in this study involves the car-following task for

one hour. Presence of strategic adaptation is examined based on the changes in driving

performances between two time points (i.e. 5 minute and 55 minute). Despite that possible

driving simulator sickness can be avoided, one-hour drive may not be sufficient to induce

driver fatigue. In the extended study, it is possible to investigate the strategic adaptation

behaviors of long-haul drivers using naturalistic driving study (Mahajan et al., 2019).

Moreover, car following behavior of professional drivers at work could be influenced by

time pressure and market competition. However, it is not possible to incorporate the effect

of work pressure in the driving simulator study. In the extended study, it is worth

exploring the effect of work pressure when comprehensive information is available using

naturalistic driving study. Also, robustness of the results can be improved when the

sample size of each experimental group increases.

4.7 Concluding remarks

This simulator study investigated the effectiveness of the compensatory strategy adopted

by older professional drivers as compared with older non-professional drivers.

Specifically, the safety effects of compensatory behaviors on the rear-end conflict risk

were examined. Two modified traffic conflict measures: time exposed time-to-collision

(TET) and time integrated time-to-collision (TIT) were adopted to indicate the risk of

severe rear-end traffic conflict in the car-following tasks. Possible changes in the conflict

risk could indicate the effectiveness of compensatory strategy. Results reveal the longer

brake reaction time and greater variability in lateral position of older drivers as compared

with the mid-aged drivers, while the time headway of older drivers is longer. This

demonstrates the degradation in driving capability and the presence of compensatory

behavior among older drivers. More importantly, the effectiveness of compensatory

strategy is more profound among the older professional drivers, as compared to the older

non-professional drivers, given that the reduction in conflict risk among professional

drivers is more remarkable. The focus of existing research has tended to be on the

compensation mechanism of older drivers while few have considered that of the older

professional drivers. As anticipated, older professional drivers are able to adopt more

effective compensatory strategy to reduce the rear-end crash risk by capitalizing on their

rich experience. In the near future, the proportion of older drivers in the transportation

industry would continue to increase and older drivers would become a major cohort.

Findings of this research provide useful insights into the driver management strategies

tailored for older drivers. For example, not only the regular health checks, but also the

comprehensive assessments of cognitive performance and driving skills for older drivers

should be introduced. Furthermore, special training program that can improve the hazard

identification and defensive driving skills of professional drivers will be of essence. Yet,

it is worth investigating the effectiveness of driver education and training in improving

the safety of older professional drivers in the long run.

Chapter 5 Perceptions and attitudes of professional drivers

5.1 Introduction

In this study, we examine the effectiveness of a fixed ASEC system in Hong Kong to

deter speeding. While Hong Kong employs a combination of human agent-based mobile

speed enforcement mechanisms as well as a fixed ASEC system, the focus will be on a

fixed ASEC system in this paper. In Hong Kong, the shares of speed enforcement

prosecutions based on human agent-based mobile speed enforcement and a fixed ASEC

system are about the same (Hong Kong Police Force, 2018). From time to time, strong

public sentiment has been expressed to expand the ASEC system as a means not only to

enhance the deterrent effect, but also to reduce the costs associated with police human

resources. In this context, it become particularly imperative to evaluate the impacts of

alternative designs for such an expanded ASEC system. While there may be benefits to

supplementing an expanded automation-based ASEC speed enforcement mechanism

with a much smaller base (relative to today) of human-based enforcement mechanisms,

examining the possible optimal combination of investments in such fused mechanisms is

not considered here. In any case, society has consistently moved closer to automation in

traffic operations, and it is not inconceivable at all that there will be a time in the near

future when no human-based resources (police personnel) will be invested on the task of

field monitoring of speed for enforcement purposes.

Four main attributes associated with threat and coping appraisals related to an ASEC

system are evaluated in the paper: DOP penalty, fine levels, camera-to-housing ratio

(explained in detail later), and the placement of the warning sign. Among these four

attributes, the first three may be considered to be associated with threat appraisal, while

the last may be considered to be associated with coping appraisal (for instance, if a

warning sign is placed farther away from the camera location, it may provide individuals

with more time to absorb the information and act to adjust their speed to comply with the

speed limit before arriving within the range of the camera detection zone). A stated

preference experiment is conducted by developing scenarios that combine the attribute

levels of the four attributes just identified. The scenarios are presented to professional

drivers, who are asked to respond by choosing a speed level at which they would travel

on a 50 km/h road at each of three sections of a roadway (corresponding to a standard

section with no enforcement and no warning, a warning section that starts from 23 meters

ahead of the placement of a warning sign, and the camera housing section itself in which

a camera detects speeding violations).

Driver perceptions regarding speeding consequences and driving history (current level of

DOP points, whether received a speeding ticket in the past 12 months, and exposure to

ASEC systems when driving), as well as driver demographic characteristics and

employment characteristics, are also collected in the survey. These variables are

considered as direct influencers of travel speed as well as moderating the impact of the

four main attributes of the SP experiment (to capture inter-individual differences in

perceptions of threat appraisal and coping appraisal of speed enforcement, as well as

overall intentions to speed or not and general attitudes toward the risks travel speeding

poses to society). In doing so, we attempt to recognize the direct and moderating effects

of driver characteristics on travel speed levels, and contribute further to the literature on

the effectiveness of speeding enforcement mechanisms. Many earlier studies of

enforcement mechanisms, on the other hand, have considered drivers as a single

monolithic group or considered variations across drivers in a relatively limited manner.

In addition, unlike many other earlier studies on professional driver speed decisions, we

consider unobserved individual-specific heterogeneity to accommodate unobserved

individual factors that are likely to influence speed choices. Such heterogeneity is

important to consider in travel choice and safety studies to ensure consistent estimation

of model parameters (see, for example, Mannering et al., 2016).

In this chapter, Section 5.2 and 5.3 discusses the stated preference survey used for data

collection as well as the methodology for our analysis, respectively. Section 5.4 provides

a description of the sample used in the analysis. Section 5.5 presents the results. Section

5.6 concludes the paper with a summary of the findings and policy implications.

5.2 Stated preference survey design

The data used in the current analysis is drawn from a face-to-face survey conducted

during the period from October 2018 to February 2019 (months inclusive). Our emphasis

on a face-to-face survey is to avoid respondent biases that may accrue from less expensive

web-based and other social media-based surveys. The professional driver participants

were approached either at on-road parking areas (e.g. public bus, taxi, and public light

bus stations) or outside the licensing offices of the Hong Kong Transport Department.

The inclusion criteria were (1) having valid licences of bus, minibus, taxi or goods (cargo)

vehicles, and (2) driving for income, either full-time or part-time. Prior to the survey, the

ethical approval from the Human Subjects Ethics Sub-committee (HSESC) of the Hong

Kong Polytechnic University was obtained.

The questionnaire had three sections: (1) SP questions regarding speed choices, (2)

Driving history and safety perceptions, and (3) Demographics and employment

characteristics of professional drivers. The SP part is discussed in the next section. The

second section collected information on the involvement with traffic offences and

crashes, attitudes towards different speed enforcement measures, and actual experience

with speed enforcement. The third section collected information on driver demographics

(gender, age, education, marital status, and income) and employment characteristics

(salary system, driving hours per day etc.)

In this study, drivers’ perceptions and attitudes towards the deterrent effect of

enforcement and penalty against speeding was gauged using their stated speed choices in

an SP survey design. SP surveys have been widely applied to evaluate the effects of

enforcement strategies and speeding penalties on the propensity for traffic offences by

measuring the driver’s response under hypothetically constructed conditions (Hössinger

and Berger 2012; Li et al., 2016; Ryeng, 2012; Wong et al., 2008). The SP questions in

the current paper are based on the scenario of driving on an urban road with a speed limit

of 50km/h. For each question, three speed choices are presented to drivers for each of

three location sections. The location sections are defined as follows: (1) a standard section,

defined as one with neither ASEC-based speed enforcement and nor warning signs of

such enforcement, (2) a warning section, defined as the road section indicating the

presence of speed camera housing unit ahead (this section starts 23 meters ahead of the

warning sign and ends at the location of warning sign; the design of the section length is

based on the vision standard for the driver licensing requirement in Hong Kong), and (3)

a camera section, defined as being within the range of speed violation detection by the

camera (this section starts 23 meters ahead of a camera housing unit and ends at the

location of the housing; see Figure 5.1). The three speed choices (one to be selected) are:

(1) comply with the prescribed speed limit; (2) exceed the prescribed speed limit by 15

km/h or less (traveling at 51-65 kms./hour, corresponding to speeding range 1); and (3)

exceed the prescribed speed limit by more than 15 km/h but less than or equal to 30 km/h

(traveling at 66-80 kms./hour, corresponding to speeding range 2). Thus, for each SP

question presented, the respondent makes a speed choice at each of the three location

sections, providing three choices.

In each of the SP questions presented to respondents, four attributes are used to

characterize the choice context: (1) Driving Offence Points (DOP) for different ranges of

speeding infractions, (2) Monetary fines for different ranges of speeding infractions, (3)

Camera-to-housing ratio, and (4) placement of the warning sign that determines the

distance of the warning section. A screenshot of the content and format of a sample SP

question is provided in Figure 5.1.

The levels of the first attribute - DOP – were set by pivoting off the current DOP for each

of the two speed infraction ranges (of course, there are no DOPs for being within the

speed limit). The current DOPs are zero for speeding range 1 and three for speeding range

2. We used these base DOPs and also introduced a higher DOP level of two for speeding

range 1 and a DOP level of five for speeding range 2. Thus, for each speeding range, there

are two possible DOP levels, and across the two speeding ranges, there are a total of four

possible DOP levels.

The levels of the second attribute – monetary fine – were also set based on the current

fine levels of 320 HKD (about US $40) for speeding range 1 and 450 HKD (about US

$57) for speeding range 2. Again, we used these base fine levels, and also introduced

increased levels of 420 HKD (about US $54) for speeding range 1 and 550 HKD (about

US $ 70) for speeding range 2. Across the two speeding ranges, there are a total of four

possible fine levels.

In Hong Kong, not all the camera housings necessarily contain a speed camera, to save

on costs (both installation and operating costs). Thus, while Hong Kong laws require that

citizens be informed of any camera locations, it is not required that all the announced

camera locations necessarily have an actual functional camera. Dummy camera housing

boxes are allowed to be installed. However, the ratio of actual speed cameras to camera

housings must be publicized. The current ratio of speed camera-to-housing is 1:6. In

particular, there are 20 speed cameras and 120 housings across the entire territory of Hong

Kong (Audit Commission of HKSAR, 2013). Four levels of the third attribute -- camera-

to-housing ratio -- are set out by either increasing the number of housings or increasing

the number of cameras: 20:240, 20:120 (status quo), 40:120, and 60:120. An analysis of

how Hong Kong professional drivers respond to different camera-to-housing levels can

inform speed enforcement strategies considering the economic constraints of the transport

authority.

Finally, four levels of the fourth attribute associated with the placement of the warning

sign are considered: 50 meters, 100 meters, 150 meters, and 200 meters upstream of the

speed camera housing (see Figure 5.1). According to the Transport Department, to

remind drivers of the presence of a speed enforcement camera ahead, the distance

between the camera housing unit and the warning sign is about 100m for most of the

enforcement sites. Exploring the effect of the placement of the warning sign helps better

understand alternative coping mechanisms, and can provide insights regarding the

optimal placement of the warning sign that can minimize the “Kangaroo effect”

associated with speed cameras.

Scenario 1:

If you are at the Standard section, at which speed range would you travel? (choose one option

from below)

 < 50 km/h  51-65 km/h  66-80 km/h

If you are at the Warning Section, at which speed range would you travel? (choose one option

from below)

 < 50 km/h  51-65 km/h  66-80 km/h

If you are at the Camera Housing Section, at which speed range would you travel? (choose

one option from below)

 < 50 km/h  51-65 km/h  66-80 km/h

Background information

Speed（km/h）

< 50

51 - 65

66 - 80

Penalty for speed

violation

➢ DOPs

➢ Fine

HK$ 320

HK$ 450

Camera-to-Housing ratio

20 cameras in 240 housing units

Location of the warning sign

Warning sign placed 50 meters ahead of housing unit

Figure 5. 1 Illustration of the location type and a hypothetical scenario for the stated

preference game

All the levels for each of the attributes were tested extensively for reasonability in pilot

surveys, and several changes were made before arriving at the final levels. In all, the SP

experiments have four factors, each with four levels. If the full factorial design were

considered, there would be 256 (4  4  4  4) combinations of factor attributes in

total for the SP question. It is however not efficient and feasible to gauge the drivers’

perceptions and attitudes if all the 256 combinations of scenarios are used. Therefore, an

orthogonal fractional factorial design (Bhat and Sardesai, 2006; Hössinger and Berger,

2012; Lavieri and Bhat, 2019; Li et al., 2014) was adopted to reduce the number of

combinations from 256 to 16. Further, our design enabled us to estimate models that are

more general than the multinomial logit model by maintaining factor orthogonality within

and between alternatives. Our design allowed for the estimation of main effects of

attributes, as well as two-way interaction effects between attributes and respondent

characteristics. Next, we developed a block design of four sets of four SP scenarios,

because it would be too much burden to ask each respondent to answer 16 SP questions.

Each participant was then presented with one of the four blocks of four SP scenarios in

the survey. The entire survey instrument is available at http://www.baige.me/v?i=RxE.

5.3 Data collection and sample used

A total of 401 professional drivers completed the questionnaire survey. Therefore, the

dataset has a total of 40112=4,812 SP choice occasions, with 1,604 choice occasions at

each of the three location sections (standard, warning, and camera). The distribution of

the dependent variable was as follows within the 1,604 choice occasions, as also shown

in Table 5.1: (1) Standard section – Not speeding (14.1%), Speeding Range 1 (71.2%),

and Speeding Range 2 (14.7%), (2) Warning section – Not speeding (57.2%), Speeding

Range 1 (40.0%), and Speeding Range 2 (2.8%), (3) Camera housing section – Not

speeding (99.8%), Speeding Range 1 (0.2%), and Speeding Range 2 (0%). As can be

observed from these descriptive statistics, drivers combine their threat and coping

appraisals due to which a large proportion of them are generally willing to speed at the

standard section (at least at speed range 1), but are more likely to adhere to the speed limit

at the camera housing section. Indeed, there is literally no variation in adherence at the

camera housing section regardless of the levels of DOP, monetary fine, camera-to-

housing ratio, warning sign placement, as well as driver characteristics. Thus, we drop

the 1,604 choice occasion observations corresponding to the camera housing section in

our analysis, because they do not contribute to understanding the effects of independent

variables on speeding ranges. The final sample for analysis includes the 3208 choice

occasions at the standard and warning sections.

Table 5. 1 Distribution of speed choices by location type

Section

Speed choice

Speed compliance

Speeding range 1

Speeding range 2

(<50 km/h)

(51-65 km/h)

(66-80 km/h)

Count

Standard

226

14.1

1142

71.2

236

14.7

Warning

918

57.2

641

40.0

2.8

Camera Housing

1600

99.8

0.2

Table 5.2 shows cross-tabulations of the four SP attributes with speed choice percentages

at each of the standard and warning sections. As expected, increasing the DOP penalty

by two points decreased the percentage of drivers choosing for speeding range 1 and

speeding range 2 in the standard section. Further, increasing the DOP penalty by two

points for both speeding levels led to a greater percentage of drivers complying with speed

limit. The descriptive statistics do not show a clear trend of the speed choices with respect

to increasing monetary fines. Interestingly, in the standard section, it seems that a greater

proportion of drivers choose to speed when the fine is increased. More discussion on this

will follow in the model results section. In the context of camera-to-housing ratio values,

an increase in the ratio from status quo (20:120) to 40:120 shows a greater decrease in the

percentage of drivers choosing speed ranges 1 or 2 than that from increasing the ratio

further to 60:120. It appears that the bang per buck is greater for increasing the ratio from

20:120 to 40:120 than that to 60:120. As for the placement of warning sign, there is a

monotonous trend of increasing percentage of speed compliance choice with decreasing

distance between the warning sign and the camera housing location.

Of course, the discussion above does not consider differential effects of the SP attributes

based on observed and unobserved driver characteristics, which is the focus of the

multivariate model results in Section 5.4.

Table 5. 2 Crosstabulation of SP attributes with speed choices at plain and warning sections

Factor

SP attribute level for different

speeds (<50kmph, 51-65kmph,

66-80kmph)

Speed choice %

Standard section

Warning section

Speed

compliance

Speeding

range 1

Speeding

range 2

Speed

compliance

Speeding

range 1

Speeding

range 2

<50 51-65 66-80 km/h

DOPs

0 0 3 (status quo)

13.2%

69.6%

17.2%

54.6%

2.6%

2.8%

0 0 5

12.0%

73.6%

14.4%

52.5%

3.8%

3.7%

0 2 3

13.2%

68.6%

18.2%

53.7%

4.3%

2.0%

0 2 5

15.5%

73.1%

11.4%

68.1%

9.2%

2.7%

Monetary fine (HK$)

0 320 450 (status quo)

11.8%

73.3%

14.9%

54.8%

1.3%

3.9%

0 320 550

15.2%

64.6%

20.2%

57.4%

9.9%

2.7%

0 420 450

14.5%

78.6%

6.9%

59.9%

8.7%

1.4%

0 420 550

15.0%

67.3%

17.7%

56.9%

0.1%

3.0%

Camera-to-Housing ratio

20:240

12.7%

71.8%

15.5%

57.9%

8.7%

3.4%

20: 120 (status quo)

13.5%

67.3%

19.2%

55.6%

0.1%

4.3%

40:120

13.7%

72.6%

13.7%

58.4%

8.7%

2.9%

60:120

16.2%

72.1%

11.7%

56.1%

2.4%

1.5%

Placement of warning sign

50m upstream

77.8%

2.2%

100m upstream (status quo)

65.3%

1.2%

3.5%

150 m upstream

43.1%

3.9%

200m upstream

42.6%

1.6%

5.8%

5.3.1 Driver demographics and employment characteristics

Table 5.3 presents the demographic and other characteristics of the 401 participants,

beginning with the demographic characteristics in the first set of rows. All participants of

this study are male. This is consistent with the distribution of employed persons by

occupation and gender in the population census dataset, which indicates that 97% of

workers in the machine operation sector are male (Census and Statistic Department,

2018a). Although the information on the official registry of professional drivers in Hong

Kong is not available, male drivers are believed to dominate the transport sector. The age

distribution of our sample is close to that of the driving licensing record of general drivers

in Hong Kong (Transport Department, 2017a). In terms of educational background, 79%

of the drivers in our sample have attained at least secondary education (the closest

possible comparison at the Hong Kong-wide level is that 89% of male workers in Hong

Kong have attained secondary education (Census and Statistic Department, 2018b). In

our sample, 73% of the drivers were married (the closest possible comparison is the most

updated marital status statistics in Hong Kong, which indicates that 62% of the males are

married (Census and Statistic Department, 2018c). Interestingly, almost all (395 of the

401) drivers provided their monthly income values. For the remaining six drivers who did

not provide this information, we imputed the income values based on the procedure

discussed in Bhat (1997). A little over 31% of the drivers have a monthly income below

HK$ 15,000 and a little over 21% of the sample earn over HK$ 20,000.

Table 5. 3 Distribution of the sample

Variable

Count

Demographics

Gender (Male)

401

100%

Age

Older (>55 years old)

24.4

Younger (<45 years old)

151

37.7

Mid-aged (46-55 years old)

152

37.9

Education

Primary or below

20.9

Secondary or above

317

79.1

Marital status

Married

293

73.1

Unmarried

108

26.9

Monthly income

less than 15K

127

31.7

Between 15K and 20K

183

45.6

More than 20K

21.2

Operational characteristics

Salary system

Trip-based

136

33.9

Monthly-based

126

31.4

Others (hourly or shift based)

139

34.7

Daily driving hours

More than 9 hours

168

41.9

Less than 8 hours

9.7

8 to 9 hours (normal working hours)

194

48.4

Work time per week

less than or equal to 48 hours

184

45.9

more than or equal to 63 hours

9.2

Others

179

44.9

Vehicle type

Bus

16.7

Taxi and Red Minibus

157

39.2

Green minibus

14.0

Goods vehicle

121

30.2

Driver history and safety perceptions

Received speeding ticket(s)

Yes

24.7

302

75.3

Perceive speeding as a cause of injury

High

281

70.1

Low

1.5

Neutral

114

28.4

Perceive speeding cameras are effective

High

270

67.3

Low

6.0

Neutral

107

26.7

Frequency of sighting cameras

High (7-10 times in 10 trips)

250

62.3

Medium (4-6 times in 10 trips)

24.4

Low (0-3 times in 10 trips)

13.3

Drivers’ employment characteristics are presented in the next set of rows in the table. The

salary system of professional drivers is stratified into three categories: (i) trip-based (34%

of the sample), (ii) monthly-based (31%), and (iii) others (hourly or shift based, 35%).

The trip-based drivers are self-employed, and their incomes vary greatly with the number

and distance of trips made (e.g. taxi, red minibus and light van drivers). The drivers who

are paid on a monthly basis are usually regular employees of a large corporation or

transport operator, such as the franchised bus companies and logistic firms. The hourly

or shift based drivers are usually (full-time or part-time) employees of small transport

operators, such as the green minibus. Their salaries vary greatly with the daily working

time. As for the daily driving hours, 8% of our sample drive for less than or equal to 7

hours per day while 42% of them drive for more than 9 hours daily. The corresponding

statistic from official reports is not accessible. The closest possible comparison is that 51%

of bus drivers in Hong Kong drive for more than 9 hours daily (Legislative Council of

HKSAR, 2018). In terms of weekly working hours, 46% of drivers in our sample work

for 48 hours or less per week, which is comparable to the 50% of employees in the

transport sector who work for less than or equal to 48 hours a week. However, only 9%

of our sample work for more than or equal to 63 hours per week, while the corresponding

percentage in the transport sector is close to 25% (Census and Statistic Department,

2018b). The commercial vehicles driven by our sample are categorized into four types –

bus, green minibus, taxi and red minibus, and goods vehicles (accounting for 17%, 14%,

39%, and 30% of the sample respectively). The official distribution for the vehicle types

of the commercial vehicle fleet in Hong Kong is not accessible.

Overall, the characteristics of drivers in the sample are reasonably close to general

expectations for Hong Kong professional drivers, at least based on the latest statistics

gleaned from the Census. Of course, one cannot be conclusive of the true

representativeness of our sample because there is no official registry of professional

drivers in Hong Kong, and the closest comparison we are able to make is with the

population census demographics for people employed in the transport sector.

5.3.2 Driver history and safety perceptions

The last set of rows in Table 5.3 report the descriptive statistics for driving history and

safety perceptions of the 401 participants, which might influence how they would respond

to the SP choice questions. As can be observed from these rows, 25% of the interviewed

professional drivers have received at least one speeding ticket in the recent past. 70% of

the drivers perceived speeding as a cause of injury while only 1.5% perceived a small

effect of speeding on traffic injuries. As for the perception on effectiveness of cameras,

67% of drivers believed that speeding cameras are effective in catching offenders, while

a smaller percentage (6%) perceived low effectiveness of this enforcement technique. The

frequency of drivers sighting camera housings was also collected in terms of the number

of times a driver would sight camera housings in 10 trips. It appears that a majority (62%)

of the drivers do usually visually locate camera housings at a frequency of at least 7 times

in 10 trips.

All the above driver history and perception variables are likely to influence drivers’

responses to the SP choice questions. Also, while we make no claim of our sample being

representative of the population of professional drivers, there is no reason to believe that

the individual-level relationship we develop between speed range choices and SP

attributes/driver characteristics would not be applicable for the general population of

professional drivers.

5.4 Statistical method

In this study, we formulate a panel mixed multinomial logit (or MMNL) model for the

speed choice of professional drivers. The panel MMNL model formulation

accommodates heterogeneity across individuals due to both observed and unobserved

individual attributes, while also recognizing correlations among the different observations

of a same individual. In the following discussion of the model structure, we will use the

index q (q = 1, 2, …, Q) for the decision-makers, i for the speed alternative (i = 1, 2, …,

I) and k for the choice occasion, i.e. SP choice occasions for a particular decision-maker,

(k = 1, 2, …, K). In the current study I = 3 (as indicated earlier, the choice alternatives are

speed compliance, or speeding range 1, or speeding range 2) and K = 43 = 12 for all q.

Within each of the four SP attribute scenarios presented, the respondents were asked to

state their speed range choice in three different sections – standard, warning, and camera

housing sections.

In the usual tradition of utility maximizing models of choice, we write the utility or

valuation that an individual q associates with the alternative i (speed range) on

choice occasion k as follows:

(1)

where is a -column vector affecting the valuation of individual q for

alternative i at the k

choice occasion, and that includes the following: (1) choice-

occasion specific attributes (that is, the four attributes varied in the SP experiments), (2)

alternative-specific constants for speeding ranges 1 and 2 (with no speeding being the

base category), (3) individual-specific attributes (driving history and perception, driver

demographics and employment characteristics), and (4) interactions within each of the

choice-specific and individual-specific variables, as well as across the two sets of

variables. β is a corresponding -column vector of the mean effects of the

coefficients of on speeding range valuations, and is another -column

vector with its m

element representing unobserved factors specific to individual q that

moderate the influence of the corresponding m

element of the vector . A natural

assumption is to consider the elements of the vector to be independent realizations

from a normal population distribution; . represents a choice-

occasion specific idiosyncratic random error term assumed to be identically and

independently standard Gumbel distributed. is assumed to be independent of .

For a given value of the vector , the probability that individual q will choose speed

range i at the k

choice occasion can be written in the usual multinomial logit form

(McFadden, 1978):

(2)

qik

qikqikqqik

xvU





= )(

qik

)1( M

qik

)1( M

qik

),0(~

mqm



qik



qik







xvx

qqik

qjkqqjk

qikqqik



The unconditional probability can then be computed as:

(3)

where F is the multivariate cumulative normal distribution and σ is a vector that stacks

up the elements across all m. The reader will note that the dimensionality in the

integration above is dependent on the number of elements in the vector.

In the MMNL model, marginal effects are calculated for variables . For the marginal

effects of dummy variables, they are calculated as the differences in the estimated

probabilities when the variables change from 0 to 1, while the means of other variables

are used for the computation. The marginal self and cross effects are computed as

(Shaheed and Gkritza, 2014; Xie et al., 2012; Washington et al., 2020):

P (1 P )

qik

i qik qik

qik





=−



(4)

qjk

i qik qjk

qik





=−



(5)

where Eq. (4) represents the effect that one unit change in has on the probability

for the decision-maker q to choose the speed alternative i on the choice occasion k

(denoted by P

qik

). Eq. (5) shows the effect of one unit change in the variable m of speed

alternative i (i≠j) on the probability (P

qjk

) for the decision-maker q to choose the speed

alternative j on the choice occasion k.

The parameters to be estimated in the model of Equation (3) are the β and σ vectors. To

develop the likelihood function for parameter estimation, we need the probability of each

individual's sequence of observed SP choices. Conditional on , the likelihood function

for individual q’s observed sequence of choices is:

)|()|(



qqikqik

vdvPP



= F



qik

(6)

where is a dummy variable taking the value of 1 if the q

individual chooses the i

speed range in the k

occasion, and 0 otherwise. The unconditional likelihood function

for individual q’s observed set of choices is:

(7)

The log-likelihood function is . We apply quasi-Monte Carlo

simulation techniques to approximate the integrals in the likelihood function and

maximize the logarithm of the resulting simulated likelihood function across all

individuals with respect to the parameters β and σ. Under rather weak regularity

conditions, the maximum (log) simulated likelihood (MSL) estimator is consistent,

asymptotically efficient, and asymptotically normal (see Hajivassiliou and Ruud, 1994;

Lee and Carter, 1992; McFadden and Train, 2000).

In the current study, we use Halton sequences to draw realizations for from its

assumed normal distribution. Details of the Halton sequence and the procedure to

generate this sequence are available in Bhat (2001, 2003).

5.5 Results and discussion

Table 5.4 presents the results of a panel mixed multinomial logit model estimated on the

afore-mentioned 3,208 observations – 1604 for the standard section and 1604 for the

 

 

= =













qqikqq

qik

vPvL

1 1

|)|(





qik





qqqq

vdFvLL )|()|(),(



),(ln),(



LL =

warning section

– with normal distributed random coefficients

. The dependent variable

is speed choice (i.e. speed compliance, speeding range 1, or speeding range 2; with speed

compliance considered as the base alternative). For each independent variable, a common

coefficient was estimated for both standard and warning sections as well as a difference

coefficient was introduced to account for the differential effect of that variable on the

warning section compared to the standard section. In Table 5.4, the parameter estimates

reported under the “Standard section” column are that of the common coefficients, which

may also be interpreted as coefficients for the standard section. The parameter estimates

under the “Difference between Warning and Standard section” column are the difference

coefficients. For a given variable, a sum of its common coefficient and the difference

coefficient would give its coefficient for the warning section. The parameter estimates

are interpreted and discussed next in Sections 5.4.1-5.4.4. The coefficients on the

constants indicate a general aversion to speeding, especially at level 2, at both the standard

and warning sections. This aversion is typically higher in the warning section than in the

standard section, though there is unobserved heterogeneity (captured by the significant

standard deviation estimates on the constants) in these general trends (the panel nature of

the data allows us to estimate the standard deviations on the constants in the table).

An important note is in order here. All results in this paper pertain to the influence of

variables on the reported speed choices in our stated experiments, not actual speed choices

in the real world. But, for presentation ease and tightness, we do not belabour over this

distinction in the rest of this paper and use the general word “speeding”. However, all our

statements should be viewed in the context of stated speed choices, not actual speed

choices.

Recall from the descriptive analysis of the SP choice data for the camera housing section that only a

single alternative (speed compliance) was chosen 99.8% of the times. So, these data were not included in

the model as the speed choice is deterministic in the camera housing section. This observation is

consistent with the findings of previous studies that drivers would slow down when they notice or are

warned of cameras (De Pauw et al., 2014a; De Pauw et al., 2014b; Elvik, 1997; Marciano and Norman,

2015).

We also explored alternative distributional assumptions such as log-normal for the random coefficients,

but the model with normal distribution provided the best fit. Besides, other distributions did not offer

substantive interpretations that were very different from the model with normal distributions.

Table 5. 4 Parameter estimates of a panel MMNL model for the speed choice of professional drivers*

Explanatory variables

Standard section

Difference between warning and

standard sections

Speed

compliance

Speed range 1

Speed range 2

Speed range 1

Speed range 2

Constants

Mean

0 (Fixed)

-1.35 (-2.52)

-10.14 (-1.77)

-5.59 (-3.71)

-13.16 (-4.64)

3.82

(10.73)

1.24 (2.46)

7.55 (6.5)

1.41 (1.67)

1.91 (2.18)

Stated Preference (SP) attributes

DOP

Mean

-0.17 (-1.96)

-0.48 (-1.80)

0.15 (1.82)

1.19 (5.20)

DOP x drivers with recent speeding ticket

-0.17 (-2.17)

Fines (in HK$ 100)

Mean

Dropped

-0.11 (-1.76)

0.14 (2.13)

Fine (in HK$ 100) x drivers with trip-based salary

-0.13 (-2.43)

Fine (in HK$ 100) x drivers with recent speeding

ticket

-0.23 (2.05)

Camera-to-

Housing ratio

(Base case:

status quo (20/120)

and (20/240)

Minor increase (40/120)

Mean

-1.92 (-2.7)

1.33 (2.8)

Major Increase (60/120)

Mean

-2.03 (-3.6)

Distance of

warning sign from

the camera section

(Base case: 100m)

50 m

-2.09 (-6.44)

-3.91 (-1.72)

150 m

2.42 (6.64)

4 (1.96)

200 m

2.31 (6.5)

9.2 (4.6)

Drivers' operational and perception characteristics

Age

(Base case: Mid-

age 46-55 years)

Older drivers (> 55

years)

-0.70 (-2.13)

Young drivers (<45

years)

3.75 (2.1)

Education status

(Base case:

secondary and

tertiary education)

Up to primary level

1.81 (2.99)

Marital status

(Base case:

unmarried)

Married

-0.45 (-2.56)

-2.59 (-2.68)

Monthly income

(in HK$)

(Base case: less

than 15K)

Between 15K and 20K

More than 20K

1.78 (2.49)

7.86 (1.94)

Salary system

(Base case: others)

Trip-based

1.37 (2.07)

10.32 (5.90)

Monthly

Daily driving

hours

More than 9 hours

-2.41 (-1.91)

-1.16 (-1.98)

-11.36 (-2.4)

Less than 8 hours

0.57 (2.71)

2.45 (2.32)

Vehicle Type

(Base case: Bus)

Green minibus

5.19 (5.77)

-2.71 (-2.6)

Goods vehicle

7.30 (7.87)

5.02 (2.35)

-3.06 (-3.4)

-3.31 (-2.58)

Red minibus and Taxi

5.77 (4.41)

-2.23 (-1.95)

Dummy if the driver recently received

speeding ticket

7.16 (2.93)

10.74 (3.35)

-5.02 (2.16)

Perception on

speeding as a cause

of injury

(Base case:

Neutral and high)

Low

6.00 (2.84)

Perception on

effectiveness of

cameras

(Base case: Low

and neutral)

High

-2.30 (-1.79)

Frequency of

sighting cameras

(Base case:

Medium (4-6/10)

and low (0-3/10)

High (>7 times per 10

trips)

-3.61 (-3.25)

-7.15 (-1.95)

Goodness of fit measures:

Number of cases

3208

Number of parameters

Log-likelihood of constants only model

-2911.09

Log-likelihood at convergence

-1279.21

Bayesian Information Criterion

3010.53

Base alternative: Speed compliance; IS: Statistically Insignificant at 90% confidence level.

5.5.1 Effects of penalty level and enforcement strategy

Among the SP attributes for penalty and enforcement, the DOP variable shows a

statistically significant deterrence on speeding in both standard and warning sections,

with higher deterrence in the warning section than in the standard section. Professional

drivers are indeed, generically speaking, sensitive to the increase in DOPs since incurring

DOPs may lead to disqualification of driving license, which is the source of their

livelihood (Wong et al., 2008). However, there is significant heterogeneity in the

influence of the DOP variable both due to observed and unobserved factors. Specifically,

drivers who were recently issued a ticket are more likely than their peers to be deterred

by DOPs when traveling in the warning section. Considerable unobserved heterogeneity

also exists in the influence of DOPs on drivers’ speeding choices in both standard and

warning sections. Interestingly, the standard deviation of the DOP coefficient in the

warning section is higher than in the standard section, implying that the deterrent effect

of an increased DOP penalty tends to be more diverse in the warning section despite its

greater deterrent effect on average. This finding could be attributed to the heterogeneity

in driver’s threat and coping appraisals of the warning messages (Kergoat et al., 2017),

as well as the effects of drivers’ characteristics on the comprehension of traffic signs (Ng

and Chan, 2008). For example, different drivers may perceive the self-efficacy of

avoiding the speeding penalty differently when forewarned about camera enforcement.

Thus, some drivers may actually initially increase their speeds as soon as they encounter

the warning section (to compensate for the fact that they have to reduce speeds at the

downstream camera section) because they feel confident in their ability (self-efficacy) to

estimate where the camera section will begin and in their ability to decelerate at the right

time to avoid speeding penalties in the camera section. Other drivers may immediately

reduce their speed upon encountering the warning section because they feel less confident

in their ability to take evasive speed reduction actions later downstream to avoid penalties

in the camera section. Such variations in self-efficacy are likely to get magnified as the

DOP penalty increases in the camera section, leading to the higher speed variance in the

warning section as the DOP penalty increases.

Unlike the deterrent effect of DOPs, the monetary fines variable turned out to have a

marginally positive coefficient in the standard section suggesting an increase in the

propensity for speeding with an increase in fines. While this may be a coping mechanism

to “make up” time in the standard section in anticipation of lost time due to adherence to

speed limits in the warning section, we noted that this effect had a strong interaction with

the length of the warning section. Thus, we chose to drop this variable and include the

length of the warning section as the primary determinant variable in our model (more on

this warning section length effect later). In the warning section itself, monetary fines are

associated with a negative coefficient for a majority of the sample (obtained from the

mean and standard deviation of the corresponding random coefficient), suggesting a

deterrent effect of monetary fine when it is combined with a warning of speed

enforcement ahead for a majority of the drivers. Furthermore, there is heterogeneity in

response to fines in the warning section based on driver characteristics. Specifically, in

warning sections, monetary fines have a larger deterrent effect (in the context of speeding)

for drivers who are paid on a per-trip basis and those with a recent speeding ticket relative

to other drivers. These results are again an illustration of the interplay between drivers’

threat and coping appraisal mechanisms, where drivers respond to the threat of a monetary

fine when they are made aware of the cameras that will increase the likelihood of them

being fined. And such interplay appears to vary across drivers based on both observed

and unobserved factors.

In the context of camera-based enforcement strategy, reducing the camera-to-housing

ratio from the status quo (i.e., from 20:120 to 20:240 camera-to-housing ratio) did not

show a statistically significant effect on the drivers’ stated speeding choices. However,

drivers were less likely to opt for severe speeding (range 2) in both the standard and

warning sections when the camera to housing ratio was increased from the status quo.

This is presumably because an increase in the number of camera installations would result

in an increased “threat” of being apprehended for speed limit violations. Interestingly, the

standard deviation associated with the coefficient of a minor increase in camera-to-

housing variable suggests that a small fraction (9%) of the drivers tend to choose speeding

with an increase in camera-to-housing. This result may be attributed to the risk-taking

behaviors of such individuals as well as heterogeneity in perceiving a threat of

apprehension due to a minor increase in the number of cameras. However, with a major

increase in the camera-to-housing ratio, this risk-taking behavior reduces, perhaps due to

a greater perception of the threat of apprehension.

The placement of the warning sign – that is, the distance of the warning sign from the

camera housing location – exhibits an influence on speeding in the warning section.

Specifically, reducing the distance between the warning sign and the housing unit leads

to lower speeding tendencies (for both speeding ranges). This is intuitive as individuals

may want to start slowing down (or at least not speed) to avoid sudden decelerations just

before arriving at the camera housing. In fact, the presence of a warning sign (upstream

of a fixed speed camera) has been found to be associated with reductions in mean driving

speed and proportion of more severe speeding (Retting et al., 2008; Høye, 2014). Kergoat

et al. (2017) postulated that the distance between warning sign and speed camera should

be increased to weaken the “Kangaroo effect”. However, the parameter estimates for

speeding range 2 suggest a heightened increase in the propensity to choose that speeding

range when the warning sign is installed 150m or 200m upstream of a camera housing.

That is, our results suggest that the deterrent effect of a warning sign could in fact be

diminished when the distance between the warning sign and the housing unit increases

excessively. That is, as drivers learn that the warning signs are placed farther away from

the housing, they speed up because they know they have a larger cushion to decelerate

and they also want to make up some time in anticipation of slowing down closer to the

actual camera housing location. Basically, as warning signs are placed farther away from

the camera housing, professional drivers start to view the early part of the warning section

as a “standard” section. This indicates a need for optimal placement of warning sign that

can tradeoff between the “Kangaroo effect” and effectiveness of the warning sign in

deterring speeding behavior.

5.5.2 Effects of demographic characteristics of professional drivers

Driver age does not have a strong association with speeding behavior in the standard

section. This could be because all professional drivers, regardless of age, tend to be more

aggressive when there is no speed enforcement and no warning (Öz et al., 2010a, Wong

et al., 2008). In contrast, in the warning section, older drivers are less likely to speed up

to range 1 and younger drivers are more likely to speed up to range 2. These results

suggest that the likelihood of speeding offences decreases with driver age, perhaps

because older drivers tend to be more cautious (Ram and Chand, 2016; Rosenbloom and

Shahar, 2007) but younger people are more likely to be sensation- and thrill-seeking

(Delhomme et al., 2012; Fernandes et al., 2010; Tseng, 2013). In the context of education

background, individuals with up to primary level education are more likely to speed up

to range 1 in both standard and warning sections. Previous studies also suggest that

professional drivers with higher education attainment are less likely to commit traffic

offences (Mallia et al., 2015; Mehdizadeh et al., 2018; Tronsmoen, 2010). Married drivers

(relative to those who are single) are less likely to speed in both the standard and warning

sections (see Mehdizadeh et al., 2018 and Wong et al., 2008 for similar findings), perhaps

because married individuals, due to their familial responsibilities, tend to be more

responsible in driving than single individuals.

Individuals with high monthly income (>20K), ceteris paribus, are more likely than

others to choose to violate speed limits in warning sections. This is perhaps because they

can afford to pay the fines. Also, recall from earlier discussion that the maximum fine of

HK$550 for speeding range 2 is a rather small percentage of HK$ 20K per month. In

contrast, the maximum monetary fine for speeding can reach 50% of average monthly

incomes of taxi drivers in the United States (United States Department of labor, 2018)

and 35% in the United Kingdom (Sentencing Council, 2017), respectively. In road safety

research, deterrence theory is widely used to investigate driver’s perception of the

sanctions (in terms of severity, certainty and celerity) for traffic offences (Kergoat et al.,

2017; Li et al., 2014; Tay, 2005a, 2005b, 2005c, 2009). It is based on the idea that people

avoid committing a crime due to the threat and fear of being legally punished, which also

involves an evaluation of the costs and benefits of the crime (Gibbs, 1985). In this sense,

the ratio of the cost (monetary fine) to the benefits (possible income) of speeding offence

is indeed quite low in Hong Kong.

5.5.3 Effects of operational characteristics of professional drivers

As discussed earlier, drivers who earn on a per-trip basis (i.e., trip-based salary) are more

likely to be deterred by monetary fines in the context of speeding in warning sections.

Regardless of the level of monetary fines, the coefficients of the trip-based salary dummy

variable suggest that such drivers are more likely than others to commit speeding offences

in both the standard and warning sections. Since their earnings depend on the number and

distance of the trips made, trip-based salaried drivers have a higher incentive to speed up

to arrive at the destination quickly. In Hong Kong, trip-based drivers (these are typically

drivers of taxis, light vans, red minibuses etc.) are generally self-employed and are not

well-regulated (Meng et al., 2017; Wong et al., 2008). In contrast, the monthly-salaried

drivers are typically regular employees of large transport operators and logistics firms

with good safety culture and driver management systems (Newnam et al., 2004; Öz et al.,

2010b, 2013) including GPS-based tracking of vehicle speeds. These factors also have a

bearing on the salary system-based differences in speeding choices.

Individuals who drive for more than nine hours per day have a lower inclination than

others to violate speed limits. This could be attributed to the possible driver fatigue caused

by a prolonged driving time. Drivers may adopt a compensation strategy by reducing their

speed to lower their risk of fatigue-related crashes (Williamson et al., 2002). In contrast,

individuals who drive for less than eight hours per day are associated with a greater

likelihood (than others) of violating speed limits in the warning section. This finding will

need further investigation to assess its robustness.

In the context of vehicle type, drivers of all types of vehicles other than buses have a

higher tendency of speeding up in both standard and warning sections, albeit they are

relatively less likely to speed up in warning sections than in standard sections. Indeed,

minibus drivers and taxi drivers in Hong Kong have been recognized as problematic and

risk-taking groups (Meng et al., 2017; Wong et al., 2008). On the other hand, goods

vehicle drivers are paid to drive for the transport of goods while bus drivers are to drive

for the transport of passengers. A greater sense of social responsibility on bus drivers

might make them less aggressive (at least in a stated preference setting) than the drivers

of other types of vehicles (Paleti et al., 2010).

5.5.4 Driver history and safety perceptions

Driving history and safety perceptions have a substantial influence on the participants’

stated speed choices. For instance, drivers who recently received a traffic ticket are

associated with a greater likelihood of speeding in both standard and warning sections

(albeit the tendency for speeding range 2 is lower in warning sections than that in standard

sections). Further, as discussed earlier in the context of interaction between this variable

with the SP attributes, increasing fines or DOP appears to reduce the speeding tendency

of these drivers in warning sections. However, even at the highest level of fine and DOP

values presented in the SP experiment, these drivers show a higher tendency (than others

without recent tickets) to violate speed limits. These results suggest that risk-taking

behavior and aggressive driving styles of these drivers overshadow any deterrent effect

from receiving a speeding ticket (Sagberg and Ingebrigtsen, 2018). It appears that simply

imposing fines or DOPs might not suffice to reduce the aggressive driving traits of such

drivers. This result suggests a need for additional investigations to assess the effectiveness

of combining DOPs and fines with driver training programs aimed to reduce risk-taking

and aggressive driving traits.

Individuals who perceive that speeding does not cause injuries have a higher tendency of

opting for speed range 2 in both standard and warning sections. This aligns with the

previous findings that drivers with lower risk perception tend to be associated with

aggressive driving behaviors (Cestac et al., 2011; Rosenbloom, 2003). In addition, drivers

who perceive that cameras are highly effective in catching offenders are associated with

a lower tendency of speeding in speed range 2 in the warning section, while their

disposition for speed range 1 is not statistically different from compliance. Individuals

who sight speed enforcement camera housings more frequently (in at least 7 out of 10

trips) have a lower tendency of speeding in range 2 (in both standard and warning

sections). This could be attributed to the perceived higher level of enforcement, which

may contribute to the decrease in driver’s speeding intention (Blincoe et al., 2006;

Hössinger and Berger, 2012) at least in the high-speed range.

5.5.5 Marginal effects due to changes in SP attributes

The model was applied to estimate marginal effects on market shares (of speed choice)

in response to changes in the SP attributes. As shown in Table 5.5, the marginal effects

were computed for both the standard and warning sections. According to these results, an

increase in the DOP by 1 point resulted in greater than 4% increase in compliance in both

the sections. In the context of monetary fines, a 10% increase resulted in only a 1.73%

increase in compliance. Such a low marginal effect is consistent with the discussion of

model estimation results that monetary fines alone might not significantly deter

professional drivers from speed violations. Note that the percentage reduction in the share

of drivers who would opt for speeding range 2 is high (13.02%). However, such a high

percentage reduction is an artifact of a rather small proportion of drivers choosing this

option in the base case.

Increasing camera-to-housing ratio from the status quo (20:120) to 40:120 shows a

considerable (at least 29%) decrease in the share of drivers choosing speed range 2.

However, the decrease is not substantial when the ratio is increased to 60:120. This

suggests that the marginal benefit from increasing the camera-to-housing ratio beyond

40:120 might not be substantial. Furthermore, since the proportion of drivers choosing

speed range 2 is itself very small (1%), even a 32% decrease in this share due to increasing

the ratio to 60:120 does not appear to hold practical effectiveness.

In the context of the placement of warning sign, increase in the distance between the

warning sign from 100m is associated with a substantial increase in the proportion of

drivers choosing to speed in the warning section. Even if we neglect these increases for

speed range 2 (due to a rather small base market share for this alternative), the increases

in the proportion of people choosing speed range 1 is substantial when the distance is

increased. These results suggest the need for an optimal placement of warning sign that

can trade-off between the “Kangaroo effect” and effectiveness of the warning sign in

deterring speeding behavior.

100

Table 5. 5 Marginal effects due to changes in the SP attributes

Variables

Percentage change in market shares

Standard section

Warning section

Speed

Compliance

Speed

range 1

Speed

range 2

Speed

Compliance

Speed

range 1

Speed

range 2

DOP

Market share in base case

16.96%

78.64%

4.40%

66.27%

32.88%

0.85%

Change in market share upon

increment by 1 point

4.22%

-0.64%

-4.85%

4.63%

-9.01%

-12.74%

Fines

Market share in base case

16.96%

78.64%

4.40%

66.27%

32.88%

0.85%

Change in market share upon

increment by 10 percent

0.00%

1.73%

-3.16%

-13.02%

Camera-to-Housing ratio

(Base case: 20/120)

Market share in base case

16.89%

77.78%

5.34%

66.20%

32.80%

1.00%

Change in market share upon change

from base case to minor increase

(40/120)

0.83%

2.11%

-33.39%

0.20%

0.50%

-29.52%

Change in market share upon change

from base case to major increase

(60/120)

0.96%

2.40%

-38.06%

0.22%

0.55%

-32.80%

Distance of warning sign from

camera housing unit

(base case: 100m)

Market share in base case

18.01%

80.91%

1.08%

73.50%

26.41%

0.09%

Change in market share upon change

from base case to 50m

0.00%

14.18%

-34.37%

-50.09%

Change in market share upon change

from base case to 150m

0.00%

-23.02%

63.56%

143.85%

Change in market share upon change

from base case to 200m

0.00%

-22.81%

59.00%

1277.90%

101

5.6 Concluding remarks

This study applied a stated preference survey and a panel mixed logit model to evaluate

the deterrent effects of penalty and enforcement strategies – DOP penalty, monetary fines,

and speed enforcement cameras along with a warning of such enforcement – on the

propensity and severity of speeding among professional drivers. In doing so, the study

controlled for the effects of driver demographics and operational characteristics as well

as driver history and safety perceptions. As importantly, observed and unobserved

heterogeneity were incorporated in drivers’ responses to penalty and enforcement

strategies. A panel mixed logit model is estimated and applied to understand the

effectiveness of penalties and enforcement strategies on driver’s speeding behaviors.

The results indicate that an increase in DOP penalty is more effective as a deterrent against

speeding than increasing monetary fines. This could be attributed to the higher sensitivity

of professional drivers to the increase in DOPs since incurring more DOPs may lead to

disqualification of the driving licence. Monetary fines were not found to be very effective,

perhaps because the monetary fine levels were very low relative to the income levels of

the drivers. It remains to be explored if increasing the quantity of fines combined with

appropriate warning messages (such as “Check speed––fines up to $1000”) can help

increase the effectiveness of monetary fines. Significant heterogeneity was found in the

influence of the DOP variable both due to observed and unobserved factors. Specifically,

while increasing DOP deters all drivers from speeding, doing so when combined with a

warning (i.e., in the warning sections) appears to more strongly deter those who recently

received a speeding ticket than others. However, the unobserved variation in the warning

section is greater than that in the standard section, perhaps because of differences in

drivers’ threat and coping appraisals of the warning messages, as discussed in section

5.4.1.

In the context of camera-based enforcement strategy, increasing the ratio from status quo

(20:120) to 40:120 showed a considerable effect (29%) on reducing the percentage of

drivers opting for severe speeding, albeit it should be noted that the base percentage of

drivers in this category is only 1%. Increasing it further to 60:120 did not show a

102

substantial effect in the policy simulations we conducted. Further, reducing the ratio from

the status quo (20:120) to 20:240 did not show a significant effect on the drivers’ stated

speeding choices.

The placement of the warning sign – that is, the distance of the warning sign from the

camera housing location – exhibits an influence on speeding behaviors in the warning

section. Placing it close to the camera housing location decreases the likelihood of

speeding but can potentially increase the “kangaroo” effect. And placing it too far from

the camera location would substantially increase the percentage of speeding behaviors.

These findings suggest a need for the optimal location of warning signs. Alternatively,

information on the penalty level can be added to the warning signs to increase the threat

appraisal of the driver for reducing speeding behaviors in warning sections.

The demographic characteristics of drivers such as age, education, income have an

influence on how drivers respond to strategies aimed at increasing speed compliance.

Similarly, the drivers’ operational characteristics, driving history and perceptions have a

substantial bearing on the efficacy of speed compliance strategies. Therefore, targeted

driver educational and training campaigns might help increase the speed compliance rates

in the population. For example, drivers with a recent history of traffic tickets continue to

demonstrate a greater tendency for speeding even for high levels of DOP and monetary

fines. It appears that simply imposing fines or DOPs might not suffice to reduce the

aggressive driving traits of such drivers. A combination of DOPs and fines with driver

training programs aimed at addressing risk-taking and aggressive driving traits may be

needed to increase safe driving tendencies among these drivers. Further, higher penalties

may be considered for repeat offenders to enhance the deterrent effect of the penalties

(Watson et al., 2015). Similar penalty strategies have been applied for repeat offenders of

drink driving in Hong Kong (Li et al., 2014).

Speeding and other traffic offences may be attributed to drivers’ goals of travel time

saving and revenue maximization (Cestac et al., 2011; Peer, 2010; Tarko, 2009), while

safe driving performance and social responsibility may be lower in the hierarchy of

professional drivers’ goals (Hatakka et al., 2002). Therefore, inclusion of positive motives

and goals in the education/training and licensing of professional drivers may be beneficial.

103

In addition, technology-based interventions, such as GPS-based automated speed

surveillance and related automated speed enforcement mechanisms, may aid in reducing

speeding behaviors.

The results from this study help enhance the current understanding and effectiveness of

penalties and speed-enforcement strategies (i.e. penalties, warning signs, camera

housings, etc.). Yet, this study is limited to the assessment of a few demographics and

operational characteristics of professional drivers. It would be worth exploring the

possible effects of latent characteristics on speeding propensity and severity, when more

comprehensive information on the physiological and psychological metrics of the

participants is available. Moreover, results of this questionnaire survey are derived from

a scenario of a typical city road with a speed limit of 50km/h. It would be interesting to

explore the effect of other road environments, such as an expressway with a speed limit

of 70 km/h or higher, on the speeding behavior of professional drivers. Further, it would

be helpful to undertake a study that evaluates the effectiveness of combining speeding

penalties with driver education/training campaigns in reducing risk-taking and aggressive

driving. Also, the separation between the placement of a warning sign and the camera

housing unit was expressed as a distance in the current study. Perhaps a time separation

rather than a space separation would be a better approach to capture how individuals

respond to warning signs before entering monitored roadway section. Yet another line of

research would be to investigate whether fixed ASEC systems, when complemented with

a small human police force, would have a higher impact in reducing speeding than a fixed

ASEC system alone. And, if so, what may be the optimal combination of investment in

human-based and machine-based enforcement mechanisms. Perhaps most importantly,

all the results and recommendations in this study are based on self-reported speed

indications within stated experiments, which clearly can influence the reliability and

accuracy of the relationships estimated. A study based on an actual field experimental

design and field observations of speed at different sections would be more credible.

104

Chapter 6 Effects of commercial vehicle mix and multivariate

analysis of crash rates by vehicle type

6.1 Introduction

In Hong Kong, commercial vehicles (buses, taxis, light-goods vehicles, and medium- and

heavy-goods vehicles) constitute only 20% of total vehicle fleet but are involved in over

70% of road crashes. A plausible reason for such disproportionate crash share of

commercial vehicles could be their relatively higher travel amounts (per-vehicle distance)

(Pei et al., 2012; Transport Department, 2019). For example, recently-published statistics

indicate that the annual per-vehicle distances travelled (million km) by commercial

vehicles (licensed taxi (0.14), bus (0.06), light goods vehicle (0.03), medium and heavy

goods vehicle (0.03)) are all significantly higher compared to the private car (0.01)

(Transport Department, 2019). In addition, it has been determined that traffic violation

rates and crash involvement are higher for commercial vehicle drivers compared to

private car drivers (Chen et al., 2020a; Öz et al., 2010; Tang et al., 2018; Wong et al.,

2008).

In this chapter, the commercial vehicle proportion (CVP) refers to the ratio of commercial

vehicles to all vehicles in the traffic stream. Interactions between road safety factors,

including road user behaviour, weather, and road geometric factors, have been examined

in previous studies. However, this has rarely been done for the CVPs and roadway

features. While consideration of the interactions between these two specific factors may

be a contribution that seems to be only incremental, it is important to address this gap in

the literature because both roadway features and commercial vehicle proportion have

been found to be significant factors of urban road crash propensity. Therefore, in this

study, we hypothesize that these two factors have some interaction effects on road safety.

If this is affirmed, then one of the two factors (the prevailing roadway features) potentially

mediates the safety effect of the second factor (CVP). In that case, the urban road design

and operations policies related to commercial vehicle operations that fail to consider such

mediating effects, could lead to increased urban crashes involving commercial vehicles

105

(Bao et al., 2019). This is a critical issue, given the realization that in several urban areas

including the city used in the methodology demonstration (Hong Kong), commercial

vehicles account for a significant fraction of all trips.

Against the background presented in the previous section, the goal of this paper is to

measure the association between the commercial vehicle proportion and crash rate; more

importantly, to examine the mediating (moderating or magnifying) effects of roadway

attributes on this association; and ascertain how this association and the moderating

effects vary by the commercial vehicle type, considering the crash severity level and road

attribute type. It is anticipated that the results will help guide policy development by the

road authorities and transport operators regarding the management and regulation of

commercial vehicles and their drivers.

The scope of this study is such that it addresses commercial vehicles at urban areas. The

reason for this is the higher travel amounts (per-vehicle mileages) and crash propensities

of this vehicle class compared to other vehicle classes in urban areas, as evidenced in past

research. The role of commercial vehicle operations in urban road safety has come under

increasing scrutiny in recent years, and the resolution of the problem of high crash rates

for commercial vehicles at such areas, continues to be an important roadway safety issue

and public relations concern for urban road agencies. Therefore, city road authorities and

transport operators seek to regularly review crash propensities and to revise safety

countermeasures, including operations policies and physical roadway interventions, to

enhance commercial vehicle safety. The commercial vehicle classes are: (i) public buses,

including single- and double-decker buses, and light buses, (ii) taxi, (iii) light van or light-

goods vehicle (5.5 tonnes maximum gross vehicle weight, and (iv) medium- and heavy-

goods vehicles (24 and 38 tonnes, respectively). This study analyses the safety effects of

the interaction between commercial vehicle percentage and roadway attributes.

This chapter is organized as follows. Section 6.2 describes the method of analysis. Section

6.3 presents the method of data collection. The analysis results and interpretations for the

overall crash rates at different severity levels are given in Section 6.4. Section 6.5

provides the results and discussion on the multivariate analysis of crash rates by vehicle

type. Finally, findings and recommendations are summarized in Section 6.6.

106

6.2 Data collection

6.2.1 Overall crash rates by injury severity

This study used comprehensive crash and traffic data from eighty-eight (88) road

segments in the study area (the City of Hong Kong) spanning a four-year period (2014–

2017) (Figure 6.1). The road segments under investigation are widely distributed

spatially over the study area. The traffic count data were collected from the Annual Traffic

Census (ATC) database which was established by the road agency primarily for transport

planning purposes. The ATC database provided road geometry data (e.g., the number of

lanes, lane width, and road type), and the traffic data. This included the annual average

daily traffic by vehicle type, hourly variation, and weekday/weekend distribution), and

the proportions of vehicle classes including public bus, taxi, light-, medium-, and heavy-

goods vehicles. The source of the crash data is the Hong Kong Transport Department’s

Transport Information System (TIS) which includes accident data (e.g., injury severity,

date and time, and location), vehicle attribute data (e.g., vehicle type and year of

manufacture), and casualty characteristics (e.g., injury severity, casualty role, age, and

gender). The crash severity levels are: killed (fatal), severe injury, and slight injury. Due

to the paucity of fatal and severe injuries, these two levels were combined to form a single

level: killed and severe injury (KSI).

The Hong Kong Road Network Dataset provided data on the length, number of

intersections, presence of on-street parking, and speed limits of the road segments. The

data on hourly variations and percentage distribution across the commercial vehicle

classes, were available for a 16-hour period (7AM to 11PM) during weekdays; therefore,

crashes that occurred within 11PM-7AM and on weekends are not included in the data.

The traffic and crash data were aggregated into eight 2-hour periods: 7AM–9AM, 9AM–

11AM, 11AM–1PM, 1PM–3PM, 3PM–5PM, 5PM–7PM, 7PM–9PM, and 9PM–11PM.

The traffic, crash and road network characteristics data were mapped to the corresponding

road segments using a geographical information system (GIS) platform.

In the study, the overall weekday crash rate (crash count per million vehicle-kilometers

travelled) of crash severity level k at road segment i in period p of year t, is specified as:

 







 





 





 



 

(1)

107

Where: i = 1,2,…88; k = 1,2; p = 1,2,…,8; t = 2014, 2015, 2016, 2017;  





is the number of weekday crashes at severity level k at road segment i in period p of year

t;  





is the two-hour traffic flow of road segment i in period p of year t

(calculated by hourly variation of the weekday AADT on road segment i in year t);





is the length of road segment i.

The crash data are aggregated into eight 2-hour periods to mitigate the problem of

excessive zero (crash) observations. Of the 2,816 observations, 1,018 observations had

zero slight-injury crashes and the remaining 1,798 had at least one slight-injury crash;

2,322 observations had no KSI crashes and the remaining 494 had at least one KSI crash.

Figure 6.2 presents the temporal distribution of the various vehicle classes at the road

segments under study, and Table 6.1 defines and presents the descriptive statistics of the

variables considered in this study.

(a) Average annual KSI and slight-injury crash counts of the selected road segments, 2014-2017

108

(b) Average annual KSI and slight-injury crash rates of the selected road segments, 2014-2017

Figure 6. 1 Study area (Hong Kong) showing the road segments studied and safety

trends

Figure 6. 2 Variation of commercial vehicle percentage by time of day

109

Table 6. 1 Descriptive statistics of the variables

Variable

Mean

S.D.

Min.

Max.

KSI crash rate

0.17

1.12

0.00

30.72

Slight-injury crash rate

1.16

4.14

0.00

151.43

Length (km)

3.00

3.48

0.18

19.08

Logarithm of 2-hour traffic flow

8.11

1.01

4.42

9.94

Average lane width (m)

3.59

0.45

2.70

5.25

Wide roadway

(more than four traffic lanes: yes = 1, no = 0)

0.48

0.50

High intersection density

( 3 intersections per km, yes = 1, no = 0)

0.21

0.41

Speed limit (km/h)

64.79

13.07

100

Presence of on-street parking

(yes = 1, no = 0)

0.06

0.23

Proportion of taxi (%)

22.20

11.18

0.00

69.75

Proportion of public buses (%)

8.08

7.34

0.00

60.15

Proportion of light-goods vehicles (%)

14.69

8.09

0.20

40.70

Proportion of medium and heavy (M&H) goods vehicles (%)

5.80

8.07

0.00

57.95

Figure 6.3 attempts to better demonstrate possible relationship between commercial

vehicle mix and crash rates by injury severity. The z-axis of Figure 6.3 is the normalized

crash rates using the equation (2) below (see Ulak et al., 2018), in order to get the same

scale for KSI and slight injury crash rates. However, it should be noted that the crash rates

used in statistical modeling are  





from equation (1).

 







 





 







(2)

110

Figure 6. 3 Normalized overall crash rates by injury severity under different commercial

vehicle mix

6.2.2 Crash rates by vehicle type

In this study, comprehensive crash and traffic data of 100 road segments in Hong Kong

in the 3-year period from 2015 to 2017 are used. For instances, crash data are obtained

from the Transport Information System (TIS) of Hong Kong Transport Department

during the period from 2015 to 2017. TIS database consists of three profiles: accident

profile (e.g. injury severity, date and time, location, speed limit, road type, etc.); vehicle

attributes (e.g. vehicle type, year of manufacture, driver age, etc.), and casualty

characteristics (e.g. injury severity, casualty role, age, gender, etc.). Crashes are stratified

into three categories in accordance to injury severity: killed, severely injured and slightly

injured. Since the numbers of killed crash and severely injured crash are too small for

efficient analysis, they are combined into ‘killed and severely injured’ crash (known as

‘KSI’).

On the other hand, traffic count (e.g. annual average daily traffic by vehicle type and

hourly variation) and road geometry data (e.g. number of lanes, lane width, road type,

etc.) are obtained from the Annual Traffic Census (ATC) database. Vehicle classes

111

considered are private car, bus, light bus, taxi, light goods vehicle, and medium and heavy

goods vehicles. Furthermore, information on road segment length and number of

intersections are obtained from the Hong Kong Road Network Dataset. Since the

information on hourly variation and percentages of different vehicle classes is available

only for 16-hour period from 7 am to 11 pm. Crashes occurred in the period from 11 pm

to 7 am are not considered in the subsequent analysis. Also, the sample would be too

small if the crashes are disaggregated at hourly level. Then, the traffic and crash data are

aggregated to four-hour period as: 7 am – 11 am (morning), 11 am – 3 pm (noon), 3 pm

– 7 pm (afternoon) and 7 pm – 11 pm (evening). The abovementioned traffic, crash and

road network characteristics data are mapped to the corresponding road segments using

the geographical information system (GIS) technique.

Furthermore, crashes are classified into three categories by vehicle type -private car, light

commercial vehicle, and heavy commercial vehicle. In Hong Kong, road traffic

regulations (see chapter 374A, construction and maintenance of vehicles) has specified

the maximum gross vehicle weights of private car (3.0 tonnes), taxi (3.0 tonnes), light bus

(5.5 tonnes), light goods vehicle (5.5 tonnes), bus (24 tonnes), medium goods vehicle (24

tonnes), and heavy goods vehicle (38 tonnes). In this study, light commercial vehicles

include taxis, light goods vehicles, and light buses, while heavy commercial vehicles

include medium and heavy goods vehicles, and buses. A total of 2323, 2655, and 1024

crashes were identified to involve private cars, light commercial vehicle crashes, and

heavy commercial vehicle crashes, respectively. Among these 6002 crashes, 1233 only

involved private cars, 2114 only involved light commercial vehicle, and 458 only

involved heavy commercial vehicle. The remaining crashes could be multiple-vehicle

crashes involving different vehicle types. It is likely that private car often collides with

other vehicle types, so does the heavy commercial vehicle. It should be noted that the

multiple-vehicle crashes involving a motorcycle (that is, a private car/light commercial

vehicle/heavy commercial vehicle collides with a motorcycle) were not excluded in this

study. Figure 6.4 presents the histograms of crashes involving private cars, light

commercial vehicles, and heavy commercial vehicles across observations.

112

113

Figure 6. 4 Histograms of private car, light commercial vehicle, and heavy commercial

vehicle crashes

In this study, crash rate (per million vehicle-kilometre travelled) of vehicle type v at road

segment i in period p of year t is specified as,

 







 





 





 



 

(3)

i = 1,2,…100; v = 1,2,3; t = 1, 2, 3; p = 1,2,3,4

Table 6.2 summarizes the statistics of considered variables. Prior to parameter estimation,

multicollinearity test is conducted to assess the correlation between independent variables.

Results indicate that VIF (variance inflation factor) of independent variables are all less

than 3, hence, there is no multicollinearity between independent variables.

114

Table 6. 2 Descriptive statistics of the variables

Variable

Mean

Min.

Max.

Private car crash count

1.94

2.48

Light vehicle crash count

1.89

2.21

Heavy vehicle crash count

0.85

1.28

Private car crash rate

1.69

11.58

274.85

Light vehicle crash rate

1.62

7.47

195.28

Heavy vehicle crash rate

0.75

7.75

262.86

Length (km)

2.71

3.35

0.08

19.08

Average lane width (m)

3.75

0.89

2.70

10.00

Number of lanes

4.58

2.35

Intersection density (number per km)

2.08

3.24

0.00

14.36

Speed limit (km/h)

100

Logarithm of four-hour traffic flow

1.74

1.11

-3.30

3.66

Presence of on-street parking (yes = 1,

no = 0)

0.12

0.33

6.3 Model specification and goodness-of-fit

In this section, the formulations of univariate random parameter Tobit model for overall

crash rates by injury severity (KSI and slight injury), and multivariate Tobit model for

crash rates by vehicle types (private car, light commercial vehicle and heavy commercial

vehicle) are specified under the Bayesian framework.

6.3.1 Random parameter Tobit model

In conventional safety literature, crash frequencies are typically modeled using count-

data approaches. In this paper, however, we examine the crash experience using an

alternative outcome – the crash rate, that is, the number of crashes per million vehicle-

kilometer travelled (Anastasopoulos et al., 2008). The advantages of crash-rate analysis

have been discussed by several recent studies (Zeng et al., 2017b, 2019; Guo et al., 2019).

For example, crash rate represents a standardized measure of the relative safety

performance of road entities as it neutralizes the crash exposure. Crash rates have always

been a common feature of government agency safety reports, and cash rate analysis

currently has several common applications including the identification of hotspots. The

crash rate variable is continuous in nature and left-censored at zero, as some road

segments in the study dataset have zero crashes; therefore, as recommended by previous

115

studies (Anastasopoulos et al., 2008; Guo et al., 2019; Zeng et al., 2017a, 2017b, 2018),

this study uses a Tobit regression approach.

The dependent variable is the crash rate, and the analysis is carried out for each level of

crash severity. The crash rate is a non-negative and continuous variable that is censored

at zero (meaning that there could exist road segments where no crash is observed during

a specific period). As such, in the analysis, we used a Tobit model (an econometric

technique originally proposed by Tobin (1958)) to resolve the problem of left- or right-

censoring of the dependent variable. Anastasopoulos et al. (2008) first applied the Tobit

approach in road safety research. It is often recommended to develop crash prediction

models separately for each level of crash severity because underreporting is often more

prevalent for less severe crashes (Anastasopoulos et al., 2012b; Pei et al., 2016).

Additionally, separate development of models by crash severity level helps eliminate

estimation bias that may arise from any shared but unobserved heterogeneity across

observations. To address this issue, previous researchers including Guo et al. (2019),

Chen et al. (2017a), Zeng et al. (2018, 2017a, 2017b), Anastasopoulos et al. (2012a,

2012b), and Anastasopoulos (2016) have used advanced modelling approaches including

random parameter Tobit model, multivariate Tobit model and multivariate random-

parameter Tobit model.

In this study, a random-parameter Tobit model

was developed to account for unobserved

shared effect among crashes and any heterogeneity in the effects of certain crash factors

across the observations. This was done for different levels of crash severity. The analysis

helped examine the associations between the crash rate and the commercial vehicle mix

(i.e., the respective proportions of the five commercial vehicle classes), and other

prospective explanatory factors including year, time of the day, road geometry, traffic

control, traffic flow were examined. The model proposed for the analysis has the form

(Equation (2) and Equation (3)):

To account for the possible unobserved shared effect among crashes at different crash severity levels,

multivariate Tobit model was also considered. However, results of goodness-of-fit assessment suggest that

multivariate Tobit model did not outperform the univariate one. Besides, our preliminary results suggest

that there is no evidence for significant correlation between KSI and slight crash rates.

116













 











 







































(2)















(3)

Where: i = 1, 2, …, N; p = 1, 2, …, P; k = 1, 2, …, K; 





denotes the latent variable

linking the expected crash rate of severity level k at segment i during period p; 





denotes the observed crash rate; N, P and K refer to the total number of road segments,

time periods and crash severity levels, respectively; 





denotes the value of j

explanatory variable at segment i during period p; 



refers to a normally and

independently distributed random error term with zero mean and variance 



; 





is a

constant; 





is the normal distributed random parameter

with a mean vector of





(that is, the coefficient of the j

explanatory variable

corresponding to crash severity

level k). 



refers to a normally and independently distributed random error term with

zero mean and variance 





. It should be noted that 





is set to be random only when

its variance is statistically significant at 5% level, otherwise the parameter is set to be

fixed.

With regard to the non-zero crash case, the marginal effect (i.e., effect of per unit increase

in an independent variable on the expected crash rate) can be determined using

methodologies established in the literature (Anastasopoulos et al., 2008, 2016; Roncek,

1992):

[ ] ( ) ( )

( ) ( )

E Y f z f z

x F z F z









=  −  −







For the zero-crash case, the marginal effect can be specified as,

Alternative distributional assumptions such as log-normal for the random parameters were also explored.

Nevertheless, the model with normal distribution provided the best fit. For the considerations about the

random-parameter density functions, detailed discussions can refer to Anastasopoulos and Mannering

(2009). Moreover, the model with normal distribution can provide substantive interpretations that were

very different from models with other distributions.

117

( ) ( )

F z f z







=



Where: 













denotes the change in the expected crash rate for non-zero crash

case; 



denotes the change in the cumulative probability of having a crash for

the cases with no crashes; β

denotes the coefficient of the j

explanatory variable; F(z) is

the area bounded by the normal curve (i.e., the normal distribution function) for the

propensity of the crash occurrence; z denotes the normalized score; f(z) denotes the

standard normal density function;  is the standard deviation of the error term 



6.3.2 Multivariate Tobit model

To capture common unobserved factors across different crash types, multivariate analysis

is commonly adopted in previous research for a better modelling performance. For

examples, Lee et al., (2015) presented a significant unobserved shared effect between the

numbers of motor vehicle, bicycle, and pedestrian crashes, using multivariate Poisson

log-normal model. Zeng et al. (2017b) and Anastasopoulos et al. (2012b) accommodated

the significant correlation between crash rates at different severity levels, using

multivariate Tobit model. Guo et al. (2019) investigated the rear-end, sideswipe, and

angle crashes at freeway diverge areas and revealed the presence of unobserved risk

factors that jointly affect different crash rates by collision type. As the dependent

variables of the proposed analysis are crash rates by different vehicle type (private car,

light commercial vehicle, and heavy commercial vehicle), a multivariate Tobit model is

specified as follows:











 









 









































(6)

i = 1,2,…,N; p = 1,2,…,P; v = 1,2,3

where 





denotes the latent variable linking the expected crash rate of vehicle type v

118

(1,2,3 represents for private car, light commercial vehicle, and heavy commercial vehicle

crashes per million vehicle-kilometer traveled in road segment i, respectively) at segment

i during period p, while 





denotes the observed crash rate. 



is the constant, 



is the estimated parameter for j

explanatory variable

corresponding to vehicle type v.







refers to a multivariate normally and independently distributed random error term

with zero mean, variance , and correlation , which can be expressed as 





~ N (0, ).

The covariance matrix is,

































































































































(7)

where 





, 



, 





represents the variance of error term 





, 





, 





, respectively.

6.3.3 Goodness-of-fit

To assess the goodness-of-fit of the proposed model using Bayesian approach, deviance

information criteria (DIC) and Bayesian R

are commonly used (Huang et al., 2016; Zeng

and Huang, 2014; Zeng et al., 2017a, 2017b, 2018; Wen et al., 2018). DIC is essentially

a generalization of Akaike's Information Criterion (AIC) and is given by (Spiegelhalter

et al., 2002),

DIC = 



+ pD

(8)





denotes the posterior mean of deviance, pD is the complexity term for the effective

number of parameters in the model. Model with the lowest DIC value, among the

candidate models, is considered to have the best prediction performance. Yet, difference

in DICs between models should also be considered. Specifically, difference should be

preferably greater than ten (Spiegelhalter et al., 2005).

For the multivariate analysis of crash rates by vehicle type, another goodness-of-fit

measure – Bayesian R

is also considered (Ahmed et al., 2011). Bayesian R

refers to the

ratio of explained sum of squares to total sum of squares. Bayesian 





for model of crash

119

severity k is specified as:









 







 

















 







 

















, k = 1, 2

(9)











 

 













(10)























 

















 

























 













(11)

where 





denotes the estimated crash rate of severity level k at segment i during period

p, and 





represents the mean observed crash rate of severity level k.

In addition, for the multivariate Tobit model, 





and 





represent the Bayesian R

values of crash rates by vehicle type v and all observations. They are specified as:







 

 







 

















 







 

















, v = 1, 2, 3

(12)









  







 





















  







 



















(13)





















 













 





















 











(14)











 

 













(15)









  

  

















(16)

120

where 





denotes the estimated crash rate of vehicle type v at segment i during period

p, 





and 



represent the mean observed crash rate of vehicle type v and all

observations, respectively.

In this study, WinBUGS software was used to specify the formulations of the random-

parameter Tobit models under the Bayesian framework. Markov chain Monte Carlo

(MCMC) simulations were used to sample the posterior distribution of the model

parameters. Prior distributions of 





were specified as being diffusely and normally

distributed N(0, 10

) (Guo et al., 2019; Lee et al., 2015; Zeng et al., 2017b, 2019).

For each model, a chain of 200,000 iterations of Markov chain Monte Carlo (MCMC)

simulation were established. In particular, the first 5,000 iterations served as burn-ins and

therefore were discarded. MCMC convergence was assessed by visually inspecting the

time-series plots of the estimated parameters, and the ratios of MC error to the

corresponding standard deviation of the estimates –specifically, the ratios should be less

than 0.05 (Ahmed et al., 2011; Guo et al., 2019; Wen et al., 2018; Zeng et al., 2017a,

2017b).

6.4 Results and discussion on the effects of commercial vehicle mix

Prior to parameter estimation, a multicollinearity test was conducted to assess the

correlations between the independent variables. The results indicated that the VIFs

(variance inflation factor) of the pairs of independent variables are all less than 5, hence,

there is no multicollinearity between the independent variables. The random parameter

Tobit models were used to identify the crash factors associated with crash rates by severity

level. Two broad categories of analysis were carried out: 1) basic models to reaffirm the

main effects of commercial vehicle proportions (CVP) on crashes, and 2) refined models

to closely examine the mediating effect of road environment crash factors on the CVP-

crash rate relationship. In the basic models, variables that were found not statistically

significant at the 5% level were eliminated using a backward stepwise regression

technique (Abdel-Aty et al., 2004; Bose et al., 2013; Huo et al., 2020). However, variables

121

including speed limit, public buses%, taxi%, and segment with wide roadways were

retained in the model. This is to allow for the possible confounding effects on the crash

rates. Then, refined models were developed to consider any interactions between the

significant variables representing the CVP and the other crash factors related to the

roadway environment, particularly, road geometry and traffic control facilities. In the

refined models, only the variables significant at the 5% level were included. Table 6.3

presents the goodness-of-fit results for the models developed. The refined models were

generally superior to the basic models in terms of DIC value, and the interactions between

the variables were manifest to a greater degree.

Table 6. 3 Results of the goodness-of-fit tests

Slight-injury crash rate

KSI crash rate

Basic model

Refined model

Basic model

Refined model



–4,185

–3,925

–9,254

–9,417

6,353

6,071

3,865

3,784

DIC

2,167

2,145

–5,389

–5,633

DIC

basic

-DIC

refined

244

Note: Model with lower DIC (difference in DICs exceeding 10) has superior prediction

performance.

Tables 6.4 and 6.5 present the estimation results of random-parameter Tobit models for

slight-injury crash rate and KSI crash rate, respectively. The basic model (Table 6.4)

confirmed that the factors representing traffic flow, road geometry, traffic control,

commercial vehicle proportion, and time of day all influenced the slight-injury crash rate.

In particular, the [log of 2-hour] traffic flow (coefficient = –1.009) and the proportion of

medium and heavy goods vehicle (–0.026) were associated with lower slight-injury crash

rate. On the contrary, a wide roadway or 5 or more traffic lanes (0.926), high intersection

density (1.120), presence of on-street parking (1.818), proportions of taxi (0.029) and

light goods vehicle (0.061), and the time of day from 5PM to 9PM (0.630), were found

positively associated with slight-injury crash rate, at 5% significance level. Also, the

heterogenous effect of the average lane width (mean of 0.479 and standard deviation of

1.151) on slight-injury crash rate was found to be statistically significant. With regard to

the basic model in Table 6.4, also, log of 2-hour traffic flow (–0.092) and the proportion

of medium- and heavy-goods vehicle (–0.006) were associated with lower KSI crash rate.

In contrast, presence of on-street parking (0.392), proportions of public buses (0.009) and

light-goods vehicle (0.005) were found to be positively associated with the KSI crash rate.

122

Also, it was shown that the effects of average lane width (mean of 0.124 and standard

deviation of 0.035) and high intersection density (mean of 0.168 and standard deviation

of 2.920) on KSI crash rate varied across the observations.

The refined models for slight-injury and KSI crash rates (Table 6.5) present the interaction

effects of the commercial vehicle proportion which were found to be statistically

significant in the basic model and the other potential crash factors. The interaction terms

that were significant at the 5% level were included in the final set of the refined models.

A comparison of the estimation results of the basic and refined models showed that most

of the contributory factors showed consistent safety effects across these two groups of

models. The exception is that the insignificant variable (wide roadways) becomes

statistically significant when the interaction terms are considered for the KSI crash rate

(see Table 6.5). Thus, only the interaction effects in the refined models are discussed here.

As shown in Table 6.5, the interactions between (i) the proportion of taxi and high

intersection density, and (ii) the proportion of light goods vehicle and presence of on-

street parking, were significantly associated with slight-injury crash rate at the 5% level.

On the other hand, interactions between (iii) the proportion of medium- and heavy-goods

vehicle and high intersection density, and (iv) the proportion of medium and heavy goods

vehicle and a wide roadway (5 or more lanes) were significantly associated with KSI

crash rate at the 5% level (see Table 6.5).

Table 6.6 presents the marginal effects of the refined models. For example, for the cases

above the limit (with crashes), a 10% increase in the proportion of taxi is expected to

contribute to an increase in the slight-injury crash rate by 0.07, while it contributes to an

increase in the slight-injury crash rate of road segments with high intersection density by

0.14. Also, it is expected that a 10% increase in the proportion of light-goods vehicle

generally increases the slight-injury crash rate by 0.21, while it increases the slight-injury

crash rate of road segments with on-street parking areas by 0.53. Moreover, a 10%

increase in the proportion of medium- and heavy-goods vehicle is expected to increase

the KSI crash rate of road segments with high intersection density by 0.16, while it

decreases the KSI crash rate of road segments with more traffic lanes by 0.02.

On the other hand, for the cases at the limit (no crashes), the probability of having a slight-

123

injury crash is expected to be increased by 2.9%, due to a 10% increase in the proportion

of taxi on the road segments with high intersection density. The probability of having a

slight-injury crash is expected to increase by 10.5% for the road segments with on-street

parking areas, due to 10% increase in the proportion of light-goods vehicle. For the road

segments with high intersection density and zero-crash observation, the probability of

having a KSI crash is expected to increase by 6.8%, due to a 10% increase in the

proportion of medium- and heavy-goods vehicle.

124

Table 6. 4 Results of random-parameter Tobit model

Variable

Basic model

Refined model

Mean

S.D.

BCI

Mean

S.D.

BCI

2.5%

97.5%

2.5%

97.5%

Constant

5.844

1.050

4.018

7.948

5.736

0.873

4.145

7.646

Ln (2-hour traffic flow)

–1.009

0.088

–1.183

–0.841

–1.030

0.097

–1.214

–0.857

Average lane width

0.479

0.177

0.105

0.728

0.531

0.121

0.238

0.725

S.D. of Average lane width

1.151

0.035

1.089

1.215

1.149

0.031

1.089

1.212

Wide roadway (nr. of traffic lanes >4)

0.926

0.199

0.559

1.328

0.925

0.158

0.624

1.248

Speed limit

–0.008

0.007

–0.022

0.006

High intersection density (>3 per km)

1.120

0.219

0.717

1.545

Presence of on-street parking

1.818

0.352

1.169

2.501

Variables for commercial vehicle proportions

Public buses %

0.000

0.012

–0.022

0.024

Taxi %

0.029

0.008

0.013

0.046

0.019

0.010

0.001

0.038

Light-goods vehicle %

0.061

0.013

0.037

0.089

0.055

0.013

0.028

0.079

M&H-goods vehicle %

–0.026

0.013

–0.050

–0.001

–0.045

0.013

–0.073

–0.023

Time effect variables

5PM-9PM

0.630

0.177

0.302

0.987

0.554

0.211

0.134

0.905

Interaction effect variables

Taxi % × High intersection density

0.038

0.006

0.025

0.048

Light goods vehicle % × Presence of on-street parking

0.138

0.023

0.091

0.190

BCI refers to Bayesian credible interval

Boldface indicates statistical significance at the 5% level

Insignificant variables are retained in the basic model to account for the possible confounding effects on crash rates

In the refined model, only significant variables are retained

125

Table 6. 5 Random parameter Tobit model for KSI crash rate

Variable

Basic model

Refined model

Mean

S.D.

BCI

Mean

S.D.

BCI

2.5%

97.5%

2.5%

97.5%

Constant

0.435

0.201

0.055

0.844

0.425

0.258

–0.068

0.952

Ln (2-hour traffic flow)

–0.092

0.019

–0.130

–0.055

–0.108

0.024

–0.156

–0.062

Average lane width

0.124

0.036

0.040

0.189

0.121

0.057

0.022

0.232

S.D. of Average lane width

0.035

0.001

0.033

0.038

0.071

0.002

0.067

0.075

Wide roadway (nr. of traffic lanes >4)

0.041

0.037

–0.032

0.113

0.121

0.052

0.023

0.226

Presence of on-street parking

0.392

0.065

0.267

0.522

0.487

0.087

0.320

0.662

High intersection density (>3 per km)

0.168

0.079

0.012

0.323

S.D. of High intersection density

2.920

0.214

2.525

3.365

Speed limit

–0.001

0.001

–0.004

0.001

Variables for commercial vehicle proportions

Taxi %

–0.003

0.002

–0.006

4.13E-04

Public buses %

0.009

0.002

0.004

0.014

0.006

0.003

8.62E-05

0.011

Light-goods vehicle %

0.005

0.002

4.03E-04

0.009

0.005

0.002

1.26E-04

0.010

M&H-goods vehicle %

–0.006

0.002

–0.011

–0.002

Interaction effect variables

M&H-goods vehicle % × High intersection density

0.084

0.020

0.045

0.123

M&H-goods vehicle % × Wide roadway

–0.010

0.003

–0.016

–0.005

Boldface indicates statistical significance at the 5% level

Insignificant variables are retained in the basic model to account for the possible confounding effects on crash rates

In the refined model, only significant variables are retained.

126

Table 6. 6 Marginal effects results for the refined models

Slight-injury crash rate

KSI crash rate















Zero-crash

sensitivity















Zero-crash

sensitivity

Ln (2-hour Traffic flow)

-0.392

-7.83%

-0.020

-0.87%

Average lane width

0.202

4.04%

0.023

0.98%

More traffic lanes (>4)

0.352

7.03%

0.023

0.98%

Presence of on-street parking

0.091

3.94%

Public buses %

0.001

0.05%

Taxi %

0.007

0.14%

Light goods vehicle %

0.021

0.42%

0.001

0.04%

M&H goods vehicle %

-0.017

-0.34%

5pm-9pm

0.211

4.21%

Taxi % * High intersection density

0.014

0.29%

Light goods vehicle % * Presence of on-street parking

0.053

1.05%

M&H goods vehicle % * High intersection density

0.016

0.68%

M&H goods vehicle % * More traffic lanes

-0.002

-0.08%

Note: Zero crash sensitivity is determined as multiplying 



by 100%

6.4.1 Effects of geometric factors

The model results suggest that the average lane width is positively associated with slight-

injury crash rate and KSI crash rate. Specifically, when the average lane width increases,

the rate of slight-injury crashes increases at 66.1% of the road segments and the rate of

KSI crashes increases at 100% of the road segments. This finding is generally consistent

with that of recent studies (Zeng et al., 2017b). This could be because drivers tend to be

less cautious and speed up when the traffic lane is wider. Therefore, potential crash and

injury risks both increase (Gross and Jovanis, 2007). Furthermore, heterogeneity for the

effect of lane width can be attributed to the variation in pavement surface condition and

driver response. In other words, the relationship between lane width and crash propensity

is not necessarily monotonic. As our results also suggest that 33.9% of the road segments

would experience reduction in slight-injury crash rate with an increase in average lane

width. Indeed, it has been reported that crash propensity increases when the lane width

first increases (known as ‘driver extreme cautious zone’), then decreases when the lane

width further increases (‘driver normal zone’), and eventually bounce back again

(‘reckless driving zone’) (Labi et al., 2017; Chen S., 2019b; Chen S. et al., 2019c; Chen

et al., 2020b). As such, it is worth exploring the non-linear relationship between lane

width and crash propensity when comprehensive information on driving behavior is

available in a future extended study. The analysis results also suggest that a wider

roadway (with more traffic lanes) is generally positively associated with slight-injury

127

crash rate, and this results could be attributed to increased lane-changing opportunities

when the number of lanes increases (Pei et al., 2012, 2016; Zeng et al., 2017b; Chen S. et

al., 2019c).

6.4.2 Effects of Traffic flow and traffic control

With regard to the effect of traffic flow, the two-hour overall traffic flow (logarithmic

form) is negatively associated with slight-injury and KSI crash rates. This result, which

suggests that the increased traffic volume would reduce the average travel speed of the

road segment and thereby decrease the likelihood of crash occurrence and severity, is

consistent with the findings of previous studies (Anastasopoulos et al., 2012a, 2012b;

Huang et al., 2016; Zeng et al., 2017a, 2017b, 2018). With regard to the effect of traffic

control, the results suggest that the presence of on-street parking increases slight-injury

and KSI crash rates. This aligns with the previous findings that higher crash propensity is

associated with more frequent roadside activities near the on-street parking areas (Pei et

al., 2016).

In addition, high intersection density was found to be positively associated with slight-

injury and KSI crash rates, which could be attributed to the effect of prevalent traffic

conflicts typically experienced at intersections (Wong et al., 2007; Sze and Wong, 2007).

However, our results also suggest that for 47.7% of the road segments, high intersection

density is associated with lower KSI crash rates. Such heterogeneous effect of intersection

density on KSI crash rate can be explained by the risk compensation theory (Mannering

and Bhat, 2014; Chen et al., 2017) where drivers adopt more cautious driving behavior to

compensate for the increased crash propensity arising from a complex driving

environment such as frequent intersections In particular, since pedestrian crashes at

intersections are more likely to be fatal or have serious injury (Zhai et al., 2019; Sze et

al., 2019), drivers generally may pay more attention to the pedestrian’s location and

behavior when driving through an intersection. As such, the lower KSI crash rate found

in some road segments with high intersection density could be attributed to the risk

compensation by drivers at these locations (Zeng et al., 2017b).

128

6.4.3 Temporal effect

The study did not find any temporal variations in the slight-injury and KSI crash rates

over the years of study. With respect to the time of day, it was determined that in the

period 5PM –9PM, the slight-injury crash rate is significantly higher than that of other

time periods at 5% level of significance. This is not surprising since such period covers

the evening peak hours that the city residents are off work and are engaging various

activities including homebound trip, shopping, and gathering with friends. In particular,

drivers tend to drive less cautiously during their off-work hours; therefore, at this period,

violation behaviours are relatively more prevalent, thus are characterized by higher crash

risk (Chin and Huang, 2009). Moreover, driving under the influence of fatigue

(particularly among commercial vehicle drivers) is more likely to occur during this period

(Boufous and Williamson, 2006). In contrast, no significant evidence was found for the

effect of time on KSI crash rate, suggesting that the temporal variation in crash rates are

different across the different levels of crash severity.

6.4.4 Main effects of commercial vehicle proportions (CVPs)

There exist anecdotal reports that the drivers of Hong Kong taxis and light-goods vehicles

tend to be risk-prone and aggressive because they are self-employed and their income

levels depend on the number and distance of their trips (Chen et al., 2020a; Meng et al.,

2017). It has also been found that taxi drivers in general, in several countries including

China, are more likely to involved in texting while driving, speeding, dangerous

overtaking and red light running (Wang et al., 2019a, 2019b; Nguyen-Phuoc et al., 2020).

The situation is exacerbated further by the taxi driver demographics: the aging taxi driver

population contributes to elevated crash propensity, as their driving performance is more

likely to be impaired plausibly due to deteriorating health, fatigue and slow response time

(Chen et al., 2019a, 2019b; Meng et al., 2017). Hence, it is not surprising that increases

in the proportions of taxi and light goods vehicle both are associated with higher rates of

slight-injury crashes.

Moreover, an increase in the proportion of light-goods vehicle is associated with an

increase in the KSI crash rate. Indeed, it could also be attributed to the difference in sense

of social responsibility across various types of professional drivers (Paleti et al., 2010).

129

For example, light-goods vehicle drivers who transport goods, presented a higher

tendency to commit traffic offenses and a higher injury risk, i.e., fatal and severe injury

(see Zhang et al., 2013, 2014), whereas a higher proportion of medium- and heavy-goods

vehicles is associated with lower rates of slight-injury and KSI crash rates. This result

could be due to the stricter regulation of the driving speed of medium- and heavy-goods

vehicles (Transport Department, 2020a). Besides the cognizance of the regulations,

heavy-goods vehicle drivers themselves tend to drive at a lower speed to compensate for

the elevated injury risk resulted from their higher vehicle weights (Saifizul et al., 2011).

On the other hand, it was expected that bus is a relatively safe transportation mode (Feng

et al., 2016). Bus drivers tend to be more risk averse because they typically possess a

stronger sense of social responsibility and lower preference to commit traffic offense

(possibly due to greater enforcement of regulations for heavy-vehicles operators) (Paleti

et al., 2010; Chen et al., 2020a; Öz et al., 2010, 2013). Surprisingly, our results showed

that the higher proportion of public buses is generally associated with a higher KSI crash

rate. This could be attributed to the exposure of commuters, the passenger capacity, the

size and weight of public buses, as well as the determination of crash severity levels

(Chimba et al., 2010; Feng et al., 2016; Tsui et al., 2009). In Hong Kong, crash severity

is determined based on the observations by police at scene and the follow-up hospital

records for up to 30 days. A severe injury crash refers to a traffic accident in which one

or more persons injured and detained in hospital for more than twelve hours. In this

context, larger capacity of passengers on the bus and higher exposure of commuters on

the road segments would contribute to the increase in KSI crash rate, as the severity level

is mainly determined by the people involved in the crash. Specifically, the passenger

capacity of public buses is much higher than that of other passenger vehicles (e.g., the

maximum capacity of a double-decker bus can reach 150 passengers, while that of a taxi

is 5 in Hong Kong). In addition, public buses in Hong Kong (including franchised bus

and public light bus) constitute 73% of overall road-based public transport patronage

(Transport Department, 2014). They are operated on fixed routes and schedules by

sizeable operators, which are regulated by the Hong Kong Transport Department. Road

segments with higher proportion of public buses tend to be located in Central Business

District, where the exposure of commuters tend to be very high on weekdays.

130

6.4.5 Interaction effects of CVPs and roadway attributes

Tables 6.4 and 6.5, which present the interaction effects, suggest that traffic control and

road geometry influence the relationship between commercial vehicle mix and slight-

injury crash rate. In particular, the increasing effect of taxi proportion on slight-injury

crash rate is magnified at road segments with high intersection density. This could be

attributed to the prevalence of traffic violations and reckless driving among taxi drivers

at intersections as evidenced in the literature. For example, Wu et al. (2016) revealed that

non-professional drivers generally tend to be more careful when driving through

intersections while taxi drivers are prone to committing red-light running and other

violations. Also, Xu et al. (2014) found that taxis are more likely to be involved in traffic

conflicts at intersections compared with other vehicle types. As such, greater emphasis

could be placed on enforcement strategies to combat the traffic violation behaviours by

taxi drivers at intersections. For example, Hong Kong presently has very few (195)

intersections with digital red-light cameras in operation (out of a total of 1,916 signalized

intersections) (Transport Department, 2017b, 2020b) and these could be significantly

increased. Based on our current finding, it is suggested that for expanding the red-light

camera network, priority could be given to road segments with relatively high proportions

of taxis. Besides, it is recommended that taxi drivers should be carefully regulated in

accordance with the licensing requirements. For example, the licensing office may invite

the taxi drivers (particularly those with a record of red-light running) to attend educational

program aimed at addressing risk-taking behaviour at intersections.

Similarly, the increasing effect of light-goods vehicle percentage on slight-injury crash

rate is magnified at road segments with on-street parking. In Hong Kong, on-street

parking is typically provided at the urban roads to facilitate direct access to the buildings.

Hence, it is likely that the road segments with on-street parking would have more frequent

roadside pick-ups, drop-offs, and loading/unloading activities involving light goods

vehicles (Sze and Wong, 2007). To address the higher crash propensity at on-street

parking areas, police patrols could be enforced at these areas, particularly at those urban

road segments that tend to have a higher proportion of light-goods vehicles. With regard

to medium- and heavy-goods vehicles on the other hand, the association between the CVP

percentage and the KSI crash rate seems to be moderated by the roadway width (number

of traffic lanes). Specifically, the decreasing effect of medium- and heavy-goods vehicle

131

percentage on KSI crash rate would be magnified by the increase in the roadway width.

A previous study revealed a dichotomous effect of heavy truck percentage on lane-

changing frequency across different traffic phases (e.g., free flow, synchronized flow and

congestion); however, the frequency of lane-changing events was found to decrease

remarkably with effective lane control measures for heavy trucks (Li et al., 2016). Hong

Kong road traffic regulations specify that medium- and heavy-goods vehicles are not

allowed to use the rightmost lane of expressway with three or more lanes in each

direction, and lane control measures implemented at such roadways are effective in

separating traffic flows by vehicle class. Thereby, such, possible conflicts between heavy

truck and other light vehicles, as well as the likelihood of fatal or serious-injury crashes,

are generally reduced (Mooren et al., 2014).

Finally, the association between medium- and heavy-goods vehicle percentage and KSI

crash rate was found to be moderated by intersection density (number of intersections

within a given unit length of road). In particular, an increase in medium- and heavy-goods

vehicle percentage would contribute to the increase in KSI crash rate of the road segments

with high intersection density. A possible explanation is the design of intersections. Over

the past decades, the dimension and weight of heavy vehicles have increased

substantially. Therefore, Dong et al. (2014) aptly questioned whether intersections

designed using earlier standards is capable of serving vehicles with various dimensions

and weights. The authors developed count models for intersection crashes and found that

an increase in the percentage of heavy trucks in the traffic stream contributes to the

increase in truck-involved crashes. Another possible explanation is the dilemma zone

driver behaviour. It was revealed that heavy truck drivers are less likely to decelerate in

response to a yellow stage of the traffic signal, thus contributing to a higher rate of red-

light running (Gates et al., 2007, 2010). Therefore, it is not surprising that in this study,

the KSI crash rate was found to be sensitive particularly to the interaction between high

intersection density and medium- and heavy-goods vehicle percentage. This result

suggests that the existence of a need for government road agencies to review the service

capability of the intersections at roads that serve a significant fraction of medium- and

heavy-goods vehicles; that way, it may be possible to reduce crashes caused by or related

to such vehicle classes at road intersections. On the other hand, to eliminate the risk-prone

behaviours of heavy-truck drivers in dilemma zones, trucking employers and freight

132

carriers could provide tailored training programs for their drivers to ensure enhanced

driving decisions and responsible driving behaviour.

6.5 Results and discussion on crash rates by vehicle type

Two models were considered to analyse the crash rates of three vehicle types: (i)

univariate Tobit model, and (ii) multivariate Tobit model. Results of DIC shown in Table

6.7 suggest that multivariate Tobit model outperforms the univariate one (difference in

DIC more than 10). Table 6. 8 presents the error parameters and goodness-of-fit of the

multivariate Tobit model. Significant correlation between crash rates across vehicle types

are found. The positive correlation between private car and light commercial vehicle, as

well as the negative correlation between private car and heavy commercial vehicle could

be attributed to the unobserved common risk factors affecting the crash rates across

vehicle types simultaneously. Moreover, it is likely that the significant correlations come

from the multiple-vehicle crashes. For example, the crash count used for calculating the

crash rate of private car includes private car only, private car-light commercial vehicle,

private car-heavy commercial vehicle, and private car-motorcycle crashes. Therefore,

when the number of private car-light commercial vehicle crashes increases, both the crash

rates of private car and light commercial vehicle increase. In either explanation,

multivariate analysis should be employed to accommodate the correlations between crash

rates.

Table 6. 7 Comparison of model performance

Multivariate Tobit

Univariate Tobit

Private car

Light commercial

vehicle

Heavy commercial

vehicle



25351

9168

8084

8286

DIC

25393

9181

8097

8299

133

Table 6. 8 Multivariate Tobit model error parameters and goodness-of-fit

Mean

95% BCI



private car

121.80

5.00

(113.8, 130.2)



light commercial vehicle

49.35

2.03

(46.12, 52.77)



heavy commercial vehicle

58.38

2.39

(54.56, 62.43)



(private car vs. light commercial vehicle)

0.379

0.025

(0.338, 0.420)



(private car vs. heavy commercial vehicle)

-0.053

0.029

(-0.101, -0.005)



(light commercial vehicle vs. heavy commercial vehicle)

-0.017

0.029

(-0.065, 0.031)







0.104

0.005

(0.095, 0.112)







0.126

0.005

(0.117, 0.133)







0.036

0.005

(0.027, 0.043)







0.084

0.002

(0.079, 0.087)

*BCI refers to Bayesian credible interval

Boldface indicates statistical significance at the 5% level

The estimation results of multivariate Tobit model are presented in Table 6.9. It is found

that number of lanes, average lane width, traffic flow, presence of on-street parking, and

year are associated with the crash rates of private car and light commercial vehicle, all at

the 95% credibility level. However, effect of intersection density is significant only for

light commercial vehicle. Time period is revealed to be associated with the crash rates of

private car and heavy commercial vehicle. For heavy commercial vehicle, two more risk

factors - number of lanes and traffic flow are found significantly affect its crash rate.

Difference in the risk factors to crash rates by vehicle type is discussed in the following

paragraphs.

134

Table 6. 9 Results of multivariate Tobit model

*BCI refers to Bayesian credible interval

Boldface indicates statistical significance at the 5% level

Insignificant variables are retained to account for the possible confounding effects on crash rates

Variable

Private car

Light commercial Vehicle

Heavy commercial Vehicle

Mean

95% BCI

Mean

95% BCI

Mean

95% BCI

Constant

-1.339

2.284

(-5.075, 2.431)

0.928

1.455

(-1.438, 3.332)

0.669

1.568

(-1.934, 3.211)

Number of lanes

0.540

0.207

(0.202, 0.881)

0.359

0.132

(0.143, 0.576)

0.289

0.143

(0.054, 0.524)

Average lane width

1.533

0.434

(0.819, 2.246)

0.572

0.279

(0.112, 1.025)

0.245

0.300

(-0.249, 0.738)

Density of intersection

-0.023

0.082

(-0.157, 0.111)

0.252

0.052

(0.167, 0.338)

-0.025

0.056

(-0.118, 0.067)

Presence of on-street parking

3.053

1.274

(0.955, 5.148)

2.719

0.811

(1.384, 4.052)

1.100

0.883

(-0.356, 2.553)

Ln(4-hour traffic flow)

-2.501

0.521

(-3.363, -1.644)

-1.466

0.333

(-2.015, -0.923)

-1.317

0.362

(-1.909, -0.719)

Speed limit

-0.020

0.028

(-0.067, 0.026)

-0.017

0.018

(-0.046, 0.012)

-0.010

0.020

(-0.043, 0.021)

Temporal effect (reference: 7pm-11pm)

Time period 1 (7am-11am)

0.395

0.929

(-1.133, 1.926)

0.152

0.594

(-0.823, 1.130)

0.478

0.645

(-0.584, 1.538)

Time period 2 (11am-3pm)

1.938

0.925

(0.414, 3.455)

0.721

0.593

(-0.254, 1.695)

0.511

0.645

(-0.553, 1.569)

Time period 3 (3pm-7pm)

1.857

0.928

(0.332, 3.381)

0.440

0.595

(-0.537, 1.421)

1.549

0.648

(0.482, 2.614)

Year (reference: year 2015)

Year 2016

-1.108

0.778

(-2.386, 0.171)

-0.730

0.496

(-1.548, 0.086)

-0.193

0.538

(-1.078, 0.694)

Year 2017

-1.525

0.777

(-2.805, -0.250)

-1.422

0.495

(-2.236, -0.608)

0.494

0.537

(-0.389, 1.378)

135

6.5.1 Effects of geometric factors

For the effect of road geometry, average lane width is positively associated with private

car and light commercial vehicle crash rates. However, such finding is not applicable to

the crash rate of heavy commercial vehicle. This could be attributed to the differences in

gross vehicle weight and dimensions (mainly the vehicle width) across vehicle types. In

particular, the weight of heavy commercial vehicle (e.g. 24 tonnes) is much higher than

that of private car (e.g. 3.0 tonnes) or light commercial vehicle (e.g. 5.5 tonnes). Indeed,

the consequences of crashes involving professional drivers, especially drivers of buses

and heavy goods vehicles, are often mortality and severe injury of other innocent road

users (Alvarez and Fierro, 2008). Therefore, to control vehicle trajectories and avoid lane

departure events, various systems have been developed and installed to stabilize the

lateral position of heavy vehicles (Imine et al., 2011; Netto et al., 2004; Hingwe et al.,

2002). On the other hand, crash rates of private car and light commercial vehicle would

increase with the increase in average lane width of road segments. This could be attributed

to driver’s reckless driving on the wider lane (Labi et al., 2017). The increase in average

lane width appears to provide relatively more space for light vehicles, which may imbue

a false sense of security to drivers, thus leading to degraded driving performance.

Moreover, wider lane in general leads to more overtaking events, in which private car,

taxi, and light goods vehicle drivers tend to overtake aggressively (Shackel and Parkin,

2014).

Number of lanes is associated with a positive coefficient for the crash rates of private car,

light commercial vehicle, and heavy commercial vehicle. This finding suggests that crash

risk would increase with the increase in lane-changing opportunities (Pei et al., 2012,

2016; Zeng et al., 2017b). Furthermore, effect of intersection density is found significant

only for the light commercial vehicles. This could be attributed to the prevalence of traffic

violations among light commercial vehicle drivers (Chen et al., 2020; Meng et al., 2017).

For example, light bus drivers have a higher tendency of red light running, speeding,

and/or stopping at intersections (Wong et al., 2008), while bus drivers tend to drive in a

cautious manner with a greater sense of social responsibility (Chen et al., 2020; Paleti et

al., 2010). Consistently, Wu et al. (2016) revealed that non-professional drivers are more

careful when driving through the intersection while taxi drivers are inclined to commit

136

red-light running. Moreover, taxis are more likely to be involved in traffic conflicts at

intersections compared with private cars (Xu et al., 2014). Hence, crash rate of light

commercial vehicle increases when intersection density of the road segment increases.

Presence of on-street parking increases the crash rates of private car and light commercial

vehicle. This aligns with the previous findings that higher crash risk is associated with

more frequent roadside activities near the on-street parking areas (Pei et al., 2016). In

Hong Kong, on-street parking areas are normally provided for the urban roads giving

direct access to the buildings. Hence, it is likely that the road segments with on-street

parking would have more frequent roadside pick-ups, drop-offs, and loading/unloading

activities involving private cars and light commercial vehicles (Sze and Wong, 2007).

Moreover, these roadside activities could translate into more frequent incidences of errant

pedestrian behavior (such as reckless crossing), thus increasing pedestrian-vehicle crash

frequency (Kim et al., 2017; Granié et al., 2014; Ukkusuri et al., 2012).

6.5.2 Effects of traffic condition and time

Regardless of the crash types, crash rates are generally lower when the overall traffic flow

increases. This could be because the mean travel speed of the road segment reduces when

the traffic volume increases (Huang et al., 2016; Zeng et al., 2017a, 2017b). However,

previous study also revealed that increase in exposure would lead to higher crash risk (Pei

et al., 2012, 2016). In this study, the four-hour traffic flow (in a logarithmical form) is

also considered as one of the exposure measures to the risk involved. As such, it would

be worth introducing other exposure measures such as the flows of different vehicle types

in the proposed crash prediction models. In addition, although it is expected that speed

limit would present different effects on crash rates by vehicle type, no significant

evidence could be established based on the current results.

As for the temporal variation, the crash rates of private car and light commercial vehicle

are found to be statistically lower in year 2017 compared with 2015 (see Table 6.9). Such

reductions in crash risk appear to be a favourable consequence of implementing various

road safety measures (e.g. penalty and enforcement strategies against traffic offences,

driver educational and training programs, and road remedial works, etc.). Moreover, as

137

discussed in Pei et al. (2016), the relationship between year and crash risk could possibly

resulted from the effects of other confounding factors over time. As such, it would be

more appropriate to say, effect of the crash year is used as a proxy to the effects of

unobserved risk factors.

With respect to the time of day, crash rate of private car is higher during noon (11am-

3pm) and afternoon (3pm-7pm), as compared to the evening period (7pm-11pm). This is

not surprising since such periods cover the afternoon peak hours that private car drivers

are off work for various activities (e.g. back home, shopping, gathering with friends, etc.).

Higher exposure of private cars on road during afternoon peak hours would contribute to

their higher crash risk. Moreover, private car drivers tend to drive less cautiously during

their off-work hours, when violation behaviours are prevalent, thus increasing the crash

risk (Huang and Chin, 2009). Also, it is revealed that heavy commercial vehicle would

have higher crash risk during the afternoon period (3pm-7pm). In contrast, no significant

evidence was found for temporal variation in the crash rate of light commercial vehicle.

Nevertheless, effects of time on crash rates are revealed to be different across vehicle

type, suggesting different temporal patterns for crash risks across vehicle types.

6.6 Concluding remarks

This chapter proposes two research objectives, 1) to throw more light on the moderating

(mitigating) or magnifying (exacerbating) safety effects of the interaction between

commercial vehicle proportion and road features, and to provide some indication of the

variation of these mediating effects across the commercial vehicle classes, 2) to

simultaneously estimate the crash rates of private car, light commercial vehicle, and

heavy commercial vehicle using the same set of possible risk factors. Effects of road

geometry, traffic control and time period are also considered. Such knowledge could help

urban road authorities develop more context-sensitive and effective roadway design and

commercial traffic operations policies for overall urban road safety enhancement. The

Bayesian random parameter Tobit model and multivariate Tobit model are adopted

respectively for the two objectives, to account for unobserved heterogeneity across

observations and unobserved shared effects across crashes of different vehicle types.

138

For the first objective, the influence of the proportions of each class of commercial

vehicles (bus, taxi, light-goods vehicle, and medium- and heavy-goods vehicles) on crash

rates, for different levels of crash injury severity, were examined. The effects of road

geometry, traffic control and time of day were also investigated. The study also

investigated whether the association between the commercial vehicle percentage and

crashes is moderated by prevailing roadway attributes. The study used random parameters

techniques within a Bayesian Tobit modelling framework, to accommodate possible

heterogeneous effects of the crash factors across the observations. Results suggest that

increases in the proportions of taxi and light-goods vehicle contribute to higher rates of

slight-injury crashes, while the proportion of medium- and heavy-goods vehicles showed

the opposite effect. Also, KSI crash rate decreases with the increase in the proportion of

medium and heavy goods vehicle, while the proportions of public buses and light-goods

vehicle impose an increasing effect. More importantly, significant interaction effects of

commercial vehicle proportions and roadway attributes were revealed in this study. First,

the increasing effect of taxi proportion on slight-injury crash rate is magnified at road

segments that have high intersection density, second, the increasing effect of light-goods

vehicle proportion on slight-injury crash rate is magnified at road segments with on-street

parking, and third, the association between the medium- and heavy- goods vehicle

proportion and KSI crash rate is moderated by the roadway width (number of traffic lanes).

Fourth, an increase in the proportion of medium- and heavy-goods vehicles contributes

to the increase in KSI crash rate of the road segments with high intersection density.

This study bridges the gap in the literature on the interaction between roadway attributes

and commercial vehicle percentages on crash rates, for different levels of crash severity.

The findings of this research provide the transport authority some policy implications in

terms of the further expansion of the red-light camera system, licensing requirements,

arrangement of human police patrols, lane control measures, and review of roadway

design. As there exist limited options to reduce the crash exposure of commercial vehicles,

it is necessary to mitigate their crash risk by improving the safety climate associated with

the operations of trucking companies and regulating the behaviour of their professional

drivers. Therefore, the results of this study can help enhance driver education and training

139

programs that can enhance the social responsibility and safe driving behaviours of

professional drivers.

For the second objective, results show that significant correlations exist between crash

rates across vehicle types. In particular, positive coefficient is found for the correlation

between private car and light commercial vehicle, while negative coefficient is found for

the correlation between private car and heavy commercial vehicle. Effects of average lane

width and presence of on-street parking are significant only for the crash rates of private

car and light commercial vehicle. This could be attributed to the difference in vehicle

characteristics (e.g. gross weight and dimensions), as well as more frequent roadside

activities involving private cars and light commercial vehicles. On the other hand,

previous studies reveal that drivers of light commercial vehicle show higher probability

of traffic violations at intersections, compared with private car and heavy vehicle drivers.

Consistently, our results show that intersection density has statistically significant

increasing effect only for the crash rate of light commercial vehicle. In addition,

significant associations between risk factors including number of lanes, traffic flow, and

temporal variation, and crash rates across three vehicle types are revealed.

To summarize, contribution of this study is twofold. First, it enhances the understanding

on the interaction between commercial vehicle percentages and roadway attributes on

crash rates. Second, risk factors associated with the crash rates across different vehicle

types are identified using the multivariate Tobit regression approach. As it is not viable

to reduce the crash exposure of commercial vehicles, it is necessary to mitigate their crash

risk by moderating the behaviour of professional drivers and crashworthiness of

commercial vehicle fleets. Yet, this study is limited to the major roads that have

continuous and detailed traffic count data. In the extended study, it is worth exploring the

effects of other road environment factors like road class and traffic control when

comprehensive vehicle trajectory data from the transport agency or operators is available.

140

Chapter 7 Conclusions and recommendations

7.1 General conclusions

In this study, attempts have been made to assess safety of professional drivers in Hong

Kong from the behavioural, psychological, and empirical perspectives. First, we

conducted simulators studies to evaluate the interaction between better driving skills of

professional drivers and age-related impairments on driving performance of elderly

drivers, using both conventional performance indicators and surrogate safety measures.

Second, we applied deterrence theory and stated preference survey to investigate the

perceptions and attitudes of professional drivers towards the traffic legislation, in terms

of the penalty and enforcement strategies. Lastly, we developed prediction models for

crash rates by injury severity and vehicle type, incorporating the factors of commercial

vehicle mix, road geometry, traffic control, and time period. Overall, findings of the

driving simulator studies, the perception survey of professional drivers, and the crash risk

analysis of commercial vehicles would provide useful insights into the training and

education, driver management strategies of the transport operators, as well as the driver

licensing policy and enforcement effectiveness of the authorities. Therefore, safety

performance of professional drivers would be enhanced in the long run.

Chapter 2 reviews the literature on the safety of professional drivers, with respect to

driving performance, perception and attitude, and crash risks. Driving under the influence

of fatigue, compensatory and adaptive driving of professional drivers can be measured

using driving simulator approach. Conventional driving performance indicators such as

lateral, steering and speed controls over the vehicle, as well as surrogate safety measures

such as time-to-collision, headway and braking time, are commonly used to assess the

safety of drivers from the behavioural perspective. From the psychological perspective,

questionnaire survey is widely used to measure drivers’ perceptions, attitudes, and self-

reported behaviours. More importantly, emphasis was given to the drivers’ perceptions

towards traffic legislation in terms of penalty and enforcement. Also, findings of the

empirical studies indicating the crash risks of various commercial vehicle types are

presented. An obvious research gap existed for the effects of increasing elderly drivers in

141

the transport sector on road safety. Effects of age, road environment and traffic condition

on the driving performance have been attempted in previous research. However, the

moderating effects by better driving skill and task familiarity of professional drivers have

yet to be explored. Another research gap lays in the evaluation of professional drivers’

perceptions and attitudes towards traffic legislation. As revealed by previous studies,

professional drivers show higher traffic violation rates and propensity to aggressive

driving. It is necessary to apply perception survey to enhance the understanding and

effectiveness of penalty and enforcement strategies. Last but not least, driver behaviour

and crashworthiness of different vehicle types are different. A study examining the safety

effects of the commercial vehicle proportions is thus urgently needed. Overall, the

aforementioned deficiencies motivated us to bridge the gap in the safety assessment of

professional drivers.

Of the road crashes involving personal injury in Hong Kong, over 70% involved at least

one commercial vehicle. Besides, the proportion of older drivers in the transport sector

has been increasing due to the shortage of labor and aging population. It is believed that

increase in age can have adverse impact on driving performance, even that professional

drivers may possess better driving skill than non-professional drivers and the age-related

impairment can be reduced by task familiarity of professional drivers. In Chapter 3, a

driving simulator experiment was conducted to address this question. Additionally,

possible factors that affect the driving performance of professional drivers and that of

non-professional drivers were examined. A total of 50 participants were recruited and 94

tests were completed. Driving performance was assessed in terms of standard deviation

of lateral position (SDLP), standard deviation of heading error (SDHE), mean heading

error (MeanHE) and standard deviation of speed (SDspeed). Results of random intercept

models indicate that lateral and speed control performances of mid-aged drivers were

better than those of older drivers. Then, disaggregated models were established for

professional and non-professional drivers respectively based on the results of market

segmentation analysis. It is found that lateral and speed control performance of mid-aged

professional drivers were better than that of older professional drivers. In contrast, older

non-professional drivers were more likely to have degraded steering performance under

high traffic conditions. Results of this study are indicative to the driver management

142

strategies of the transport operators for sustained safety improvement of commercial

vehicle fleet.

In Chapter 4, we attempt to examine the compensatory behavior and its safety effect

amongst older professional drivers, as compared to those of older non-professional

drivers, using the driving simulator approach. It has been a controversial issue for the

effect of ageing population on driving safety. Apparently, drivers’ physiological and

cognitive performances deteriorate with age. However, older drivers may compensate for

the elevated risk by adjusting their behaviors, known as compensatory strategy. Despite

the extensive research on this topic, the compensatory strategy of older professional

drivers is not well understood since many studies focused on the differences in

compensatory behavior between older and young drivers. Professional drivers tend to be

more skillful and able to cope with the unfavorable driving environments, thus presenting

a higher capability to mitigate the risk. In the driving simulator experiment, participants

were asked to follow a leading vehicle for one hour, and two sudden brake events were

presented. 41 (mid-aged and older) drivers completed the driving tests. Each participant

was required to complete a car-following test, either under high or low traffic flow

conditions. Performance indicators include driving capability (i.e. lateral control,

longitudinal control, and brake reaction time) and compensatory behavior (i.e. average

speed, and time headway). Additionally, two modified traffic conflict measures: time

exposed time-to-collision (TET) and time integrated time-to-collision (TIT) are applied

to indicate the traffic conflict risk. The random parameter Tobit models were estimated

to measure the association between conflict risk and driver attributes, and random

intercept models were used to assess other driving performance indicators. Results show

that despite the impaired lateral control performance and longer brake reaction time of

older drivers, the likelihood of severe traffic conflict of older drivers is lower than that of

mid-aged drivers. Furthermore, though both older professional and older non-

professional drivers adopted longer time headway, the reduction in the risk of severe

traffic conflict is more profound among the older professional drivers. Such findings

suggest that older professional drivers are more capable of mitigating the possible

collision risk by adopting the compensatory strategy, as compared to older non-

professional drivers. This justifies the existence of compound effect by the compensatory

strategy of older driver and better driving skills of professional driver. This research

143

provides useful insights into driver training and management strategies for employers, as

older drivers would become a major cohort in the transportation industry.

Next, for the drivers’ perceptions and attitudes, Chapter 5 presents an evaluation of the

effectiveness of different penalty and camera-based enforcement strategies in curbing

speeding offences by professional drivers in Hong Kong. A stated preference survey

approach is employed to measure the association between penalty and enforcement

strategies and drivers’ speed choices. Data suggest that almost all drivers comply with

speed limits when they reach a camera housing section of the road. For other road

sections, a panel mixed logit model is estimated and applied to understand the

effectiveness of penalties and enforcement strategies on driver’s speeding behaviors.

Driving-offence points (DOPs) are found to be more effective than monetary fines in

deterring speeding offences, albeit there is significant heterogeneity in how drivers

respond to these strategies. Warning drivers of an upcoming camera-based enforcement

section increased speed compliance. Several demographic and employment

characteristics, driving history and perception variables also influence drivers’ choices of

speed compliance. Finally, besides penalty and enforcement strategies, driver education

and training programs aimed at addressing aggressiveness/risk-taking traits might help

reduce repeated speeding offences among drivers.

Finally, in Chapter 6, we aim to examine the main and interaction effects of commercial

vehicle proportion (CVP) and roadway attributes on the overall crash rates, and to conduct

a multivariate analysis of crash rates by vehicle type. First, a random-parameter Tobit

model was used to measure the relationships between the CVP and crash rates by severity

level. Second, a multivariate Tobit model is applied to identify the risk factors affecting

the crash rates across different vehicle types (including private car, light commercial

vehicle, and heavy commercial vehicle). A comprehensive database integrating the crash,

traffic and road inventory data of the target road segments in Hong Kong was used. The

results suggest that the CVP of each class significantly and directly affect the crash rates,

for the various crash severity levels. The results also suggest that the interaction between

CVP and roadway attributes is significant enough to mediate the effect of CVP on crash

rates, and the magnitude and direction of such mediation varies across the vehicle classes,

crash severity levels, and roadway attribute type in four ways. First, the increasing effect

144

of taxi proportion on slight-injury crash rate is magnified at road segments with high

intersection density. Second, the increasing effect of light-goods vehicle proportion on

slight-injury crash rate is magnified at road segments with on-street parking. Third, the

association between the medium- and heavy-goods vehicle proportion and killed/severe

injury (KSI) crash rate, is moderated by the roadway width (number of traffic lanes).

Finally, a higher proportion of medium- and heavy-goods vehicles generally contributes

to increased KSI crash rate at road segments with high intersection density. Furthermore,

the results of crash rates across three vehicle types show that correlations exist between

crash rates across vehicle types. Crash risks of private car and light commercial vehicle

are affected by number of lanes, average lane width, traffic flow, presence of on-street

parking, and year. While effect of intersection density is significant only for the light

commercial vehicle. For the crash risk of heavy commercial vehicle, three risk factors -

number of lanes, traffic flow, and time period are significant. Overall, the findings of this

research are expected not only to help guide commercial vehicle enforcement strategy,

licensing policy, and lane control measures, but also to review existing urban roadway

designs to enhance safety.

7.2 Main findings and contributions

This thesis has assessed the safety of professional drivers in Hong Kong by addressing

five research questions from the behavioural, psychological, and empirical perspectives:

1) whether age-related impairments on driving performance can be reduced by the driving

experience and task familiarity of professional drivers, 2) whether the compensatory

strategies of older drivers are different between professional and non-professional drivers,

3) how the penalty and enforcement strategies deter professional drivers from traffic

violations, 4) whether the relationship between commercial vehicle proportions and

crashes can be moderated by roadway attributes, and 5) whether the effects of risk factors

vary across crashes categorized by vehicle type. The main findings are concluded below.

1) Driving performance of professional drivers

 Results of the driving simulator experiment evidence the better driving skill of

professional drivers and impaired driving performance of older drivers. It is revealed

145

that age-related impairments on driving performance (i.e. steering control) could be

reduced by the driving experience and task familiarity of professional drivers.

2) Conflict risk of professional drivers

 Two modified traffic conflict measures: time exposed time-to-collision (TET) and

time integrated time-to-collision (TIT) are applied to indicate the traffic conflict risk.

 Though both older professional and older non-professional drivers adopted longer

time headway for risk compensation, a more profound reduction in conflict risk is

found for the older professional drivers.

 Older professional drivers are more capable of mitigating the possible collision risk

by adopting the compensatory strategy, as compared to older non-professional

drivers.

3) Perceptions and attitudes of professional drivers

 A stated preference approach was employed to evaluate the deterrent effects of

penalty and enforcement strategies on the speeding behaviours of professional

drivers.

 Driving-offence points are found to be more effective than monetary fines in

deterring speeding offences among professional drivers. Also, warning drivers of an

upcoming camera enforcement section increased speed compliance.

4) Safety effect of commercial vehicle mix

 This study examines the moderating effects of roadway attributes on the association

between commercial vehicle proportion and crash rate.

 Small commercial passenger vehicle (taxi) proportion: effect on slight-injury crash

rate is moderated by intersection density.

 Light-goods commercial vehicle proportion: effect on slight-injury crash rate is

magnified by the presence of on-street parking.

 Medium- and heavy-goods vehicle proportion: association with KSI crash rate is

moderated by the number of traffic lanes and intersection density.

146

5) Risk factors to crash rates of different vehicle types

 Vehicles are categorized into three types — private car, light commercial vehicle,

and heavy commercial vehicle.

 A Bayesian multivariate Tobit model was used to estimate the crash rates of private

car, light commercial vehicle, and heavy commercial vehicle simultaneously.

Significant correlations exist between crash rates across vehicle types.

 Effects of geometric factors and traffic control on crash rates vary among different

vehicle types.

Based on the results from the proposed research questions, this thesis is able to make

contributions to driver recruitment and management strategies, effective penalties and

enforcement strategies against traffic violations, as well as safety countermeasures

tailored for professional drivers. Here suggest some potential implications derived from

the above findings. For examples, (i) in the context of the driver recruitment and

management in an aging society like Hong Kong, transport operators should not make

decisions solely based on driver age. Instead, rigorous assessment of driving skills and

training programs should be provided for the older drivers. (ii) For the effectiveness of

penalties, combining the quantity of fines with appropriate warning messages could

probably increase the deterrent effect of monetary fines. Further, higher penalties could

be considered for repeat offenders. Moreover, equipping all dummy camera housing

boxes with actual speed cameras may not necessarily enhance the effectiveness of

camera-base enforcement. (iii) Increases in the proportions of taxi, light bus, light goods

vehicle is associated with increase in the crash rates of the studied road segments.

Transport authorities or operators should provide taxi, light bus and light goods vehicle

drivers with additional training programs aimed to reduce risk-taking and aggressive

driving traits. (iv) Effects of geometric factors and traffic control vary across the crash

rates of private car, light commercial vehicle, and heavy commercial vehicle. This would

provide useful insights for the roadway safety design. For instance, engineers would

consider whether a particular design (e.g. lane width, intersection) is capable of serving

heavy vehicles and private cars simultaneously.

147

7.3 Limitations

Despite the contributions to the literature described in the above paragraphs, this research

should be interpreted in the context of the limitations. Firstly, with respect to the simulator

studies, effect of gender on driving performance is not attempted. The driving

performances of male drivers only are assessed in the driving simulator experiment, given

the relatively small sample size. In addition, although many previous studies have

demonstrated the absolute and relative validity of the simulator experiment, the ability of

simulator studies to reflect realistic driving is often questioned. Naturalistic driving

studies could aid in verifying the findings of our simulator studies. In particular, car

following behaviour of professional drivers at work could be influenced by time pressure

and market competition. However, it is not possible to incorporate the effect of work

pressure in the driving simulator study. Moreover, compensatory strategy could be

prevalent for the drivers who have known cognitive impairment and crash involvement

records. However, the cognitive ability of older drivers is not examined in current study.

Secondly, for the stated preference survey, the study merely investigates a few

demographics and operational characteristics of professional drivers, while the

information on the psychological metrics of the participants is not collected. In addition,

our scenarios are limited to a typical city road with a speed limit of 50km/h. Moreover,

for the attribute associated with the placement of the warning sign, distance of 50 meters,

100 meters, 150 meters, and 200 meters upstream of the speed camera housing are

considered in the survey setting. Perhaps a time separation rather than a space separation

(between the placement of a warning sign and the camera housing unit) would be better

capture how individuals respond to warning signs before entering monitored roadway

section. It is also important to note, findings of this study are based on self-reported speed

choices within stated experiments, which could influence the reliability and accuracy of

the relationships estimated. A study based on an actual field experimental design and field

observations of speed at different sections would be more credible. Finally, for sensitive

questions related to social desirability, a self-administered or internet survey could

improve the reliability of the survey, despite that face-to-face interviews help receive

more serious responses from the participants.

148

There were limitations for the crash rate analysis as well. First, its scope is limited to the

major roads that have continuous and detailed traffic count data. Also, due to the data

availability, this study only considered weekday crashes. Moreover, the effects of other

road attributes such as road functional class, horizontal alignment and the mean speed of

traffic are not attempted. Last but not least, consistent with the positions of Zhai et al.

(2019), Xing et al. (2019), and Gao et al. (2020), it will be worth exploring, when

comprehensive weather information is available, the moderating effects of weather

conditions on the association between commercial vehicle percentage and crash rates.

7.4 Recommendations for future research

Section 7.1 and 7.2 has outlined the contributions of this thesis to the safety assessment

of professional drivers through the behavioural, psychological, and empirical studies. Yet,

the current work can be further extended in the future. The recommendations for future

research in four aspects are listed.

7.4.1 Conventional driving performance measured by simulator

Further studies could explore the possible effect of personal characteristics on the level

of aggressiveness, and thus the driving behaviour, when more comprehensive information

on the psychological metrics of participants is available. Similarly, the effects of driving

history (e.g. crash involvement, traffic offense) and risk perception of professional drivers

on the driving performance should be attempted. The effect of gender on the temporal

change in driving performance of professional drivers could be explored. Moreover,

cognitive measures for the elderly drivers could be adopted, as well as the subjective or

physiological measures for mental workload and fatigue. Furthermore, effects of

environmental factors such as road design, lighting condition and weather on the driving

performance of professional drivers can be revealed in extended study.

7.4.2 Traffic conflict analysis using driving simulator

Driving simulator can capture the driving behaviour and vehicle trajectory data in a

controlled driving environment. Thus, it could be a promising approach for traffic conflict

149

analysis in the future. In addition, other surrogate measures such as headway, post

encroachment time, lateral distance to departure, etc., can be adopted in driving simulator

experiments to explore their applicability. Moreover, other critical events (e.g. pedestrian

crossing; car incursion from roadside) can be incorporated into experimental design to

assess the driver’s conflict risk in different situations. Furthermore, effects of other driver

characteristics (e.g. income, vehicle type, salary system), driving history, and safety

perception on drivers’ conflict risks certainly constitute important avenues for future

research as the findings from these studies could be indicative to the safety management

of professional drivers.

7.4.3 Perception survey

In addition to a few demographics and operational characteristics of professional drivers,

it would be worth exploring the possible effects of latent characteristics on the propensity

and severity of traffic offense, when more comprehensive information on the

physiological and psychological metrics of the participants is available. Moreover, other

driving environments could be incorporated in the stated preference design. It would be

interesting to explore the effects of different road environments on the violation behaviour

of professional drivers. On the other hand, the perceptions and attitudes of private car

drivers towards the penalty and enforcement strategies could be revealed in the extended

study for a comparable result. Also, similar approach can be applied to examine the

deterrent effect of police surveillance, as drivers would perceive differently when

encountering automated enforcement systems and manual enforcement.

7.4.4 Crash risk analysis for commercial vehicles

In future studies, it will be worthwhile to explore the safety effects of other road attributes

such as road class, horizontal alignment and average vehicle speed of traffic when more

comprehensive data are available. Furthermore, the study did not include the crash

pattern, and therefore, in a future study of this kind, it will be illuminating to consider that

factor using a multivariate approach. For the multivariate analysis, possible

heterogeneous effects of risk factors across observations should be carefully considered

by introducing random parameters. Also, defining an exposure measure is critical in crash

150

risk assessment. As such, it would be worth introducing other exposure measures such as

the flow of vehicles by different modes in crash prediction models.

Moreover, the safety effects of behavioural attributes of commercial vehicle drivers could

be insightful to the practical effectiveness of traffic control and management strategies

geared towards commercial vehicle safety enhancement and could be considered in future

studies. Last but not least, it may be worthwhile collecting comprehensive weather

information and including such data in similar future analysis. That way, the moderating

effects of weather conditions on the association between commercial vehicle percentage

and crash rates. Prospectively, the information to be earned from all such future research

could help the road agency refine existing driver regulations and streamline urban traffic

control and management strategies related to commercial vehicle operations and safety.

151

References

Abdel-Aty, M., Uddin, N., Pande, A., Abdalla, M. F., & Hsia, L. (2004). Predicting

freeway crashes from loop detector data by matched case-control logistic

regression. Transportation Research Record, 1897(1), 88-95.

Abdel-Rahim, A., & Sonnen, J. (2012). Potential safety effects of lane width and shoulder

width on two-lane rural state highways in Idaho (No. FHWA-ID-12-200). Idaho.

Transportation Department.

Aguero-Valverde, J., & Jovanis, P. P. (2008). Analysis of road crash frequency with

spatial models. Transportation Research Record, 2061(1), 55-63.

Ahlström, C., Anund, A., Fors, C., & Åkerstedt, T. (2018). Effects of the road

environment on the development of driver sleepiness in young male

drivers. Accident Analysis and Prevention, 112, 127-134.

Ahmed, M. M., Abdel-Aty, M., & Yu, R. (2012). Assessment of interaction of crash

occurrence, mountainous freeway geometry, real-time weather, and traffic

data. Transportation Research Record, 2280(1), 51-59.

Ahmed, M., Huang, H., Abdel-Aty, M., & Guevara, B. (2011). Exploring a Bayesian

hierarchical approach for developing safety performance functions for a

mountainous freeway. Accident Analysis & Prevention, 43(4), 1581-1589.

Ajzen, I. (1991). The theory of planned behavior. Organizational behavior and human

decision processes, 50(2), 179-211.

Alarifi, S. A., Abdel-Aty, M., & Lee, J. (2018). A Bayesian multivariate hierarchical

spatial joint model for predicting crash counts by crash type at intersections and

segments along corridors. Accident Analysis & Prevention, 119, 263-273.

Albertsson, P., & Falkmer, T. (2005). Is there a pattern in European bus and coach

incidents? A literature analysis with special focus on injury causation and injury

mechanisms. Accident Analysis & Prevention, 37(2), 225-233.

Ali, Y., Haque, M. M., Zheng, Z., Washington, S., & Yildirimoglu, M. (2019). A hazard-

based duration model to quantify the impact of connected driving environment on

safety during mandatory lane-changing. Transportation Research Part C: Emerging

Technologies, 106, 113-131.

152

Alvarez, F. J., & Fierro, I. (2008). Older drivers, medical condition, medical impairment

and crash risk. Accident Analysis & Prevention, 40(1), 55-60.

Álvarez, P., Fernández, M. A., Gordaliza, A., Mansilla, A., & Molinero, A. (2020).

Geometric road design factors affecting the risk of urban run-off crashes. A case-

control study. Plos One, 15(6), e0234564.

Anastasopoulos, P. C. (2016). Random parameters multivariate tobit and zero-inflated

count data models: addressing unobserved and zero-state heterogeneity in accident

injury-severity rate and frequency analysis. Analytic Methods in Accident

Research, 11, 17-32.

Anastasopoulos, P. C., Mannering, F. L., Shankar, V. N., & Haddock, J. E. (2012a). A

study of factors affecting highway accident rates using the random-parameters tobit

model. Accident Analysis & Prevention, 45, 628-633.

Anastasopoulos, P. C., Shankar, V. N., Haddock, J. E., & Mannering, F. L. (2012b). A

multivariate tobit analysis of highway accident-injury-severity rates. Accident

Analysis & Prevention, 45, 110-119.

Anastasopoulos, P. C., Tarko, A. P., & Mannering, F. L. (2008). Tobit analysis of vehicle

accident rates on interstate highways. Accident Analysis & Prevention, 40(2), 768-

775.

Anastasopoulos, P.C., & Mannering, F. L. (2009). A note on modeling vehicle accident

frequencies with random-parameters count models. Accident Analysis &

Prevention, 41(1), 153-159.

Andrews, E. C., & Westerman, S. J. (2012). Age differences in simulated driving

performance: Compensatory processes. Accident Analysis & Prevention, 45, 660-

668.

Arnedt, J. T., Geddes, M. A. C., & MacLean, A. W. (2005). Comparative sensitivity of a

simulated driving task to self-report, physiological, and other performance measures

during prolonged wakefulness. Journal of Psychosomatic Research, 58(1), 61-71.

Audit Commission of Hong Kong Special Administrative Region. (2013, March).

Administration of Road Safety Measures. Retrieved July 10, 2018, from

https://www.aud.gov.hk/eng/pubpr_arpt/aud_TraDep.htm

Autey, J., Sayed, T., & Zaki, M. H. (2012). Safety evaluation of right-turn smart channels

using automated traffic conflict analysis. Accident Analysis & Prevention, 45, 120-

130.

153

Azimi, G., Rahimi, A., Asgari, H., & Jin, X. (2020). Severity analysis for large truck

rollover crashes using a random parameter ordered logit model. Accident Analysis

& Prevention, 135, 105355.

Ballesteros, M. F., Dischinger, P. C., & Langenberg, P. (2004). Pedestrian injuries and

vehicle type in Maryland, 1995–1999. Accident Analysis & Prevention, 36(1), 73-

81.

Bao, J., Liu, P., & Ukkusuri, S. V. (2019). A spatiotemporal deep learning approach for

citywide short-term crash risk prediction with multi-source data. Accident Analysis

& Prevention, 122, 239-254.

Barua, U., & Tay, R. (2010). Severity of urban transit bus crashes in Bangladesh. Journal

of Advanced Transportation, 44(1), 34-41.

Behnood, A., & Mannering, F. (2019). Time-of-day variations and temporal instability of

factors affecting injury severities in large-truck crashes. Analytic Methods in

Accident Research, 23, 100102.

Bélanger, A., Gagnon, S., & Yamin, S. (2010). Capturing the serial nature of older drivers’

responses towards challenging events: A simulator study. Accident Analysis &

Prevention, 42(3), 809-817.

Bhat, C. R. (1997). Work travel mode choice and number of non-work commute

stops. Transportation Research Part B: Methodological, 31(1), 41-54.

Bhat, C. R. (2001). Quasi-random maximum simulated likelihood estimation of the mixed

multinomial logit model. Transportation Research Part B: Methodological, 35(7),

677-693.

Bhat, C. R. (2003). Simulation estimation of mixed discrete choice models using

randomized and scrambled Halton sequences. Transportation Research Part B:

Methodological, 37(9), 837-855.

Bhat, C. R., & Sardesai, R. (2006). The impact of stop-making and travel time reliability

on commute mode choice. Transportation Research Part B: Methodological, 40(9),

709-730.

Blincoe, K. M., Jones, A. P., Sauerzapf, V., & Haynes, R. (2006). Speeding drivers’

attitudes and perceptions of speed cameras in rural England. Accident Analysis &

Prevention, 38(2), 371-378.

Blower, D. F., Campbell, K. L., & Green, P. E. (1993). Accident rates for heavy truck-

tractors in Michigan. Accident Analysis & Prevention, 25(3), 307-321.

154

Borowsky, A., & Oron-Gilad, T. (2013). Exploring the effects of driving experience on

hazard awareness and risk perception via real-time hazard identification, hazard

classification, and rating tasks. Accident Analysis & Prevention, 59, 548-565.

Bose, D., Arregui-Dalmases, C., Sanchez-Molina, D., Velazquez-Ameijide, J., &

Crandall, J. (2013). Increased risk of driver fatality due to unrestrained rear-seat

passengers in severe frontal crashes. Accident Analysis & Prevention, 53, 100-104.

Boufous, S., & Williamson, A. (2006). Work-related traffic crashes: a record linkage

study. Accident Analysis & Prevention, 38(1), 14-21.

Boyle, L. N., & Lee, J. D. (2010). Using driving simulators to assess driving safety.

Accident Analysis & Prevention, 42(3), 785-787.

Boyle, L. N., Tippin, J., Paul, A., & Rizzo, M. (2008). Driver performance in the moments

surrounding a microsleep. Transportation Research Part F: Traffic Psychology and

Behaviour, 11(2), 126-136.

Braitman, K. A., Kirley, B. B., Ferguson, S., & Chaudhary, N. K. (2007). Factors leading

to older drivers' intersection crashes. Traffic Injury Prevention, 8(3), 267-274.

Briscoe, S. (2004). Raising the bar: can increased statutory penalties deter drink-

drivers?. Accident Analysis & Prevention, 36(5), 919-929.

Brookhuis, K. A., & De Waard, D. (1993). The use of psychophysiology to assess driver

status. Ergonomics, 36(9), 1099-1110.

Brown, I. D. (1997). Prospects for technological countermeasures against driver fatigue.

Accident Analysis & Prevention, 29(4), 525-531.

Bunn, T. L., Slavova, S., Struttmann, T. W., & Browning, S. R. (2005). Sleepiness/fatigue

and distraction/inattention as factors for fatal versus nonfatal commercial motor

vehicle driver injuries. Accident Analysis & Prevention, 37(5), 862-869.

Cameron, M. H., Newstead, S. V., Diamantopoulou, K., & Oxley, P. (2003). The

interaction between speed camera enforcement and speed-related mass media

publicity in Victoria, Australia. In Annual Proceedings/Association for the

Advancement of Automotive Medicine (Vol. 47, p. 267).

Cantin, V., Lavallière, M., Simoneau, M., & Teasdale, N. (2009). Mental workload when

driving in a simulator: Effects of age and driving complexity. Accident Analysis &

Prevention, 41(4), 763-771.

Carnis, L., & Blais, E. (2013). An assessment of the safety effects of the French speed

camera program. Accident Analysis & Prevention, 51, 301-309.

155

Census and Statistic Department of Hong Kong Special Administrative Region. (2018a,

July).Women and Men in Hong Kong - Key Statistics. Retrieved March 6, 2019, from

https://www.censtatd.gov.hk/hkstat/sub/sp180.jsp?productCode=B1130303

Census and Statistic Department of Hong Kong Special Administrative Region. (2018b,

March). Report on Annual Earnings and Hours Survey. Retrieved March 6, 2019,

from https://www.censtatd.gov.hk/hkstat/sub/sp210_tc.jsp?productCode=B1050014

Census and Statistic Department of Hong Kong Special Administrative Region. (2018c,

March). Report on Marriage and Divorce Trends in Hong Kong, 1991 to 2016.

Retrieved July 20, 2019, from

https://www.censtatd.gov.hk/hkstat/sub/sp160.jsp?productCode=FA100055

Census and Statistics Department of Hong Kong Special Administrative Region,

Demographic Statistics Section. (2017, September 8). Hong Kong Population

Projections. Retrieved April 26, 2018, from

http://www.censtatd.gov.hk/hkstat/sub/sp190.jsp?productCode=B1120015

Cestac, J., Paran, F., & Delhomme, P. (2011). Young drivers’ sensation seeking,

subjective norms, and perceived behavioral control and their roles in predicting

speeding intention: How risk-taking motivations evolve with gender and driving

experience. Safety Science, 49(3), 424-432.

Chang, X., Li, H., Qin, L., Rong, J., Lu, Y., & Chen, X. (2019). Evaluation of cooperative

systems on driver behavior in heavy fog condition based on a driving

simulator. Accident Analysis & Prevention, 128, 197-205.

Charlton, J. L., Oxley, J., Fildes, B., Oxley, P., Newstead, S., Koppel, S., & O’Hare, M.

(2006). Characteristics of older drivers who adopt self-regulatory driving

behaviours. Transportation Research Part F: Traffic Psychology and

Behaviour, 9(5), 363-373.

Chen, F., Ma, X., & Chen, S. (2014). Refined-scale panel data crash rate analysis using

random-effects tobit model. Accident Analysis & Prevention, 73, 323-332.

Chen, S. (2019b). Safety implications of roadway design and management: new evidence

and insights in the traditional and emerging (autonomous vehicle) operating

environments (Doctoral dissertation, Purdue University Graduate School).

Chen, S., Saeed, T. U., Alqadhi, S. D., & Labi, S. (2019a). Safety impacts of pavement

surface roughness at two-lane and multi-lane highways: accounting for

156

heterogeneity and seemingly unrelated correlation across crash severities.

Transportmetrica A: transport science, 15(1), 18-33.

Chen, S., Saeed, T. U., Alqadhi, S.D., & Labi, S. (2017b). Safety impacts of pavement

surface roughness at two-lane and multi-lane highways: accounting for

heterogeneity and seemingly unrelated correlation across crash severities.

Transportmetrica A: Transport Science, 15(1), 18-33.

Chen, S., Saeed, T.U., & Labi, S. (2017a). Impact of road-surface condition on rural

highway safety: A multivariate random parameters negative binomial approach.

Analytic Methods in Accident Research, 16, 75-89.

Chen, S., Saeed, T.U., Alinizzi, M., Lavrenz, S., Labi, S. (2019c). Safety sensitivity to

roadway characteristics: A comparison across highway classes. Accident Analysis &

Prevention, 123, 39-50.

Chen, T., Bai, L., & Sze, N. N. (2019b). Factors affecting the severity of rear-end conflicts:

a driving simulator study. In 2019 5th International Conference on Transportation

Information and Safety (ICTIS) (pp. 1182-1187). IEEE.

Chen, T., Sze, N. N., & Bai, L. (2019a). Safety of professional drivers in an ageing

society–A driving simulator study. Transportation Research Part F: Traffic

Psychology and Behaviour, 67, 101-112.

Chen, T., Sze, N. N., Saxena, S., Pinjari, A. R., Bhat, C. R., & Bai, L. (2020). Evaluation

of penalty and enforcement strategies to combat speeding offences among

professional drivers: a Hong Kong stated preference experiment. Accident Analysis

& Prevention, 135, 105366.

Chen, T., Sze, N.N., Chen, S., Labi, S. (2020b). Urban road space allocation incorporating

the safety and construction cost impacts of lane and footpath widths. Journal of

Safety Research, In press.

Chen, T., Sze, N.N., Saxena, S., Pinjari, A.R., Bhat, C.R., & Bai, L. (2020a). Evaluation

of penalty and enforcement strategies to combat speeding offences among

professional drivers: a Hong Kong stated preference experiment. Accident Analysis

& Prevention, 135, 105366.

Chimba, D., Sando, T., & Kwigizile, V. (2010). Effect of bus size and operation to crash

occurrences. Accident Analysis & Prevention, 42(6), 2063-2067.

Chin, H.C., & Huang, H.L. (2009). Safety assessment of taxi drivers in

Singapore. Transportation Research Record, 2114(1), 47-56.

157

Choudhary, P., & Velaga, N. R. (2017). Modelling driver distraction effects due to mobile

phone use on reaction time. Transportation Research Part C: Emerging

Technologies, 77, 351-365.

Cnossen, F., Meijman, T., & Rothengatter, T. (2004). Adaptive strategy changes as a

function of task demands: a study of car drivers. Ergonomics, 47(2), 218-236.

Comte, S. L., & Jamson, A. H. (2000). Traditional and innovative speed-reducing

measures for curves: an investigation of driver behaviour using a driving

simulator. Safety Science, 36(3), 137-150.

Cooper, P. J. (1984). Experience with traffic conflicts in Canada with emphasis on “post

encroachment time” techniques. In International calibration study of traffic conflict

techniques (pp. 75-96). Springer, Berlin, Heidelberg.

Damm, L., Nachtergaële, C., Meskali, M., & Berthelon, C. (2011). The evaluation of

traditional and early driver training with simulated accident scenarios. Human

Factors, 53(4), 323-337.

Davidse, R. J., Hagenzieker, M. P., van Wolffelaar, P. C., & Brouwer, W. H. (2009).

Effects of in-car support on mental workload and driving performance of older

drivers. Human Factors, 51(4), 463-476.

De Craen, S., Twisk, D. A., Hagenzieker, M. P., Elffers, H., & Brookhuis, K. A. (2008).

The development of a method to measure speed adaptation to traffic complexity:

Identifying novice, unsafe, and overconfident drivers. Accident Analysis &

Prevention, 40(4), 1524-1530.

De Pauw, E., Daniels, S., Brijs, T., Hermans, E., & Wets, G. (2014a). An evaluation of

the traffic safety effect of fixed speed cameras. Safety Science, 62, 168-174.

De Pauw, E., Daniels, S., Brijs, T., Hermans, E., & Wets, G. (2014b). Behavioural effects

of fixed speed cameras on motorways: Overall improved speed compliance or

kangaroo jumps?. Accident Analysis & Prevention, 73, 132-140.

De Raedt, R., & Ponjaert‐Kristoffersen, I. (2000). Can strategic and tactical compensation

reduce crash risk in older drivers?. Age and Ageing, 29(6), 517-521.

De Waard, D., & Rooijers, T. (1994). An experimental study to evaluate the effectiveness

of different methods and intensities of law enforcement on driving speed on

motorways. Accident Analysis & Prevention, 26(6), 751-765.

158

Delhomme, P., Chaurand, N., & Paran, F. (2012). Personality predictors of speeding in

young drivers: Anger vs. sensation seeking. Transportation Research Part F: Traffic

Psychology and Behaviour, 15(6), 654-666.

Desapriya, E., Subzwari, S., Sasges, D., Basic, A., Alidina, A., Turcotte, K., & Pike, I.

(2010). Do light truck vehicles (LTV) impose greater risk of pedestrian injury than

passenger cars? A meta-analysis and systematic review. Traffic injury

prevention, 11(1), 48-56.

Diakaki, C., Papageorgiou, M., Papamichail, I., & Nikolos, I. (2015). Overview and

analysis of vehicle automation and communication systems from a motorway traffic

management perspective. Transportation Research Part A: Policy and Practice, 75,

147-165.

Dinu, R. R., & Veeraragavan, A. (2011). Random parameter models for accident

prediction on two-lane undivided highways in India. Journal of safety

research, 42(1), 39-42.

Discetti, P., & Lamberti, R. (2011). Traffic sign sight distance for low-volume

roads. Transportation Research Record, 2203(1), 64-70.

Dong, C., Clarke, D.B., Yan, X., Khattak, A., & Huang, B. (2014). Multivariate random-

parameters zero-inflated negative binomial regression model: An application to

estimate crash frequencies at intersections. Accident Analysis & Prevention, 70,

320-329.

Dowling, K. W., & Holloman, E. (2008). The effects of conspicuous traffic enforcement

on speeding behaviors: A study of speed reduction response. International Social

Science Review, 83(3/4), 181-188.

Du, H., Zhao, X., Zhang, X., Zhang, Y., & Rong, J. (2015). Effects of fatigue on driving

performance under different roadway geometries: a simulator study. Traffic Injury

Prevention, 16(5), 468-473.

Duke, J., Guest, M., & Boggess, M. (2010). Age-related safety in professional heavy

vehicle drivers: A literature review. Accident Analysis & Prevention, 42(2), 364-371.

Dumbaugh, E. (2006). Design of Safe Urban Roadsides: An Empirical Analysis,

Transportation Research Record 1961, 74–82.

Efron, B. (2014). Estimation and accuracy after model selection. Journal of the American

Statistical Association, 109(507), 991-1007.

Efron, B., & Tibshirani, R. J. (1994). An introduction to the bootstrap. CRC press.

159

Elvik, R. (1997). Effects on accidents of automatic speed enforcement in

Norway. Transportation Research Record: Journal of the Transportation Research

Board, (1595), 14-19.

Elvik, R. (2019). The more (sharp) curves, the lower the risk. Accident Analysis &

Prevention, 133, 105322.

Elvik, R., & Christensen, P. (2007). The deterrent effect of increasing fixed penalties for

traffic offences: the Norwegian experience. Journal of Safety Research, 38(6), 689-

695.

Farahmand, B., & Boroujerdian, A. M. (2018). Effect of road geometry on driver fatigue

in monotonous environments: A simulator study. Transportation Research Part F:

Traffic Psychology and Behaviour, 58, 640-651.

Farrow, A., & Reynolds, F. (2012). Health and safety of the older worker. Occupational

Medicine, 62(1), 4-11.

Feng, S., Li, Z., Ci, Y., & Zhang, G. (2016). Risk factors affecting fatal bus accident

severity: Their impact on different types of bus drivers. Accident Analysis &

Prevention, 86, 29-39.

Fernandes, R., Hatfield, J., & Job, R. S. (2010). A systematic investigation of the

differential predictors for speeding, drink-driving, driving while fatigued, and not

wearing a seat belt, among young drivers. Transportation Research Part F: Traffic

Psychology and Behaviour, 13(3), 179-196.

Filtness, A. J., Reyner, L. A., & Horne, J. A. (2012). Driver sleepiness—Comparisons

between young and older men during a monotonous afternoon simulated

drive. Biological Psychology, 89(3), 580-583.

Fitzpatrick, C. D., Rakasi, S., & Knodler Jr, M. A. (2017). An investigation of the

speeding-related crash designation through crash narrative reviews sampled via

logistic regression. Accident Analysis & Prevention, 98, 57-63.

Fleiter, J. J., Lennon, A., & Watson, B. (2010). How do other people influence your

driving speed? Exploring the ‘who’and the ‘how’of social influences on speeding

from a qualitative perspective. Transportation Research Part F: Traffic Psychology

and Behaviour, 13(1), 49-62.

Friswell, R., & Williamson, A. (2013). Comparison of the fatigue experiences of short

haul light and long distance heavy vehicle drivers. Safety science, 57, 203-213.

160

Gao, K., Tu, H., Sun, L., Sze, N. N., Song, Z., & Shi, H. (2020). Impacts of reduced

visibility under hazy weather condition on collision risk and car-following behavior:

Implications for traffic control and management. International Journal of

Sustainable Transportation, 14(8), 635-642.

Gargoum, S. A., & El-Basyouny, K. (2018). Intervention analysis of the safety effects of

a legislation targeting excessive speeding in Canada. International Journal of Injury

Control and Safety Promotion, 25(2), 212-221.

Gates, T.J., & Noyce, D.A. (2010). Dilemma zone driver behavior as a function of vehicle

type, time of day, and platooning. Transportation Research Record, 2149(1), 84-93.

Gates, T.J., Noyce, D.A., Laracuente, L., & Nordheim, E.V. (2007). Analysis of dilemma

zone driver behavior at signalized intersections. Transportation Research Record,

2030, 29-39.

Gibbs, J. P. (1985). Deterrence theory and research. Nebraska Symposium on

Motivation, 33, pp. 87-130

Gitelman, V., Balasha, D., Carmel, R., Hendel, L., & Pesahov, F. (2012). Characterization

of pedestrian accidents and an examination of infrastructure measures to improve

pedestrian safety in Israel. Accident Analysis & Prevention, 44(1), 63-73.

Glendon, A. I., & Cernecca, L. (2003). Young drivers’ responses to anti-speeding and

anti-drink-driving messages. Transportation Research Part F: Traffic Psychology

and Behaviour, 6(3), 197-216.

Granié, M. A., Brenac, T., Montel, M. C., Millot, M., & Coquelet, C. (2014). Influence

of built environment on pedestrian's crossing decision. Accident Analysis &

Prevention, 67, 75-85.

Gross, F., & Jovanis, P. P. (2007). Estimation of the safety effectiveness of lane and

shoulder width: Case-control approach. Journal of Transportation

Engineering, 133(6), 362-369.

Gu, X., Yan, X., Ma, L., & Liu, X. (2020). Modeling the service-route-based crash

frequency by a spatiotemporal-random-effect zero-inflated negative binomial model:

An empirical analysis for bus-involved crashes. Accident Analysis &

Prevention, 144, 105674.

Guo, Y., Li, Z., Liu, P., & Wu, Y. (2019). Modeling correlation and heterogeneity in crash

rates by collision types using full Bayesian random parameters multivariate Tobit

model. Accident Analysis & Prevention, 128, 164-174.

161

Hajivassiliou, V. A., & Ruud, P. A. (1994). Classical estimation methods for LDV models

using simulation. Handbook of Econometrics, 4, 2383-2441.

Haque, M. M., & Washington, S. (2015). The impact of mobile phone distraction on the

braking behaviour of young drivers: a hazard-based duration model. Transportation

Research Part C: Emerging Technologies, 50, 13-27.

Haque, M. M., & Washington, S. (2014). A parametric duration model of the reaction

times of drivers distracted by mobile phone conversations. Accident Analysis &

Prevention, 62, 42-53.

Harless, D. W., & Hoffer, G. E. (2007). Do laboratory frontal crash test programs predict

driver fatality risk? Evidence from within vehicle line variation in test

ratings. Accident Analysis & Prevention, 39(5), 902-913.

Hatakka, M., Keskinen, E., Gregersen, N. P., Glad, A., & Hernetkoski, K. (2002). From

control of the vehicle to personal self-control; broadening the perspectives to driver

education. Transportation Research Part F: Traffic Psychology and Behaviour, 5(3),

201-215.

Hauer, E. (1988). Case for Science-Based Road Safety Design and Management, in

Highway Safety: At the Crossroads, Ed: R.E. Stammer, 241-267, American Society

of Civil Engineers, Reston, VA.

Hayward, J. C. (1972). Near miss determination through use of a scale of danger (traffic

records 384). Highway Research Board, Washington, DC

Health and Safety Commission. (2001). Reducing at-work road traffic incidents. UK,

London: Her Majesty’s Station.

Hingwe, P., Tan, H. S., Packard, A. K., & Tomizuka, M. (2002). Linear parameter varying

controller for automated lane guidance: experimental study on tractor-trailers. IEEE

Transactions on Control Systems Technology, 10(6), 793-806.

Hole, G., 2007. The Psychology of Driving. Lawrence Earlbaum Associates, Inc.,

Mahwah, New Jersey.

Hong Kong Planning Department (Ed.) (2016, May). Hong Kong Planning Standards

and Guidelines: Chapter 8 Internal Transport Facilities. Retrieved March 25, 2019,

from https://www.pland.gov.hk/pland_en/tech_doc/hkpsg/full/pdf/ch8.pdf

Hong Kong Police Force (2018, June). Traffic Annual Report. Retrieved June 27, 2018,

from https://www.police.gov.hk/info/doc/statistics/traffic_report_2015_en.pdf

Hong Kong Trade Development Council (2019) Title: Logistics Industry in Hong Kong.

162

Hong Kong, Retrieved May 13, 2020, from

https://hkmb.hktdc.com/en/1X0018WG/hktdc-research/Logistics-Industry-in-

Hong-Kong

Horne, J., & Reyner, L. (1999). Vehicle accidents related to sleep: a review. Occupational

and Environmental Medicine, 56(5), 289-294.

Hössinger, R., & Berger, W. J. (2012). Stated response to increased enforcement density

and penalty size for speeding and driving unbelted. Accident Analysis &

Prevention, 49, 501-511.

Høye, A. (2014). Speed cameras, section control, and kangaroo jumps–a meta-

analysis. Accident Analysis & Prevention, 73, 200-208.

Hu, L., Wu, X., Huang, J., Peng, Y., & Liu, W. (2020). Investigation of clusters and

injuries in pedestrian crashes using GIS in Changsha, China. Safety science, 127,

104710.

Huang, H., & Chin, H. C. (2009). Disaggregate propensity study on red light running

crashes using quasi-induced exposure method. Journal of Transportation

Engineering, 135(3), 104-111.

Huang, H., Siddiqui, C., & Abdel-Aty, M. (2011). Indexing crash worthiness and crash

aggressivity by vehicle type. Accident Analysis & Prevention, 43(4), 1364-1370.

Huang, H., Song, B., Xu, P., Zeng, Q., Lee, J., & Abdel-Aty, M. (2016). Macro and micro

models for zonal crash prediction with application in hot zones

identification. Journal of transport geography, 54, 248-256.

Huo, X., Leng, J., Hou, Q., Zheng, L., & Zhao, L. (2020). Assessing the explanatory and

predictive performance of a random parameters count model with heterogeneity in

means and variances. Accident Analysis & Prevention, 147, 105759.

Hydén, C. (1987). The development of a method for traffic safety evaluation: The

Swedish Traffic Conflicts Technique. Bulletin Lund Institute of Technology,

Department, (70).

Imine, H., Madani, T., & Fridman, L. (2011, December). Sliding-mode controller for

heavy vehicle lane departure avoidance. In 2011 9th IEEE International Conference

on Control and Automation (ICCA) (pp. 372-377). IEEE.

Islam, S., & Mannering, F. (2006). Driver aging and its effect on male and female single-

vehicle accident injuries: some additional evidence. Journal of Safety

Research, 37(3), 267-276.

Jacobs, B. A. (2010). Deterrence and deterrability. Criminology, 48(2), 417-441.

163

Jamson, A. H., & Merat, N. (2005). Surrogate in-vehicle information systems and driver

behaviour: Effects of visual and cognitive load in simulated rural

driving. Transportation Research Part F: Traffic Psychology and Behaviour, 8(2),

79-96.

Keall, M. D., Povey, L. J., & Frith, W. J. (2002). Further results from a trial comparing a

hidden speed camera programme with visible camera operation. Accident Analysis &

Prevention, 34(6), 773-777.

Kergoat, M., Delhomme, P., & Meyer, T. (2017). Appraisal of speed-enforcement

warning messages among young drivers: Influence of automatic versus human speed

enforcement in a known or unknown location. Transportation Research Part F:

Traffic Psychology and Behaviour, 46, pp. 177-194.

Kim, K., Brunner, I. M., & Yamashita, E. Y. (2006). Influence of land use, population,

employment, and economic activity on accidents. Transportation Research

Record, 1953(1), 56-64.

Kim, M., Kho, S. Y., & Kim, D. K. (2017). Hierarchical ordered model for injury severity

of pedestrian crashes in South Korea. Journal of Safety Research, 61, 33-40.

Knipling, R. R., P. F. Waller, R. C. Peck, R. Pfefer, T. R. Neuman, K. L. Slack, and K.

K. Hardy. (2004). NCHRP Report 500: Guidance for Implementation of the

AASHTO Strategic Highway Safety Plan; Volume 13: A Guide for Addressing Heavy

Truck Crashes. Transportation Research Board of the National Academies,

Washington, D.C.

Konduri, S., Labi, S., & Sinha, K.C. (2003). Incident Occurrence Models for Freeway

Incident Management. Transportation Research Record 1856(1), 125-135.

Kontogiannis, T. (2006). Patterns of driver stress and coping strategies in a Greek sample

and their relationship to aberrant behaviors and traffic accidents. Accident Analysis

& Prevention, 38(5), 913-924.

Labi, S. (2011). Efficacies of roadway safety improvements across functional subclasses

of rural two-lane highways. Journal of Safety Research, 42(4), 231-239.

Labi, S., Chen, S., Preckel, P. V., Qiao, Y., & Woldemariam, W. (2017). Rural two-lane

highway shoulder and lane width policy evaluation using multiobjective

optimization. Transportmetrica A: Transport Science, 13(7), 631-656.

Lamptey, G. (2004). Development of a Safety Management Systems for Indiana, MS

Thesis, Purdue University, W. Lafayette, Indiana, USA.

164

Langford, J., Methorst, R., & Hakamies-Blomqvist, L. (2006). Older drivers do not have

a high crash risk—A replication of low mileage bias. Accident Analysis &

Prevention, 38(3), 574-578.

Langlais, E. (2008). Detection avoidance and deterrence: some paradoxical

arithmetic. Journal of Public Economic Theory, 10(3), 371-382.

Lavieri, P. S., & Bhat, C. R. (2019). Modeling Individuals’ Willingness to Share Trips

with Strangers in an Autonomous Vehicle Future (No. 19-01775).

Lawpoolsri, S., Li, J., & Braver, E. R. (2007). Do speeding tickets reduce the likelihood

of receiving subsequent speeding tickets? A longitudinal study of speeding violators

in Maryland. Traffic Injury Prevention, 8(1), 26-34.

Lee, H. C., Cameron, D., & Lee, A. H. (2003). Assessing the driving performance of older

adult drivers: on-road versus simulated driving. Accident Analysis & Prevention,

35(5), 797-803.

Lee, J., Abdel-Aty, M., & Jiang, X. (2015). Multivariate crash modeling for motor vehicle

and non-motorized modes at the macroscopic level. Accident Analysis &

Prevention, 78, 146-154.

Lee, R. D., & Carter, L. R. (1992). Modeling and forecasting US mortality. Journal of the

American Statistical Association, 87(419), 659-671.

Lee, Y.C. (2018) Average age of occupational drivers reached sixty: Road accident and

aging population. Hong Kong 01 (28 January 2018, Chinese only). www.hk01.com

Last accessed 3 February 2020.

Legislative Council of Hong Kong Special Administrative Region. (2018, June).

Measures managing long working hours and occupational fatigue of bus drivers

(ISE07/17-18). Retrieved July 20, 2019, from https://www.legco.gov.hk/research-

publications/english/essentials-1718ise07-measures-managing-long-working-hours-

and-occupational-fatigue-of-bus-drivers.htm

Li, M. K., Yu, J. J., Ma, L., & Zhang, W. (2019). Modeling and mitigating fatigue-related

accident risk of taxi drivers. Accident Analysis & Prevention, 123, 79-87.

Li, X., Li, X., Xiao, Y., & Jia, B. (2016). Modeling mechanical restriction differences

between car and heavy truck in two-lane cellular automata traffic flow model.

Physica A: Statistical Mechanics and its Applications, 451, 49-62.

165

Li, X., Yan, X., & Wong, S. C. (2015). Effects of fog, driver experience and gender on

driving behavior on S-curved road segments. Accident Analysis & Prevention, 77,

91-104.

Li, Y. C., Sze, N. N., Wong, S. C., Yan, W., Tsui, K. L., & So, F. L. (2016). A simulation

study of the effects of alcohol on driving performance in a Chinese

population. Accident Analysis & Prevention, 95, 334-342.

Li, Y. C., Sze, N.N., & Wong, S.C. (2014). Effective measures for combating drink-

driving offenses: an attitudinal model for Hong Kong. Transportmetrica A: Transport

Science 10.8 (2014): 722-739.

Li, Z., Wang, W., Liu, P., Bai, L., & Du, M. (2015). Analysis of crash risks by collision

type at freeway diverge area using multivariate modeling technique. Journal of

Transportation Engineering, 141(6), 04015002.

Lord, D., & Mannering, F. (2010). The statistical analysis of crash-frequency data: a

review and assessment of methodological alternatives. Transportation research part

A: policy and practice, 44(5), 291-305.

Lord, D., Washington, S. P., & Ivan, J. N. (2005). Poisson, Poisson-gamma and zero-

inflated regression models of motor vehicle crashes: balancing statistical fit and

theory. Accident Analysis & Prevention, 37(1), 35-46.

Lord, D., Washington, S., & Ivan, J. N. (2007). Further notes on the application of zero-

inflated models in highway safety. Accident Analysis & Prevention, 39(1), 53-57.

Lu, G., Liu, M., Wang, Y., & Yu, G. (2012). Quantifying the severity of traffic conflict

by assuming moving elements as rectangles at intersection. Procedia-Social and

Behavioral Sciences, 43, 255-264.

Lundberg, C., Hakamies-Blomqvist, L., Almkvist, O., & Johansson, K. (1998).

Impairments of some cognitive functions are common in crash-involved older

drivers. Accident Analysis & Prevention, 30(3), 371-377.

Lyman, J. M., McGwin Jr, G., & Sims, R. V. (2001). Factors related to driving difficulty

and habits in older drivers. Accident Analysis & Prevention, 33(3), 413-421.

Machin, M. A., & De Souza, J. M. (2004). Predicting health outcomes and safety

behaviour in taxi drivers. Transportation Research Part F: Traffic Psychology and

Behaviour, 7(4-5), 257-270.

Maddala, G. S. (1986). Limited-dependent and qualitative variables in

econometrics (No.3). Cambridge University Press.

166

Makinen T.E. (2001) Halo effect of automatic speed enforcement. Techn. Res. Centre

Finland, pp. 341-345

Mallia, L., Lazuras, L., Violani, C., & Lucidi, F. (2015). Crash risk and aberrant driving

behaviors among bus drivers: the role of personality and attitudes towards traffic

safety. Accident Analysis & Prevention, 79, 145-151.

Mannering, F. (2018). Temporal instability and the analysis of highway accident

data. Analytic Methods in Accident Research, 17, 1-13.

Mannering, F. L., & Bhat, C. R. (2014). Analytic methods in accident research:

Methodological frontier and future directions. Analytic Methods in Accident

Research, 1, 1-22.

Mannering, F. L., Shankar, V., & Bhat, C. R. (2016). Unobserved heterogeneity and the

statistical analysis of highway accident data. Analytic Methods in Accident

Research, 11, 1-16.

Marciano, H., & Norman, J. (2015). Overt vs. covert speed cameras in combination with

delayed vs. immediate feedback to the offender. Accident Analysis & Prevention, 79,

231-240.

Martchouk, M., Mannering, F., & Bullock, D. (2010). Analysis of freeway travel time

variability using Bluetooth detection. Journal of Transportation

Engineering, 137(10), 697-704.

Martin, A. (2006). Factors influencing pedestrian safety: a literature review (No. PPR241).

Wokingham, Berks: TRL Limited.

Martin, J. L. (2002). Relationship between crash rate and hourly traffic flow on interurban

motorways. Accident Analysis & Prevention, 34(5), 619-629.

Matthews, G., Tsuda, A., Xin, G. U., & Ozeki, Y. (1999). Individual differences in driver

stress vulnerability in a Japanese sample. Ergonomics, 42(3), 401-415.

Maycock, G. (1996). Sleepiness and driving: the experience of UK car drivers. Journal

of Sleep Research, 5(4), 229-231.

McFadden, D. (1978). Modeling the choice of residential location. Transportation

Research Record, (673).

McFadden, D., & Train, K. (2000). Mixed MNL models for discrete response. Journal of

Applied Econometrics, 15(5), 447-470.

Mehdizadeh, M., Shariat-Mohaymany, A., & Nordfjaern, T. (2018). Accident

involvement among Iranian lorry drivers: Direct and indirect effects of background

167

variables and aberrant driving behaviour. Transportation Research Part F: Traffic

Psychology and Behaviour, 58, 39-55.

Meng, F., Li, S., Cao, L., Li, M., Peng, Q., Wang, C., & Zhang, W. (2015). Driving fatigue

in professional drivers: a survey of truck and taxi drivers. Traffic Injury

Prevention, 16(5), 474-483.

Meng, F., Wong, S. C., Yan, W., Li, Y. C., & Yang, L. (2019). Temporal patterns of driving

fatigue and driving performance among male taxi drivers in Hong Kong: A driving

simulator approach. Accident Analysis & Prevention, 125, 7-13.

Meng, F., Xu, P., Wong, S. C., Huang, H., & Li, Y. C. (2017). Occupant-level injury

severity analyses for taxis in Hong Kong: A Bayesian space-time logistic model.

Accident Analysis & Prevention, 108, 297-307.

Merat, N., Anttila, V., & Luoma, J. (2005). Comparing the driving performance of

average and older drivers: The effect of surrogate in-vehicle information systems.

Transportation Research Part F: Traffic Psychology and Behaviour, 8(2), 147-166.

Minderhoud, M. M., & Bovy, P. H. (2001). Extended time-to-collision measures for road

traffic safety assessment. Accident Analysis & Prevention, 33(1), 89-97.

Mohamed, M. G., Saunier, N., Miranda-Moreno, L. F., & Ukkusuri, S. V. (2013). A

clustering regression approach: A comprehensive injury severity analysis of

pedestrian–vehicle crashes in New York, US and Montreal, Canada. Safety

Science, 54, 27-37.

Mollenhauer, M. A., Lee, J., Cho, K., Hulse, M. C., & Dingus, T. A. (1994, October).

The effects of sensory modality and information priority on in-vehicle signing and

information systems. In Proceedings of the Human Factors and Ergonomics

Society Annual Meeting (Vol. 38, No. 16, pp. 1072-1076). Sage CA: Los Angeles,

CA: SAGE Publications.

Molnar, L. J., & Eby, D. W. (2008). The relationship between self-regulation and driving-

related abilities in older drivers: an exploratory study. Traffic Injury Prevention, 9(4),

314-319.

Mooren, L., Grzebieta, R., Williamson, A., Olivier, J., & Friswell, R. (2014). Safety

management for heavy vehicle transport: A review of the literature. Safety

Science, 62, 79-89.

Navarro, J. E. J. B., Sanz, J. L. G., Del Castillo, J. M., Izquierdo, A. C., & Rodríguez, M.

M. (2007). Motivational factors and physician advice for physical activity in older

168

urban adults. Journal of Aging and Physical Activity, 15(3), 241-256.

Netto, M. S., Chaib, S., & Mammar, S. (2004, June). Lateral adaptive control for vehicle

lane keeping. In Proceedings of the 2004 American Control Conference (Vol. 3, pp.

2693-2698). IEEE.

Newnam, S., Blower, D., Molnar, L., Eby, D., & Koppel, S. (2018). Exploring crash

characteristics and injury outcomes among older truck drivers: an analysis of truck-

involved crash data in the United States. Safety Science, 106, 140-145.

Newnam, S., Koppel, S., Molnar, L. J., Zakrajsek, J. S., Eby, D. W., & Blower, D. (2019).

Older truck drivers: How can we keep them in the workforce for as long as safely

possible?. Safety Science, in press.

Newnam, S., Watson, B., & Murray, W. (2004). Factors predicting intentions to speed in

a work and personal vehicle. Transportation Research Part F: Traffic Psychology and

Behaviour, 7(4-5), 287-300.

Neyens, D. M., & Boyle, L. N. (2007). The effect of distractions on the crash types of

teenage drivers. Accident Analysis & Prevention, 39(1), 206-212.

Ng, A. W., & Chan, A. H. (2008). The effects of driver factors and sign design features

on the comprehensibility of traffic signs. Journal of Safety Research, 39(3), 321-328.

Nguyen-Phuoc, D. Q., De Gruyter, C., Nguyen, H. A., Nguyen, T., & Su, D. N. (2020).

Risky behaviours associated with traffic crashes among app-based motorcycle taxi

drivers in Vietnam. Transportation Research Part F: Traffic Psychology and

Behaviour, 70, 249-259.

Ni, R., Kang, J. J., & Andersen, G. J. (2010). Age-related declines in car following

performance under simulated fog conditions. Accident Analysis & Prevention, 42(3),

818-826.

Noland, R. B., & Quddus, M. A. (2004). A spatially disaggregate analysis of road

casualties in England. Accident Analysis & Prevention, 36(6), 973-984.

Oron-Gilad, T., & Ronen, A. (2007). Road characteristics and driver fatigue: a simulator

study. Traffic Injury Prevention, 8(3), 281-289.

Oron-Gilad, T., Ronen, A., & Shinar, D. (2008). Alertness maintaining tasks (AMTs)

while driving. Accident Analysis & Prevention, 40(3), 851-860.

Otmani, S., Rogé, J., & Muzet, A. (2005). Sleepiness in professional drivers: effect of age

and time of day. Accident Analysis & Prevention, 37(5), 930-937.

169

Öz, B., Özkan, T., & Lajunen, T. (2010a). Professional and non-professional drivers’

stress reactions and risky driving. Transportation Research Part F: Traffic

Psychology and Behaviour, 13(1), 32-40.

Öz, B., Özkan, T., & Lajunen, T. (2010b). An investigation of the relationship between

organizational climate and professional drivers’ driver behaviours. Safety

Science, 48(10), 1484-1489.

Öz, B., Özkan, T., & Lajunen, T. (2013). An investigation of professional drivers:

Organizational safety climate, driver behaviours and performance. Transportation

Research Part F: Traffic Psychology and Behaviour, 16, 81-91.

Paleti, R., Eluru, N., & Bhat, C. R. (2010). Examining the influence of aggressive driving

behavior on driver injury severity in traffic crashes. Accident Analysis & Prevention,

42(6), 1839-1854.

Park, E. S., Carlson, P. J., Porter, R. J., & Andersen, C. K. (2012). Safety effects of wider

edge lines on rural, two-lane highways. Accident Analysis & Prevention, 48, 317-

325.

Peer, E. (2010). Speeding and the time-saving bias: How drivers’ estimations of time

saved in higher speed affects their choice of speed. Accident Analysis &

Prevention, 42(6), 1978-1982.

Pei, X., Sze, N. N., Wong, S. C., & Yao, D. (2016). Bootstrap resampling approach to

disaggregate analysis of road crashes in Hong Kong. Accident Analysis &

Prevention, 95, 512-520.

Pei, X., Wong, S. C., & Sze, N. N. (2012). The roles of exposure and speed in road safety

analysis. Accident Analysis & Prevention, 48, 464-471.

Poulter, D. R., Chapman, P., Bibby, P. A., Clarke, D. D., & Crundall, D. (2008). An

application of the theory of planned behaviour to truck driving behaviour and

compliance with regulations. Accident Analysis & Prevention, 40(6), 2058-2064.

Qin, X., Ivan, J. N., & Ravishanker, N. (2004). Selecting exposure measures in crash rate

prediction for two-lane highway segments. Accident Analysis & Prevention, 36(2),

183-191.

Raihan, M. A., Alluri, P., Wu, W., & Gan, A. (2019). Estimation of bicycle crash

modification factors (CMFs) on urban facilities using zero inflated negative

binomial models. Accident Analysis & Prevention, 123, 303-313.

170

Ram, T., & Chand, K. (2016). Effect of drivers’ risk perception and perception of driving

tasks on road safety attitude. Transportation Research Part F: Traffic Psychology and

Behaviour, 42, 162-176.

Retting, R. A., Farmer, C. M., & McCartt, A. T. (2008). Evaluation of automated speed

enforcement in Montgomery County, Maryland. Traffic Injury Prevention, 9(5), 440-

445.

Retting, R. A., Ulmer, R. G., & Williams, A. F. (1999). Prevalence and characteristics of

red light running crashes in the United States. Accident Analysis & Prevention, 31(6),

687-694.

Ritchey, M., & Nicholson‐Crotty, S. (2011). Deterrence theory and the implementation

of speed limits in the American states. Policy Studies Journal, 39(2), 329-346.

Rogers, R. W. (1983). Cognitive and psychological processes in fear appeals and attitude

change: A revised theory of protection motivation. Social Psychophysiology: A

sourcebook, 153-176.

Rohani, M. M., Wijeyesekera, D. C., & Karim, A. T. A. (2013). Bus operation, quality

service and the role of bus provider and driver. Procedia Engineering, 53, 167-178.

Roncek, D. W. (1992). Learning more from tobit coefficients: Extending a comparative

analysis of political protest. American Sociological Review, 503-507.

Rosenbloom, T. (2003). Risk evaluation and risky behavior of high and low sensation

seekers. Social Behavior and Personality: An International Journal, 31(4), 375-386.

Rosenbloom, T. (2011). Professional drivers. In Handbook of traffic psychology (pp. 389-

399). Academic Press.

Rosenbloom, T., & Shahar, A. (2007). Differences between taxi and nonprofessional

male drivers in attitudes towards traffic-violation penalties. Transportation Research

Part F: Traffic Psychology and Behaviour, 10(5), 428-435.

Rosenbloom, T., Shahar, A., Elharar, A., & Danino, O. (2008). Risk perception of driving

as a function of advanced training aimed at recognizing and handling risks in

demanding driving situations. Accident Analysis & Prevention, 40(2), 697-703.

Ryeng, E. O. (2012). The effect of sanctions and police enforcement on drivers’ choice

of speed. Accident Analysis & Prevention, 45, 446-454.

Sagberg, F., & Ingebrigtsen, R. (2018). Effects of a penalty point system on traffic

violations. Accident Analysis & Prevention, 110, 71-77.

171

Saifizul, A. A., Yamanaka, H., & Karim, M. R. (2011). Empirical analysis of gross vehicle

weight and free flow speed and consideration on its relation with differential speed

limit. Accident Analysis & Prevention, 43(3), 1068-1073.

Savolainen, P., & Mannering, F. (2007). Probabilistic models of motorcyclists’ injury

severities in single-and multi-vehicle crashes. Accident Analysis & Prevention, 39(5),

955-963.

Sayed, T., Zaki, M. H., & Autey, J. (2013). Automated safety diagnosis of vehicle–bicycle

interactions using computer vision analysis. Safety Science, 59, 163-172.

Shackel, S. C., & Parkin, J. (2014). Influence of road markings, lane widths and driver

behaviour on proximity and speed of vehicles overtaking cyclists. Accident Analysis

& Prevention, 73, 100-108.

Shaheed, M. S., & Gkritza, K. (2014). A latent class analysis of single-vehicle motorcycle

crash severity outcomes. Analytic Methods in Accident Research, 2, 30-38.

Shams, M., Shojaeizadeh, D., Majdzadeh, R., Rashidian, A., & Montazeri, A. (2011).

Taxi drivers’ views on risky driving behavior in Tehran: a qualitative study using a

social marketing approach. Accident Analysis & Prevention, 43(3), 646-651.

Shankar, V. N., Ulfarsson, G. F., Pendyala, R. M., & Nebergall, M. B. (2003). Modeling

crashes involving pedestrians and motorized traffic. Safety Science, 41(7), 627-640.

Shanmugaratnam, S., Kass, S. J., & Arruda, J. E. (2010). Age differences in cognitive and

psychomotor abilities and simulated driving. Accident Analysis & Prevention, 42(3),

802-808.

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 & Prevention, 33(1), 111-116.

Shinar, D., Tractinsky, N., & Compton, R. (2005). Effects of practice, age, and task

demands, on interference from a phone task while driving. Accident Analysis &

Prevention, 37(2), 315-326.

Silcock, D., Smith, K., Knox, D., & Beuret, K. (2000). What limits speed. Factors That

Affect How Fast We Drive, Final Report: AA Foundation for Road Safety Research.

Spiegelhalter, D. J., Best, N. G., Carlin, B. P., & Van Der Linde, A. (2002). Bayesian

measures of model complexity and fit. Journal of the Royal Statistical Society:

Series b (Statistical Methodology), 64(4), 583-639.

Spiegelhalter, D., Thomas, A., Best, N., Lunn, D., (2005). WinBUGS User Manual. MRC

172

Biostatistics Unit, Cambridge, United Kingdom.

Srinivas, C., Dinu, R. R., & Veeraragavan, A. (2007). Application of Poisson and negative

binomial regression for modeling road accidents under mixed traffic conditions.

In Proceedings of the 86th Annual Meeting of the Transportation Research Board,

Washington, DC.

Sullman, M. J., Meadows, M. L., & Pajo, K. B. (2002). Aberrant driving behaviours

amongst New Zealand truck drivers. Transportation Research Part F: Traffic

Psychology and Behaviour, 5(3), 217-232.

Summala, H., Lamble, D., & Laakso, M. (1998). Driving experience and perception of

the lead car's braking when looking at in-car targets. Accident Analysis &

Prevention, 30(4), 401-407.

Sze, N. N., & Wong, S. C. (2007). Diagnostic analysis of the logistic model for pedestrian

injury severity in traffic crashes. Accident Analysis & Prevention, 39(6), 1267-1278.

Sze, N.N., & Christensen, K.M. (2017) Access to urban transportation system for

individuals with disabilities. International Association of Traffic and Safety Sciences

(IATSS) Research, 41(2), 66-73.

Sze, N.N., Su, J., & Bai, L. (2019) Exposure to pedestrian crash based on household

survey data: Effect of trip purpose. Accident Analysis and Prevention 128, 17-24.

Szeto, W. Y., Wong, R. C. P., Wong, S. C., & Yang, H. (2013). A time-dependent logit-

based taxi customer-search model. International Journal of Urban Sciences, 17(2),

184-198.

Tang, Z., Chen, S., Cheng, J., Ghahari, S. A., & Labi, S. (2018). Highway design and

safety consequences: A case study of interstate highway vertical grades. Journal of

Advanced Transportation, 2018. Article ID 1492614

Tarko, A. P. (2009). Modeling drivers’ speed selection as a trade-off behavior. Accident

Analysis & Prevention, 41(3), 608-616.

Tarko, A. P. (2012). Use of crash surrogates and exceedance statistics to estimate road

safety. Accident Analysis & Prevention, 45, 230-240.

Tay, R. (2003). Marginal effects of changing the vehicle mix on fatal crashes. Journal of

Transport Economics and Policy (JTEP), 37(3), 439-450.

Tay, R. (2005a). Drink driving enforcement and publicity campaigns: are the policy

recommendations sensitive to model specification?. Accident Analysis &

Prevention, 37(2), 259-266.

173

Tay, R. (2005b). The effectiveness of enforcement and publicity campaigns on serious

crashes involving young male drivers: Are drink driving and speeding

similar?. Accident Analysis & Prevention, 37(5), 922-929.

Tay, R. (2005c). General and specific deterrent effects of traffic enforcement: do we have

to catch offenders to reduce crashes?. Journal of Transport Economics and Policy

(JTEP), 39(2), 209-224.

Tay, R. (2006). Ageing drivers: Storm in a teacup?. Accident Analysis & Prevention, 38(1),

112-121.

Tay, R. (2008). Marginal effect of increasing ageing drivers on injury crashes. Accident

Analysis & Prevention, 40(6), 2065-2068.

Tay, R. (2009). The effectiveness of automated and manned traffic

enforcement. International Journal of Sustainable Transportation, 3(3), 178-186.

Tay, R., & Choi, J. (2016). Factors associated with crashes involving taxi owners and non-

owners: A case of moral hazard and adverse selection?. Accident Analysis &

Prevention, 87, 78-82.

Teh, E., Jamson, S., Carsten, O., & Jamson, H. (2014). Temporal fluctuations in driving

demand: The effect of traffic complexity on subjective measures of workload and

driving performance. Transportation Research Part F: Traffic Psychology and

Behaviour, 22, 207-217.

Thiffault, P., & Bergeron, J. (2003). Monotony of road environment and driver fatigue: a

simulator study. Accident Analysis & Prevention, 35(3), 381-391.

Ting, P. H., Hwang, J. R., Doong, J. L., & Jeng, M. C. (2008). Driver fatigue and highway

driving: A simulator study. Transportation Research Part F: Traffic Psychology &

Behavior, 94(3), 448-453.

Tobin, J. (1958). Estimation of relationships for limited dependent

variables. Econometrica: journal of the Econometric Society, 24-36.

Transport and Housing Bureau (2017) Title: Public Transport Strategy Study. Hong

Kong. http://www.td.gov.hk/filemanager/en/publication/ptss_final_report_eng.pdf

Transport Department of Hong Kong Special Administrative Region (2014). Travel

Characteristics Survey 2011 - Final Report. Hong Kong, p.14. Retrieved July 20,

2019, from https://www.td.gov.hk/filemanager/en/content_4652/tcs2011_eng.pdf

   
174 
Transport Department of Hong Kong Special Administrative Region. (2017a, January). 
Road  Traffic  Accident  Statistics.  Retrieved  June  27,  2018,  from 
http://www.td.gov.hk/tc/road_safety/road_traffic_accident_statistics/index.html 
Transport Department of Hong Kong Special Administrative Region. (2017b, January). 
Red Light Cameras and Speed Enforcement Cameras. Retrieved September 10, 2020, 
from 
https://www.td.gov.hk/en/transport_in_hong_kong/its/its_achievements/red_light_
cameras_and_speed_enforcement_cameras/index.html 
Transport Department of Hong Kong Special Administrative Region. (2018b, January). 
Road  Traffic  Accident  Statistics.  Retrieved  November  16,  2019,  from 
http://www.td.gov.hk/tc/road_safety/road_traffic_accident_statistics/index.html 
Transport Department of Hong Kong Special Administrative Region. (2018a, February). 
Driving-offence  points  system.  Retrieved  June  27,  2018,  from 
http://www.td.gov.hk/en/road_safety/safe_motoring_guides/driving_offence_points
_system/index.html 
Transport Department of Hong Kong Special Administrative Region. (2019, January). 
Road  Traffic  Accident  Statistics.  Retrieved  September  10,  2020,  from 
http://www.td.gov.hk/tc/road_safety/road_traffic_accident_statistics/index.html 
Transport  Department  of  Hong  Kong  Special  Administrative  Region.  (2020a,  May). 
Commercial  Vehicles.  Chapter  374A,  construction  and  maintenance of  vehicles, 
Retrieved  September  1,  2020,  from 
https://www.td.gov.hk/tc/road_safety/road_users_code/index/chapter_6_for_profes
sional_drivers/commercial_vehicles_/index.html 
Transport Department of Hong Kong Special Administrative Region. (2020b, February). 
Area  Traffic  Control  Systems,  Retrieved  October  15,  2020,  from 
https://www.td.gov.hk/en/transport_in_hong_kong/its/its_achievements/area_traffi
c_control_systems/index.html 
Trick, L. M., Toxopeus, R., & Wilson, D. (2010). The effects of visibility conditions, 
traffic  density,  and  navigational  challenge  on  speed  compensation  and  driving 
performance in older adults. Accident Analysis & Prevention, 42(6), 1661-1671. 
Tronsmoen,  T.  (2010).  Associations  between  driver  training,  determinants  of  risky 
driving behaviour and crash involvement. Safety Science, 48(1), 35-45. 

175

Truelove, V., Freeman, J., Szogi, E., Kaye, S., Davey, J., & Armstrong, K. (2017).

Beyond the threat of legal sanctions: What deters speeding

behaviours?. Transportation Research Part F: Traffic Psychology and Behaviour, 50,

128-136.

Tseng, C. M. (2013). Speeding violations related to a driver’s social-economic

demographics and the most frequent driving purpose in Taiwan’s male

population. Safety Science, 57, 236-242.

Tsui, K. L., So, F. L., Sze, N. N., Wong, S. C., & Leung, T. F. (2009). Misclassification

of injury severity among road casualties in police reports. Accident Analysis &

Prevention, 41(1), 84-89.

Tulu, G. S., Washington, S., Haque, M. M., & King, M. J. (2015). Investigation of

pedestrian crashes on two-way two-lane rural roads in Ethiopia. Accident Analysis

& Prevention, 78, 118-126.

Tulu, G. S., Washington, S., King, M. J., & Haque, M. (2013). Why are pedestrian crashes

so different in developing countries? A review of relevant factors in relation to their

impact in Ethiopia. In 36th Australasian Transport Research Forum (ATRF):

Transport and the New World City, 2-4 October 2013, QUT Gardens Point, Brisbane,

Australia.

UK Sentencing Council (2017, April). Road Traffic Regulation Act 1984, s.89(1).

Retrieved March 6, 2019, from

https://www.sentencingcouncil.org.uk/offences/magistrates-court/item/speeding-

revised-2017/

Ukkusuri, S., Miranda-Moreno, L. F., Ramadurai, G., & Isa-Tavarez, J. (2012). The role

of built environment on pedestrian crash frequency. Safety Science, 50(4), 1141-

1151.

Ulak, M. B., Ozguven, E. E., Vanli, O. A., Dulebenets, M. A., & Spainhour, L. (2018).

Multivariate random parameter Tobit modeling of crashes involving aging drivers,

passengers, bicyclists, and pedestrians: Spatiotemporal variations. Accident Analysis

& Prevention, 121, 1-13.

Ulfarsson, G. F., & Mannering, F. L. (2004). Differences in male and female injury

severities in sport-utility vehicle, minivan, pickup and passenger car

accidents. Accident Analysis & Prevention, 36(2), 135-147.

176

United States Department of labor, (2018, May). Quick Facts: Taxi Drivers, Ride-Hailing

Drivers, and Chauffeurs. Retrieved March 6, 2019, from

https://www.bls.gov/ooh/transportation-and-material-moving/taxi-drivers-and-

chauffeurs.htm

Van der Hulst, M., Meijman, T., & Rothengatter, T. (2001). Maintaining task set under

fatigue: a study of time-on-task effects in simulated driving. Transportation

Research Part F: Traffic Psychology and Behaviour, 4(2), 103-118.

Vogel, K. (2003). A comparison of headway and time to collision as safety

indicators. Accident Analysis & Prevention, 35(3), 427-433.

Washington, S., Karlaftis, M. G., Mannering, F., & Anastasopoulos, P. (2020). Statistical

and Econometric Methods for Transportation Data Analysis. CRC press.

Walter, L., Broughton, J., & Knowles, J. (2011). The effects of increased police

enforcement along a route in London. Accident Analysis & Prevention, 43(3), 1219-

1227.

Wang, K., Ivan, J. N., Ravishanker, N., & Jackson, E. (2017). Multivariate poisson

lognormal modeling of crashes by type and severity on rural two lane

highways. Accident Analysis & Prevention, 99, 6-19.

Wang, Q., Fan, Z., Gui, L., Chen, Z., & Song, H. (2004). Crashworthiness analysis of a

minibus body in white through reverse engineering. Tsinghua Science and

Technology, 9(3), 338-344.

Wang, Y., Li, L., & Prato, C.G. (2019a). The relation between working conditions,

aberrant driving behaviour and crash propensity among taxi drivers in

China. Accident Analysis & Prevention, 126, 17-24.

Wang, Y., Zhang, Y., Li, L., & Liang, G. (2019b). Self-reports of workloads and aberrant

driving behaviors as predictors of crash rate among taxi drivers: A cross-sectional

study in China. Traffic Injury Prevention, 20(7), 738-743.

Watson, B., Watson, A., Siskind, V., Fleiter, J., & Soole, D. (2015). Profiling high-range

speeding offenders: Investigating criminal history, personal characteristics, traffic

offences, and crash history. Accident Analysis & Prevention, 74, 87-96.

Wen, H., Sun, J., Zeng, Q., Zhang, X., & Yuan, Q. (2018). The effects of traffic

composition on freeway crash frequency by injury severity: A Bayesian multivariate

spatial modeling approach. Journal of Advanced Transportation, 2018.

177

Wen, H., Zhang, X., Zeng, Q., & Sze, N. N. (2019). Bayesian spatial-temporal model for

the main and interaction effects of roadway and weather characteristics on freeway

crash incidence. Accident Analysis & Prevention, 132, 105249.

Westerman, S. J., & Haigney, D. (2000). Individual differences in driver stress, error and

violation. Personality and Individual Differences, 29(5), 981-998.

WHO. (2018, February). Road traffic injuries. Retrieved June 27, 2018, from

http://www.who.int/mediacentre/factsheets/fs358/en/

Williamson, A. M., Feyer, A. -M., Friswell, R., & Finlay-Brown, S., (2002). Driver

Fatigue: A survey of long distance heavy vehicle drivers in Australia. Australian

Transport Safety Bureau report, CR198.

Williamson, A., & Boufous, S. (2007). A data-matching study of the role of fatigue in

work-related crashes. Transportation Research Part F: Traffic Psychology and

Behaviour, 10(3), 242-253.

Williamson, A., Friswell, R., Olivier, J., & Grzebieta, R. (2014). Are drivers aware of

sleepiness and increasing crash risk while driving?. Accident Analysis & Prevention,

70, 225-234.

Wong, S C., Wong, C. W., & Sze, N.N. (2008). Attitudes of public light bus drivers to

penalties to combat red light violations in Hong Kong. Transport Policy, 15(1), 43-

54.

Wong, S. C., Sze, N. N., & Li, Y. C. (2007). Contributory factors to traffic crashes at

signalized intersections in Hong Kong. Accident Analysis & Prevention, 39(6),

1107-1113.

Wu, H., Han, Z., Murphy, M. R., & Zhang, Z. (2015). Empirical Bayes before–after study

on safety effect of narrow pavement widening projects in Texas. Transportation

Research Record, 2515(1), 63-69.

Wu, J., Yan, X., & Radwan, E. (2016). Discrepancy analysis of driving performance of

taxi drivers and non-professional drivers for red-light running violation and crash

avoidance at intersections. Accident Analysis & Prevention, 91, 1-9.

Xie, Y., Zhao, K., & Huynh, N. (2012). Analysis of driver injury severity in rural single-

vehicle crashes. Accident Analysis & Prevention, 47, 36-44.

Xing, F., Huang, H., Zhan, Z.Y., Zhai, X., Ou, C., Sze, N.N., & Hon, K.K. (2019) Hourly

associations between weather factors and traffic crashes: non-linear and lag effects.

Analytic Methods in Accident Research. 24, 100109.

178

Xu, C. C., Liu, P., Wang, W., Jiang, X., & Chen, Y. G. (2014). Effects of behavioral

characteristics of taxi drivers on safety and capacity of signalized

intersections. Journal of Central South University, 21(10), 4033-4042.

Xu, X., Wong, S.C., & Choi, K. (2014). A two-stage bivariate logistic-Tobit model for

the safety analysis of signalized intersections. Analytic Methods in Accident

Research, 3, 1-10.

Yan, X., & Radwan, E. (2006). Analyses of rear-end crashes based on classification tree

models. Traffic Injury Prevention, 7(3), 276-282.

Yan, X., Abdel-Aty, M., Radwan, E., Wang, X., & Chilakapati, P. (2008). Validating a

driving simulator using surrogate safety measures. Accident Analysis & Prevention,

40(1), 274-288.

Yan, X., Radwan, E., & Abdel-Aty, M. (2005). Characteristics of rear-end accidents at

signalized intersections using multiple logistic regression model. Accident Analysis

& Prevention, 37(6), 983-995.

Yau, K. K. (2004). Risk factors affecting the severity of single vehicle traffic accidents in

Hong Kong. Accident Analysis & Prevention, 36(3), 333-340.

Zegeer, C.V., Reinfurt, D.W., Hummer, J., Herf, L., Hunter W. (1988). Effects of Cross-

section design for two-lane roads, Transportation Research Record 1195, 20-32.

Zeng, Q., & Huang, H. (2014). Bayesian spatial joint modeling of traffic crashes on an

urban road network. Accident Analysis & Prevention, 67, 105-112.

Zeng, Q., Guo, Q., Wong, S. C., Wen, H., Huang, H., & Pei, X. (2019). Jointly modeling

area-level crash rates by severity: a Bayesian multivariate random-parameters

spatio-temporal Tobit regression. Transportmetrica A: Transport Science, 15(2),

1867-1884.

Zeng, Q., Huang, H., Pei, X., & Wong, S. C. (2016). Modeling nonlinear relationship

between crash frequency by severity and contributing factors by neural

networks. Analytic Methods in Accident Research, 10, 12-25.

Zeng, Q., Wen, H., Huang, H., & Abdel-Aty, M. (2017a). A Bayesian spatial random

parameters Tobit model for analyzing crash rates on roadway segments. Accident

Analysis & Prevention, 100, 37-43.

Zeng, Q., Wen, H., Huang, H., Pei, X., & Wong, S. C. (2017b). A multivariate random-

parameters Tobit model for analyzing highway crash rates by injury

severity. Accident Analysis & Prevention, 99, 184-191.

179

Zeng, Q., Wen, H., Huang, H., Pei, X., & Wong, S. C. (2018). Incorporating temporal

correlation into a multivariate random parameters Tobit model for modeling crash

rate by injury severity. Transportmetrica A: Transport Science, 14(3), 177-191.

Zhai, X., Huang, H., Sze, N.N., Song, Z., & Hon, K.K. (2019). Diagnostic analysis of the

effects of weather condition on pedestrian crash severity. Accident Analysis &

Prevention, 122, 318-324.

Zhang, G., Yau, K.K., & Chen, G. (2013). Risk factors associated with traffic violations

and accident severity in China. Accident Analysis & Prevention, 59, 18-25.

Zhang, G., Yau, K.K., & Gong, X. (2014). Traffic violations in Guangdong Province of

China: speeding and drunk driving. Accident Analysis & Prevention, 64, 30-40.

Zhao, S., Wang, K., Liu, C., & Jackson, E. (2019). Investigating the effects of monthly

weather variations on Connecticut freeway crashes from 2011 to 2015. Journal of

Safety Research, 71, 153-162.

Zheng, L., Ismail, K., & Meng, X. (2014). Traffic conflict techniques for road safety

analysis: open questions and some insights. Canadian Journal of Civil

Engineering, 41(7), 633-641.

Zhou, T., & Zhang, J. (2019). Analysis of commercial truck drivers’ potentially

dangerous driving behaviors based on 11-month digital tachograph data and

multilevel modeling approach. Accident Analysis & Prevention, 132, 105256.

180

Appendix

Questionnaire survey for speeding violation

Section 1 State Preference

Comprehensive strategy has been adopted by the Hong Kong Government to tackle the

speeding problem and to improve road safety. The application of camera systems is one

of the major approaches for the speed enforcement since 1999. As of 2012, the transport

department has installed 120 camera housing units for speed enforcement, of which

only 20 have actual cameras within these units (that is, the camera-to-housing ratio

today is 20:120). However, in the future, this ratio may be changed to enhance the

effectiveness by increasing the number of cameras or the housings. Meanwhile,

adequate advance warning signs will be erected to alert drivers of the presence of this

speed enforcement camera systems. Also, the current penalties for speeding are as

given below.

Table 1 Current penalties for speeding in Hong Kong

Level

Speeding Offences in Excess of Speed Limit by

Monetary Fine

DOP

15 km/h or less

320

15 + km/h ~ 30 km/h

450

30 + km/h ~ 45 km/h

600

Attention：There are three types of locations on the roadway, as identified below.

Table 2 three types of locations

Location Type

Characterization

Plain Section

No enforcement and no warning signs in roadway

section

Warning Section

Warning sign indicating the presence of speed camera

housing unit X meters ahead

Camera housing

Section

Camera housing unit present in roadway section

Figure 1 Illustration of the road sections

With the above as background, we will present you with four scenarios, reflecting:

181

(a) different levels of penalty for speed violations;

(b) different camera-to-housing ratios;

Consider a roadway with a speed limit of 50 km/h.

Please consider each scenario carefully, and make a choice of the speed at which you

would travel for each of the three location types on the roadway.

If you are in location Type A (Plain section), at which speed range would you travel ?

(choose one option from below)

 < 50 km/h  51-65 km/h  66-80 km/h

If you are in location Type B (Warning section), at which speed range would you

travel ?

 < 50 km/h  51-65 km/h  66-80 km/h

If you are in location Type C (Camera housing section), at which speed range would

you travel ?

 < 50 km/h  51-65 km/h  66-80 km/h

If you are in location Type A (Plain section), at which speed range would you travel ?

(choose one option from below)

 < 50 km/h  51-65 km/h  66-80 km/h

If you are in location Type B (Warning section), at which speed range would you

travel ?

 < 50 km/h  51-65 km/h  66-80 km/h

Background information

Speed（km/h）

< 50

51 - 65

66 - 80

Penalty for speed

violation

➢ DOP

➢ Fine

320 HKD

450 HKD

Camera-to-Housing ratio

20 cameras in 240 housing units

Location of the warning sign in

section B

Warning Sign placed 50 meters ahead of

housing unit

Background information

Speed（km/h）

< 50

51 - 65

66 - 80

Penalty for speed

violation

➢ DOP

➢ Fine

320 HKD

450 HKD

Camera-to-Housing ratio

40 cameras in 120 housing units

Location of the warning sign in

section B

Warning Sign placed 200 meters ahead of

housing unit

Scenario ○

182

If you are in location Type C (Camera housing section), at which speed range would

you travel ?

 < 50 km/h  51-65 km/h  66-80 km/h

If you are in location Type A (Plain section), at which speed range would you travel ?

(choose one option from below)

 < 50 km/h  51-65 km/h  66-80 km/h

If you are in location Type B (Warning section), at which speed range would you

travel ?

 < 50 km/h  51-65 km/h  66-80 km/h

If you are in location Type C (Camera housing section), at which speed range would

you travel ?

 < 50 km/h  51-65 km/h  66-80 km/h

If you are in location Type A (Plain section), at which speed range would you travel ?

(choose one option from below)

 < 50 km/h  51-65 km/h  66-80 km/h

If you are in location Type B (Warning section), at which speed range would you

travel ?

 < 50 km/h  51-65 km/h  66-80 km/h

If you are in location Type C (Camera housing section), at which speed range would

you travel ?

 < 50 km/h  51-65 km/h  66-80 km/h

Background information

Speed（km/h）

< 50

51 - 65

66 - 80

Penalty for speed

violation

➢ DOP

➢ Fine

320 HKD

450 HKD

Camera-to-Housing ratio

20 cameras in 120 housing units

Location of the warning sign in

section B

Warning Sign placed 150 meters ahead of

housing unit

Background information

Speed（km/h）

< 50

51 - 65

66 - 80

Penalty for speed

violation

➢ DOP

➢ Fine

420 HKD

550 HKD

Camera-to-Housing ratio

40 cameras in 120 housing units

Location of the warning sign in

section B

Warning Sign placed 100 meters ahead of

housing unit

Scenario ○

183

Section 2 Driving history and perception

1. Have you been involved in any traffic accidents in the past 12 months?

 No

 Yes［involved in time(s)］

2. Have you received tickets for violating the speed limit in past 12 months?

 No ［please turn to question 3］

 Yes［How many tickets have you received in past 12 months?］

○ 1~3

○ 4~6

○ 7~9

○ > 10

3. In how many out of 10 trips have you noticed the orange housings?

○ 0

○ 1~3 time(s)

○ 4~6 times

○ 7~9 times

○ 10 times

4. I think speeding cameras are effective to catch offenders.

○1(Strongly disagree)

○2

○3

○4

○5(Strongly agree)

5. If you are speeding now, how likely will you decelerate when you see a:

i. Warning sign for alerting motorists of the presence of speed camera system

○1(Not at all)

○2

○3

○4

○5(Very much likely)

ii. Orange housing for speeding camera

○1(Not at all)

○2

○3

○4

○5(Very much likely)

6. Speeding violation is a very risky behavior that leads to property damage only.

○1(Strongly disagree)

○2

○3

○4

○5(Strongly agree)

7. Speeding violation is a very risky behavior that leads to injuries.

○1(Strongly disagree)

○2

○3

○4

○5(Strongly agree)

Section 3 Personal information

1. Gender:  Male  Female

2. Age: 18-25 26-35 36-45 46-55 56-65 > 66

3. Education: Primary Secondary Tertiary or above

4. Marital status:

 Single Married/Cohabiting Divorced/Separated Widowed

5. Personal monthly income:

 < 10,000 HKD

 10,000-14,999 HKD

 15,000-19,999 HKD

 20,000-24,999 HKD

 25,000-29,999 HKD

 > 30,000 HKD

6. How many years have you obtained your driving license? _______ year(s)

7. What is your current driving-offence points on driving license? (0-15

point(s))

8. How many hours do you usually drive per day? hour(s)

9. How many days do you usually drive per week？ day(s)

10. What is the type of your vehicle?

Taxi

Private Bus

Franchised Bus

Public Light Bus

Light Van

Medium/heavy Goods vehicle

Others

11. What is your current employment status?

 Self-employed  Permanent  Contract

12. What is your current salary system?

Trip based  Hourly  Shift based  Monthly