2007 Zheng PHD

The Impact of Localized Road Accident Information
on Road Safety Awareness
Yunan Zheng
A Thesis submitted for the degree of Doctor of Philosophy
TO
The University of Glasgow
April 2007
Abstract
The World Health Organization (WHO) estimatethat road traffic accidentsrepresentthe

third leading causeof 'death and disease' worldwide. Many countries have, therefore,
launchedsafetycampaignsthat are intendedto reduceroad traffic accidentsby increasing
public awareness.In almost every case, however, a reduction in the total number of
fatalities has not beenmatchedby a comparablefall in the total frequencyof road traffic
accidents. Low severity incidents remain a significant problem. One possibleexplanation
is that these road safety campaignshave had less effect than design changes.Active
safety devices, such as anti-lock braking, and passive measures,such as side impact
protection, serveto mitigate the consequences of those accidentsthat do occur. A number
of psychologicalphenomena,such as attribution error, explain the mixed successof road
safety campaigns. Most drivers believe that they are less likely to be involved in an
accident than other motorists. Existing road safety campaignsdo little to addressthis
problem; they focus on national and regional statisticsthat often seemremote from the
local experiencesof road users.Our argumentis that localized road accidentinformation
would have better impact on people's safety awareness.This thesis,therefore,describes
the design and developmentof a software tool to provide the generalpublic with access
to information on the location and circumstancesof road accidentsin a Scottishcity. We
also presentthe results of an evaluation to determinewhether the information provided
by this software has any impact on individual risk perception. A route planning
experiment was also carried out. The results from the experiment gives more positive
feedback that road users would consider accident information if such information was
availablefor them.
Acknowledgments
I would like to thank both my first and secondsupervisorsChris W Johnsonand

Philip Gray.Togetherthey havemanagedto provide far more thanjust the required
supervisory feedbackand guidanceover the last 4 years. They have supported
everything I have endeavouredto achieve throughout my PhD and nurturedall my
ideasandmethods.
I would also like to thank the following for being supportive, inspirational and
downright delightful: Marilyn McGee-Lennon, Roderick Murray-Smith, Richard L
Cooper, Prof John Davis.
Finally, I would not have survived my PhD time without my magnificent family, who
have made sure I kept my chin up and remained focused. They endureme at my worst
becausethey have faith that I will always triumph in the end, and become a little
better in the process.Thank you for believing that I could do this even when I didn't
believe in myself This thesis is therefore dedicated to my wife Yan You, my mother
Fengxia Li, my father Taisen Zheng, my mother-in-law Yuan Shu Ying, my
father-in-law You Guang Duo, and the last but not the least my brother-in-law You
Wei.
Table of Content
Chapter 1 Introduction I
1.1 An Introduction to Road Traffic Accident 1

1.2 An Introduction to Road Safety Research 3
1.3 Potential users of accident data 6
1.4 Research Objectives 7
1.5 Structure of the thesis 9
Chapter 2 Identified Problems and Research methodology 12
2.1 Human Error and Road accidents 12

2.2 Practices of influencing drivers' attitude 17
2.3 Identifled problems 20
2.4 Research Methodology 27
Chapter 3 Evaluation of Existing Road Safety Websites 29
3.1 Existing road accident related information systems 29

3.1.1 Simulating individual road accident 29
3.1.2 Real-time Road Traffic Accidents Reporting System 31
3.1.2.1 TxDOT Expressway 31
3.1.2.2 MATTISSE 32
3.1.2.3 AA 33
3.1.3 Road Traffic Accident Analyse Systems 34
3.1.3.1 KeyAccident 34
3.1.3.2 AIS (Accident Information System) 35
3.2 Evaluation of Existing Road Safety Websites 37
3.3 Information on the websites 38
3.4 Risk perception evaluation 41
3.4.1 Evaluation Method 42
3.4.2 Evaluation Results and Discussion 45
3.5 Conclusion 51
I
Chapter 4 Data Sources and Requirement Analysis 53
4.1 Data Sources for Road Accident Information System 53

4.2 System Development Process 55
4.3 Requirement Analysis 56
4.4 Conclusion 63
Chapter 5 Theories for visualizing multivariate data 65
5.1 Statistical graphics for visualizing multivariate data 66

5.1.1 Multiple-code plotting symbols 66
5.1.2 Histogram Plots 67
5.1.3 Star Plots 68
5.1.4 The Scatterplot Matrix 70
5.1.5 Conditioning Plots 71
5.2 Interactive Information Display Theories 73
5.3 Conclusion 78
Chapter 6 System Design and Implementation 80
6.1 Interface Design 81

6.1.1 Initial Design 84
6.1.2 Revised Design 88
6.2 System Architecture 91
6.3 Class Classification and Implementation 93
6.4 Conclusion 96
Chapter 7 System Evaluation 98
7.1 Method 98
7.2 Results of Usability Questionnaire 99
7.3 Results of Psychometric Study 104
7.4 Focus GrouP Discussion 107
H
7.5 Discussion and Conclusion 109
Chapter 8 Route Planning Experiment ill
8.1 Motivation and Goals ill

8.2 Experiment Design 113
8.3 Measurement and tests 114
8.4 Results and Analysis 115
8.5 Conclusion 128
Chapter 9
Route Planning Experiment with University Staff and Students 129
9.1 Experiment Design 130

9.2 Measurement and tests 131
9.3 Results and Analysis 131
9.4 Conclusion 138
Chapter 10 Conclusions and Future Work 141
10.1 Conclusions 141

10.2 Future Work 142
Appendix A: A brief Review of road safety research 151

Appendix B: Collection of Road Accident Information 159
Appendix C: Questionnaire for Road Accident Prediction 163
Appendix D: Questionnaire for gathering experts' information and initial
system requirements 165
Appendix E: User Interface Evaluation Form 170
Appendix F: An example of calculating sign test 176
References 177
III
List of Figures and Tables
Table 1.1 Accidental deaths: by cause 1971-1998

2
............................. ooo
... o.....................................................
Table 1.2 Road accident casualties: by road user type and severity 1993-2002
2
.... 0............................ .................................................
Figure 1.1 Percentage contributions to road accidents

9
.............. 0.............. *....................................... 0...... 0.00000
Figure 1.2 Snapshot of ReCap system

16
........................ 0................................................................
Table 2.1 The major error-shaping factors at each level of performance

14
.........................................................................................
Figure 2.1 Dynamics of generic error-modelling system (GEMS)

15
.........................................................................................
Figure 2.2 The theory of Planned behaviour

16
.........................................................................................
Figure 2.3 Different kinds of road accident information

22
.........................................................................................
Figure 2.4: UK Department of Transport Road Safety Site

23
.........................................................................................
Table 2.2: Scottish Executive Accident Frequencies for Urban Junctions

(1998-2002)
24
.........................................................................................
Figure 2.5: Excerpt from Glasgow City Council's Review of Road Safety Strategy
25
.......................................................................................
Figure 3.1 Screenshotsof HVE

30
- ... ---ooo o..o.ooo.
... o.... -o ... o..... o... o........... o..oooooooooo.. o... oo..o..
Figure 3.2 Screen shot of TxDOT

............. - ... oo.000 ...... 000.0-0.0..... 0........... 00....... 0..... 0...031
... 0000.00
I
Figure 3.3 Screen shot of MATTISSE
32
.......................................................................................
Figure 3.4 Screen shot of the AA

*........................................................................................ 33
Figure 3.5 An example result of KeyAccident system

. *.. **............................ 6.0- .......... 0.00.0000
.... 0....... ...... o...... o35
Figure 3.6 Snapshots of AIS
36
.........................................................................................
Figure 3.7 Facts and Figures from Think Road Safety Website
38
.......................................................................................
Table 3.1 Statistical analysis of the evaluation questions

......................................................................................... 46
Figure 3.8 How safe do you feel in terms of the likelihood that you could be
injured in a road traffic accident?
....................................................................................... 47
Figure 3.9 Are you worried about yourself being injured in a traffic accident?
47
.......................................................................................
Figure 3.10 How probable do you think it is for yourself to be injured in a traffic
accident?
48
.................................... 0........................................................
Figure 3.11 How concerned are you about traffic risks and are thinking that you
yourself could be victimized?
48
........................................................................................
Figure 3.12 Please rate the equity of this risk, in terms of whether those who
receive the benefits are the same people who carry the risks.
49
........................................................................................
Figure 3.13 Please rate how much the risk of being in a road traffic accident
affects you
49
........................................................................................
H
Figure 3.14 Please rate the extent to which this risk is chosen voluntarily by the
people affected
47
............... 0........................................................................
Figure 3.15 Please rate the immediacy of this risk, in terms of how soon possible
harmful effects may occur
50
.................................. 0.......................... 6....... 0............... **...
Figure 3.16 Please rate how controllable this risk is.

51
......................... *.. 0....... 04.0
..................................................
Figure4.1 Format for Accident Records in UK Road Traffic Accident Data

54
............... 0.... 0...................................................................
Figure 4.2 Sample Accident Information from the UK Data Archive

................................................................ 0.0..............
Figure 4.3 Waterfall Model

........................................................................................ 56
Figure 4.4 Some results of experts background

60
.......................................................................................
Figure 5.1 Multi-code plotting symbols

67
.......................................................................................
Figure 5.2 Examples of Histogram Plots

68
.......................................................................................
Figure 5.3 Example of Star Plots

69
.......................................................................................
Figure 5.4 Example of Scatterplot Matrix for Acedient data

70
.......................................................................................
Figure 5.5 Example of conditioning plot for accident information

72
............................... o.......................................................
Figure 5.6 Pre-computer creation of a visualization tool

73
.......................................................................................
Figure 5.7 The creation and use of a computer-based visualization tool

74
.......................................................................................
III
Table 5.1 Interaction Techniques
75
.......................................................................................
Table 6.1 Relationships between task categories, requirements and techniques

82
.................................. 0....................................................

83
.... 0...................................................................................
Figure 6.2: Overview of the first version Map-Based Interface

86
........................................................................................
Figure 6.3: Reviewing Details for Several Selected Incidents

88
........................................................................................
Figure 6.4 Second version of the User Interfaces

....................................................................................... 89
Figure 6.5 Filtering control dialog

....................................................................................... go
Figure 6.6 Option panel for more filtering functions

90
.......................................................................................
Figure 6.7 Details for selected accident

91
.......................................................................................
Figure 6.8 System architecture

92
.......................................................................................
Figure 6.9 Use Case Diagram

94
.......................................................................................
Figure 6.10Relationshipsbetweenclasses
95
...................................................... ................................
Figure 6.11 an example of the system flow chart

96
.................................................................................. o....
Figure 7.1 Overlaps on the map

101
.......................................................................................
IV
Figure 7.2 Detailed accident information
102
............................... 0......................................................
Figure 7.3: Quantitative Results from the Initial Usability Evaluation

104
......................................................................................
injured in a road traffic accident?
105
..................... o..o.... .......... 0..................... 0..0....
Figure 7.5 How controllable is the risk of involvement in a road traffic accident?
105
.... 0.0........ 0......................... 00.0*0
..................... 0000
..............
Figure 8.1 Glasgow road map with time and accident information
116
....................................................................................
Table 8.1 Preferences for the Glasgow route finding task

117
....................................................................................
Table 8.2 Sign test result for the Glasgow route finding task
117
....................................................................................
Figure 8.2 The experiment map of Manchester

118
....................................................................................
Table 8.3 Preferences for the Manchester route finding task

119
.....................................................................................
Table 8.4 Sign test result for the Manchester route finding task
120
.....................................................................................
Table 8.5 Second Group of Taxi Drivers' Preferences on Glasgow route finding
task
122
........................ 0.0..........................................................
Table 8.6 Sign test result for the Second Group's Glasgow route finding task
123
.....................................................................................
Table 8.7 Second Group's Preferences on Manchester route finding task

123
.....................................................................................
Table 8.8 Sign test result for the Second Group's Manchester route finding task
124
........................ 00...........................................................
V
Table 8.9 Summary of Routes in the Glasgow Map
125
.....................................................................................
Table 8.10 Summary of Routes in the Manchester Map

127
................................................ 0.... O..*.*o............. 0.0.........
131
....... 0...............................................................................
Table 9.110 university staff's preferences on Glasgow route finding task

132
.......................................................................................
Table 9.2 Sign test result for the Glasgow route finding task (university staff)
132
.......................................................................................

133
.......................................................................................
Table 9.3 Preferences for the Manchester route finding task (university staff)
134
.......................................................................................
Table 9.4 Sign test result for the Manchester route finding task (university staff)
135
.......................................................................................
Table 9.5 Preferences for the Manchester route finding task (university students)
136
.......................................................................................
Table 9.6 Sign test result for the Glasgow route finding task (university students)
136
.......................................................................................
Table 9.7 Preferences for the Manchester route finding task (university students)
137
.......................................................................................
Table 9.8 Sign test result for the Manchester route finding task (university
students)
137
........................................................................................
Figure 10.1 Advices on "Think! " website

144
.......................................................................................
Figure 10.2 Pictures in the system.

145
.......................................................................................
vi
Figure 10.3 Annual Average Daily Flows and Major Road Network Information
in Glasgow
149
.......................................................................................
Table A. 1 Injuries comparison between unbelted and belted occupants

157
......... o.... - .... o........... oo.....................
.......... - ..... oo..... oo.ooooo
Figure B. 1 STATS19 -Accident Record

161
.............. 0... 0.................. 0..... 0............. ................ o.o.
Figure B. 2 STATS19 - Vehicle Record

161
...................................................................... 0..................
Figure B.3 STATS19 - Casualty Record

162
.........................................................................................
Vil
The Impact of Localized RoadAccident Infonnation on Road Safety Awareness
Chapter 1 Introduction
An Introduction to Road Traffic Accidents
In 1896,a 44 yearsold womanwas hit by a car travelling at "tremendous
speed"(4 mph) and died minutes later of head injuries. This is the first
recordedroad traffic deathin the world [ 102].The coronersaid,"This must
neverhappenagain".More than a centurylater, road traffic accidentsare a
seriousproblemfor almosteverycountry.
In the UK, road traffic accidents are the leading injury-related cause of
death among people aged 15-44 years. Table 1.1 shows accidental death
statistics from 1971 to 1998 in Great Britain.
Accidental deaths:by cause
United Kingdom Numbers
1971 1981 1991 1998
Road accident 8,302 5,133 5,276 3,421

1
Railway accident 213 97 90 47
Other transport accident 235 146 115 87
Other accident
At home or in communal
establishments 7,224 4,865 3,763
Elsewhere 3,930 2,569 4,319
All other accidents 11,207 10. 73,493 8,519

1 %4141
I
The Impact of Localized Road Accident Information on Road Safety Awareness
All accidental deaths 19,957 15,790 12,974 12,154
Table 1.1 Accidental deaths:by cause 1971-1998(from National
Statistics[65])
As we can seefrom the table, road accidentscausedabout 40 percentof

deaths
accidental in 1971 and around 30 percent of deaths
accidental in
1998.In the last decade,around3,500peoplewerekilled on Britain'sroads
and 40,000were seriouslyinjured every year (SeeTable 1.2). The direct

costof roadaccidentsinvolving deathor injury is thoughtto be in the region
of Million a year. This doesnot include consequentiallossesin termsof
employment,etc.
Road accident casualties: by road user type and severity
Great Numbers
Britain
Year 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002
Killed 3814 3,650 3,621 3,598 3,599 3,421 3,423 3,409 3,450 3,431
Killedor 48,834 50,190 49,154 48,097 46,583 44,255 42,545 41,564 40,560 39,407
seriously
injured
All
306,135
severities 315,359 310,687 320,578 327,803 325,212 320,310 320,283 313,309 302,605
Table 1.2 Road accident casualties: by road user type and severity
1993-2002 (From National Statistics)
From table 1.2, we can see that the number of people killed in road
accidents has been decreasing.From 1994 to 1997, there were about 3,600
2
people killed every year. From 1998 to 2002, this number was down to
3,400. At the same time, the number of people killed or seriously injured
was falling every year from 50,190 in 1994 to 39,407 in 2002. That is a fall
of 21.5 percent during the 9 years. The improvement of road infrastructure
and of car design and government's safety campaign are helping to achieve
this. However, the reduction in the total number of fatalities has not been
matched by a comparable fall in the total number of all severities. From
1993 to 2002, the number of all severities remains high, over 30,000 every
year.
Road traff ic accidents also have significant impact on world economics.
According to the WHO's (World Health Organization) "World report on
road traffic injury 'Prevention in 2004", the cost of road crash injuries is
estimated at roughly 1% of gross national product (GNP) in low-income
countries, 1.5% in middle-income countries and 2% in high-income
countries. The direct economic costs of global road crasheswere estimated
at US$ 518 billion in 2004. The estimated annual economic cost of injury
sustained in road accidents in China is equivalent to US$12.5 billion ---
almost four times the total public health servicesbudget for the county.
1.2 An Introduction to Road Safety Research
Since road traffic accidentsare a seriousproblem for all countriesin the
world, the questionof how to effectivelyreduceroad has

accidents become
a moreandmoreimportantissueconsideredby a largerangeof scientists.
After an individual road accident,there is a concernto find out what the
causewasandwho wasto blame. Sometimes

peopleblamealcohol[4,54],
speed [20,13], bad weather [32], poor road surface conditions [75],
3
TheImpactof LocalizedRoadAccidentInformationon RoadSafetyAwareness
inadequate traffic signs [27], vehicle defects [21]..., and so on. These
accidents cannot therefore be addressedby any single solution. They are
causedby a combination of lots of factors [72,45]. This is why different

disciplines are engagedin the searchfor a solution.
The main aim of road safety researchis to use scientific ways to study the
road and traffic system in any of its aspectsto find suitable solutions for
reducing the number of road accidents or their severity. Assessing the
effectivenessof these solutions is another researchpurpose. However, road
traffic accidents are a dynamic phenomenon. An advance in one direction
can raise new problems. 'Risk Homeostasis' is the theory that humans
behave in such a way that if a risk is identified in a given system, and is
reduced by design, then a compensatory increase in risk-taking will occur
somewhereelse in the system [98]. Therefore, solutions are changing all the
time and road safety researchinvolves a very large scientific area. It can be
divided into the following four fields [72]:
1.) the study of occurrenceof accidents(statistics);
2.) the detailed study of any aspect of the accident processwhich may
be a factor in accidentproduction;
3.) the consequencesof accidents,i. e. injury and damage;
4.) the economic aspectsof accidentsand safety measures;
Detailedinformationaboutresearchin theseareascanbe foundin Appendix

A: A brief review of roadsafetyresearch.
According to the researchwork by Sabeyin 1975 [80], the accident process
can be divided into three categories: vehicle, road environment and road
users, and road users make the greatest contribution to road accidents.
He attempted to identify the "main contributory factors" responsible for
4
The Impact of Localized Road Accident Information on Road Sat'cty Awareness
2130 accidents which were investigated in great detail. Evidence was
obtained by observations of roads, vehicles and road users, by interviews,
and by assessing errors which were made by the road users, by examining
defects in vehicles, and by noting adverse features of vehicle and road. The
research team formed certain opinions. They categorized these contributory
factors into 3 areas (road environment, road user, vehicle), some accidents
being caused by a single factor, some caused by multiple factors. The result
shows that 8.5 percent of the causes could be assigned to vehicles, 28
percent to the road environment, and 95 percent to road users; these add up
to more than 100 percent because many accidents were considered to be the
result of more than one contributory factor. Figure 1.1 shows this result.
Road
environment Road uscr Vehicle
Single factor 2.5 65 2.5
Double factors 24 4.5
Treble factors 1.5
Finding solutions for accidentscausedby road users IS Much more difficult
than for those accidents caused by vehicle or road environment defects.

Changing road users' attitudes toward safety is a common response to
accidents causedby road users. Road safety campaigns,education and law

5
enforcementare ways to influence people's safety awareness.Chapter 2 will

discussmore human factors involved in road accidents.
With the development of computer technology, there are more and more
road accident related systems available for people to analyze massive
accident information. These systemsprovide a lot of help for experts to find

the solutions for accidents causedby vehicle or road environment defects.
Chapter 3 will discusssome of thesesystems.
1.3 Potential Users of Accident Data
Whatever the solution is, road accident data is essential for all road safety
research. Appendix B describes how UK government collects their road

accident data. Road accidents are reported to police. Police record the
information (e.g. STATS19form), such as time, location, weather, casualties,
and so on. Different user groups are interestedin road safety information for
different purposes.
Local authorities use the information for their road planning and
maintenance [24]. For example, from our interview with the Land Service
Department in Glasgow City Council, we know that road safety engineers
need historical data to identify accident black spots in local areas.They then
look into the data in detail so that they can find out the causesfor these
black spotsand take some countermeasures.
Vehicle engineers are interested in accident data because they can find
vehicle defects and ideas for improvements from accidents. For example,
research found that head related injuries accounted for most of the injury
cases,therefore, helmets for moto cyclists and seatbelts for car drivers are
6
TheImpactof Localized onRoadSafetyAwareness
RoadAccidentInformation
designed for protecting people from injury [64].
Insurancecompaniesalso care about road accidents[85]. Their accident

data sometimesare more completethan the police's becausepeoplereport
non-injury accidentsto their insurancecompanyto claim compensation
rather than report them to the police. Insurancecompaniesalso carry out
many surveysabout traffic accidentsto collect relevantinformation.They
usethe informationto determinetheir insurancepremiums.
Road safety campaignswhich are launched by governmentsneed the

informationto supporttheir aimsandremindpeopleof the potentialdangers
on the road. Meanwhile, local governmentsand schools can use the
information to educate pupils and students in the right attitude and
behaviourtowardsroadsafety[67,3 1].
The general public is also interested in road accident information for their
own safety during their travel. However, unlike the above user groups, the
general public doesn't have much of this information. Governments show
them tables, figures, and diagrams which are highly aggregatedstatistics.
From real time accident report systems,the general public can know where
accidents are happening now. But they don't have any access to the
historical and local road accident information. However, we think this
information can be helpful for changing people's safety awareness.
1.4 Research Objectives
From the previous sections, we can see that researchon historical accident
data can find injury patterns of victims and propose protective measuresin
the light of thesepatterns, therefore can reduce the number of people killed
or seriously injured. Furthermore, historical accident data can show defects
7
The Impact of Localized RoadAccident Information on Road SafetyAwareness
of vehicles and identify accident hot zones. But the statistics show that
although the number of people killed or seriously injured in accidents is
falling, the total number of accidents still remains high (See tablel. 2).
Increaseroad users' safety awarenessand attitude has been a popular way to
improve road safety for many years. There are three most common ways to
influence road users' attitudes, 1) Driver training, school programs or
education; 2) Road safety campaign; 3) Legislation and enforcement.These
methods will be discussedin detail in chapter 2. Our research,however, is
related to localized accident information.
Our research focuses on how the historical accident information can
affect the general public's safety awareness. We believe that

information which is customized to a person's characteristics and
situation is more likely to affect risk perception or safety awareness.

Therefore, localized accident information would have a better impact
than high level national statistics.
Using localized information to find out certain patterns which can't be
obtained from high level is

statistics not a new idea. Figure 1.2 is a snapshot
from Donald Brown's Regional Crime Analysis Program (RECAP). He
believes analyzing regional crime data at the level of individual districts or
even blocks is increasingly important when police initiatives to address
in
criminal activities one area or district can displace those activities to other
areas.
8
The Impact of Localized Road Accident Infon-nation on Road Safety Awareness
r-71rim 101xl
1ý eý 0 F7 CD L;
porgel Zoom In Pan copy RadSeledRect Select clustef Layefs Prt Map
W,
Zoom
out
T
Figure 1.2 Snapshot of ReCap system
Therefore the research objective is to find out how to affect the general
public's safety awareness by using localized historical road accident

information. The thesis will start with evaluating the existing road safety
campaign websites in order to see whether they have a significant impact on
people's safety awareness. Then it will describe our own localized road
accident information system and evaluate it. A route planning experiment is
also described to see whether such localized accident infon-nation would
affect people's route choices. Positive results from the experiment would
suggest such information does have an impact on people's safety awareness.
1.5 Structure of the Thesis
As we said in previous sections, most contributions to road accidents are
from road users. Therefore, how to improve the general public's safety
awareness and their behaviour become more and more important. However,
9
as we found, there is no such road accident analysis system for general
public so that they can know more detailed information about accidents
happening near them. We developed such a system and compared it with the
existing road safety campaign website to find out what kind of accident
infon-nation is helpful for improving their awareness. The thesis structure is
as follows.
Chapter2 discussesthe relationship betweenhuman error and road accident.
It also analyzes some problems of current road safety measurements,
education, legislation and road safety campaigns.Then it describesthe user
group of our study and our researchmethodology.
Chapter 3 evaluates of existing road safety information system ('Think' and
Scottish road safety campaign websites) and their effect on general public's
safety awareness by using psychometric questionnaire. At the end of the
chapter, the reasons why our results do not show a consensus are discussed.
In order to build our localized accident inforination system, chapter 4

discusses the available road traffic accident data and their structure. It also
describes our requirements elicitation and analysis. A full list of functional
and non-functional requirements is provided.
Chapter 5 discussessome statistical graphics for visualizing multivariate
data, their advantagesand disadvantages.It also explains why we chose
map-basedvisualization technique. In order to enhanceour system's effect,
interactive information display theories, their advantagesand disadvantages
are also discussed.
Based on the requirements listed in chapter 4, chapter 6 discusses the
interface design, system structure design, class classification and the system
10
implementation.
Chapter 7 discussesthe systemusability evaluation. The samepsychometric
questionnaireas in chapter 3 was used to find the impact of our system on

In
people's safety awareness. addition, a focus group discussion is used as
an alternative measurefor accessingsuch impact.
In chapter8, we describehow we designeda routeplanningexperiment.In
this experiment,we provided taxi drivers with local accidentinformation.
Wewantedto find out thatwhetheraccidentinformationwould changetheir
routeplanning.The Sign test was usedto test our hypothesis.

At the endof
this chapter,we discussthe findings from this experimentand possible
applicationareas.
In chapter 9, we carried out the similar route planning experiment with
university staff and students.
Chapter 10 discussesthe future work on the impact of localized accident
information and its applications. Conclusionswill be offered as well.
11
The Impact of Localized RoadAccident Information on Road Safety Awareness
Chapter 2
Problems Identified and Research methodology
2.1 Human Error and Road accidents
As mentioned in Appendix A: A brief review of road safety research,
perception, decision and action are the three main aspects involved in
driving. Any error in thesephasesmay causean accident. For example, the
following accident happenedon Saturday,October 13,2001 (NTSB Number
HAR-04/01, NTIS Number P132004-916201).A school bus entered a work
zone and struck road side barriers. It then rode up onto a bridge's sidewall,
and rolled 270 degreesclockwise as it fell about 49 feet. It landed on its left
side in a 1-foot-deep creek below the bridge. The NTSB determined that
the probable cause of this accident was "the failure of the Nebraska
Department of Roads to recognize and correct the hazardous condition in
the work zone createdby the irregular geometry of the roadway, the narrow
lane widths, and the speed limit. Contributing to the accident was the
accident bus driver's inability to maintain the bus within the lane due to the
perceived or actual threat of a frontal collision with the approaching
eastboundmotorcoach and the accident bus driver's unfamiliarity with the
accident vehicle". In this accident, the bus driver had to perceive the
hazardouscondition in the work zone and the threat of a collision with the
approachingcoach. The bus driver's inability to maintain the bus within the
lane and unfamiliarity with the accident vehicle could indicate errors in
decision and actions.
Rasmussen[44] thinks that the whole of human behavioural control can be

12
categorised into three levels, skill-, rule- and knowledge-basedbehaviour.

Skill-based behaviours are auto-matedroutine actions during familiar work
situations, such as make a 'U' turn or take a right (left) turn. At the level of
rule-basedbehaviour, people act on a stored rule which is the integration of
a sequenceof subroutines,such as overtake a vehicle. The rule or control is
selected on the basis of previous successful experiences.Knowledge-based
behaviour focuses on problem solving when there is no suitable rule
available for the unfamiliar situation, such as handing emergencysituations.

JamesReason'sGeneric Error-Modelling System analyzes common human
error forms based on Rasmussen's framework. Figure 2.1 shows the

dynamics of this system.Reasonalso gives error-shapingfactors for eachof
these levels. Table 2.1 shows these factors. According to the NTSB report,
the above road accident involves errors in all these three levels. First, the
Nebraska Department of Roads failed to recognize and correct the
hazardouscondition in the work zone. This error may becausethey thought
the work zone was safe as others for the traffic, which is an example of
rule-based error-shaping factor No. 1 'mind set' or No. 5 'over-confidence;
or because they unable to analyze the danger based on the available data,
which is an example of knowledge-based error-shaping factor No. 7
'Incorrect/incomplete mental model'. When bus driver saw the approaching
motorcoach, he may not have thought that would be a problem for him,
which is an example of the skill-based error-shaping factor No. 2: fail to
recognize 'environmental control signals'; or he may have been thinking
about something else, which is an example of skill-based error-shaping

factor No. 4 'concurrent plans'; or he trusted his driving skills, which is an
example of rule-based effor-shaping factor No. 5 'over-confidence'. Then,
when collision happened,he was unable to maintain the bus within the lane
because of his unfamiliarity with the accident vehicle. This should be an
example of knowledge-based error-shaping factor No. 5
'Incomplete/incorrect knowledge' becausethe driver may have the required
13
skill to control the bus. But his knowledge (unfamiliarity with the accident
vehicle) prevents him from performing the correct skills.
PERFORMANCELEVEL ERROR-SHAPINGFACTORS
SKILL-BASED 1. Recencyand frequencyof previoususe

2. Environmentcontrol signals
3. Sharedschemaproperties
4. Concurrentplans
1. Mind set ('It's alwaysworkedbefore')

RULE-BASED 2. Availability (Tirst comebestpreferred')
3. Matching bias ('like relatesto like')
4. Over-simplification
5. Over-confidence('I'm sureI'm right')
1. Selectivity
KNOWLEDGE-BASE
2. Working memory overload
3. Out of sight out of mind
4. Thematic 'vagabonding' (Flitting from
issue to issue quickly, treating each one.
superficially) and 'encysting' (Can't see
wood for the trees)
5. Memory cueing/reasoning by analogy
6. Matching bias revisited
7. Incomplete/incorrect mental model
Table 2.1 The major error-shapingfactors at eachlevel of performance [44]
14
Routine Actions III Attentional checks upon progress

Familiar Environment
GOAL
-----------------I
STATE
SKILL-BASED (SLIPS) NO
------------------
--------------------------- * ----------------------- ;,ý -----------------------
PROBLEM
.4
CONSIDFR
RULE-BASED LOCAL
(MISTAKES) STATF
INFORMATION
APPLY STORED
RULE
'4; IF (situation)
THEN (action)
I
I NO
-------------------------- f --- ------------------------------------ I
CONSIDER MORE
ABSTRACTRELATIONS
BETWEEN STRUCTURE
&FUNCTION
--------------------------
Subsequent
KNOWLEDGE-BASED attempts
(MISTAKES) ------------------------- *
L -------------------------- i
Etc.
Figure 2.1 Dynamics of generic error-modelling system(GEMS)
Other branches of psychology have also been used in analysing human
errors and road accidents. Rothengatter [79] defined the study of the
behaviour of road users and the psychological processes underlying that
behaviour as traffic psychology. The basic psychological theory used in road
safety research is the theory of planned behaviour, which was originally
proposed by Ajzen [3]. The diagram in Figure 2.2 shows a simplified
version of the theory.
15
Figure 2.2 The theory of Planned behaviour
According to this theory, intentions are the main dctcri-ninants of behaviour.
intentions are deten-nined independently by attitudes towards a behaviour
(global positive or negative evaluations about performing the behaviour),
subjective norms (perceived social pressure concerning performance of the
behaviour), and perceived control (the perceived ease or difficulty of
performing the behaviour). In addition, perceived behaviour control might
inhibit or facilitate performance of the behaviour. For example, drivers
speed because they have certain intentions. These intentions may come from
the following judgments: "Speeding can really save my time and I can bear
the consequences (getting a ticket from the police)" (attitude); "Everyone
speeds" (subjective norm); "If there is a danger, I can control my car under
that speed. I did it many times and it is difficult for me to keep the speed
under the limit on such quiet road" (Perceived behaviour control).
The common feature from the above theories is that they are all trying to
analyze the relationships between human being's behaviour (errors) and
16
The Impact of Localized Road Accident Information on Road SafetyAwareness
road accident. The differences are Dynamics of Generic error-modelling
system (GEMS) and the theory of Skill-, rule- and knowledge-based

behaviour are focused on how to make road users choosecorrect behaviours
to avoid potential dangers. However, nowadays, some drivers choose the
wrong behaviours even they know those are dangerous and wrong. The
theory of planned behaviour is to analyze human errors from psychology
point of view. It identifies the factors which could affect drivers' intention
which ultimately affect drivers' behaviour. It try to explain why drivers
choosedangerousbehaviours even they know they are wrong.
2.2 Practices in Influencing Drivers'Attitude
Attitude plays an important role in the theory of planned behaviour. There is
evidenceto suggestthat 'people drive as they live' [90]. The term "attitude"
is a key concept in social psychology, referring to a hypothetical mental
structure which determines actions to in

or prepares a person act a certain
way [41]. It is generally used to characterize a person's convictions or

beliefs about objects regarded as good or bad, acceptableor unacceptable,
and with which the person feels in compliance or not, as manifested in
various kinds of behaviour. Here is an example from Fildes' study "Speed

Behaviour and Drivers' Attitudes to Speeding" [9]. The study found "A
surprisingly high number of motorists at all speedlevels did not believe it to

be dangerous to travel 30km/h above the posted speed limits and most
thought the chanceof being stoppedby the police for speedingat these sites
to be low". These drivers think travelling over 30km/h is acceptable,even
when that is above the speedlimit.
There is still little practical experience of how to identify, measure and
modify attitudes. As mentioned in the previous chapter, driver training,

17
school programmes or education [67,31], road safety campaign [87,70],
legislation, enforcement [22] are all used to influence behaviour and attitude.
Driver training and school programmesare aiming at establishingnew kinds
of traffic safety behaviour and attitudes. The other methods are targeting at
behavioral or attitude changefor existing drivers.
Driver training and examination can provide necessarydriving skills. It can
also influence driving behaviour and attitude. For example, the UK driving
test includes both a theory test and a practical test. The theory test has
multiple choice questionsand a hazard perception test. In hazard perception,
participants are required to view 14 hazard video clips on the computer

for
screen approximately one minute each.There are 15 hazardsto find. The
practical test covers a wide variety of different road conditions, from quiet
low speedroads to busy high-speed roads and town or city centre driving.
Drivers are also required to undertake several set maneuvers. These tests
provide drivers with information about dangerous environments and
situations on the road. At the same time, they learn how to cope with those
situations, using the skills received from training.
Finland and Norway have carriedout evaluationsof a new driver training

The
course. coursehas two Skid
phases. training is one componentof the
second phase. The result shows that the drivers trained by the new
curriculumwere less afraid of driving on a slipperyroad. However,

young
driverstrainedby the new curriculumhad a greaterproportionof accidents
on a slippery road than the older ones and those trained by the old
curriculum.It was recommendedthat exercisesin skid training should

be developed so that the perception of risks and anticipation are
stressedmore than training of skilled performance [49].
Traffic education in schools has helped pupils to learn basic rules,
18
understandtraffic processesand provides basic instructions on behaviour. It

is intended to establish positive attitudes towards safety standards.But the
effectiveness of the education can decline over time when the course is
finished. Establishing long-lasting good attitudes needs help from other
sources.
Road safety campaigns remind people of the dangers of road traffic
accidents, what kinds of situations they should pay attention to and what
skills could be used to cope with such situations. A variety of techniques

have been used to provide information to the general public about the
dangers associatedwith road use. These include media broadcast, leaflet
campaigns, community-based schemes, and so on. The key point in the
success of a campaign is that the information provided must catch the
attention of the target audience. They must also be willing to listen to the
information. Another problem is that safety campaigns involve significant
costs to the agencies that use them. The production costs associatedwith
leaflet production must be added to the logistics involved in disseminating
the publications to the generalpublic. Even when theseleaflets are delivered
to individual householdsa significant proportion of them will be discarded
before they are read [91].
Legislation and enforcement are other means to influence road users'
behaviour and attitude. Legislation dictates behaviour and in theory makes
the behaviour of all road users predictable and safe. However, when the
goals of legislation appearto be unreasonable,

people may be less willing to
adapttheir behaviour, or changetheir attitudes.
In many cases, legislation and regulation are not enough to change
behaviour. Police enforcement is necessary.But enforcement is an external
motivator for behaviour. It makes certain behaviour disadvantageous

so that
19
drivers or riders may consider the consequence of their decisions and
hopefully do not take any dangerousactions. Enforcement often leads to a
change in behaviour, but in many cases only for as long as the perceived
chance of being caught is high enough. Changesin attitudes do not always
result from enforcement. A combination of enforcement and information
provides better support for lasting behaviour change[41].
2.3 Identiried Problems
As discussedabove, efforts have been made to changethe attitudes of road
users, e.g. by public information campaigns and driver education. Such
measureshave so far had limited successin terms of reducing accidents[17,

37, and 68]. Charles Goldenbeld, Peter Bernadus, and Jelle Heidstra [16]
suggestpossible reasonsfor this situation. They believe attitude changedoes
not necessarilylead to behaviour changebecause:

1) A person may not have the skills or knowledge to behave in accordance
with his/her attitude;

2) Situational factors hamperthe showing of attitude-compatiblebehaviour;
3) At a particular moment, other attitudes or preferencesare stronger.
However, this doesn't mean influencing attitude is not important. A negative
attitude towards particular behaviour can act as a barrier to performing that

behaviour. It is always important to remove such barriers, even if it does not
immediately lead to a behaviour change. In fact, they believe changing
attitudes towards traffic behaviours by providing road users with
attitude-relevant information is an important tactic to achieve
behavioural change. They also provide some recommendations:
1) The information should address the needs of road users in order to
increase the probability that the information will be noticed and used;
20
2) "We should be precise as to the distinct character of the traffic
behaviour we aim to change and the context of the behaviour. " For
example, researches about speeding [2] showed that the importance of
attitude and social norm can differ depending on the type of road considered.
Therefore, road safety campaign information about driving speed should
make a distinction between the different categories of road, inside and
outside built-up areas.
From the aboverecommendations

and previousroad safety practises,we
can seethat road accidentrelated information in
can play a role reducing
road accidents.Driver training coursesneed this information to identify
potential risks and provide appropriateadvice or instructions. School
educationneedsthis informationto inform pupils. Roadsafetycampaigns
need this information to support their plans and show evidenceto the
generalpublic. Legislationand enforcementneedthis informationto help

peopleunderstandthe rulesandthereforeobeythe laws.
Road accident information is presented in various format including text,
images and video. The presentation of this information can be divided into
1-dimensional, 2-dimensional and multi-diMensional representations.Figure
2.3 shows some examples of different dimensional road accident
information data.
RoadIncident:
A4148: Lane(s) closed between West Bromwich Road: Fullbrook

andBirminghamRoadin Anti-clockwise direction.
Onedimensional(text)
21
The Impact of Localized Road Accident Information on Road Satlety Awareness
Tpe
, asLA,
Two dimensional (charts, images, graphs, et.)
Multi dimensional (video)
Figure 2.3 different kinds of road accident information
I-dimensional infori-nation provides a literal description of the accident.
Those infori-nation are usually come from traffic police. They recorded the
accident process according to the description of witnesses. 2-dimensional
information, such as charts and graphs, can show trends in accidents,Those
22
The Impact of Localized Road Accident Information on Road Sal'ety Awareness
infori-nation are usually generated by road safety experts or researchers.
They use such information to identify potential accident patterns so that they
may find countermeasures. Multi-dimensional information, such as
accidents simulations, can help experts or researchers get a better
understanding of the single accident. However, at the moment there are few
guidelines for use of I D, 2D, or multi -dimen sional data for road safety.
In order to provide a cost effective meansof accessinginforination about the
causesand consequencesof road accidents,many road safety agencieshave

developed on-line inforination sources. Figure 2.4 illustrates the main UK
Department of Transport portal for road safety information. As can be seen,
this web site provides access to information about a range of issues

including the use of speedcameras,child safety and alcohol limits. National
level statistical information is also provided.
ONEWW"M7. I
tf- fl, 1,
- 1
-jaý
Transport
Ia
TM lance tonsi*. IlpIls aal..., Md 11014 CIIM*MS. I Ml. 4 wo. . dll. Dfl. al. F*I rflpdI4ldSy liMp au .e n$oc*ddpkS M9S100701pw
AI. MMIIII
T*l ISIS lISIkI I4(y000 US OSSeWIMU *iMfl tO . vd0 1144SliMy - ItS flut *0.6.
(4,4.11.. *14.110*111.6Th... . 04 bId..

"I*c*a0 '144*1,0*. . Id.. c*.. q (MO.., 11*41106 (
S4IOOM11I
1* **fl't . 04 aMt
WaM. bI . 'telo, n Old 014*111.0.11,. *060*0* 01,

frOl t*k.
I 050 4111cia, dl. powsecw .4 'o06 coMaS,.. led 40.1 . 54,4.11. - .1. 14. tIflPp, *lS 611440114*01
L*1M#. 400'lT fl14'*04i01111*k, iKa. 0 *01*4,
L-1 T,. 10.1 . S. ty . *. %d 5.541.0. « 4*. d p.*I, l*, '«. 4.411,4..
tfl, l4,4.11 11,1440 ll01tS I*001 401.1411« eId..,., od te1*, *4 an, *41,40*10550*40tH 411*4554041110911M1'0*06.4*4
'4* *0405
r-, F
p W.
ýj
tjL J; -;;;::- - jsý2ý1 in
!
Figure 2.4: UK Department of Transport Road Safety Site
23
Within Scotland, it is possible to accesslocal accident data published by the
Scottish office. There are, however, a number of usability issuesassociated
with this information source. Members of the public must use a series of
interlinked Excel workbooks to navigate between different views of the
regional statistics (there are html and pdf versions available now for the
users). This data has the benefit of providing detailed information about the
frequency of accidents in particular, generic types of road geometry. Table
2.2 illustrates the Scottish Executive data from mean accident frequenciesat
different severity levels for built-up (urban) junctions between 1998 and
2002.
All
Fatal Serious Slight Severities
Junction 45 772 2,995 3,812
Roundabout 3 60 521 583

Mini-roundabout 6 42 48
-
T/Y/staggered junction 31 596 2,994 3,621
Slip road 9 55 64
-
Crossroads 10 172 1ý014 1,197

Multiple junction - 22 115 138
Private drive 1 18 93 112
Otherjunction 2 40 202 244
Total 93 1,695 8,032 9,820

1 -1
Table 2.2: Scottish Executive Accident Frequencies for Urban Junctions
(1998-2002).
24
The Impact of Localized Road Accident Int-orniation on Road Safety Awareness
At a local level, individual city and municipal organizations will also Issue
information about road safety issues. For example, GkISgOWCity Council
developed a road safety action plan. This publicized the various direct steps
that they were taking to reduce the accident rate within the areas under their
control. This argued "In 1997,22 people were killed, 496 were seriously
injured and 2597 were slightly inured on the roads in Glasgow. Although,
this was the lowest total of casualties in modem times, it is 22 deaths too
many". Figure 2.5 shows how the review of their strategy presented
aggregate local accident statistics to members of the public.
C-Ity H-L- 1990,

-10.16001-05 A--g.
1--ý
-0
TAJWWTLiNE
1ý ----
G__.
.--3. aaS
-1 -
20DO
"600
nec SQi ltfl an gee im *97 *99 1999
--im-MET uNIE
en,
AGUT LIMU
40
1"1-85 IGW ýl 1092 1910 W" Ma in* IM OM TM RODO
Figure 2.5: Excerpt from Glasgow City Council's Review of Road
Safety Strategy
From the previous examples,we can see these on-line information sources
only provide high level aggregate national or regional statistics. These
statistics provide an overview of road accidents and certain relationships

between accidents and factors, such as road geometry, accident locations,
25
accident times. But this kind of information does little to cope with the
problem of fundamentalattribution error.
FundamentalAttribution Error (FAE) is a psychological phenomenon.This
refers to the fact that whenever people are making attributions about an
action, they tend to over-emphasizedispositional factors about the actor, and
under-emphasizesituational factors [77]. Because it appearsas if people
generalize from the actor's behavior and ignore the situational context in
which behavior occurs, the FAE is often described as a tendency to
underattribute the cause of behavior to situations and overattribute it to

dispositional traits [77]. For example, most drivers believe that they are less
likely to be involved in an accident than other motorists. Accidents occur
becauseother motorists are careless. They believe that they would never
drive like that.
Even if drivers learn that speed, drink-driving, close following are
dangerousfrom accident information, they may not changetheir behaviour.
As we mentioned at the beginning of this section, Situational factors can
hamper showing attitude-compatible behaviour. They think they know their
traffic environment well. They have traveled many times in their local area
and never had an accident. Therefore they feel that the rules are not
applicable for them in this area. Researchshows an average car driver can
be expected to be involved in a personal-injury accident once in about 60
years [29]. So it is normal for a driver not to have an accident for long
periods of time. But if drivers drive dangerously 10 times without an
accident, this doesn't mean they won't get hurt the next time. If they know
someonein their area who has had an accident with the samesituation they
had, they may be more alert to that situation.
2.4 Research Methodology

26
From previous sections, we can see increase road users' safety awareness
and attitude has been a popular way to improve road safety for many years.
However, the existing methods have their limitations. Driver training,
school programs or education only have short term impacts on road users'
safety awarness.Road safety campaign could be affected by Fundamental
Attribution Error (FAE). Enforcement often leads to a changein behaviour,
but in many casesonly for as long as the perceived chanceof being caught
is high enough. Our research focuses on different aspects of accidents
information. We argued that more localized precise road accidents
information may have a significant impact on the general public's safety
attitudes. In order to test our hypothesis, we decided to develop a software
application, which will provide a map-based road accident information
system for the Glasgow area. Using localized information is not a new idea.
For example, Don Brown's group at the University of Virginia has recently
implemented a Regional Crime Analysis Program (ReCAP) for use in small
cities and towns. This system uses a client-server architecture so that local
officers can monitor patterns in crime reports across time and geographical
locations. This tool enableslaw enforcementagenciesto analyze crime data
down to the level of individual districts. This is increasingly important when
police initiatives to address criminal activities in one area or district can

displace those activities to other areas. As with ReCAP and unlike route
planning systems,such as the AA navigation tool, our proposed application
will provide localized historic accident information for the generalpublic.
Our research had three main stages. First, we used risk perception
assessmenttechniques to find out how existing on-line road accidents

information affects the general public's safety attitudes. This is discussedin
detail in Chapter 3. Secondly, we developed a map-based road accident
information systemfor the Glasgow area.Through our systemuserscan find
27
out detailed information about every accident that has happened over the
last few years. Additional tools will also be provided so that users can find
accident patterns in the area. We then used the same psychometric
techniquesto assessthe impact of our systemon people's safety attitude.
By comparing existing on-line information system and our system, we
hoped we could identify which information and information presentation
tools are critical in changing the generalpublic's attitudes to road safety.
Chapter 3
28
Evaluation of Existing Road Safety Websites
3.1 Existing Road Accident Related Information Systems
3.1.1 Simulating Individual Road Accidents
The simulationof individual road accidentsis an importantaspectin road

It
safetyresearch. can show the interactionsbetweenhumans,
vehiclesand
the environmentbasedon data collected at the sceneof an accidentor
incident. The Human Vehicle Environment (HVE) is such a system,
developedby the EngineeringDynamics Corporation.This system can
simulate the of
response multiple occupantsduring a rollover collision
including contactbetweenthe occupants.It can simulatea crashsequence
and changethe initial conditions to study avoidability. It can route the

results directly to a video showing multiple views including thoseof the
driver of eachvehicle, use virtual thermocouplesto monitor brake lining
on a downhill grade,createa complex3-D terrain meshand

temperatures
drive a vehicle on it, and so on [39]. Figure 3.1 showssomescreenshots
from the system.The left hand side figure simulatesa van driver's view
beforethis van collideswith the comingtruck. The whole accidentprocess
canbe seenfrom the simulationwith HVE's virtual cameraattachedto the

witnessesand driver of each vehicle. The right hand side figure shows
simulation
another of two car From
crash. the simulation,
we can during
see,
the accident,not only did the occupantsstrike the vehicleinterior,they also
struckeachother.This kind of informationcould help expertsto explainthe

injury patternsof the victims. However,this softwarehaslimitations. Such
simulationsonly help researchersunderstandindividual road accidentsor
generic scenariosbased on particular types of common accidents.It is
29
The Impact of Localized Road Accident Information on Road Sal'cty Awareness
difficult for researchers to see the relationships between accidents. For
example, from one accident simulation, researchers might find that tile car
speed is too high. If this car can be driven under 60 mph, the accident will
not happen. But if conditions change a little, the conclusion may not be right.
For instance, another car, which is driven Linder 60 mph at the same place,
may still have an accident because it is raining. Therefore, the information
that comes from individual accidents is not enough for road safety research.
However, statistical tools can be used to help identify relationship between
different individual accidents.
Figure 3.1 Screenshots of HVE
CarSim (http: //www. carsim. com/) and Pam-Crash
(http: //www. esi-group. com) provide further examples of single accident
simulation tools.
3.1.2 Real-time Road Traffic Accidents Reporting System
30
The Impact of Localized Road Accident Information on Road Salety Awareness
At present, many commercial websites provide real-time road traffic
accident information using their own visualization tools. This information is
updated every hour or even every minute. People browsing these websites
can find out where and when accidents have happened. Using the
information, people can plan to avoid traffic before they go out. The
following are typical of these websites.
3.1.2.1 TxDOT Expressway
m I
ep
-I Xi
Fil Ebt V- F-Ký 1. ýd, mip
0-0-X-,
S. k F-d 5tOP -i
RefreSh F-t. %ý ýei
MeJa Hgory
Z, Edt
FVJ ý] 0
ALkýe. http jj- dý tt, t. ht. G. L.*,
m
All Road Conditions
IF'Aft =11, f-, 71,17n
-
These iepoits iefled TxDOT'sbest wilopowshon.
9 Ilse tunesfile, l. Comfibm's -1,
ChM)gelapidy. No W.VI. Wdyol amtiýv of cotislAefeliessis espleSse,J ol 1111plied
Ckk a colmed sip- withe mip to mew a specific map cooidmWe coovdittono; mew
A load C. 141olls.
d
.ýý1 !ýII C'... ýLi In,ý
N- 19,2002 04: 50AM EST
OCOPY[W 2002, TOOT DiscL*m I Comem Usage I PiNacy Poky I OpftRecwds I Acceslib&V I TXDOI Cwfl&ls
All Rights Resened. Im
-- -- - -- - ----
,o DW--- Ik-. t
0
Figure 3.2 screen shot of TxDOT
The TxDOT Expressway website provides road condition information for
Texas State [95]. The map is divided into many squares. Those that are
31
colored blue have road traffic problems. Users can select an area by clicking
the corresponding blue square. Then a new page with detailed road
condition information is displayed. Users can also use a combo box above
the map to select a specific condition, such as construction, bridge-out,
damage, and so on. Figure 3.2 shows a screen shot of TxDoT system. With
the information provided from this system, road users can be infon-ned by
the current road conditions on every road in Texas State so that they can
plan their j ourney more eff iciently.
3.1.2.2 MATTISSE
Figure 3.3 Screen shot of MATTISSE
The MATTISSE website provides traffic information covering the UK's
"crossroads" in the Midlands [59]. Using a map, users can select a small
area that they are interested in. Then a local road map will appear. When
users click on an icon, a further sub-window will come out, which gives
detailed information. An icon list is always at the right-hand side of the web
page, which gives users an explanation of each icon. This website also
provides an incidents list. Figure 3.3 shows a screen shot of MATTISSE.
Like the TxDoT system above, the information provided by MATTISSE
32
The Impact of Localized Road Accident Inforniation on Road Safety Awareness
could help road users plan theirjourney more ell-iciently and safer.
3.1.2.3 AA
k p- I- ý. p
" Es - '- - Es
.. ""
e
p..- ....... à. d. h.,.. 'A
(EN
Aacýt
u. _
I.- aim
:1
NO
Iý
btVno(Kleo ore of the saNg pbý Dabe
Eq Ia2a
7ý, W*-Aý-
r-7r-W-- zi
Figure 3.4 screen shot of the AA
The 'AA' website provides traffic news throughout the UK [881. Users
must firstly fill in a query by city name, by region or by road name. The
system will then display a map in the middle of the page. There is only one
kind of icon to indicate all situations. At the bottom of the page there is a list
of incidents. Again, like the above two systems, AA's traffic news and maps
are very helpful for road users to know what the current road conditions are,
where the road works are, where the incidents happened. With such
information, road congestion can be reduced around those areas and road
uses can save a lot of time.
Real time accident information system can give road users ideas about
to
which road avoid so that people can chooseanother better route. However,
such systemsdo not provide historic accident data. People can't know which
area is an accident hot spot and which route is safer. The next section will
talk about accident analysis systemwhich could provide such information.
3.1.3 Road Traffic Accident Analysis Systems
33
3.1.3.1 KeyAccident
KeyAccident (See Figure 3.5) is an AutoCADV based GIS application,
including a standalonedatabasemanagementtool, for storing, maintaining
and analyzing UK road traffic accident data. Enquiries can be made
graphically by selecting areas on a map using rectangular, circular or

irregular polygon search areas. Data can be displayed and arranged
automatically on OS maps. Alternatively, semi-automatic displays can be
controlled to distribute accident information around a junction. It also has

the following features:
" Setting up the SEARCH CRITERIA and extracting the selected
accidents
" Creating a FILE from the selectedaccidentsfor future use.
" Creating and drawing the STICK DIAGRAMS and sorting them.
" Creating and drawing the HISTOGRAMS.
" Creating and drawing the PIE CHARTS.
" Creating and printing out the TEXT SUMMARY.

" Ability to perform GIS searchesquickly with easeand presentRTA
data in a graphical format
GenerateaccurateCluster Site lists using recognisedalgorithms
Maintain RTA data effortlessly using comprehensivedata

managementtools
Createstatistical, narrative, comprehensiveor user defined reports
Facility to monitor sites with the automatic extraction of RTA data

to the Dfr Safety CameraPartnershipdatabase
34
The Impact of Localized Road Accident Inforniation on Road Sal'ety Awarcticss
--
!. W=
: Tr
i, , ,, Z,, - .,'ý ý- 'a, ý.. ." :-.;., .,ý Zý L, 4, ý!, :ý.!,.
"I,
I I., ýý 1r.:! "
n%11ý11 -ý = "T
111P
;ý,A'
zlr I- ýt. I... .ý,. ý...4" :.,-,",.-1
-& -I---------,
IIt 1- Xf v "I TV
---
1pumm"Mmomimmomp-q
'LA
- -3r -3r- -Ir- -Z ---
;: i i
'"ll at
"---j ot ag&tIId; IA --tl aIII 'K IIgI&ttA
:.
Irf,, rtI. - tPr.;
If, ý' " '' ýf Iw, I
,
As we shall see, the KeyAccident system is used by several local authorities
to guide road planning.
3.1.3.2 AIS (Accident Information System)
AIS was developed for police departments in California. It is a
comprehensive accident data reporting and tracking system, for traffic
enforcement personnel. In use since 1993, AIS can view, edit, compare and
combine visual representations of accident scenes (photos). And it can
generate many reports customized by users. Figure 3.6 shows some
snapshots of the system.
IN=TIUM =-.
E-1- Edý( 2. -h
,--:
EwP. dý G,. phý j-g- J.. l. t1.1p warrüm
M j4 a kL 414 v lij OZIY
Spead C-M.. In dT nNbl,
F-. T-A«my
KillWd1 c4fic., Dýle r- Tm UMn-
st. 18 ". y P. 1
-, zý , ý, (,1,3 il F,
,
Lomto
SI. et N. St.. t D.. Lt. ý DM. - Uýh C-St.. t
F--, 1,,-; Ll' FL--3
Pnma, y Colksion Factors
Code V. hC. d.., R. -k Cnedl?
S. 4. -
[,
1 -, - hA'lý ,
Týffiý C., "l, PE, V. h,d, Wh

-- 7: ] Fý ------------- 3
-111 - -1 ',,, 'f, -1r4, p
35
The Impact of Localized Road Accident Inforniation on Road Sal'ety Awareness
I tM hI !I 1'.
hL a
III
.z
4 II it
11
ýD -;5-
21
Ed. Typ. U. 10
Input I aummmy I Qdlaset
Accidents by Day of Week
Figure 3.6 Snapshots of AIS
Although all the above systemscan provide great insight into road accidents,
none of them provides historical road accident information to the general
public. Most of them are only available to road safety experts. In the UK, in
order to increasethe general public's road safety government

awareness, has
launched several websites to provide road accident infori-nation. The next
section will discuss these websites and evaluate the effectiveness of these
websiteson people's safety awareness.
36
3.2 Evaluation of Existing Road Safety Websites
As mentioned in chapter 2, many road safety agencies and governments
have developed on-line road safety information in order to provide a cost
effective meansof accessinginformation about the causesand consequences
of road accidents.In this chapter,we evaluate two such websites to find out
how the information provided on these websites can affect people's safety
awareness.
2
We chosethe Scottish Road Safety Campaign' (SRSC) and Think! websites
becausethey are two of the major on-line road accident information systems
within the UK. The SRSC was founded in 1985 and is funded by the
Scottish Executive. Its remit is to develop and co-ordinate Scotland-wide
road safety initiatives and campaigns. The SRSC works closely with all
local authority and police road safety units in an attempt to ensure a
coordinated approach to road safety in Scotland. The Think Road Safety
Website was launched by the UK Department of Transportation to provide
the news of latest campaignsand road safety advice to keep everyone safer
on the Ws roads.
1http://www. srsc.org.uk/
2 http://www. thinkroadsafety.gov.uk/index.htm
37
3.3 Information on the Websites
A variety of road safety related information can be found on these two
websites. It can be divided into four categories: road traffic accident
statistics, tips and advice on road safety, interactive multimedia infon-nation
on road safety, and road traffic safety research outcornes and publications.
The websites focus on several different kinds of road users, from young
teenagers to old persons, from road safety engineers to traffic police, and
from pedestrians to drivers and cyclists.
A number of different techniques are used to present inforination on these
websites. First are statistics, facts and figures. The information covers all
important aspectsof road safety, such as facts about drinking and driving,
seatbeltusage,motorcycle accidentsand casualties,and so on.
Figure 3.7 shows some facts and figures from the Think website.
II. r
'ý: fý IF,
r.
-44114 a,
rrrm.
ýw
w,
IV, 'I'm 'A' ý 04 '1, " ." 'lo
CooffMOM
---.. -. - "t" -
.JL,; -
- -- - -1 "1 , ýýr,wt ýI
MISMEREM
- : 1)
IN a, I
Imp
1-1- : =: 7: ý=V
1Mo=,
Figure 3.7 Facts and Figures from Think Road Safety Website
38
Thesewebsitesalso provide tips and advice on road safety. This information
focuses on specific user groups, like drivers, pedestrians, cyclists, school
pupils. For example, SRSC give some tips for older pedestrians:
" "Try to plan your day to avoid rush hour traffic. In this way you can
reduce the risk of being involved in a road accident.
" Allow extra time for journeys. When in a hurry or running late,
people tend to becomecarelessand overlook safety
" Try to keep yourself fit and healthy. Have regular check-ups,
especially for your eyesight and hearing. It is important for all
pedestriansto be able to see and hear the oncoming traffic.
" If you need to wear glassesor a hearing aid, make sure you wear
them every time you are out.
" Try to use safer crossing places such as zebra and pelican crossings
whenever possible. Many accidents involving pedestrians happen
near or on PedestrianCrossings.These crossingsare only safe when
usedproperly
" If you ever need help to cross the road don't be afraid to ask."
(http://www. thinkroadsafety.gov.uko
The Think website provides the following advice for cyclists.
" "Be visible. Ride well clear of the kerb, wear bright clothing and
always use lights after dark or in poor weather conditions.
" Show drivers what you plan to do. Always look and signal before
you start, stop or turn.
" Ride a straight line past parked cars rather than dodging between
them.
" Don't jump red lights.
" Don't ride on pavements.
39
" Don't ride the wrong way up one-way streets, unless there's a sign
saying cyclists can.
" Don't ride across pedestrian crossings."

(http://www. thinkroadsafety.gov.uk/)
However, it is difficult to determine whether such advice has been shown to
have an effect on behaviour.
In order to attractyoung road users,interactivemultimediainformationis
availableon the sites. These includes video (TV road safety campaign),
audio, gamesand Through

screensavers. the interaction,it is hoped that
young road users may understandand assimilatesome important safety

conceptandrules.
Lists of related researchresults and publications are also a part of thesesites.
They may use thesefor publicity, educationand further research.
The informationon both websitesis intendedto help road usersrealizethe
risks on the road, provide information aboutwhat they shouldand should

not do for their own safety,how they can reducepotentialrisks during the
journey."THINK! is aboutsavinglives. We campaignall year roundto get
peopleto think moreaboutroad safety,whetheryou walk, drive or ride. As

part of the Government's
strategy,
our aim is to cut the numberof deaths
and
serious injuries from road accidents by 40% by 2010."

(http://www. thinkroadsafety.gov.uký
These two websites provide lots of different kinds of accident information
for different road users. In order to improve safety awarenessand attitudes,
the most important factor for the two websites is that they can attract people
to visit them. People should be interestedin the information provided by the
websites. From our focus group discussion, which is described in detail in
40
chapter 7, people are not interestedin theseaggregatedstatistics and advices.

People are only interested in the information that relates to them. For
example, one student said "when I go to a new town, the accident

information would help me decide where I should drive carefully".
For those who did visit the websites,we need meansof assessingrisk
perceptionto find out whetherthe informationprovidedby the websiteshas

impacton their safetyawareness.
3.4 Risk Perception Evaluation
Risk perception is "one's opinion of the likelihood of risk associatedwith
performing a certain activity or choosing a certain lifestyle" [83]. Our
evaluation used a questionnairethat amalgamatedelementsof three existing
road accidentsrisk perception forms [92,69, and 57].
A campaign to promote safe driving behaviour has been carried out among
adolescentsin two Nordic counties since 1998. It focused on traffic accident
risk perception. Rundmo and Iversen present the results of the evaluation of
the effect of the campaign.They use this to examine the associationbetween
risk perception and traffic behaviour [92]. The respondentsperceived the
risk to be higher after the campaign than before. The respondentsof the
post-sample of the experimental group also reported less 'risky behaviour'

in traffic. The questionnaire measured cognition-based as well as
emotion-based risk perception. Cognition-based risk perception is about
personal belief, such as assessmentof the probability of an accident.

Emotion-basedrisk perception is about the senseof feelings, such as worry
41
and insecurity. Model tests showed that assessmentsof the probability of

traffic accidents and concern were non-significant predictors for self-report
risk behaviour. In contrast, worry and other emotional reactions related to
traffic hazardssignificantly predicted behaviour.
A similar study was carried out by Rundmo and tried to integratetwo
researchtraditions,the personalitytrait approachand the social cognition

approach,to explain individual differencesin risky driving behaviourand
traffic accidentinvolvement[69]. The cognitionapproachstudiedvariables
such as managementof attentionand information processingcapabilities.

Personalityapproachhas focusedupon the predictivevalue of personality
traits,suchasaggression,anxiety.The studywasbasedon a self-completion
questionnaire survey. The questionnaire included measures of risk

perception,attitudestowardstraffic safety and self-reportedrisk-taking in
traffic. The resultssuggestedthat the relationbetweenthe personalitytraits
andrisky driving behaviourwas mediatedthroughattitudes.On this basisit

was concludedthat personalityprimarily influencesrisky driving behaviour
indirectlythroughaffectingthe attitudinaldeterminants
of thebehaviour.
3.4.1 Evaluation Method
Basedon the aboveresearch,our resultinginstrumentaskeda total of nine
questions.Thesequestionsrelateto both cognition-based

andemotion-based
risk perception.We did not include any personalityquestionsbecauseour
purpose
research is to find out the relationshipbetweenaccidentinformation
and risk perception,not the relationship betweenpersonality and risk
A
perception. ten-pointbipolar evaluationscalewas appliedfor measuring
all typesof risk The
perception. questionsareshownbelow.
I Do you feel unsafe that you yourself could be injured in a
42
traffic accident?Answers are scalesfrom I to 10.

Safe 123456789 10 unsafe
2 Are you worried for yourself being injured in a traffic accident?
Answers are scalesfrom I to 10.
Not worried 123456789 10 worried
3 How probabledo you think it is for yourself to be injured in a
traffic accident?Answersarescalesfrom I to 10.

Unlikely 123456789 10 likely
4 How concernedare you about traffic risks and are thinking that
you yourself could be victimized? Answers are scalesfrom 1 to

10.
Not concerned 123456789 10
concerned
5 Please rate the equity of this risk, in terms of whether those
who receive the benefits are the same people who carry the
risks. Answers are scalesfrom 1 to 10.

Not equitable 123456789 10
equitable
6 Please rate how much the risk of being in a road traffic
accident affects you. Answers are scalesfrom Ito 10.

Not affect me 123456789 10 affect
me
7 Pleaserate the extentto which this risk is chosenvoluntarily

by the peopleaffected.Answersarescalesfrom I to 10.
43
Involuntary 123456789 10 voluntary
8 Please rate the immediacy of this risk, in terms of how soon
possible harmful effects may occur. Answers are scalesfrom 1

to 10.
Delayed 123456789 10 immediate
9 Pleaserate how controllablethis risk is. Answersare scales

from I to 10.
Uncontrollable 123456789 10
controllable
The first four questionsare aboutparticipants'generalattitudeand concern
for road traffic accidents.Questionfive can be used to find out whether
peoplethink thosewho carry the risks also benefit most from the road use
behaviour. Question seven asks whether people take risks voluntarily.
Similarly,questionnine asksif this kind of risk is controllable.
Evaluation Goal:
To determine whether road safety websites have a big impact on people's
safety awareness?
Evaluation Method:
(above)
Risk perceptionquestionnaire
Evaluation Participants and the procedure:
The participants were drawn from a sample of road users in and around the
Glasgow area. They included drivers, pedestrians and cyclists. 26
participants did the evaluation in three stages.First, participants were asked
to fill in the risk perception questionnaire. Then they were encouragedto
44
browse two road safety web sites for about 10 to 15 minutes. These were the
'ThinkV campaign site coordinated by the UK Department of Transport and
the web site for the Scottish Road Safety Campaign, mentioned in previous
paragraphs. During these 15 minutes, they can choose the information
whatever they like to see. There has no specific task. We then ask them to
fill out the risk perception questionnaire again and using T test to analysis
the results to find out whether their exposure to this information had any
short-term impact on their expressedattitudes towards the risks associated
with road use. We decided not to assess the relative strengths and
of
weaknesses thesetwo sites becausethey arguably presentcomplementary
information. We also restricted our evaluation to a relatively small
sample because this base-line study was partly intended to identify any
potential problems with the existing web sites and partly also to establish
the utility of psychometric techniques as means of assessingthe impact of
theseinformation sources.
3.4.2 Evaluation Results and Discussion
Table 3.1 shows the summary of statistical analysis (t-test) of all the
questions. The results show some interesting points. First, the sites seem to
have a very small impact on people's safety awareness. Many participants
did not alter their assessment of road accident risks. Those who have
different answers before and after exploring the websites only changed their
ratings by I or 2 points. Secondly, the results show there are strong
individual differences in the manner in which we perceive the risks of
involvement in road traffic accidents. Some people think that the risks of
road traffic are controllable, others think the opposite. Thirdly, some
participants have difficulty in interpreting some of the questions. They felt it
was hard to apply them to their own experience of road usage. We decided
to use existing psychometric instruments to minimize the possibility of such
45
influences.
Questions Mean Mean Sdev Sdev P value

(before) (after) (before) (after) (the two group
haveno
difference)
FeelSafe/unsafe 5.46 5.96 2.50 2.44 0.96
Worried/notworried 5.54 6.15 2.61 2.44 0.99

injured
Unlikely/likely 4.85 5.23 2.62 2.3 0.775
Notconccmed/Concemed 5.5 6.03 2.23 2.36 0.94
Equityof roadrisk/bencfits 5.27 6.08 2.43 2.30 0.929
Notaffectme/Affect
me 6.38 6.73 2.77 2.42 0.8
Involuntary/voluntary 6.69 6.35 2.36 2.01 0.655
Immediacy 6.96 7.12 1.59 1.58 0.56
6.15
Controllable/uncontrollable 5.92 2.46 2.51 0.52
Table 3.1 Statistical analysis of the evaluation questions
Figure 3.8 gives the results from question 1, how safe do you feel in terms
of the likelihood that you could be injured in a road traffic accident?Before
exploring the websites, there are 12 answersare higher than 6. After seeing
the websites, 11 people have answers higher than 6. Nearly half of the
participants (12 out of 26) felt less safe after accessing the web sites. 3
participants felt safer. From the mean, we can see there is a very little
change in their feeling about road risks. Similar results can be found in
Figure 3.9, Figure 3.10 and Figure 3.11.
46
10
9
n
8
7
S
I ftII
5
fi
ii Itt
3 H-
2
t t-
0 Before 23456789
ELLI
tO 11 12 13 11
-, MLLL
16) t7 18 19 20 21 22 23 '-'1
0 After Participant:
Figure 3.8 QuestionI: How safe do you feel in terms of the likelihood that
you could be injured in a road traffic accident?
10
9
8
IT * __ nn
7 ----p
6
5
4
iý
IMI-L ilf
i
123456789 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
0 Before
Participants
0 After
Figure 3.9 Question 2: Are you worried about yourself being injured in a
traffic accident?
47
10
()
7 F
6
S ( [Ii-
4
0 Before
123456789
I
I
I.
h 'I
ýl
I
EL'
AM-1-1
10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
PwIlCipants
0 After
Figure 3.10 Question 3: How probable do you think it is for yourself to

be injured in a traffic accident'?
10
9
8
7
6
5
4
3
2
1
I
-- 123156789 10 11 12 13 11 15 16 17 18 19 20 21 2122:1 24 21526
0 Before Participants
0 After
Figure 3.11 Question 4: How concernedare you about traffic risks and are
thinking that you yourself could be victimized?
Question 5 is trying to find out whether people think the person or group
who createsthe risk of a road traffic accident is also the most likely to suffer
any adverse For

consequences. example, a speeding motorist may increase
the risk of an accident to a pedestrian.Most participants found this question

hard to answer.They had to think of what 'equitable' means.They never use
this concept to compare such activities. Therefore, our responseswere very
48
The Impact of Localized Road Accident Intlormation on Road Safety Awareness
different. Figure 3.12 shows the results.
10
9
7 n
6
5
4
I. II
i
IlL
I
L ILI
I
M Before
123456789
11 li
10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
Participants
0 After
Figure 3.12 Question 5: Please rate the equity of this risk, in terms of
whether those who receive the benefits are the sarne people who carry the
risks.
With regard to "affect me or not", "voluntary" and "immediacy",
participants have relatively same judgments. They think people who take
risks do so voluntarily. People also think the han-nful effect may occur very
soon after taking the risks. Figure 3.13, Figure 3.14 and Figure 3.15 shows
the results.
10
9
8
7
1
111 I
ýM 123456789 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
-- --
-Before
Particoants
0 After
Figure 3.13 Question 6: Pleaserate how much the risk of being in a road
traffic accident affects you.
49
10
9
8
7
6
5
Zý
I
?
123456789 10 11 12 13 H 15 16 17 18 19 20 21 22 23 24 26
0 Before
Participants
0 After
Figure 3.14 Question 7: Pleaserate the extent to which this risk is chosen
voluntarily by the people affected.
10
9
8
7
6
5
4
i- Before
Before
0 After
2
I
1234 ;-) 6789
11 10 11 12 13 14 15 16 17 18 19 20 21 22 23 2,125 26
Pwticoants
Figure 3.15 Question 8: Pleaserate the immediacy of this risk, in terms of

how soon possible harmful effects may occur
As seen from Figure 3.16, which shows the results from question 9, several
participants (9 out of 26) felt that the risk of involvement in a road traffic
accident was more controllable after studying the on-line road safety
campaigns. Nearly the same amount of people felt the opposite; to them,
the risk seemed less controllable after exposure to the web sites.
50
10 -F
9
8
7
6
if111
4 1, fit
3 I f
-- -123456789
0 Before
I
IL I
10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
Partupants
0 After
Figure 3.16 Question 9: Pleaserate how controllable this risk is.
3.5 Conclusion
In conclusion, this lack of consensusraises a number of concernsabout the
use of psychometric risk assessmentand about the effectivenessof on-line
information sourcesin this area. One interpretation is

of our results that the
information and representationof the information on these two websites do
not have significant impact on people's safety Therefore,

awareness. as we
indicated before, we devised an alternative accident information system
which provides localized accident infori-nation to the general public. This is

describedin chapter 4 and chapter 5.
Another possible reason is, during the evaluation, we did not provide
specific tasks for participants to do. At first, we thought that different road
users may care about different road safety information. If we provide tasks
which are not interested by participants, they would ignore what they saw.
Therefore, we give them freedom to browse whatever they like to know
about these websites. But this method is not ideal either. Without specific
51
tasks, participants could be lost in this massive safety information. Future
researchcould use the way combine tasks and free browsing.
The other possible reason for the apparent lack of consensusmay be due to
problems with the psychometric approach. Participants have no problem to
answer emotion-basedrisk perception questions,such as concern and worry.

But they found it difficult to answer the cognition-based questions, such as
equity and voluntary. That is because they have never think about those
questions before and hardly apply them to their own experience of road
usage. We did not expect this because all the questions come from other
researches which they use these questions successfully to monitor the
changes of safety awareness.In chapter 7 and 8, we discuss alternative
evaluation techniqueswhich could help us to measuremore accurately the
effect of accident information on people's safety awareness. Those
alternatives include focus group discussion and road planning experiments.
52
Chapter 4
Data Sources and Requirement Analysis
This chapteris basedon a joint paper [103] with C.W Johnsonand all
here
sections arebasedon my contribution.
4.1 Data Sources for Road Accident Information System
The opening sections of this thesis argued that 'attribution error' is a
significant problem for road safety campaigns.Individuals feel that adverse
events are less likely to occur to them than they are to their peers. One way
of addressingthis bias is to provide information about those accidents and

incidents that occur in their neighborhood rather than simply providing
to
access national and regional aggregatestatistics. The UK Data Archives
provide an important source for information about individual road accidents

[96]. In contrast, to the UK Department of Transport and Scottish Executive
sites this source is not primarily intended to support public accessto road
safety information. It provides a more focused resource for 'raw' data on
road accidents from 1991 to 2004. Each entry in the UK Data Archive has
three sections. The first acts as a meta-record for information about each
accident. The other records can be used to store information about the
vehicles involved and any casualties. The resulting data structures are
capable of storing a wide range of incident data. For example, there are 30
fields in the meta-level accident record. These include timing information, a
police force reference code, and general information about the vehicles and
whether or not pedestrianswere involved. Figure 4.1 provides an overview
of the accident fields.
53
Accident Records
Variable Character Integer Variable Label

Position Wpha
ACCYR 1-4 (1) Accident Year (IrYYY)

ACCREF 5-13 (A) Accident Ref No.
12 20-21 (1) Police Force Code
A3 22 (1) Accident Severity
1.5 23-25 (1) No. of Vehicles
1.6 26-28 (1) No. of casualties
ACCDAY 29-30 (1) Accident Day
ACCMTH 31-32 (1) Accident Month
A7 33 (1) Day of Week
ASH 34-35 (1) Hour of Accident
ASM 36-37 (1) Minute of Accident
1.10 38-40 (1) Local Authority
AID 41-45 (1) Location - Easting
All 46-50 (1) Location - Nortbing
1.12 51 (1) 1st Road Class
1.13 52-55 (1) 1st Road Number
1.14 56 (1) Road Type
1.15 57-59 (1) Speed Limit
1.16 60-61 (1) Junction Detail
1.17 62 (1) Junction Control
1.18 63 (1) 2nd Road Class
1.19 64-67 (1) 2nd Road Number
1.20A 68 (1) Pedestrian Crossing - Hutrian Control
1.2013 69 (1) Pedestrian Crossing- Physical Faci lities
1.21 70 (1) Light Conditions
1.22 71 (1) Weodber Conditions
1.23 72 (1) Road Surface Conditions
1.24 73 (1) Special Conditions at Site
125 74 (1) Carriageway Hazards
1.26 75 (1) Place Accident Reported
Figure 4.1 Fonnat for Accident Recordsin UK Road Traffic Accident
Data
The vehicle and casualtyrecord sectionscan be usedto provide additional

information about each road traffic accident.24 fields are provided for
vehicle specific data.Theseinclude informationaboutwhetherthe vehicle

It
skiddedor overturned. can also recordwhetherit was towing anythingat
the time of the incident.This sectionalso providesmeansof denotingthe
compassdirection from which the to

vehicle was coming and which it
moved during the adverseevent.Vehicle data includesthe age and sex of
the driver. It alsorecordswhetherit wasa 'hit andrun' incidentandwhether
or not a breathtest was to

conducted detectalcohol Similarly,
consumption.
the casualtyrecord provides 16 fields that capturethe 'severity' of the
casualty,whetherthey were a pedestrian

or a on
passenger a bus or coach,
they
whether werewearinga safetybelt etc.
Figure 4.2 illustrates the raw fonnat used by the UK Data Archive. The first
54
The Impact of Localized Road Accident Information on Road Safety Awarcne. v,
field represents the year of the accident. The second provides the accident
reference number that acts as a key to the associated vehicle and casualty
records and so on. The key issue here is that the archive provides a valuable
resource for researchers and for road safety managers in local government.
However, additional tools must be used to analyze this data before it can
inforrn the work of these professionals far less provide usable information to
the general public.
I -I oI x1l
File Edit Format View Help
ACCYR A,-ý7REF AI-2 A3 Al-S AI-6 ACCDAY ACCMTH A7 A8H ARM Aij 0
2001 97ABOO102 97 31132 7 12 10
2001 97ABOO103 97 31123 6 11 30
2001 97ABOO111 97 2111 11 5 3 35
2001 97ABOO112 97 3211 12 7 22 30
2001 97ABOO201 97 31231 4 15 37
2001 97ABOO204 97 31114 1 1 40
2001 97ABOO205 97 31135 5 18 0
2001 97ABOO206 97 31126 7 15 45
2001 97ABOO209 97 31129 1 14 15
2001 97ABOO301 97 32151 6 17 5
2001 97ABOO303 97 31133 7 11 30
2001 97ABOO307 97 32227 2 19 20
2001 97ABOO308 97 32138 6 16 55
2001 97ABOO309 97 31129 1 23 5
2001 97ABOO311 97 3113 11 7 0 20
2001 97ABOO407 97 31147 4 15 20
2001 97ABOO411 97 3113 11 7 16 0
2001 97ABOO601 97 31281 2 15 5
2001 97ABOO602 97 22172 4 17 50
2001 97ABOO603 97 32143 1 17 50
2001 97ABOO606 97 32226 7 16 0
2001 97ABOO612 97 322 10 12 2 11 5
2001 97ABOD803 97 31153 2 13 20
2001 97ABOO805 97 32145 6 18 30
2001 97ABOO806 97 31176 5 17 20
2001 97ABOO608 97 32116 4 17
Figure 4.2 Sample Accident Information from the UK Data Archive
4.2 System Development Process
The UK data archive was used to drive our infori-nation systern. The
development process which followed the iterative process is shown in the
following figure.
55
The Impact of LocaUed Road Accident Information on Road Safety Awareness
Requirements
Analysis and
Definition Systemand
Software
Design Implementation I
IAI and Unit Testing
Integration
and System
I Opcmtion
and
maintcnancc
Figure 4.3 Iterative process
The next section of this chapter focuses on the first phase - requirements
analysis and definition.
4.3 Requirement Analysis
"The processof establishingthe servicesthe systemshouldprovideandthe
constraints under which it must operate is called Requirements

"[40]
Engineering.
As the first stage, capturing accuraterequirements is highly critical for the
whole process. Since developing software systemscan be both complex and
time-consuming, if developers, at the end of the process, find what they
have done is not the product the customeractually wants or needs,they have
56
to spendmore time and money in rebuilding the system.
For future development,a documentwhich explains clearly all of the
requirementsthat have to be fulfilled by the systemis needed.This stage

consists of three major steps: requirements capture and analysis,
requirementsdefinition,andrequirementsspecification.Theserequirements
canbe eitherfunctional("It describesa systemserviceor function" [40]) or
non-functional ("It is a constraint placed on the system or on the
developmentprocess"[40]).
A requirements definition normally uses natural language plus diagrams to
describe "what services the system is expected to provide and the
constraints under which it must operate."[40] A requirements specification

is usually a structured list which describes the services of the system in
detail.
A key aim of the final systemwas to increaseawarenessabout the causesof
road traffic accidents. Therefore, a two-step approach was adopted. In the

first step, we contacted existing road traffic 'experts' since they have more
knowledge about what constitutes more important accident information.
After developing a system prototype according to these experts' advice, in
the second step, we validated key design issues with a wider audience,
gathered more requirements from them and redesigned the system (See
chapter 6.1.2 and chapter 7.2).
The initial requirements elicitation focused on the local police force. They
provide the immediate responsein the aftermath of a road traffic accident.

The formal elicitation also focused on two agencies with a longer term
responsibility for road traffic safety: the Land Service Department in

Glasgow City Council and Amey Highways Ltd. These two organizations
57
The Impact of Localized Road Accident Infonnation on Road Safety Awareness
have responsibility for safety on minor and major roads respectively. They
also have experience of using existing geographical information tools to
review road traffic accident data. This is important becauseone of the key
concerns that we had at this early stage in the project was that the existing
forms of interaction with map-based systems were largely through SQL
interfaces or through a form of spreadsheetinteraction. Neither of thesewas
consideredsuitable for our more general user group.
We, therefore,developeda system prototype using direct manipulation
techniquesto accessincident information from the UK Data Archive

througha local mapbasedinterface.This designis describedin moredetail
sections.For now, it is sufficientto observethat we basedour
in subsequent
requirementsanalysisarounda two-part questionnaire(SeeAppendix D).

SectionI elicitedbackgroundinformationaboutthe participants.Section2
focusedon requirementsfor the system.We were particularlyconcernedto
identify the causalinformation that they felt was more important for the
potentialend-usersof the system.Also, basedon the observationof existing

road safety information system (in chapter 3), we were interestedin
functions and features which can be used to present the information
effectively. As this
mentioned, use of 'expert' advice an
represents initial
compromisebefore subsequentvalidation with a wider class of potential

We
end-users. alsoconducted
an initial studyof existingweb basedsystems,
is in
which mentioned chapter3.
The initial part of the questionnaire revealed the depth of the respondents'
expertise. There are 8 experts that answered the questionnaire. Six of the
analysts had spent more than five years working with road traffic data. One
had less than a year experienceand another had between one and three years.
All of them used a computer in their everyday activities. Sevenrespondents
stated that their main task was the analysis of historical data about road
58
The Impact of Localized Road Accident Infonnation on l0ad Sately Awareness
traffic accidents. One respondent reported that their occupation was
centered on preparing reports for higher (regional) government so that it
could perform further analysis of the data. Question I asked ý111014t

their Job
title, the results are as follows.
" Accident Investigation & Protection (A. I. P) Officer
" Project Manager--- A. I. P & Development Control
" Road Safety Engineer (3 persons)
" Road Planning Engineer (2 persons)
" Technician
Figure 4.4 shows results from other questions in this section.
working period
6
5
Number of 4
PersonS 3
2
I
0 II
less than a year 1-3 years 3-5 years more than 5
years
Question 2: How long have you worked at this job?
59
The Impact of Localized Road Accident Information on Road Saf'ety Awareness
8
7
6
5
2
I
II
Record new data Classify and Analysis historical other
subrnit to higher data
government
Question 3: How do you deal with accident data?
Frequency
10
8
6
Persons
4
2
almost everday once a week few times a severaltimes a

month year
Question 4: How often do you deal with accident data?
Figure 4.4 Some results of experts background
The secondpart of the questionnairerevealed a number of requirementsthat
the road safety professionals felt to be important in the successof any tool
that would provide local accident infori-nation to a wider audience. Three
main types of questionsare asked in this part, which are:
0 What kinds of accident information do you think are important and
should be displayed in the map-based system? (choose top 5 answers

from all the columns in the accident inforination)
60
" What information do you think necessaryto be incorporated in prepared
queries to accessaccident data?(such as by time of the day, by accident
severity, by road surface condition, and so on)

" Which features in the system do you think should be editable by end
users? (such as road color and width, accident point size and color,
background color, and so on)
The reasonwhy we ask road safety experts about information presentationis
that they are very familiar with existing road accident systems.They know
what feature they like, what features are useful for them to identify
information quickly. Through the questionnaire and discussion, we found
out that firstly, their experiencewith the existing information sourcesmade

them advocate the use of geographical interaction techniques, for instance
based on road maps, rather than spreadsheetsor traditional histograms. It
was also argued that users should be able to distinguish between different
types of roads. As mentioned accident frequencies typically correlate with
traffic flow and this, in turn, can differ between motorways and UK
A B
category and roads. The requirementselicitation exercisealso identified
the need to filter the data given the large number of low severity incidents in
the UK data set. One means of doing this would be to initially provide the
users with access to the most recent data but with the option to provide
additional years as required.
The open-endedquestions,suchas"If you have alreadyusedsomesimilar
could
system, you point out somelimitations of that systemor the features
that you wish to improved?", failed to identify specific information

from
requirements amongthe massof data providedby the UK archives.
Insteadit was felt that potential end users should have some meansof
all
accessing of the availableinformation in the threerecordsdescribedin
the previous section. As we shall see, however, initial field trials later
identifiedthe needto tailor this massof datato supportparticulargroupsof
61
end users. The road safety professionals also advocated the use of prepared
queries. They argued that it can be difficult for many users of existing
systems to form their own information requests until they have gained a
significant degree of expertise with an application. They said "Prepared
queries can make their tasks easy and quicV. This requirement raised the
difficulty of determining what queries to provide for novice users. Some
seemed relatively uncontroversial, such as a request to display all fatal
accidents. Beyond this, it was less clear whether the general public might
want, for example, to display all accidents that happened on a Monday or

those adverse events involving casualties over 60 years old. Initial focus
group discussions revealed that these queries were important to particular

individuals but it is less clear that they will have a wider significance for the
potential user population.
The following is a list of requirements identified from the questionnaireand
discussion.
Functional requirements:
1) Users can seea Glasgow local road map
2) Users can distinguish different types of road (motorway, A, B...
3) Users can seeaccidentson the road map
4) Users can get brief information for each selectedaccident
5) Users can seeall detailed information for each selectedaccident
6) Users can searchfor specific accidentsby using preparedqueries
7) Users can manipulate the map, such as zoom in and out, move, reset,
and so on
8) Users can get a list of accidentsaccording to query results
9) Users can see tooltips for road and accident to show basic information,
such as road name and accident reference

10) Users can choosewhich type of road they want to see
11) User selectedroad section or accident will be highlighted.
62
12) Users can set up which infonnation they want to see for each selected
accident
Non-functional requirements:
1) The system should be web based application so that users could easily
access.
2) Using sensible words to describe the filter function or prepared queries
so that users should easily understand them without having the

knowledge of SQL language.
3) The query should be quick and without too much delay, otherwise,
users should get feedbackof the query progress.

4) Use reasonablefonts and layout to avoid infon-nationoverlapping.
5) Show the accident information in sensible words other than predefined
codesand types.
6) Users should always know where they are and can go back to previous
step.
4.4 Conclusion
In chapter 3, we evaluatedthe existing road safety websites. We found that
the information and the representationof the information in those sites have
little influence on the public's safety awareness.In this chapter, I described
our concept a localized road accident information system which might have
better effects. Three things have been done. First, we gathered 'raw'data on
road accidents from UK Data Archive which is essential for our system
development.Secondly,we consulted and discussedwith severalroad safety
experts. They provided useful advices on the accident information and their
experiencesabout the usage of existing road accident information system.

63
Thirdly, we elicited a list of requirements which would guide us to develop
the system. We finally chose the map-basedinteraction system to represent
the massive accident data becausewe believe this is an easy way for the
general public to understand such information. Actually, there are many
different ways to represent accident data, and the next chapter discusses
some of them. In addition, the next chapter discusses information
visualization techniqueswhich could be used in our system.
64
Chapter 5
Theories of Visualizing Multivariate Data
As we arguedin previouschapters,we were trying to find an easyway for
the generalpublic to understanda massof road accidentinformationso that

this information would influence their safety awareness.Information
is to Although
visualization a good way meet our requirement. we finally
chosemapbasedvisualization,thereare many otherwaysto visualizesuch

information.In this chapterI discussthesemethodsandexplainwhy we did
not considerthemin our system.The secondpart of this chaptertalks about

techniqueswhich canbe usedin the system.
As William S. Cleveland mentions [100], "Visualization is an approach to
data analysis that stressesa penetrating look at the structure of data". Robert
Spencealso points out that information visualization is a cognitive activity
which can help people get insight into data [76]. There are many different
tools for and theories about visualizing information. Normally, these tools
are categorized by type of data. Spence argues that the difficulty of
designing an interactive display is strongly influenced by the number of
attributes or variables.
Univariate data deals with a single attribute, such as people's height or
weight. Bivariate data and trivariate data are data with two attributes and
three attributes respectively. For example, we would need a bivariate
representation to analyze the relationship between people's height and
weight. In our research,we were concernedwith multivariate data. Traffic
accident data include weather, speed,road geometry features,and so on. The

following pages, therefore, focus on theories and tools for visualizing
65
multivariate data.
5.1 Statistical Graphics for Visualizing Multivariate Data
Statistical graphics is a major way to represent multivariate data. Such
graphics can contain a lot of information. They also can be used to analyse
the relationships between different attributes. These static representationsof
multivariate data be
can used to carry out further analysis of a group of road
accidents which generatedfrom predefined queries. They can also be used
to generate summary reports on certain types of road accident data. The

following are severaltypical statistical graphics.
5.1.1 Multiple-code Plotting Symbols
The problem with visualizing multivariate data is they require
multi-dimensions to fully represent all of the variables. One strategy for
dealing with this is to representthe data objects with symbols or icons that
contain multiple components.Information is then encoded into the features
of the icons themselves. Multiple-code plotting symbols can be used to
obtain graphical representationsof multivariate data. For example, we use
color to code accident time (e.g. green for morning, orange for noon, black
for night), use shapeto code accident severity (e.g. round for slight accident,
square for serious, diamond for fatal accidents), and so on. For example,
figure 5.1 shows two icons to representtwo accidents.The first one (a) is a
green diamond shaped icon and it means this is a fatal accident which
happenedin the morning. The secondone (b) is a black squareshapedicon
it
and meansthis is a serious accident which happenedat night.
66
a b
Figure5.1 Multi-codeplotting symbols
However, there are also some potential weaknesses.One is that when
variablesdivide the data into a larger numberof discrete it

subgroups, is
difficult to use different symbolsto representall the categoriesin a single

display,for example,one accidentdata item containsabout 70 data fields
(30 in accidentrecord,24 in vehicle record,and 16 in casualtyrecord,see
4).
chapter It is very difficult to find symbolswhich can representall these
variables.Another difficulty is severeoverplotting will inhibit the visual
resolutionof the multivariateinformation.That meansthat evenif we find
suitablesymbolsto represent
all accidentvariables,it is very difficult for
to all
users remember the meaningsof attributesin the Therefore,
symbols.
this is not an easy way for the general public to understandaccident
information and it does not meet our requirements.But we can use this
techniquefor displaying certain featuresof accidentsin our system.For
we
example, used different shapesto different
represent accidentseverity
levels.
5.1.2 Histogram Plots
A histogram plot represents each observation as a set of k concatenated
vertical bars [28]. Each bar corresponds to one of the k variables in the
profile. For example, we want to find out whether overloaded vehicles will
have more casualties than non-overloaded vehicles in accidents. We can
draw a histogram plots (Figure 5.2), which includes the number of casualties
67
(c), the number of passengersin the vehicle which had the accident (p), the
maximum number of passengersallowed in the vehicle (mp).

A £ A
6 6 6
4 4 4
2 2 2
00 0
MP pC MP p C MP p C
Vehicle I Vehicle 2 Vehicle3
Figure 5.2 Examples of Histogram Plots
Histogram plots can help people to find out certain relationships between
variables. However, histogram plots are not suitable for visualizing large
amounts of information, especially location based information. Each road
accident has several kinds of data, and they can not be drawn on one single
graphic. But we can use Histogram plots to do some fin-theranalysis only on
certain types of road accident data, for example, use histogram plots to
compare accident rate on each day of a week to find out which day is the
most dangerousin this area.
5.1.3 Star Plots
In a starplot, eachobservationis shownas64a

collectionof k raysemanating
from a centralpoint. Theserays are equally spacedaroundthe point; the
anglebetweenanypair of adjacentrays is therefore360/k degrees.Eachray

is assignedto a particularvariable,and the lengthof the ray is proportional
to the value of the variablefor that observation"[15]. Starplots are more
suitablefor largeprofiles than histogrambecausethoseicons are restricted

to a fairly narrow width so that all n icons can be includedwithin a single
display.Figure 5.3 showsan exampleof star plots. It is aboutautomobile
68
analysis in 1979 in the US. The following are definitions of the 9 rays. I
Price, 2 Mileage (MPG), 3 1978 Repair Record (I = Worst, 5= Best), 4
1977 Repair Record (1 = Worst, 5= Best), 5 Headroom, 6 Rear SeatRoom,
7 Trunk Space,8 Weight, 9 Length. [66]
1979AUTOM091LEANALYSIS
4
2
A-
tj
7a
AMCCONCORD AMCPACGR AMCSPIRrr AU 01mm
AUD1FCX BMW3201 DUCKCENTURY BUCKELECTRA
:::'4...
BUICKL9 SAOR9 BUICKOPCL DUCK RMAL 0U ICK RIV19RA
0 UICK8 KYLALR
K CAD.0 EVILLE CAD.GLOORADO CAD.99VILLS
Figure 5.3 Example of Star Plots
Star plots can representmore variables than histogram plots. They are useful
tools to make a comparison between certain types of road accident. Users
could get more information about the relationships between these variables.
Therefore star plots are suitable for further analyzing after users filtered out
a group of road accidents.For example, users can use star plots to compare
accident patterns on different roads or areas.Each star representsa road or
area. Different rays will represent different accident information, such as
averageage of the driver, averagetraffic volume, and so on.
69
7be Impact of Localized Road Accident Information on Road Safety Awareness
5.1.4 The Scatterplot Matrix
A scatterplot matrix is defined as a square, symmetric table or "matrix" of
bivariate scatterplots[ 18]. This table has k rows and columns, with eachone
to
corresponding a different variable. Each of the table's cells formed by the
intersection of row i and column j contains a scatterplot showing Xi as the
vertical axis variable and Xj as the horizontal axis variable. Because the
scatterplot matrix is symmetric about its diagonal, these named variables
also appearin panel ji, with their vertical and horizontal positions reversed.
The most important feature of the scatterplot matrix is that its panels show
all possible bivariate relationships that exist within the multivariate data
[99]. Figure 5.4 shows an example of the scatterplot matrix. Three variables
are chosen, speed (mph), driver's age (years), and driver's drink frequency
(drinks per week). From this figure, we can see the younger drivers are, the
faster they drive. The younger drivers are, the more frequently they drink.
0 20 40 60
120
Speed
60
(mph)
60
0
Driver's age
40
(yearsold) 20
9
Drinkfrequency
6
(drinksperweek)
3
0
0 60 120 9 12
Figure 5.4 Example of Scatterplot Matrix for Accident data.
70
The scatterplot matrix does have two potential disadvantages.One is the
limitation of the number of variables. That means it is not suitable for
displaying a large number of variables otherwise the table would be too
huge to look at. The other one is it cannot really show a multivariate
structure, because the scatterplot within each panel is constructed
completely independently of the information in any other panel. Figure 5.4
shows all possible bivariate relationships among speed, driver's age and
drink frequency, each panel was drawn without the information in the other
panel. Therefore, Scatterplot Matrix can be used to analyze relationships

between certain accident data.
5.1.5 Conditioning Plots
The conditioning plot can show how a dependent variable is affected by
several other variables, simultaneously. The values of each conditioning
variable are divided into a series of m. intervals. Then, each interval of the
conditioning variable is allocated to one of m separate panels in the
conditioning plot. There are two other variables in each panel which shows
the bivariate scatterplot for them. Each panel only includes the data points
that fall within the interval corresponding to that panel [99]. The variable
that corresponds to the vertical axis in each display panel is called
"dependent variable". The variable that is shown on the horizontal axis of
eachpanel is called the "panel variable" [ 100]. Figure 5.5 shows an example
of using conditioning plots to analyze accident information. The table is

divided into different age groups. Then table shows relationships between
the number of years holding licence (years) and driving speed (mph). The
data for this table are not real. This is merely to demonstratehow to use a
conditioning plot to analyze accident information. From the figure, we can
71
The Impact of Localized Road Accident Inforniation on Road Safety Awareness
see that although the younger drivers (age<20) have fewer years ot'holding
a licence than the older drivers (agc>60), they more likely drive faster than
old drivers. However, in each age group, those who have more expericnce
(more years of holding a driving licence) are among the people who like to
drive fast.
Specd
() 200
20
Age <20 Age: 20-30 Age: 30-40
0
20
Agc: 40-50 Age: SO-60 Age >60
0 200 0 200
Figure 5.5 Example of conditioning plot for accident information
The most important feature of a conditioning plot is that it shows
multivariate data in a form that is highly organized through visual inspection.

"They enable a researcher to discern patterns within a complex dataset
without requiring stringent assumptionsabout the nature and details of any
underlying structure that may exist among the data points". [99]
Conditioning plot is suitable for analyzing accident patterns. Our system
could use this function to provide users better understanding of the
relationships between different accident infort-nation,such as above figure.
72
5.2 Interactive Information Display Theories
Robert Spence argues that [76], in the pre-computer age, the author of' an
image had to perform selection, representation and presentation according to
his or her understanding of the task to be performed or the inessage to be
conveyed: author and viewer were two different people (Figure 5.6). Now,
with the availability of powerful computers (Figure 5.7), interactive control
by a user, who is thereby to some extent the author of external izations, can
influence all these activities within a freedom defined by the architect of the
visualization tool.
Figure 5.6 Pre-computer creation of a visualization tool
73
Figure 5.7 The creation and use of a computer-based visualization tool
The statistical graphics discussedin section5.1 are all static representations
of multivariate data. In other words, they are pre-cornputer creations of
visualization tools. They are not suitable for displaying location based
infori-nation. But we could use them as additional functions of the system to
show certain features of accident infori-nation. In the light of the
requirements, our preferred system was a map-based interactive
visualization tool so that users can choose what kind of information they
want to display.
There are many theories available for the visualization tool designer.
Generally, interaction techniques can be divided into three categories:
interacting with data transformations; interacting with visual mappings;
interacting with view transfori-nations. Each category makes use of several
techniques. Table 5.1 lists these techniques.
74
Modifles data Modifies visual Modifles view

transformation mappings transformation
Dynamic queries Dataflow Direct selection
Direct walk Pivot tables Cameramovement
Details-on-demand etc. Magic lens

Attribute walk Overview + detail
Brushing Zoom
Direct Manipulation etc.
etc.
Table 5.1 Interaction Techniques 186]
"A dynamic query involves the interactive control by a user of visual query
parametersthat generatea rapid (100 ms update), animated, visual display
of databasesearch results" [86]. In particular, sliders or radio buttons are
used to select value ranges for variables in the Data Table. The cases for
which all the variables fall within the specified ranges are displayed. The
other cases are hidden. For example, in road accident analysis, this
technique could help users to limit the display of accidents according to
specified conditions which are set up in dynamic queries.
Direct walk proceedsby a set of linkages from case to case [14]. A web
browser is an example of direct walk. Users can link from page to page.
Through a series of clicks on visualizations, users can search for
information or modify it. In our system, we could use direct walk to link
accidents which are displayed in the map to the detailed information about
each accident. That means when users click one accident spot on the map,
the systemcan show them relevant information.
Details-on-dernand expands a small set of objects to reveal more of their
75
variables. It allows more of the variables of the case to be mapped to the
visualization. This technique is similar to direct walk. Users' request for
detailed information of a particular accident can be considered as a
details-on-demand.
Attribute walk is where users select some case and then search for other
cases with similar attributes. The technique was developed in a system
called Rabbit for searchingdatabases[93]. When analyzing accidents,users
can use attribute walk to find certain accident patterns. But this technique is
not suitable for our system because the general public, our targeted end
users, don't have enough knowledge to pick up useful attributes. Therefore,
the pattern which they find out would not be meaningful for them.
Brushing is used with multiple visualizations of the same objects [19].
Highlighting a casefrom the Data Table in one view selectsthe samecasein
the other views. For the same reason discussedabove, this technique is not
suitable for our system.
Direct manipulation can be used to modify transformations. For example,
Hendon and Myer at Brown University [36] have developed31) widgets that
can be embedded in visualizations, allowing multiple parameters to be
In
adjusted. our system, this technique can be to
used set up the display of
accidents and roads. For example, users can set up the color and the width
of roads in order to seethem clearly.
Dataflow is a common technique used by commercial visualization systems
to map data to visual form. The basic idea is to use an explicit representation
such as node-link diagrams to represent the mapping. Our system is to
76
display accidents on a local road map. Dataflow is not suitable for such a
system.
A pivot table is a technique found in modem spreadsheetprograms that lets
the user rapidly manipulate the mapping of data to the rows and columns of
They
a spreadsheet. representa domain-specific example of interacting with
visual mappings. Our system is not a spreadsheetbased system. Therefore,
we rejected this technique.
Direct selection refers to the set of schemesthat have evolved for selecting
and highlighting objects and groups of objects. They enhance the
appearanceof a Visual Structure in some way, often to identify the set of
objects that will be the argumentsto some action. This technique is similar
to direct walk and detail-on-demand. We can use this technique to display
accident information when users select their target accident(s).
Camera movement is the change of position of the observer,especially in
3D space. For example, information landscapes may sometimes benefit
from allowing the user to view them from another angle, avoiding
occlusions. Swinging the camera around could reveal parts of the
visualization not otherwise visible. We rejected this technique since our
systemis not a 3D representationof accidents.
Magic lenses select objects according to the X, Y position of their marks,
and then apply fin-ther selection techniques,such as dynamic queries [10]. In
addition, they can apply data or view transformations to the items selected.
Becausemultiple lensescan be placed atop each other, they can be used to
createmore complex Boolean queries.
77
Overview + detail use two or more levels of linked visualizations [7 1]. One
visualization displays either all of the objects or at least some visual

framework that spansall of the objects, such as a map of the world or key
nodes in a tree. Another window shows a more detailed view of the object.
The nodesin the detail view are marked as a region that can be moved in the
overview. This technique is useful for a map-basedsystem, especially when

the map is very large and users may feel lost while they look into small
areas.
Zooming involves reducing the number of objects that are visible, but
possibly increasing the number of variables per object that are shown. In
addition, less compressedtechniquesmay be used to view the objects. When
analyzing accidents, this technique can help users to focus on a particular
area on the map, such as a junction or city center. Especially when the area
involves many accidents,zoom in can help usersto identify eachaccident.
5.3 Conclusion
This chapter reviews two major categories of information representation,
static statistical graphics and interactive information presentationtechniques.

Both can provide us with insights into the road accidents. Interactive
techniques could help users to find certain types of accidents quickly and
statistical graphics can help users to analyze the data and find the
relationships between different variables in multivariate data. However,
since the general publics are our end users, they may find it difficult to
interpret statistical graphics since they don't have much professional
knowledge about statistics. Moreover, the requirements from chapter 4
suggestthat our system does not need complex statistical tools. Therefore,
78
interactive techniquesare the main tools in our system. What we are trying
to do is to provide the general public with a tool which can allow them to
accesslocal road accident information easily and efficiently. Based on the
requirements from chapter 4 and the available techniques which discussed

in this chapter, next chapter will discuss the system design and
implementation process.
Chapter 6
79
System Design and Implementation
In chapter4, a numberof key observationsemergedfrom the resultsof our

initial requirementselicitationwith roadsafetyprofessionals.In particular,it
wasclearthat the massof dataavailableaboutroadtraffic incidentswithin a

single city centre created considerableproblems. For example, as
describedin Chapter4, each traffic accident record has three separate
sections.The first actsasa meta-recordfor informationabouteachaccident.

The other records can be used to store information about the vehicles
involved and any casualties.Thereare 30 fields in the meta-levelaccident
record, 24 fields are provided for vehicle specific data and 16 fields for
casualtyrecord.Thus,it is very difficult for usersto find out the information

they need. Furthermore,even if they find the information, most of it is
codedby police and usersneeda checklist to interpretit. In consequence,

many public information campaigns only provide access to highly
aggregatedregional and national statistics. In chapter5, we analysed

severaldifferentways of representingsuchmultivariatedata.We concluded
that staticstatisticalgraphicsare not suitablefor our system.We decidedto
visualisationtechniquesto provideuserswith meansof

usecomputer-based
interactivelyexploringmoredetailedinformationaboutadverseroad events
in their local area.The impactof localizedaccidentinformationon people's
risk perceptionis the main purposeof this Therefore,

research. as long as a
userinterfacemeetsour requirementsin chapter4, suchascandisplaymaps
andaccidents,andprovidepredefinedqueries,it is what we need.Although
it may be not the best,the following evaluationin the chapter7 showsthat
theusabilityof this systemis quite good.
I now turn to the issuesof systemdesign and implementation.
80
6.1 Interface Design
Many different approachesto information visualization were mentioned in
Chapter 5. Both Tufte [25] and Spence [76] have developed theories that
support the visual display of quantitative information. However, we choseto
focus on the more general guidelines provided by Shneiderman'sTTT (Task
by Data Type Taxonomy for Information Visualizations) theory [8].
According to the TTT theory, seven broad categories can be used to
distinguish between the various tasks that users perform with information
visualization systems. Table 6.1 shows the relationships among those task
categories, our system requirements (See chapter 4) and available

techniques.
SystemRequirements Task Categories Techniques/Functions
DisplayGlasgowlocal Overview Zoom out, resetmap
roadmapandroad
accidents
Manipulate the map Zoom Zoom in
Users can searchfor Filter Queries,checkbox,radio
specific accidents box
Users can seeall Click anaccidentspot,users
detailed information Details-on-demand will getall theinfonnation

for each selected aboutthisaccident
accident
n/a Relate n/a

Users should always History Undo
know where they are
and can go back to
81
previous step.
n/a Extract Save,print
Table6.1 Relationshipsbetweentaskcategories,requirements
and
techniques
The first task is to an overview of the entire collection. Overview strategies
include 'zoomed-out' views of each data type to see the entire collection
plus an adjoining detail view. The second set of common tasks with any
visualization involves 'zooming-in' on items of interest. Users typically
have an interest in some portion of a collection. They need tools to
enable them to control the zoom focus. Smooth zooming can helps users
preserve their senseof position and context. A third class of tasks is to
filter out irrelevant information. Our initial requirements elicitation also
confirmed this as an important attribute of any implementation. The
fourth of Shneiderman's items describes tasks that enable users to access
additional details on demand. Once a collection has been trimmed to a few
dozen items it should be easy to browse the details about the group or
individual items. A fifth set of tasks should help users to view
relationships between the items in a data set. The sixth focuses on the
maintenance of a history or log to support undo, replay, and progressive
refinement. Information exploration can involve iterative refinement of an

initial search,retaining the history of actions and allowing users to retrace
their steps is important. The final class of tasks focuses on the extraction
of sub-collections and of the query parameters. Once users have obtained
the item or set of items they desire, it would be useful to be able to extract
that set and save it to a file in a format that would facilitate other uses such
as sending by email, printing, graphing, or insertion into a statistical or
presentationpackage.
These guidelines together with the initial requirements elicitation helped to
82
inforrn the interface design Of Our Visualization system. Our first version
design was based on a number of recent software tools that have been
developed by professional road accident investigators where each incident is
displayed on a map by the location in which it occurred. For example, as
discussed in chapter 1, KeyAccident (See Figure 6.1) is an ALItoCAD(R)
based GIS application. Users use AutoCAD to show the map and use
KeyAccident to select, display and analyse accidents. However,
KeyAccident is only available for road safety experts and engineers. This
system is not intended to be used by the general public.
..................
b
IN * 1A
ijial.... __ln.. IIflfl -'
IZ. -
- -K -X -77 -y-
ILTT 141ö lfl"IV IS IöflICI I Ill III II I£ It 11111111
In chapter 1,1 mentioned how real time accident report systemsalso show
accidents on the map and are available for the general public, such as The
AA website, The MATTISSE websites, and the TxDOT Expresswaywebsite.
Such a system helps people to know what happenedin the immediate past.
For example, in the TxDoT, highway conditions expected to exist for more
than two hours and create unsafe driving conditions (i. e. major accidents)
are posted.
However, this makes it difficult to accessthe risks of a future journey. When
conditions are changed, such information will not be useful. But historical
accident data can provide an insight into certain risk patterns. For instance,
if drivers know how many accidents happenedin fog in the past two years
83
and where, they would raise their attention when they encounter these
conditions. Therefore, in the light of our requirements, we neededa system
which can provide historical accident data and techniqueswhich allow users
to manipulate the display of accident information freely.
6.1.1Initial Design
Based on the observation on the existing road accident system

(KeyAccident)and the discussionwith road safetyexperts,we had drawn
simple
several picturesas our very first prototypeof the The
system. system
will be a mapbasedroad information

accident On
system. the centerof the
interfacewill be a Glasgowroadsmap.Onceusersmadequeries,the system
will display from

accidents the query results.Each will
accident havebrief
informationdisplayedon the right handsideof the system.Whenusersclick
an accident,anotherwindow will be shownwith all detailedinformationof

this accident.This raised a number of important issues;in particular we
were to
concerned ensurethat we could log the positionof eachaccidentas
based
accuratelyas possible, on the data provided in the nationalarchive.
We, therefore,engineeredthe systemusing the 10-meterOrdnanceSurvey
Grid Reference(OSGR) referenceformat rather than the more familiar
conceptsof Longitudeand Latitude. OSGR can be thoughtof as a flat
grid overlaid on Britain. The grid is in

measured metersand the point of
origin is definedto be southwest of the Isles of Scilly. Pointson a mapare
given Eastingand Northing values.These denote
co-ordinates the distance
betweena location and the origin. For example,the centreof Manchester
city is approximately383,000meterseastand 398,000metersnorth of the

origin. This referencefonnat acts as a meansof moving between the
location dataprovidedin the national accidentarchiveand the position of
84
the Geographical Information System that we developed for the centre of

Glasgow.
We decided to focus the development of the system around the area of
Glasgow that falls between the middle of national grid references(OSGR)
NS56 and the middle of NS66. This decision was partly determined by
pragmatics. We wanted to focus our initial evaluation within a precise
geographical location where we had accessto a wide range of potential end
users and to road safety professionals who can validate our findings.
Linking this geographical area back to the national data archive using the
OSGR indexing system yielded a total of 1486 incidents for 2001 and 1471
incidents for 2000. In 2001, those incidents resulted in 3568 vehicle records
and 2731 casualty records; recall that each incident can involve several
vehicles and casualties. In 2000, those incidents resulted in 3649 vehicle
records and 2724 casualty records. The widespread problem of
underreporting led us to exclude damage-only incidents. However, the
available data could be included in subsequentversions of the system given
the relatively modest size of the data files that we have compiled from the
national data sources. In the first version of our system, we created a

databaseby using MySql to store the accident data. When users make a
query, the systemneedsto connect to the databaseand retrieve results of the
query. Because of this design, our system runs slow when users use the
query function. Therefore, in the revised design, we stored all the accident
information in separateJava Object files. These files are generatedby a
small Java application. This application can read one year's accident data
from the database.Then it will create objects from these data. Finally, it
writes these objects into a file. Each time, when the system is loaded, all the
accident information will be loaded into the memory as well from these files.
By doing this, we increase the query response speed.
85
APIM -ON, Controlpanelfor

manipulating the
view of the map.
Tool tip providing

accident identifier of
selected event.
Information panel
showing
contributory
factors for the
selected accident.
Menus and buttons used to control the

filtering of accidents shown on the map.
Figure 6.2: Overview of the first version Map-Based Interface
Figure 6.2 illustrates the resulting interface. Accidents are shown by the dots.
In the interface, these dots are coloured red. The user can select an
individual incident by moving the mouse over one of these dots. A tool
tip then appears with the individual incident identifier to provide feedback
on their selection. If there is no record of an incident under the mouse
location then the tool tip provides contextual information by giving the
name of the nearest road. The system uses these techniques to enable the
user to access 'details on demand'. A summary of the accident
information about each selected incident is presented in a panel on the right
hand side of the screen.
The system enables the user to filter infori-nation by selecting a number of
preformed queries through the menus and buttons at the bottom of the
screen. In particular, it is possible to specify the severity of the incidents
to be displayed. This filters the mass of red dots that would otherwise
overwhelm the user. Our initial requirements analysis with road safety
professionals also identified the need to distinguish between incidents that
86
occur on different categories of roads: A and B roads or motorways.

Users can select the category of roads that they arc interestedin by sclccting
check boxes.
In addition to these filtering features, the design also provides facilities for
users to explore accident statistics over a wider area. The map display
can be panned to examine incidents in adjacent locations. This is
combined with an area based zoom facility. A dragged left-click on the
mouse will zoom into the map at a rate that is proportionate to the area of
the drag. A long and sustaineddrag will result in a larger change than a
short drag. The direction of the zoom is determined by the direction of the
drag. A movement from bottom left to top right zooms into the map
while the opposite movement will zoom out. We have also included a reset
facility so that any actions should be easily reversible back to a known
initial state.
As mentioned, users can exploit the menu options and checkboxesprovided
by the systemto filter the massof incident data held about road accidentsin
the Glasgow area. However, even the successive refinement of these
queries can still yield a significant number of 'hits. It would be
extremely laborious if users had to manually view each set of contributory
factors using the tool tip and panel combination illustrated in Figure 6.2.
For this reason we have implemented a facility to review all of the
information that is associatedwith the currently highlighted set of incidents.
Figure 6.3 illustrates how the user can gradually read through accident,
vehicle and casualty information with each incident presentedon a different
&page'.
87
The Impact of Localized Road Accident Inforniation on Road Sallety Awareness
UTýý X1
76'. 5 -I. t. .. ti. It. Th.. ý. -1101
ý
ploI .I
A.. i d.. t1. - 97YA70910P.1-C. A. 97
10 I"Ofl.. k
19 a-t. 30 L- dA. th.. ,I 93
2M50 lkýtkig 684040 F-tl., Acl, gg 3
7- tjt. al-Lý 74 IL., Hp. 8 Sp.. dL.., t 3U
J-. t. .. D. W 13 J-t -C.. ý .14S... a.. A(3 6
P. &. ýi MC ... 0 Nd.. týi-f . ...... t.. g I
L., ktC .. Jý U I. tkýCMC UII.. As- f... I
Sp ... le-d. i i.. 0 C- ,M.. 71k.. ig 0 rl ... A. d. p. ". d I
-f-ti- . ýtl tk. .. ii-t

............... T'j. d. 1. [. ý* ................
Nekt CAM Page No,I
Figure 6.3: Reviewing Details for Several Selected Incidents
6.1.2 Revised Design
After the system evaluation which is discussedin detail in the next chapter,
we redesignedthe system according to the feedback from end users. Figure

6.4 shows the secondversion of the interface. We made several changesto
the system.First, the new system usesdifferent shapesto representdifferent

levels of accident severity. For example, round means slight, squaremeans
serious, diamond means fatal (the level of accident severity comes from
accident data directly). Other changesinclude additional filtering controls.

In the filtering control panel there are two buttons, 'Do query' and 'All
result information'. When users clicked the 'Do query' button, a new
filtering control dialog appears(See figure 6.5). The system now provides
more filtering functions in the option panel (See figure 6.6). Users can
choosewhich one or more they want to use.
The second version of the interface design also changed the layout and
88
The Impact of Localized Road Accident Information on Road Salcty Awareness
content of the detailed information for each selected accident (See figure
6.7). Feedback from the end users suggested that it is better to display all file
infon-nation for one accident in one page. They also stated that it is not
convenient for them to look up the check list to find out all the meanings of'
the coded information. Therefore, we replaced all the predefined codes and
types by their formal descriptions.

Fl. Edt Vlw F-ot. T.. 11 HeIP
Ik , S-lh F-ký 1ý
11 1-31
IA-
2001 Glasgow Road Accidents Information
The shape of the
dots represents
different severily
m.. w, e 91A010106
,
-0. t. ýM., 1 Filtering control

S-Iry
panel is reduced
into two buttons.
All Resumkdm.. t,. "
[O;j
,
Figure 6.4 Second version of the User Interfaces
89
The Impact of Localized Road Accident Inl'orniation on lWad Safety Awareness
Accident Year
-101 Xi
YedF ý2001
About Accident
,'DayofWeekF-
Severity All All
Abotd Vehicle
Driver Sex All w
AbotA Casualty
CasualtySex "Iml
1 OK II Ca"cel
IJava Applet Window
Figure 6.5 Filtering control dialog
nui rn i 1131
- X1
Accident Data Vehicle Data Casualty Data Prepared Queries
v, Severity v, CastiaRySex - Day(YfWeek ýwýDmmrSex CdSUdltyAge Month ý WeatheF Casualty Class

, I
Towing and AFficulatim
Select All Unselect All OK

fJaya Applet Window
Figure 6.6 Option panel for more filtering functions
90
The Impact of Localized Road Accident Information on Road Salety Awareness
AcciderKDili -911
AccidenlYeal 2001 AccidoitRefeietwo 97AB70610 PoliceCude SlimthOftle
NwherO(Vehicles I Wimbet0fCastiaNtes I AccideMD4v IQ Iltic oncs arc the

DayOWeek Ffiday Hour 22 Kntge it] Gidsom CNY
Easting 258410 Nolthw* 665190 H MoadClass A 812 olumn names.

Roadlype Smoe cafriageway-SpeedLimg 30 JuiichonDelad CIOSSIOddS AtAo 04IN: signal
-hinchmCorgitil
SecurxftadClass Umlassfwd Secm"adNumber ml PedesttimiCiossing No ci ussing m So rl, ýledestild"CFOS%Cg-.
PRNCMI. Ptifillýlý-.
LodCwiddim Darkmss lights M WeAherCwmMiwi [mmttighmnds tWdSuiMcm WO w damp SpciW(ultim N.m
CNna"aOlArwdsNow Plw.eAccReported At -
VehicieData.-No.I 31ack ones are

AccidetIlYedr AccidentReleierce97AD10610 VeliLleflefetence I Goodsow ISI. an lie values of
Articulated
TowintAndArticulal. Yehicle Goinqatioadulber CumpassPoint
VOKWAllanotinme Fromfast CnmP4"p d T" W-f
- he columns.
Onthemainroad
VehicieLocation-Ra. OnmainC'Waynot Junctioni-ocaliticAn
VehicleLocation-Re Notat pinction Skiddirul None
Owntoming
-
None
HdObjeclInCainage Vel"clet-edwingCair
Didnotlean cdrria.- HgObjecl(XfCantag...
None Vehiclel)i
efixtSullix-tinknuom
1stpondoll"IPWI Front OtilmrVehicleHil

Ret-2 CornhinedDimiageNone PjxgI Ind. sideOd Nodamall"
Sexof D(K" male Atie(if Dmier 36 BreathTest Neipative HOandRun Other
Data- No,1
Casuafty
AccidentYear 2001 Acciden(Reference 97AB70610 Vehiclemefence I casuawereme I
Castialtfi,
lass Pedesirmn CasuaftySex male CasualtyAge 33 casuaRyseveloy Fatal
PedestfumiLoCallon CrossRVmpedest-PedestridrltAoVefnerdClOSSHMITOMdINO..Ped"FianDoection Hembngwih SchoolPupd Ottim (Iforn 1994)
Seall3ekUsage(M 0 C& Pdsspnqff NCO

cm pass.. g. 11-OrCoachPass.... NOId bUSOf coýh- Pe-lim
Close
rJ Ap*\, (. d-
Figure 6.7 Details for selected accident
6.2 System Architecture
Since the system is intended for use by the general public, the easier our
system can be accessed, the more impact it can make. Also system help
function or tutorials should provide enough demonstration about system
usage so that users don't need face to face training. Therefore, an Internet
based system would be a good design choice. There are several techniques
can be used to build such a system, such as PHR JavaScript, JavaApplet.
However, PHP and JavaScript are scripting language, especially server-side
scripting, which mainly is used for collecting form data, to generate
dynamic page content, or to send and receive cookies. Geographical
91
Information System (GIS) is another feature of our system. We need Geo,
to
packages edit map and draw accidents on it. Our source for free GIS
packagesis from freegis.org. In that website, PHP based geo,packagesare
very limited. Also, the author has more experience of Java programming
than of PHP and JavaScript. So I choose JavaApplet to build this system.
Figure 6.8 shows the overall architecture.
(::: D
download
Mp s in D)T fonmt
ArcView
Ccnvat artdFAk
Mps in shpMefoamt
functions, such as
display maps, edit
maps, show accidents
InformationSystem Loadinto Accidmts 'N DataSources

(JavaApplet) - (STAT19)
Data File
Figure 6.8 System architecture
GIS applications link locations to information, such as people to addresses,
buildings to parcels, or streetswithin a network. They layer that information
to give you a better understanding of how it all interrelates [82]. GIS
systems have the following features: interactive (requests and feedback),
large information (database),map-based GUI (Graphical User Interface),
92
regular information update (including data structure). The Glasgow Road

Accident Information System is such a system. It will store data on several
thousands of accidents. Users can browse the map of Glasgow, set up
queries and retrieve accident information. Since the accident records are
basedon STAT 19 forms, if the police or government changethe structure of
the STAT19, then our systemneedsto be able to update the information with
minimum effort and cost.
Object-orienteddesign (OOD) and programming provide a suitable

frameworkfor developingsucha system.OOD modelsthe real world into
the softwareobjects and operations.Sanjiv [30] arguesin his book that

OOD has the following benefits: 1) Better mappingbetweenthe business
andsoftware(easyto understand);2) Copingwith change(easyto maintain);

3) Improved productivity and quality through the reuse of existing
components(reusability).Thesefeaturesare all useful in our system.For

when
example, we map the vehicle datainto an objectin the system,all the
items would becomeattributesof the object. If the structureof STAT19is
say
changed, they add one item into the vehicle data, we don't need to
rewrite the whole program.There is an attribution list for eachrecord.We
just needto add one attributeto the list and leavethe rest of the attributes
and methodsunchanged.The detailed classificationof all objects in the

systemwill be in
discussed the following section.
6.3 Class Classification and Implementation
There are a number of proven approaches for identifying objects in the
system [11], such as classical approaches, behaviour analysis, domain
analysis,Use-caseanalysis, CRC (Class, Responsibilities, and Collaborators)

93
The Impact of Localized Road Accident Information on Road Sal'ety Awarencss
cards, informal English Description, and structured analysis. We use
classical approaches plus use-case analysis to identify classes III Our system.
Classical approaches derive primarily from the principles of classical
categorization. From the perspective of database modelling, Ross offers a
list [78]: people, places, things, organizations, concepts and events. In Our
GIS system, things (infon-nation) include accident data, vehicle data, Ind
casualty data. Events include map drawing and queries. The result of a
query may generate a list of accident data, vehicle data or casualty data.
Maps include both the road map layer and the accident layer.
Use-case analysis was first formalized by Jacobson. He defines a use case as
"a particular form or pattern or example of usage, a scenario that begins
with some user of the system initiating sorne transaction or sequence of
interrelated events" [42]. Figure 6.9 shows a use case diagram of our
system.
Figure 6.9 Use Case diagram
94
Using the combined classical approach and use-caseanalysis, we identified
the following classes:Accident Data, Vehicle Data, Casualty Data, Accident

List, Vehicle List, Casualty List, My Layer, My Theme, Result Theme, Road
Accident, Application Manager, and Information Frame. Their relationships
are show in figure 6.10
Infomiaum
frme
Display
I
Accesssystem
settings ies ii
AppManage I-
RoadAccidents , ResWt
Th=e
dý
Accidad
List MyLayer 1...
My Th=e
IVI
I
List
Vehicle Accidmt
Data
1...
Data 0
Vehicle Casualty
List Casualty
Data
Figure 6.10 Relationships between classes
The system currently holds two years accident data (2001 and 2000). When
the applet starts, the system loads all static road maps and accidents
information. This may take a few minutes. The reason for this is that in the
first version of our system, we stored all the accident data in a MySQL
database.The data were not loaded into the systemuntil usersmade a query.
However, the evaluation shows users feel the system is too slow. Therefore,
95
The Impact of Localized Road Accident Information on Road Sallety Awareness
in the second version of our system, we stored all the accidcnt data in
separate Java Object files. Each file has one year's accident (hita. Although
it takes a few minutes to load all the information, after loading, queries are
very fast. In the future, if we need to include more years' accident data or
need a much large geographical area, we need to consider other ways to
balance the system memory limit and the system speed. Figure 6.11 shows
an example of the system flow chart.
Systcm Start
Load Accident Data
Load map
System Settings
Query
v
Update map with
A Zoom, Pan, Select
query results
I
Quit?
No ý
Yes
Figure 6.11 an example of the system flow chart
6.4 Conclusion
Having in mind the requirements listed in chapter 4, we designed and
developedtwo versions of our system. After completing the first version, we
carried out an initial evaluation and made several changes to the system,
96
such as using different shapes to represent different accident severity,
replacing the bottom panel by a more simplified query control panel,
replacing accident codes by the formal descriptions, and so on. The system
was designed based on object-oriented principles which make it easy to
understand and maintain. Through our system, users can see the local
accident information in Glasgow area, can find specific kinds of road
accidents,and can read the detailed accident records for every accident. The
next chapterwill discussthe evaluations of these design decisions.
97
Chapter 7 System Evaluation
Previouschapterhavearguedthat severalbiases,including attributionerror,
result in drivers the

underestimating likelihood that they will be involved in
a roadtraffic accident. It hasalso beenarguedthat previousinitiatives to

improveroad safetymight be supportedif driverswereprovidedwith more
direct informationaboutthe frequencyof accidentswithin their local area
rather than through regional or national aggregatestatistics. We are,
therefore,concernedto determinewhetherusingour applicationwould have
any measurableeffect on an individual's perceptionof their likelihood of

beinginvolvedin a roadaccident.
7.1 Method
Major evaluation goal:
Evaluate whether our system has a significant impact on people's safety
awareness.
Minor evaluation goal:
Evaluate the usability of the system.
Evaluation Methods:
Usability questionnaireand risk perception questionnaire
Evaluation participants and process:
The formal evaluation was conducted in three stages.First, each participant
completed a benchmark risk perception questionnaire. This was based on
98
three existing general road accidentsrisk perception forms, described in the
chapter 3 [92,69,57]. The secondstageprovided users with an opportunity
to interact with the system described in the previous section. We offered
a scripted introduction to the functions offered by the application and then
left them to browse the accidents. Each user was told that they could use
the system for less than half an hour. After they had finished using the
system they were again requested to complete the same risk perception
questionnaire. They were also asked to provide feedback on the user

interface to the tool (SeeAppendix E). We did not provide specific tasks for
participants to do. At first, we thought that different road users may care
about different road safety information. If we provide tasks which are not
interestedby participants, they would ignore what they saw. Therefore, we
give them freedom to browse whatever they like to know about our system.
We started with a sample of 25 potential users, although 29 users did the
usability evaluation (4 users had difficulty completing the risk perception
questionnaire). We did not want to place undue time constraints on the

familiarization period that was available for usersas they learned how to use
the system. Therefore, we used a different sample from that involved in
our risk perception study of existing web sites which was described in
chapter 3.
7.2 Results of Usability Questionnaire
The feedback from open-endedquestionsabout the usability of the software
included several negative comments. These have formed the focus for
our redevelopment efforts that resulted in the revised interface design
described in chapter 6. In retrospect, we realize that we should have
anticipated some of their criticisms earlier in the development cycle. For
99
example, accident records in the UK archive use a number of predefined

types to represent certain values. These types have a relatively clear
meaning to safety professionals. For example, in the severity column, '1'
means fatal accident, '2' means serious accident, and '3' means slight
accident. Our more general users were not, however, familiar with these
distinctions between road types and local authority codes. In
many
consequence, of our participants reported high levels of frustration
whenever they had to crosscheckthe codes in the accident records with an
explanatory table. For example, one participant wrote "It is very difficult to
always refer to the manual to interpret codes". Our first design of user
interface was served as a heuristic prototype to help us to identify the
potential problems. The above problem was identified in the first user
interface evaluation and has been fixed in our revised design of user
interface.
A number of fin-ther criticisms related to significant details in the
presentation of the user interface. For instance, several users complained
that the font used to display the detailed information about each accident
was too small. Others complained that it was difficult to distinguish
between incidents because the associated location markers on the map
began to overlap when they viewed the system at a low level of
gmagnification' or 'zoom' (See figure 7.1). Other common criticisms
echo previous studies of the usability of geographical information systems

by a relatively wide user population. Several people found it difficult to
distinguish between the pan and zoom actions. Others becamelost as they
traversedthe map and requestedthe provision of additional landmarks.
100
- ES "eta 1 'tb p
J; '. -. -
¬ lu
i
Overlaps
BI., NI-- G. - A R-d, 0-ff,. B R.. d, BI-k: 0.1.. R... h

4Ü
1 oý -1
Figure 7.1 Overlaps on the map
A final cluster of adverse comments focused more on the information
content that was provided by the system. One participant wrote "The
cause of accidents is not clear". This is a key observation and there are
several explanations. It is important to stress that the national archive
does not directly record the cause of each adverse event. This is justified
both for legal and ethical reasons. The police's view of the potential causes
might be used in legislation. Hence it is not described in the data archive.
instead, the data focuses on contextual details such as the prevailing weather
conditions and whether or not alcohol was involved. Some police
agencies do retain causal information in a format that can be linked to the
national archive data. However, we could not obtain public access to
these data sources during the initial development of the prototype interface.
Several other factors can explain end-user criticisms about the presentation
of causal information. In order to track the geographical distribution of
common contextual factors users must first select the incidents that they are
interested in, using the map-based display, and then access the individual
records associated with each accident, using the text-based pages Illustrated
in Figure 7.2. Post-evaluation debriefs helped to identify the basis for
101
current redesign where the icon used to represent an incident on the map
will be changedto reflect particular values for contCXtLMIfactors selectedby
each user. This is intended to reduce the need to accessthis massof(Ictailcd
text.
ow"t ZT,;'. ý,, -=id
A-ule, A IMt.
2001 97ABDI301 Plll.. '. ( Od. SI, Ath.: " %
A,:,. -Ilmly-
i Numbe"CastmOlies I Acnd., MDýy 30 A WWW. 01, J. "...
N[IMI)OfOfVellKleS y
DayOrWeek Twsday Hour 17 IM-1.30 lo, IAWI Ry Cmy
i .. t. tj 250MO Nortllm 665350 Mfloadcj... U-1 ... If. d I-M-dN .... b.,, n, al
R.. JT". &noe c-ageway-Spee(JUM 30

F"
j0 DOM O. "'o. d. 0... I A,Ao f, ffi,. iq,. i
SmondRoadClass Umlassdled 5ý-MoadNwnbw nul PedesIllwCaossmill N. --11 so P'. 11k,.
LightCmddton Dwk. - bghe. 10 Weath., C-dd-, F.. - NUh -I. Wut oi dmip 5.4--K In N-o
CmriagewayHazmcts Ný PWeAccReportecl Elsewhwe
VelmleDMa No.1
AccidentYem A-leatRef-ce 97ABO3301 VehicleReforetice I Wlypo Ihn'. c-, 1. jII

V"I
ImmijAMM-slax NolowfarMulaticin VehcleMWýe Twninogright cmipasspoild Ilon, smsil C To L. St
V. h. clot-atiml R.... L. -, VehWIeLocation-Re... On main c'way not... Junctuiril An Idl.

g nl- --I 0-tionAll. .1p, nt In Sk, dd q0w, Iq No. "
HdOb)eOlnCmria9s N- vehýlele"Tlgcfff Osdý I. - -m Hit0l, j. ctoffc-,., j N-
IMP-90fl. 11-( Nearside Ott-WhideHO -RL4...0 C-b,,,. dDm,. jj,, N. - lblofl 1II11t,
I IM- NIIIII. Iwj. ý
S... (D- mabe Alle f D- 40 13,ýMh T.. l N. (Im- IN pII, - (mhe"
Cýsu. ltyDM. l
-N..
A-d. AY.. 20DO kcciderytRefmeme 97ABO3301 V. h. d. Rý4.,. - i
C-myclass PW.. ft- C..,., Rys.. male C .... ItVAUe 46
PodeSUWADCat- Crossngetsewhef PedesItianMýinenICioSsin(Ifrointliý. HeadingsoLdh scl« Oth., (In lq! 14)

-Pe(itýýýtriiýrli)ilooi()II
C. PdssenUer Not cm pas-pm 11-(), ( 13-hPasse

seamemusaueoq7. - 0 -
Not A bus w "Wh C-.,. "Vlvp.
Clow
Fj--.
-Ap. - -
Figure 7.2 Detailed accident information
The evaluation also helped to elicit criticisms that the prepared queries were
too restrictive and that the data provided by the national archive was too
limited. In particular, one participant wrote "(The system) should add
more data regarding the details of accidents, e.g. the mental, physical
conditions of the driver who causes the accident". This last point raises a
number of important issues. It can be difficult for investigators to elicit
personal infori-nation, for instance about the mental state of a driver, in the
aftermath of an adverse event. Informal comments made by the
participants indicated that these sorts of details would have a profound
impact on any personal assessmentabout how likely they were to find
themselvesin a similar situation in the future.
Likert scales were used to assessthe overall 'usability' of the prototype.

102
The Impact of Locaaed RoadAccident Information on Road Safety Awareness
In an initial sample of 29 potential users, nineteen ranked the overall
usability at 4 on a scale from 1, denoting that the system was 'terrible' to 5,
denoting that the systemwas 'wonderful'. Six users scored the system as
3 on this scale while four users ranked the system in the highest category.
Similar results were obtained for a range of additional usability questions,
including the clarity of the interface and the ease with which participants
learned to interact with the system. However, a number of important
caveats were revealed by the quantitative analysis of these Likert scales.
Figure 7.3 presentsthe results of questions about the speed of the system
and about frustration levels. As can be seen, several users found that the
system was too slow, even though the dataset that we used was limited to
approximately 3,000 incidents in Glasgow within a two year period. Part
of this problem stemsfrom the relative easewith which users could generate
queries. Small changes in the area selected using the zoom function
would result in update calls being made to regenerate incident data from
within a MySQL database. In our second version system, we load the
accident information from stored files rather than from the MySQL database.
Although this increasethe loading time since when users open the system,it
will load all available data into the system, the query times are reduced and
the system is more stable than in the previous version. The right hand side
figure shows most people are satisfied with the application. But there is one
participant feel frustrating about the system. The system was crashedafter
this participant executedseveral queries. Same reason as above causedthis
problem. Frequently query the databasecaused the system unstable. After
we redesignedour system,it has not been crashedanymore.
103
20
14 In
a) 18
12 I (n 1G
10 14
12
8
lo
6
C) 8
4
' 0- 4
2 2
U-
riIIfl 0 m r--l
1 (too 3 45 (fast
12 : 5
slow) enough) (frustrating) (satisfying)
Figure 7.3: Quantitative Results from the Usability Evaluation
7.3 Results of Psychometric Study
The following figures present a sample from the results that were obtained
from the second stage of our evaluation. This was based on the same
procedure that had been initially used to sketch the impact that existing road
safety web sites might have upon the potential users which was described in
chapter 3. In this instance, the participants were given a questionnaire
that was based on three previous psychometric studies of road accident risk
perception before they interacted with our map-based visualisation systems.
They were then encouraged to use the prototype for as long as they wanted.
After they completed the more general interface evaluation questions,
described above, they were then asked to repeat the risk perception
questions. As in the previous exarnination of national and regional web
sites, the intention was to determine whether interaction with our

information source had any measurable effect on their risk perception scores.
Figure 7.4 and figure 7.5 present the results from two of the nine questions
before and after using the system.
104
The Impact of Localized Road Accident Information on Road Sat'ety Awarencss
Ql
10
ri
n
11
I]-
I
r
0 A
123456789 10 It 12 13 14 15 16 17 18 19 20 21 22 23 "1 2ý)
Before
Particoarits
0 After
injured in a road traffic accident? (p = 0.98)
Q9
10
0
M Before
EDAfter
123456789
I I I
I
iiI f-
10 11 12 13 14 15 16 17 18 19 20 21 22 2:3 21 25
Participants
Figure 7.5 How controllable is the risk of involvement in a road traffic
accident?(p = 0.98)
Interaction with the new systemincreasedthe perception of the likelihood of
an accident in 12 of the participants. In two participants it and the
remainder showed no change in their expressedperception. However, the
magnitude and frequency of this effect is broadly similar to that observed
with the existing on-line resources. The graph on figure 7.5 shows the
responses to a question about the controllability of risk in road traffic
105
accidents. Interaction with our system had no effect on 12 of the
participants. Ten felt that the risks were less controllable and three felt
that they had greater control. This diversity is again similar to the effects
noted in the pilot study. In both cases,however, further tests are required
over a larger population to determine whether these observations are not
artefacts that stem from an inappropriate sample of a far larger population.

For example, we observed several patterns in our responses that suggest
individual traits and attitudes to risk may be having more of an impact than
exposureto the systemsthat we have developed. For instance, user 10 in
Figure 7.4 felt that there was no likelihood that they would be involved in a
road accident even after they had used our system. They also felt
completely in control of the associatedrisks. Conversely, participant 8 felt
that they were extremely likely to be injured and yet also felt that they could
control the risk of involvement. It is possible to hypothesise about the

different attitudes that motivate such expressedperceptions. For example,
the attitude of participant 10 might reflect a high tolerance for risk.

Participant 8 might have a low threshold for risk. Their senseof control
might, in part, reflect steps that they have taken to limit their exposure to
road related risks. It is difficult to validate theseassertions. However, it
seems clear that further research is required if we are to understand the

individual factors that affect our perception of the risks of road related
incidents. For instance,none of the previous psychometric tests that we
adopted account for the strong effects that must be introduced when
individuals have recently been involved in an adverse event. Similarly,
they fail to account for the different risk profiles that are associatedwith
different forms of road use. Individuals who never drive may have
different attitudes towards the risks of road accidents than individuals who
regularly use high-powered motorcycles.
106
7.4 Focus Group Discussion
Our consultationsafter this evaluationled us to considera greateruse of

focus groupsto supplementpsychometrictechniques.Unlike one to one
interviews,"focus groupsgeneratedatathroughthe give and take of group
discussion.Listeningas peopleshareand comparetheir different points of
view providesa wealth of information-not just aboutwhat they think, but
why they think the way they do" [97].
We selected 8 undergraduatesin Glasgow University to form the focus
group about road traffic accidents.The reasonswhy we chose them are, first,
according to the statistics young people (17 to 22 years old) are most likely
to be involved in traffic accidents [74], and secondly, they are familiar with
traffic conditions and the local environment in Glasgow city. Several
questions were discussed. We choose these questions because we thought

they are useful for us to identify how road users think about road accidents
and accident information.
I. Do you think the modem road traffic system is safer than in the
past?
2. What do you think are the main causesof road accidents?
3. How controllable are theserisks?
4. From a road users' point of view, what kind of means can affect
people's safety awarenessor their behaviour?
5. What are the advantagesand disadvantagesof national accident
statistics and the information provided on websites?

6. Do you think local accident information is more important or
helpful than high level statistics?
7. When you plan your journey, would you consider accident
information in addition to distanceand road type information?
107
They believe that the modem road traffic system is safer than the past since
the total number of accidentsis less than decadesago. However, with more
traffic, more powerful cars, they are still concerned about road safety.
Drink-driving and speed are blamed for most road accidents, although
someone mentioned poor road and car maintenance, and one overseas
student said "roads in Glasgow are too narrow". They all think these main
risk factors are hard to control. One student argued "We have to build a
safety culture by education or law enforcement to change people's attitude
towards these risks". When they are asked about useful ways to affect
people's attitude, more severe punishment and showing people the
consequencesof accidentswere the most popular answers.They think one
of the TV road safety campaignsis very helpful, which showed four young
students driving a car after they have had a drink, when they are hit by
another car. The advert actually showed the injuries of every one in the car.
They believe if people know they cannot afford or bear the consequencesof
accidents,they won't risk them.
When we talked about national statisticsand their safety campaignwebsites,
they thought that although the information is useful, "people just don't go to
these websites". People don't care how many casualties there are on the
road last year. They are not interested in what the main reason for road
accidentsis. They think "I never have an accident and I live in a safe area".
People are only interested in the information that relates to them. For
example, one student said "when I go to a new town, the accident

information would help me decide where I should drive carefully". After
seeing our system, they said this kind of specific, localized information
would definitely be helpful for road safety, "People are interested in this
information". They would like to find out what happened in their
neighbourhood.Once they realize the area is not as safe as they thought,

they might change dangerousdriving behaviours. When talking about the
108
impact of accident information on journey planning, they thought this really
dependson a lot of factors. One said "If a safer road will cost an extra half
an hour or more, I would rather choose the dangerous one". It is hard to
compare different situations just by talking. But they said if such

information was available, they would consider this when they plan a
joumey.
In the next chapter,I describean experiment we did which helped us to find
out whether accident information would affect people'sjourney planning.
7.5 Discussion and Conclusion
Since conducting our evaluation, we have endeavored to identify other
metrics that we might use to support the validation of our localized
approach to the provision of accident information. In particular, we are

looking for techniques that might both demonstrate the utility of the
to
approach sponsoringorganizationsand also produce results that can drive
iterations
subsequent in the developmentcycle of our tools. This searchhas
involved discussionswith a number of cognitive psychologists and safety
engineers.These discussionshave raised questions about the utility of the
psychometric approach in this context. Asking users about the costs and
benefits associatedwith the risks of road usage is like 'asking someone
whether they would like a swimming pool on the moon, it has simply never
occurred to them before'. Many people find it hard to answerour questions

becausethey have never thought explicitly about the risks of road use.
The experiencegained from the developmentof the software tools showed
that generalpublic has great interestsin the localized information. However,
we have learned valuable lessons about the criteria that might be used to
109
validate these systems. We cannot simply take metrics from cognitive
science and hope that they can be directly applied to support software
design and interface development.In retrospectwe realize that we may have
been na-Ivein ever expecting that this would be straightforward.
This area raises considerablepractical and ethical problems. The ultimate
measureof successmight be to demonstratethat those who had accessto

this localized information were statistically less likely to be involved in road
traffic accidents.Ethical and practical considerationsmake it unlikely that
we will be able to conduct such tests. However, we continue to look for
alternativemeasures.
We are also increasingly aware that many of these metrics may be
inappropriate for such software. In particular, the popularity of the system
has come as a considerablesurprise. Many of the participants in the study
wanted longer-term accessso that they could monitor the impact of road
planning policy and lobby for changes in traffic managementwithin their
neighborhood. It might, therefore, be argued that this increased sense of
engagementwith road safety issues is a greater benefit for this type of
software than any transientchangesin risk perception.
Overall, the main contributions of this chapter are, firstly, usability
evaluation showed that participants are satisfied with the system and the
system met our requirements. Secondly, the psychometric questionnaire

revealed that people's responsebefore and after they saw the detailed local
road accident information showed little change. Thirdly, the focus group
discussion shows road users have great interests in localized accident
information.
110
Chapter 8
Route Planning Experiment with Taxi Drivers
8.1 Motivation and Goals
Previous chaptersargued that high-level national road accident statistic can
have little impact on people's safety awarenessand their behaviour. The
possiblereasonfor this is attribution error: most drivers believe that they are
less likely to be involved in an accident than other motorists. Previous
chaptershave arguedthat local accident information may have more impact
on people's awarenessand their behavioursthan aggregatenational statistics.

However, our questionnairerevealedthat people's responsebefore and after
they saw the detailed local road accident information showed little change
either. As discussedbefore, this could have severalexplanations.One is that
our psychometricquestionnaireforced people to answer questionsthey had
not consideredbefore.
We needed to find another way to assessthe impact of local accident

information on people's safety awareness and their behaviours. Route
planning is one of the activities drivers perform almost every day. From the
route planning function on the ANs website, we can see, traditionally,
people choosetheir route by the type of road (e.g. avoiding motorways), by

traffic condition and by cost. The website does not include accident factors,
such as 'avoid high accidentrate area'. That may not be becausethey don't
care about accident information, but they don't have ways of accessingsuch
information. We designedan experiment to find out whether, once people
know about local road accident information, they would changetheir route
choice. If the experimentshowed there is some link betweenpeople's route
III
planning and their local safety knowledge, it would support our hypothesis
which is that localized accident information can affect people's safety
awarenessand we can use local accident information to influence people's

behaviour.
We choseto adopt a within subject design and decided to focus on two sets
of user groups, taxi drivers and university staff and students.We chooseto
focus on this group becausethey were really accessiblefor our study. They
can also be a base case for ftirther comparisons. The next chapter will
discuss the experiment with university staff and students. The decision to
choose taxi drivers was justified by the observation that this group has an
extremely high-risk exposureand hencehas both a commercial and personal
motivation to consider localised accident information systems.The decision
to focus on this user group introduced a number of additional complications

into our experimentaldesign. In particular, we were forced to consider the
other types of local knowledge that might help to shape the taxi driver's
route planning. We were concernedthat their experienceof delays on certain
routes or particular environmental aspectsincluding road layout might have
a greater impact on route selection than any information about average

journey times or even localized accidentinformation. We, therefore, decided
to repeatthe test using two different route-planning tasks. One was basedin
the centre of Glasgow and hencewas familiar to the drivers. The other was
in
set central We
Manchester. ensuredthat none of the participants had ever
driven in this areabefore starting the evaluation.This method was, therefore,
intendedto supporttwo different but complementaryhypotheses*.
* This chapter describesjoint work with Chris Johnson,Phil Gray and Marilyn Mcgee-Lennon.
conductedthe experimentalwork and analysis.The study designwas the result ofjoint discussions.
112
8.2 Experiment Design
Hypotheses
Hypothesis 1: Local road accident information would affect people's
choice in the rankings of the candidateroutes.
Hypothesis 2: People's own local experiencewould affect their choice in
the rankings of the candidateroutes.
Independent Variable:
Roadaccidentinformation
Dependent Variable:
The rankings of the candidateroutes and locations
This was a within subject design. Two scenarios were involved in this
experiment. The first one provided users with time information, which
indicated how much time drivers may spend on each road section. The task
was to drive from point A to point 0 (See Figure 8.1 and Figure 8.2).
Participants needed to rank 7 candidate routes. The second condition
113
provided users with both time and accident information. This told
participants how many accidents happened on each road section last year.
The task remainedthe same.
Participants had to do the experiment on two different maps. One was a
Glasgow city map, the other was a Manchester city map. The reasonwe did
this is becausewe wanted to see when drivers made their choices on the
routes, whether they used their own experiencesrather than the information
we provided. Therefore, we had to make sure all the participants had never
or very rarely driven through the centre of Manchester, so that we could
make comparisons.
The participantswe focusedon are taxi drivers. Their daily job is driving
throughGlasgow.They arefrequentlymakingroutedecisionsbasedon road
traffic conditions and time of the day. Also they have plenty of local
Therefore
experiences. they are the ideal participantsfor our experiments.
20 driversin total did the experiment.
8.3 Measurement and Tests
Since we use rankings to measurepeople's choices on routes, it is a

test.
non-parametric This meansthat it doesnot rely on assumptionsabout
underlying population (mean,
parameters variance).Therefore,we cannot
use West or other parametrictests to test our experimentresults. After
consultingwith other in
researches the department,we choosethe Sign test.
The Sign test is one of the non-parametrictestsand it can be usedto test
The
rankings. Sign test is generallylesspowerful than the Wilcoxon test. It
is, however,simpleandeasyto use.Thegeneralprocedureis asfollows.
114
1. Inspect the difference between each pair of scores. Put a plus sign (+)
next to the differences in one direction, and a minus sign (-) next to the
differencesin the other direction.
2. Find the total number of either +s or -s, whichever occurs less
frequently. Let this number be x.
3. Find the critical value of x for the desired level of significance using the
following table. This value will depend upon (1) N, the total number of
pairs, and (2) whether the prediction is directional.
4. If x is less than or equal to the critical value, reject the null hypothesis
in favour of the alternative hypothesis that the independent variable has
had an effect on behaviour.
If a pair of scoresare equal they should be dropped from the analysis, and
the value of N reducedaccordingly [12].
8.4 Results and Analysis
Figure 8.1 showsthe experimentmap with time and accidentinformation

for Glasgow.
115
GreatWesternRd.
0.6mln
O.6mln
4A
\6xgyle SL -9. )lope SL
A
1.6min
2Accs
Our initial experiment focussedon 10 taxi drivers working in the centre of
Glasgow, the largest city in Scotland. Table 8.1 presentsthe results that were
obtained when these potential users were asked to rank different routes
across the city where they worked. The left-hand section of the table
documents the order of preference for routes before accident information
was provided. For instance,the row labelled SI representsthe first subject's
preference for route R5 above R7, which was preferred over R4 and so on.
The right-hand section of Table 8.1 illustrates the changes in preference
after the subjects were provided with the map that included information
about previous accidentson those routes.
116
Route Ranking Route Ranking

(no accidentinfornwtion) (local accident inromiation)
I 2 3 4 5 6 7 I 2 3 4 S 6 7
si R5 R7 R4 R6 R3 R2 RI R7 R5 Rl R3 R4 R2 R6
S2 R5 R6 R7 R4 R3 R2 RI R5 R6 R7 R4 R3 R2 RI
S3 R5 R6 R7 Rl I R2 R3 R4 R7 R5 Rl R4 R3 R2 I R6
S4 R5 R6 R7 R4 R3 R2 RI R7 R5 Rl R3 R4 R2 R6
S5 R5 R7 R6 R4 R2 Rl R7 R5 R4 R3 R6 RI R2
S6 R5 R6 R7 R4 R3 R2 RI R7 R5 RI R4 R6 R3 R2
S7 R5 R7 R6 R4 R3 R2 RI R7 R5 R4 RI R3 R6 R2
S8 R5 R7 I R6 R3 R4 R2 Rl R7 R5 R4 R3 R6 Rl R2
S9 R5 R7 R6 R4 R3 R2 RI R7 R5 RI R3 R4 R6 R2
sto R5 I R6 I R7 I R4 I R3 I R2 I RI II R7 I R5 I RI I R4 I R3 I R2 I R6
Table 8.1 Preferencesfor the Glasgow route finding task
The Appendix F shows an example of calculating sign test for route I (RI).
Table 8.2 shows sign test result on these 7 candidate routes on Glasgow
map.
Numberof Numberof Number P value
9+9s 4-9s unchanged
RI 9 0 1 0.004
R2 0 6 4 0.031
R3 4 1 5 0.375
R4 4 3 3 1
R5 0 9 1 0.004
R6 0 9 1 0.004
R7 9 0 1 0.004
Table 8.2 Sign test result for the Glasgow route finding task
117
Table 8.2 provides the results obtained from applying the Sign test in the
manner described in previous sections. The results show that the provision
of local road accident information did affect the rankings of all routes. If we
choose the 0.05 level of significance then the Nvalues obtained from a
two-tailed sign test show that the provision of this information had a
significant impact on the ranking for routes RI, R2, R5, R6 and R7. A
two-tailed sign test is appropriate here becausewe are interested in whether
the provision of local accident information increases the ranking of some
routes and correspondingly leads to a decrease in the ranking of other
routes.
Figure 8.2 shows the experiment map with time and accident information
for Manchester.
118
Table 8.3 provides an overview of the results from the route preference task
obtained when the Glasgow taxi drivers were asked to navigate across part
of Manchester. Recall that none of the drivers were familiar with the road
layout of this city in the North of England. As before, the left hand section
of table provides the rankings for the task when the drivers only had access
to the information available from existing route finding systems. The
right-hand section of the table, in contrast, illustrates the impact on their
preferences from providing information about previous accidents on the
routes. In order to counter balancing the experiment, half the participants
to
were asked rank the Manchester City road map first and then rank the
Glasgow road map. The other half did the other way around. The reasonwe
did this is becausewe don't want the order of the experiment affect taxi
drivers'judgements. However, in the experiment, all the participants always
saw the map with accident information after the map without accident
information. The obvious reason is when participants saw the map with
accident information first. They would not forget accident information when
they saw the map without accident information. Therefore, it will make no
difference whether the map with or without such information.
Route Ranking RouteRanking

(no accident inforniation) (local accidentinrorniation)
I 2 3 4 5 6 7 1 2 3 4 5 6
si R9 RIO R13 R14 RS RII R12 R8 RII R12 R9 R14 RIO R13
S2 R9 RIO R13 R14 R8 RII R12 RS R12 RII R9 R14 RIO R13
S3 R9 RD R14 R13 R8 R12 RII R8 R12 R11 R9 R14 RIO R13
S4 RIO R9 R13 R14 R12 R8 R11 R12 R8 RII R9 R14 RIO R13
S5 R9 RII R12 RIO R8 R13 R14 R8 RII R12 R9 R14 RIO R13
S6 R9 RIO R14 R13 RS R12 RII Rg RII R12 R9 R14 RIO R13
S7 RIO R9 R13 R14 R12 R8 RII R12 R8 RII R9 R14 RIO R13
S8 R9 RIO R14 R13 R8 R12 RII RS RII R9 R12 R14 R13 RIO
S9 R9 RII R12 RIO R8 R13 R14 R8 R12 R11 R9 R13 RIO R14
sio R9 JR11 JR121RIOIR8 JR131R141 IRS JR11 JR9 JR121RIOIR13IR14
for the Manchesterroutefinding task

Table8.3 Preferences
119
Before looking in more detail at the results of the Sign test, it is worth
noting that severalpairs of routes were chosento take the same time. These
were (R9, RIO), (RII, R12), (R13, R14). However, each clement of these
pairs was chosen to have a different historic number of accidents. The
results in Table 8.4 show that taxi drivers do consider accident information
when two routes are shown to require the same time in an unknown location.
For instance, subject SI initially preferred R13 to route R14. After they
were provided with the accident information this ranking was reversed.
SubjectsS2, S4, S5 and S7 reveal a similar pattern while subjects S3 and S6
did not changetheir initial ranking that placed route R14 before R13. Such
descriptive analysis can be supportedby the statistical measuresof the Sign
test in Table 8.4.
Number of Number of Number P value

f+ PS - gs uncbanged
R8 10 0 0 0.002
R9 0 10 0 0.002
R10 0 10 0 0.002
RII 7 1 2 0.070
R12 8 1 1 0.040
R13 0 9 1 0.004
R14 1 7 21 0.070
Table 8.4 Sign test result for the Manchesterroute finding task
Table 8.4 sununarizesthe results of the Sign test for the Manchester route
finding task. As before, the provision of local accident infonnation has an
effect on the ranking of all routes. However, the P-values indicate that the
effects on the position of routes R8, R9, RIO, R12 and R13 were statistically
120
significant. We cannot conclude that the change in ranking of RII and R 14
were due to the provision of accident information and may simply have
been due to changeexperimental effects.
It seemsclear that both hypothesis I and hypothesis 2 are supported by the
results that are summarized in Tables 8.2 and 8.4. The provision of local
accident information did have an effect on the ranking of routes for a town
that was not well known to the taxi drivers. Access to this information also
had an impact on the rankings associatedwith routes across a city that were
used by the taxi drivers every day of their working lives.
Our results showed significant differences for the rankings associatedwith
five of the seven routes in both experimental conditions. This raised
questions about the nature of those routes that were unaffected by the
provision of information. One possibility is that thesewere routes that were
never actively consideredby the drivers, they appearedto be so undesirable

that the provision of accident information was largely irrelevant. Debriefing
sessionsalso raised fin-therconcernsabout the conduct of this initial run. We
used a prepared script to ensurethat all of the taxi drivers received the same
set of instructions. These asked them to "Please rank the following seven
routes by order of preference.(From 1, which you think is the best route, to

7 which is your least favourite)." A small number of the drivers in this study
were confused by this instruction and asked whether they were being asked
to basetheir selection on speedor cost. We deliberately did not want to bias
the criteria that they might use in their ranking. We, therefore, decided to
conduct a second study using the same method but with ten more taxi
drivers. The experimental instructions were changed to "From the point of
view of your work as a taxi driver, pleaserank the following routes in order
of preference from I to 7 where the first is the route that you would
definitely use and the last is the route that you would be least likely to use."
121
We also decided to simplify the experiment slightly by marking the total
time and cumulative number of accidents next to each route. Table 8.5
illustrates the results for this second cohort of taxi drivers on the routes
across Glasgow city centre. Table 8.6 presentsthe results from the Sign test
applied to this second set of route preferences in an area that is already

familiar to the drivers.
Route Ranking RouteRanking

(no accident infonnation) (local accidentinfonnation)
1 2 3 4 S 6 7 1 2 3 4 S 6 7
Sil R5 R6 R7 R3 R4 R2 RI R7 R5 R4 R6 R3 Rl R2
S12 R5 R6 R7 R3 R4 R2 Rl R7 R5 R6 R4 R3 R2 RI
S13 R5 R6 R7 R4 R3 R2 RI R7 R5 R4 R6 R3 RI R2
S14 R5 R6 R7 R4 R3 R2 Rl R7 R5 R4 R6 R3 RI R2
S15 R5 R7 R6 R4 R3 R2 RI R7 R5 R4 RI R3 R6 R2
S16 R5 R6 R7 R3 R4 R2 RI Ri R5 R6 R4 R3 R2 RI
S17 R5 R6 R7 R4 R3 R2 Rl R7 R5 R4 R6 R3 RI R2
S18 R5 R7 R6 R3 R4 R2 LR I R7 R5 R6 R4 R3 RI R2
S19 R7 R6 R4 R3 R2 Rl R7 R5 R6 R4 R3 RI R2
S20 R5 I R6 I R7 I R3 I R4 I R2 I RI II R7 I R5 I R4 I R6 I R3 I RI I R2
Table 8.5 SecondGroup of Taxi Drivers' Preferenceson Glasgow route

finding task
122

+'s
49 G-9s unchanged
RI 8 0 2 0.008
R2 0 8 2 0.008
R3 0 5 5 0.063
R4 9 0 1 0.004
R5 0 10 0 0.002
R6 0 8 2 0.008
R7 10 0 0 0.002
Table 8.6 Sign test result for the SecondGroup's Glasgow route finding task
As can be seen, the addition of localised accident information had a
statisticallysignificantimpacton the ranking of all routesexceptfor R3. It

be
can arguedthat by clarifying the experimentalbriefing andby providing
cumulativetotals for expecteddelays and raw accident frequencies,the
provisionof this additionalsafetyinformationhashad a clear effect on the
taxi drivers' rankingsfor the routesacrossthe roadsof a city where they
work. Table 8.7 goes on to summarisethe rankings associatedwith the

Manchesterroutesfor this secondcohortof taxi drivers.
Route RaiiMng Route Ranking

(no acddent inronnation) flocal accidentinformation)
1 2 13 4 5 6 7 1 12 3 4 5 6 7
Sil R9 RIO I R13 R14 R8 R11 R12 R8 R12 R9 Rll R14 RIO R13
S12 R9 RIO R13 R14 R8 RII R12 R8 R12 R9 RII RIO R14 R13
S13 R9 RIO R14 R13 R8 RII' TI2 T8 R12. RII R9 R14
RIO R13
S14 RIO R9 R13 R14 R12 R8 RII .
R8 R121 R9 RII RIO R14 R13
S15 R9 RIO R14 R13 R8 Rll R12 R8 R12 R9 RIl RIO R14, R13
S16 R9 RIO I R14 R13 R8 RII R12 RS R12 R9 RII R14 RIO R13
S17 RIO R9 R13 R14 R12 R8 RII R8 R12 R9 RII RIO R14 R13 ,
S18 R9 RIO R14 R13 R8 RII R12 R8 R12 R9 RII RIO R141 R13
S19 R9 RIO R14 R13 R8 R12 RII R8 R12 RII R9 R14
S20 R9 R14 R8 RII R12 R8 R12 RII R9 RIO
_RIO_L_RI3_j
,
Table 8.7 SecondGroup's Preferenceson Manchesterroute finding task

123
Table 8.8 presents the results of the Sign test on this second cohort of
preferences from the Manchester study. As can be seen, the provision of

localised accident information had a significant impact on the rankings
associated with every route across this 'unfamiliar' road system. It is
important to stress that the sign test does not provide a quantitative
assessmentof the strength of any effect. If a particular route were to be

downgraded from first to seventh place then this would be registered as a
single '-' value in the Sign test. However, the unequivocal results from this
stage of our study in Table 8.8 illustrate the impact of localized accident
information on the taxi drivers in our study.

&+'s &_, s unchanged
R8 10 0 0 0.002
R9 0 10 0 0.002
RIO 0 10 0 0.002
R11 10 0 0 0.002
R12 10 0 0 0.002
R13 0 10 0 0.002
R14 0 10 0 0.002
Table 8.8 Sign test result for the SecondGroup's Manchesterroute finding
task
The results from this second group show that the rankings for six of the
seven Glasgow routes support hypothesis 1. In other words, the ten taxi
drivers changedtheir initial route preferencesonce they had been provided
with information about previous accidents on those routes. In the caseof the
124
Manchester routes, that were not previously familiar to the taxi drivers, all
of the route preferenceschanged as a result of the accident information. Of
course, neither of these findings implies any consequent change in
behaviour. There is no guaranteethat the taxi drivers will use the accident
information during their work, just as there is no guaranteethat a driver will
follow the fastest or cheapestroute recommendedby conventional planning
software. Route planning depends on the knowledge and skill of the
navigator. Our studies have, however, shown that the provision of localised
accident information can have an impact on these complex, problem solving
tasks. It is important to acknowledge, however, that our results do not
provide anything like a complete theoretical understanding of the cognitive
processesthat are involved in the exploitation of local, historical accident

information. For instance, some drivers may rate safety information more
highly than their peers. Only driver S2 from the first group of the Glasgow
users did not change any of his rankings after being presented with the
accident information surnmarisedin Table 8.9. As mentioned, however, this
may have been due to the way in which the task was presented.However,
finther studies are required to determine whether the revised procedure with
the second group would entirely eliminate this observed behaviour or
whether S2 is typical of a persistent group of road users who would not be
influenced by the previous number of accidentsover the routes that they use
every day.
Identifier Nodes Total estimatetime Cumulativenumber or accidents

(seconds)
RI ABEKNO 204 42
R2 ABEKLNO 198 52
R3 ACEKLNO 234 34
R4 ACEKNO 240 24
R5 ACFNIO 144 25
R6 ADFMO 186 28
R7 ADJMO 204 12
Table 8.9 Summary of Routes in the Glasgow Map
125
Table 8.9 summarisesthe routes from the Glasgow map that was used in
both groups of this study. As mentioned, there were 20 taxi drivers in both
groups. If accident related information had been the main determinant in
route planning then we might have expected that most would have chosen
the order R7, R4, R5, R6, R3, RI and R2. However, none of the drivers in
the first or second cohorts chose this ordering. Instead, five chose a single
permutation of R5 and R4. In other words, they chose: R7, R5, R4, R6, R3,
RI and R2. This apparent inconsistency can be explained in terms of the
similarity between R4, with 24 accidents, and R5 with 25. It seems likely
that they were trading the single additional accident against a larger time
saving with R5's estimated 144 seconds being preferred to R4's 240
seconds.
At one level it can be argued that the taxi drivers were employing a
relatively simple form of satisfying. In other words, they were willing to
trade a certain degree of speed against a higher number of accidents and
vice versa. For example, R6 (28) has had more accidents than R4 (24).
However, R6 has an estimatedjourney time of only 186 secondscompared
to 240 secondsfor R4. In the second cohort, four drivers ranked R6 above
R4. Six taxi drivers rated R4 above R6. It might, therefore, be argued that
such differences illustrate a degree of risk preference and aversion when
faced with similar routes. However, there is also evidence of more complex
forms of constraint satisfaction that go beyond simple trade-offs between
time and safety. RI was the least preferred route on the Glasgow map even
though it was not the route with the highest averagejourney time. 19 out of
the 20 drivers put it in last place in their ranking. This would seem to
indicate that the drivers relied upon outside knowledge that was not fully
being captured in the simple timings that we presented. However, when
additional information was provided about the number of local accidents
126
most drivers revised their choice. Recall that R2 had 52 accidents compared
to 42 for RI and that the difference in estimatedjourney time was only six
In thirteen
seconds. consequence, of the twenty drivers revised their ranking
to put R2 in last place.
Identifier Nodes Total estimatetime Cumulative number or accidents

(seconds)
R8 ABCDEO 20
R9 ABJDEO 540 24
RIO AHJDEO 540 34
RII AHIFEO 720 23
R12 AHEFGO 720 21
R13 AHJFEO 600 34
R14 AIUFM 600, 32
Table 8.10 Summary of Routes in the ManchesterMap
Table 8.10 provides a summary of the cumulative accident frequencies and
estimatedjourney times for the Manchester routes, which were not familiar
to either group of taxi drivers. The pattern of results for both individual
subjects and for particular routes is broadly similar to that for the Glasgow
journeys. There are, however, some important differences. Unlike the local
route-planning task, there is no individual who retained his or her original
preference list after being shown the accident information. This tends to
confirm our suggestion that the individual from the first cohort of the
Glasgow study may have shown an inelastic preferences for routes with
which they were very familiar. However, that same individual appearsmore
willing to revise their preferencesfor routes that they do not drive every day.
This is an important finding becauseit implies that there will be some value
added from the provision of accident information given that individuals are
less likely to use route-planning software for routes that they already know
very well. However, as mentioned in previous paragraphs,further work is
to
required support this additional hypothesis.
127
If taxi drivers' preferences were being determined by the number of
accidents alone then one might have expected most participants to opt for
the sequenceR8, R12, RI 1, R9, R14, RIO, R13. As with the Glasgow study,
things were not this straightforward. Only four of the taxi drivers chose this
ranking after being presentedwith the accident information. The remaining
drivers seemedto exhibit the samesatisfying behaviour that was observedin
the previous paragraphs.Trade-offs are being made between journey time

It
and accidents. can, however, be argued that the interpretation of these
is
results simpler becausethe drivers did not have any previous experience
of navigating in Manchestercity centre.
8.5 Conclusion
The main contribution of this chapter is that we carried out route planning
experiments on taxi drivers and find that road accident information has
some impact on their route planning decisions. However, the design of this
experiment does not consider other factors that may affect drivers' decision
on route choices. The sign test suggests the extra information (accident
information) may have influence on drivers' decision. But we can not
conclude that local accident information will affect drivers' safety awareness.
Further researches can be done to perform some stronger tests which
compareroad accident information against other form of information, such
as news agent locations, or pub locations. Once we find out that road
accident information have stronger impacts on drivers' route planning

decision than other form of information. We can say local accident
information will affect drivers' safety awareness.
128
Chapter 9
Route Planning Experiment with University
Staff and Students
From the above study, we can see localized accident information has a
significant impact on taxi drivers' route planning. In other words, they
consideredsafety as one factor when planning their journey during our

The
experiment. use of taxi drivers in the experimenthas advantagesand
disadvantages.It is certainly true that this subset of drivers has a
commercialandpersonalmotivationto consideraccidentinformation.This
subsetprovides an interestingcross-sectionof participantswho are very
familiar with many routes, and thus the population seemsa reasonable
choice. However,
taxi driversare a specialgroup.Few driverswould have
sameknowledgeof local roadsas taxi drivers. Few other drivers have the
samehigh-risk as
exposure taxi drivers. [1]
Research shows,"Buses,taxis
vehicles
and emergency were found to be in
over-represented accidentson
urbanroadsof all Those
classifications. driving buses,
taxis and emergency
vehicles, suffered more accidentsprimarily causedby other road users.

Their problemwas thereforepredominantlyone of exposureto dangerous
"
environments.
In order to see whether localized accident information has a significant

impact on other drivers, we repeated our experiment with university staff
and students. This group typically do not live on driving. They don't have
the same high-risk exposure as taxi drivers. They don't consider route
planning often as taxi drivers. This group usually have a higher level of
education than taxi drivers.
129
The Impact of Locaaed RoadAccident Information on Road Safety Awareness
9.1 Experiment Design
Hypotheses
1:
Hypothesis Local road accidentinformationwould affect people's
in
choice therankingsof thecandidate
routes.
Hypothesis 2: People's own local experiencewould affect their choice in
the rankings of the candidateroutes.
Independent Variable:
Road accident information
Dependent Variable:
The rankings of the candidateroutes and locations
This was a within subject design. Two scenarios were involved in this
experiment. The first one provided users with time information, which
indicated how much time drivers may spend on each road section. The task
was to drive from point A to point 0 (See Figure 9.1 and Figure 9.2).
Participants needed to rank 7 candidate routes. The second condition
provided users with both time and accident information. This told
participants how many accidents happened on each road section last year.
The task remained the same. Participants had to do the experiment on two
different maps. One was a Glasgow city map, the other was a Manchester
130
city map. We chose 10 staffs and 10 students from University of Glasgow to

do the experiment.
9.2 Measurement and Tests
Sameas chapter 8, we use Sign test to analysethe experiment.
9.3 Results and Analysis
for Glasgow.
This is the same as before but we repeat it here to help readers follow the
in
analysis this chapter.
Table 9.1 presents the results that were obtained from 10 university staff.
131
The left-hand section of the table documents the order of preference for
routes before accident information was provided. The right-hand section of

Table 8.1 illustrates the changes in preference after the subjects were
provided with the map that included information about previous accidents
on thoseroutes.
Route Ranking
(no ac ident iuform tion) Oocal accidentinfo ation)
1 2 3 4 5 6 7 11 2 3 4 5 7
Sl R5 R6 R7 R3 R4 R2 RI R7 RS R4 R6 R3 Rl R2-
S2 R5 R6 R7 R4 R3 R2 Rl R7 R5 R6 R4 R3 R2 RI
S3 R5 R6 R7 R3 R4 R2 R7 R5 R6 R4 R3 Rl R2
S4 P.5 R6 R7 R3 R4 R2 _Rl
Rl R7 R5 R6 R4 R3_ Rl R2
S5 R5 R6 R7 R3 R4 R2 RI R5 R4 R6 R3 Rl R2
R5 R6 R7 R3 R4 R2 RI _R7
R7 R5 R4 R6 R3 Rl R2
S6
S7 R5 R6 R7 R3 R4 IR2 Rl R7 R5 R6 R4 R3 R2 RI
I
Ss R5 R6 R7 R3 R4 R2 Rl R7 R5 Rr> I R4 mR3 Rl
S9 R5 R6 R7 R3 R4 R2 Rl R7 R5 R6 R4 R3 1 R2
SIO R5 R6 R7 R3 R4 R2 Rl R7 R-5 R6 RI R2
Table 9.1 10 university staff's preferenceson Glasgow route finding task
Table 9.2 shows sign test result on these 7 candidate routes on Glasgow
map.
Number
Number of Ys Number of '-'s p value
unchanged
R1 7030.016
R2 0730.016
R3 0910.004
R4 9010.004
R5 0 10 0 0.002
R6 0 10 0 0.002
R7 10 000.002
Table 9.2 Sign test result for the Glasgow route finding task (university
staff)
The resultsshow that the provision of local road accidentinformationdid

132
affect the rankings of all routes. If we choose the 0.05 level of significance
then the P-values obtained from a two-tailed sign test show that the
provision of this information had a significant impact on the ranking for all
routes.
for Manchester.
Table 9.3 provides an overview of the results from the route preferencetask
obtained when 10 membersof university staff were asked to navigate across
part of Manchester. None of them were familiar with the road layout of
Manchester.As before, the left hand section of table provides the rankings
for the task when the drivers only had accessto the information available
from existing route fmding systems.The right-hand section of the table, in
133
contrast, illustrates the impact on their preferences from providing

information about previous accidents on the routes. Same as taxi drivers'
experiment, half the participants were asked to rank the Manchester City
road map first and half the participants were asked to rank the Glasgow road
map first.
Route Ranking
(no accident information) Oocal accident information)
1 2 3 4 5 6 7 1 2 3 4 5 6 7
Sl R9 R10 R13 R14 RS Rll R12 R8 R12 R9 Rll R10 R14 R13
S2 R9 R10 R13 R14 RS Rll R12 R3 R12 R9 RlI R10 R14 R13
S3 R9 R10 R13 R14 RS RlI R12 R8 R12 R9 RIl R14 R10 R13
S4 R9 R10 R13, R14 RS Rll R12 R3 R121 R9 Rll RIO R14 R13
S5 R9 R10 R13 R14 RS Rll R12 R8 R12 R9 Rll RIO R14 R13
S6 R9 R10 R13 R14 R8 Rll R12 RS R12 R9 Rll RIO R14 R13
S7 R9 R10 R14 R13 R8 RIl R12 RS R12 R9 Rll R10 R14 R13
Ss R9 R10 R13 R14 R3 Rll R12 RS R12 R9 Rll R14 R10 R13
1 - ql2
S9 R9 R10 R13 R14 R3 RIFT R8 R12 R9 RII RIO R14 R13
R10 I R14 I R13 1 R3 I Rll - R12
R RS R121 R9 1 tR13
RIl R10
'12
_R14
Table 9.3 Preferencesfor the Manchesterroute finding task (university staff)
As we mentioned in Chapter 8, several pairs of routes were chosen to take
the same time. These were (R9, RIO), (RII, R12), (R13, R14). However,
each element of these pairs was chosen to have a different historic number
of accidents.The results in Table 9.3 show that university staff do consider
accident information when two routes are shown to require the sametime in
an unknown location. For instance, subject SI initially preferred R13 to
route R14. After they were provided with the accident information this
ranking was reversed. Subjects S2, S3, S4, S5, S6, S8 and S9 reveal a
similar pattern while subjects S7 and S10 did not change their initial ranking
that placed route R14 before R13. Such descriptive analysis can be
supportedby the statistical measuresof the Sign test in Table 9.4
134
4+'s &-ts unchanged
R8 10 0 0 0.002
R9 0 10 0 0.002
RIO 0 10 0 0.002
RII 10 0 0 0.002
R12 10 0 0 0.002
R13 0 10 0 0.002
R14 0 10 0 0.002
Table 9.4 Sign test result for the Manchester route finding task
(university staff)
Table 9.4 summarizesthe results of the Sign test for the Manchester route
finding task. As before, the provision of local accident information has an
effect on the ranking of all routes. In the case of the Manchester routes, that
were not previously familiar to the university staff, all of the route
preferenceschangedas a result of the accident information.
It seemsclear that both hypothesis1 and hypothesis

2 by
are supported the
resultsthat in
are summarized Tables9.2 and 9.4. The provision of local
information
accident did havean effect on the ranking of routesfor a town
that was not well known to the university staff. Accessto this information
also had an impacton the rankingsassociated

with routesacrossa city that
wereusedby the universitystaffseveryday.
Table 9.5 presentsthe results that were obtained from 10 university students.
Table 9.6 presentsthe results from the Sign test applied to this set of route
preference.
135
Route Ranldng
(no accident infonnation) Oocal ccident information)
1 2 3 4 5 6 7 1 2 13 4 51 6 7
Sl R5 R6 R7 R2 RI R3 R4 R7 R5 R6 R4 R3 Rl R2
S2 R5 R6 R7 R2 RI R3 R4 R7 R5 R6 R4 R3 Rl R2
S3 R5 R6 R7 R2 Rl R3 R4 R7 R5 R6 R4 R3 RI R2
S4 R5 R6 R7 R2 Rl R3 R4 R7 R5 R6 R4 R3 RI R2
S5 R5 R6 R7 R3 R4 R2 RI R7 R5 R6 R4 R3 R2 RI
S6 R5 R6 R7 R2 Rl R3 R4 R7 R5 R6 R4 R3 Rl R2
S7 R5 R6 R7 R? Rl R3 R4 R7 R-'; R6 R4 R3 R2 Rl
S8 R5 R6 R7 R2 Rl R3 R4 R7 R5 R6 R4 R3 Rl R2
S9 R5 R6 R7 R2 Rl R3 R4 R7 R5 R6 R4 R3 R R2
slo R5 R6 R7 lu Rl M R4 R7 R-'; RC, R4 M M Rl
for the Manchesterroutefinding task(university

students)
+ - ys unchanged
RI 0 9 1 0.004
R2 0 9 1 0.004
R3 9 1 0 0.004
R4 10 0 0 0.002
R5 0 10 0 0.002
R6 0 10 0 0.002
R7 10 0 0 0.002
Table 9.6 Sign test result for the Glasgow route finding task (university
students)
Resultsshow that the provision of accidentinformation had a significant

impacton the rankingfor all routes. The sameresultscanbe found in the
Manchestermapin Table9.7 and9.8.
136
Route Ranking
(no accident information) 60cal ccident info nation)
1 2 3 4 5 6 7 1 2 3 4 5 6 7
Sl R9 R10 R13 R14 RS Rll R12 R8 R12 R9 RIl R14 R10 R13
S2 R9 R10 R13 R14 RS Rll R12 R8 R12 R9 RIl R10 R14 R13
S3 R9 R10 R14 R13 RS RIl R12 RS R12 R9 RIl R14 R10 R13
S4 R9 R10 R13, R14 RS RIl R12 R8 R12 R9 Rll R10 R14 R13
S5 R9 R10 R13 R14 RS Rll R12 RS R12 R9 Rll RIO R14 R13
S6 R9 RIO R13 R14 RS Rll R12 RS R12 R9 Rll R10 R14 R13
S7 R9 R10 R13 R14 RS RIl R12 RS R12 R9 Rll R14 RIO. Rl
SS R9 R10 R13 R14 RS Rll R12 RS R12 R9 Rll R10 R14 R13
_ S9 _
R9 R10 R13 R14 RS Rll R12 RS R12 R9 Rll R10 R14 R13
-El-3 .I -R -14- FR
slo R9 R10 R14 R3 I Rll I R12 1 R8 R12 R9 Rll R10 13
,
for the Manchesterroutefinding task(university

students)
C+ts 6-9s unchanged
R8 10 0 0 0.002
R9 0 10 0 0.002
RIO 0 10 0 0.002
RII 10 0 0 0.002
R12 10 0 0 0.002
R13 0 10 0 0.002
R14 0 10 0 0.002
Table 9.8 Sign test result for the Manchesterroute finding task (university
students)
As mentioned in Chapter 8, none of these findings implies any consequent
changein behaviour. Our studies have, however, shown that the provision of
localised accident information can have an impact on these complex,
problem solving tasks. It is important to acknowledge, however, that our
results do not provide anything like a complete theoretical understandingof
the cognitive processes that are involved in the exploitation of local,

historical accident information.
137
9.4 Conclusion
Route planning is an activity in which people use their own knowledge and
experience to perform. Our experiment chose route planning to assess
people's safety awareness:in other words, do they consider road accident
information when they plan their journey? We believe the result from the
experiment supports our argument that detailed local traffic accident
information has an impact on people's safety awarenessand behaviour since
the national statistics on this problem appearto have had little impact.
From the experiment, we have gathered support for our two hypotheses.
Firstly, the provision of local accident information does have an impact on
the routes that drivers select in areas that they are already familiar with.
Secondly, this effect can also be seen when drivers have access to
information about towns that they are less familiar with. Our results were
obtained from studies with a group of taxi drivers because these users
typically have the greatest exposure to risk, illustrated by correspondingly

high insurance premiums. We also obtained results from studies with a
group of university staff and students.
As mentioned, we have already developed a web-based system for
presenting localised accident information. As we developed this tool to
provide route-planning functionality, a number of additional problems have
arisen. Many of these were not directly considered within our study. In
particular, the journeys in our system are over relatively short distances
within two city centres. Hence, we have a relatively low number of
accidents for each route within our data set and it is possible to provide the
drivers with raw accident frequenciesover a five or ten year window. As we
expand the length of a route then these frequencies become increasingly
problematic, as there may be several thousand acci4ents on a journey of an

138
hour or more over busy local roads. Previous sections have also argued that
raw cumulative accident frequenciesignore the different traffic volumes that
pass over each route. It is relatively straightforward to factor traffic flows
into the calculation of accident rates rather than frequencies. However, it can
be difficult to explain the resulting cumulative rates to a broad group of end
users. For example, comparable risk exposures can be derived from a
prolonged period of driving on a road with a medium accident rate or from a
short period driving on a road with a high accident rate followed by a longer
period driving on a 'low risk' road.
If we are to move from raw accident frequencies to consider issues of risk
exposurethen there seemsto be considerable scope for the development of
appropriate interfaces that might mask some of the underlying complexity
from end-users,such as our taxi drivers, who have a significant interest in
using this data. We are currently experimenting with extensions to existing
route planning systems.These already offer users planning criteria such as
davoid motorways' or 'avoid toll roads and congestion charging'. Further
work is required to determine whether this supports the more complex
trade-offs between time and risk that seemto be emerging from our study of
route planning behaviour by the two groups of drivers.
A naturalextensionof manyof the ideasin this thesiswould be to integrate

local accidentinformationinto an in-car GPSnavigationsystem.This would
be relativelystraightforwardwith accidentfrequencyupdatesbeingsupplied
with each map upgrade. The existing web-basedaccident information
system draws its updates directly from the national accident records
suppliedindirectly to the public by regional police forces.Local accident

informationmight be usedby in-car systemsin the mannerdescribedabove
for more conventionalroute planningsoftware.The provision of GPSalso
enablesthe use of visual or audio warnings to inform drivers when they
139
enter an accident 'hotspot'. It is important not to underestimate the

importance of such potential applications. However, the problem for such
warning system is it may cause distraction to drivers. Moreover, displayed
accident information should be detailed but brief so that drivers don't need
much time to read it during the driving. Therefore, different user interfaces
be
may required for different user groups, such as taxi drivers and university
staff and students.All these questions are all need further researchwork to
find out the best solutions.
Overall, the main contribution of this chapter is that we carried out route
planning experiments on university students and staffs and find that road
accident information has some impact on their route planning decisions.

However, as mentioned in chapter 8, the design of this experiment does not
consider other factors that may affect drivers' decision on route choices.The
sign test suggests the extra information (accident information) may have
influence on drivers' decision. But we can not conclude that local accident
information will affect drivers' safety awareness. Further researchescan
be done to perform some stronger tests which compare road accident
information against other form of information, such as news agent locations,
or pub locations. Once we find out that road accident information have
stronger impacts on drivers' route planning decision than other form of
information. We can say local accident information will affect drivers'
safety awareness.
140
Chapter 10
Conclusions and Future Work
10.1 Conclusions
Our study was basedon the hypothesisthat existing road safety advice
dependstoo muchon the useof nationalstatisticsand localizedinformation
may have more impact on people's safety We

awareness. carried out an
evaluationon existingroad safetycampaignwebsiteswhich providemainly
high level nationalstatistics.The result showsthat the information on the
made
websites little impact.
Then with advice from experts in Glasgow City Council and the Highway
Agency, we developed a localized accident information system. The same
evaluation method for government websites was carried out. The result
showed some impact on people's but

awareness, it was not as big as we had
anticipated. We therefore used other methods to assessthe impact of local
accident information on people's safety awareness.These included focus
group discussionand a route planning experiment.
The result from these two methods gave us more positive feedback and the
experiment shows that taxi drivers do consider accident information when
two routes are shown to require the same time in an unknown location.
Access to such information also had an impact on the rankings associated
with routes acrossa city that were used by the taxi drivers every day of their
working lives. But it is still early to say which is better, a high level national
safety campaign or local accident information system since as I pointed out

in chapter 7, the local accident information system has much room for
141
improvement. It seemsunlikely, furthermore, that many of the effects that
we observedfor our taxi drivers will be equally apparent for all drivers.
10.2 Future Work
The study has found that road users are interested in local accident
information and such information may affect people's safety awareness.
However, in order to find out whether local accident information has more
impact on people's safety awarenessthan high level accident statistics, we
need to include more kinds of road users in the study. We may also need to
divide these people into several sub groups, such as drivers, cyclists,
pedestrians,and so on. This is becausedifferent road user groups may have
different requirements in accident information. For example, pedestrians
may only care about pedestrian-involved accident information. Other
information would have little impact on their safety awareness.Moreover,
we need more experiments to find out ways to improve the impact of

localized road accident information on people's safety awareness.
I now discuss possible future research from following three aspects:

information,interactionandprediction.
a) Information
Local accident information can help people to assessthe safety of their
environment. However, each accident record holds lots of detailed
information. Some users may not need all of this, different road users may
be interested in some specific information. In the future, experiments are
neededto identify the information most important to road users.
Brief description. In the early stageof the study, experts from Glasgow city
142
to
council suggested giving a brief description of each accident. Currently, I
directly display all the columns from the database.It is not clear if users can
obtain a coherent overview from all of the values and codes. An experiment
is neededto decide whether 'brief description' is a good idea and what kind
of information should be included in this description.
Advice from experts. Our study shows that since people are less interested
in high level national statistics on government road safety campaign
websites, they won't go there voluntarily. However, they think that advice
and guidelines on the websites are useful. Figure 10.1 shows advice from
the "Think! " website. Therefore, adding advice to a localized accident

information system could be more effective than those websites. Moreover,
advice normally focuseson specific road user groups or locations (road type)
or special conditions (winter driving, driving when tired). When users
search for specific kinds of accidents, they could also be given advice on
how to avoid this type of accident.
143
The Impact of Localized Road Accident Information on Road Saf'ety Awareness
11Fill I IIIIIIIII! III pil III I ! Iýnm=ll

Ij,
M1T-7W '1113
Eile Edit yi- Fývorites lools Help
Jr,
0 Back - l 5earch Favorites 4Er>r
I) I 'ýl L-;
htm
1!ýj http: /Iwww. thinkruadsafety. gov. ukladvicý/cycling.
I .., II..
.4t'l -
III ntroduction THINK' Home I Advice
11ýChild road
safetyý parents THINK! advice - cyclists
.1............... ..............
b-Child road
safety: teachers Cyclists and drivers both have a rightto use our roads - but ýnn-wfir, -. ytm ii, -l t, qw#- .11,1
- ............... more thought to each other
'-Child car seats
-- ............
I-Avoiding low
DOS . 11141CIOII*ltS fOF CyCliStS
bridges
-. - .............................
OýCouriers
Be visible. Ride well clear ofthe kerb, wear bright cluthinq and flvvaVý, iqht,. . 4fl, r
...............
11-Cyclists dark or in poorweather conditions
I........................................ Show drivers whatyou plan to do. Always look and signal bofore you ýfavt, .t,, p ,r
01Driving when
tired turn.
........................................... Ride a straight line past parked cars rather than dodging between th,, m
'-Emergency Don't jump red lights
vehicles Don't ride on pavements.
IýOverseas Don't ride the wrong way up one-way stieets., unleýý therp's a ýiqi) V, 11,ýt
visitors can.
I........................................ Don't ride across pedestrian crossings
P-Mobile phones
I.............................
I-Motorcycling Dos fog ists
all4l 41011"t's Illotol
..........
li-Driving on
motorways " Expects udden movements by cyclists, e-6peci aI lyiriwirWVwk-ljwr ri (I ým h. id i. -d
- .............
..........................
... -,
0-New drivers surfaces.
.1.1 ..............
........... I............. " Watch for cyclists on the inside when you turn left
I- Older drivers look for cyclists before
" Always opening a car door.
...................
b, Seat belts " Give cyclists turning right extra consideration.
............... Don't squeeze past cyclists - give them space. at least half a rar'ý width
11-Teenagers
Don't dazzle dipped headlights, the way
........... cyclists - use you would with another car
P,Top ten tips Don't the kerb - they
get annoyed when cyclists ride awayfrom need to avoid drains
.1-............... and potholes, and be seen as they to junctions
'-Tyre safety come with side roads
.......................................
P-Winter driving
...................... .......... Our leaflet Drive safe. cvc-fe 3afefor more advice. Visit our site for vounu, i , v, li,; t,,
>
qD Internet
Figure 10.1 Advices on "Think! " website
Pictures. From our focus group discussion, some people feel that shocking
images could have a big impact on people's safety awareness. These have
already been used in a UK government TV campaign showing the
consequences of cars colliding. If pictures from accident scenes can be seen
in our system, this might be also alert people about safety issues. Pictures
from an accident location will show geometry features of that area. The
pictures can be added to the full accident information panel (see figure 10.2).
However, this raises a lot of ethical issues and requires careful consideration.
it may be very disturbing to the relatives of anyone killed in such an
accident, even if graphical reconstructions were used rather than actual
photographs.
144
The Impact of Localized Road Accident Information on Road Safety Awarencss
Accident Picturcs
infon-nation
Figure 10.2 Pictures in the system.
b) Interaction
Usability is the key issue of every Human Computer Interaction Systern.
Although the system evaluation shows people are satisfied with the current
interface, there are problems. Some symbols overlap, the query speed can be
slow, and so on. In addition, further development might support the

following:
Display accidents by area or roads. When users want to see the situation
in a specific area, such as a junction, they could use a mouse to drag a circle
and then information about all the accidents in this area would be displayed.
Similarly, if users click on a road or select a road name from a drop down
list, accidents on that road would show up.
Fish eye function. Although I provide zoom in and out, users sometimes
want to see detailed infon-nation on a large scale background. Providing
functions like a magnifier, users could quickly browse the whole rnap with
more accurate infon-nation.
Self defined icons. Currently, I use different shapes to code accidents by
their severity. In the future, users could define their own icon groups to
145
identify other relationships between accidents, such as different weather
conditions, different car type. Mixed icons could represent multi level
information. These might help users to make quick comparisons and get
more information directly.
c) Prediction
As mentionedin chapter1, Predictionis anotherway to reduceaccidentsby

identifying potentialproblemareas.If we add prediction into our system,
people can see what may happenin the future. But first, we need to do
to
research seewhetherpeopleare interestedin prediction.The experiment
may be similar to the one in this study.If the predictioncan affect people's
decisions,thenthe studycancarry on to build a predictionmodel.
Researchinto prediction of road traffic accidents has been carried out for
many years. There are many approachesto prediction. The first is accident
observation or a 'before and after' study. Accident totals or rates can be
comparedat treated sites in the periods before and after an implementation;
sites can be ranked according to their observed accident totals and/or rates
[53]. But accident counts are prone to sourcesof error [33,56,43]. Accident
rates are normally used to allow for variations in exposureto comparesafety
levels at different locations or times on the assumptionthat a lower accident
rate indicates a safer site. However, if the relationship between accidents

is
and exposure non-linear, accident rates will vary with exposure so that a
lower accidentrate would not necessarilyimply a safer site [34].
Then researchershave tried to find suitable prediction models which relate
accident occurrenceto traffic volume and a range of attributes such as road
geometry features, traffic control features, and so on. In the early years,
146
people used multiple linear regression modelling, with its assumption that
accident counts follow a normal distribution. Then people found it is far
better to model the processusing the Poisson distribution for the frequency
of accidents in a given period of time at any one site. Examples of this
model can be found in Jovanis and Chang's research [48]. Since people
realized the advantagesof the Poisson model over the standard regression
model, generalized linear models have become popular to develop the
desired forms of relationships using a Poissonmodel [47,101 ].
Although the Poisson model offers significant advances in accurate and
reliable modelling, it has some weaknesses. One is called the
"overdispersion" phenomenon. In the pure Poisson model, the variance is
equal to its mean. But in practice, actual accident variance has been found to
be greater than the mean. Recent studies have proved that the negative
binomial distribution might be more appropriate becauseit allows greater
variance in the data and thereby deals with the overdispersion [63,1].
There are other models for prediction, like Andrew and Stig's multilevel
model [5] and Matthew and Ioannis's hierarchical tree-based regression
model (HTBR) [58]. Andrew and Stig think their multilevel model helps to
quantify the various influences on casualty outcomes. HTBR model is a
non-parametricstatistical method.
After comparing these prediction models, in the future, we would use
generalised linear modelling techniques. The distributions of accident
counts will be assumedto follow a Poisson distribution. The main advantage
of the Poisson distribution is its simplicity. Although the overdispersion
phenomenon is one of the weaknesses of Poisson distribution,
overdispersion does not affect the coefficient estimatesbut does causetheir
standarderrors to be underestimated[61]. Some studies show that the values
147
of the model parameterswere almost the sameregardlessof the distribution
assumption[51].
The model structure is:
Y= aNPexpEpjXij
Y is the expectednumber of accidents,N normally is the traffic flow

(AADT), x variablesdescribingroad geometryor environmentof the road,
a,p, 0 areestimatedparameters.
The development of our prediction model would have four major steps
using the GeneralizedLinear Model.
Identify potential parametersin the prediction model. We will consider
two different ways to do this. We prepareda questionnairefor people in
the Land Service Department of Glasgow City Council. In the
questionnaire (see Appendix C), we list many different parametersthat
could be used for prediction, most of them from existing prediction
papers mentioned above. We then asked them to scale the parameters
from I to 10.10 means the most important. Alternatively, we have
considered Locally Linear Embedding. This can help us to find useful
parameters for prediction. Locally linear embedding (LLE) is an
unsupervised learning algorithm that computes low dimensional,
neighborhood preserving embeddings of high dimensional data. LLE
attempts to discover nonlinear structure in high dimensional data by
exploiting the local symmetriesof linear reconstructions(16). We could
also combine the results from these two methods to pick potential
parameters.
148
The Impact of Localized Road Accidew Information on Road Salc(y Awareness
2 Data Collection. We contacted the Department of' Transportation, and
they provided us with 10 years (1993-2002) tral'fic volunic data and
major road network information in Glasgow, such as road type, speed

limit, number of lanes, width, length, and so on. We have already
obtained 11 years accident data (1991-2001). Figure 10.3 shows in
example of traffic volume data and major road network information.
x I . 19dA
Ek.. Lcit jj- Jýt F2rmat Jools Qata tyd- ti'lp AdOLPDF -qx
-j .4 -ý -j -j
il
-j -A -
') - '-a A1
1: -L1,1 44 MA ý1(1.0111
if
MAI 1% ipý-- I
Ir",!ý *)
Al LACndý,
A c0 ýP 1ý F: C3 HIIi M-N
ZAISI.
kWOR CPI d0pened I Mosed POAOSI kýcýFj Soj_LRCat Id Name MLPý5E. I.. S Ref EIS Ref N
101ju 01/01/11AU-1 NM t4i 1NI __ rýl-
-1.11
1001
1()09
01/ul/19811
01,01/1%2
NM
NM
Nfl
AB
17H Im
IrAf) PI)
rVA
;I-t P-J, y
'IA77
" "ll I! III III,
ý'j Ira H
OllI
14.41.00
1010 01/0111982 NM M 5140 PU Street Mun,".. A// 25OWU 1,646,10
1011 011U111982 NM AB 220 PU I B"dg. g.? A89/A/49 259560 66471,f]
1027 01V/1982 NM A77 320 PU 0 Street AB AB14 2W40 FFA720
1029 0110111962 NM ABO 10 PU AB Ma 2b2900 61,6000
uld Road
1033 010/1982 NM A82 10 PU we'le'rr A,904 A804 256100 666500
1030 01YO111982 NM A89 10 PU e AB Millersion 2605W 15154770
1090 01V/1962 NM A730 20 PU Street A7 20 A8 259000 664200
1091 Olnl/1982 NM A736 100 PU Road A726 Banhead 251900 661200
1094 01)01/1982 NM A7 39 40 PU Road A1314 A82 254600 b68WO
109701M111982 NN1 A749 40 PU ck Road Boundary ýV, l 26 1220 663320
1106 01/01/1962 NIVI AD04 10 PU Geo'ge's A014 A82 258" 666300
1112 OM111982 NM A814 120 PU WA MIYAW4 B'o-ela 258400 6651()0
1143 01101/1982 NM A879 10 PU street M8 Hayston 259100 615,7800
10013 01V/1982 NM KAO 200 TM NIA `13 '12 263000 666600
ý3 '52
10814 01,UI[1982 NNi Ma 120 TIVI N/A Als 22 256000 664200
10819 01MI/1982 NM AS 100 PU Road A736 A739 2,52650 666"
10820 01MI/1982 NM AB 170 PV Street A730 A814 259200 664600
10821 OlMl/1982 NM A8 230 PU Street of A39 AW3 260000 665330
10836 01101/1982 NM A74 321 PR Road M74 Mount 267500 662500
241 9260 10918 01V/1982 NM A804 40 PU Road MBIA81 MO/A879 258800 666374
9260 10923 01/0111982 NM AB14 90 PU Pa,k D-e Kingsway A739 254000 667100
9260 10924 01)0111982 NM AB14 ISO PU jannes A77 258500 664930
ýZ? 1 9260 11020 01YO111902 NM A739 20 PU od Road Me AB 253600 665000
lb "14 261000 665900
291 9260 20809 01,13111AI-2 NM NIU 180 TM N/A
'25 '2A 2154000 664600
9260 20811 Oljül/1982 NM hm 100 TM N/A
921.0 20815 OlU1/1q02 NM m1 PU EI-e E3768 B768 A761 255500 664550
210
31 9260 20816 01A]111982 NM m INI PU A814 Saltmarki, 2,S9400 6647(it)
. -1 ýnpi7 nini/icn) NIM AR

an Pi i 'Zý -1, C. 01.1ý1 ?Fn win F4,', Aln
11 p
Ready NLIM
Figure 10.3 Annual Average Daily Flows and Major Road Network
Information in Glasgow
3 Using statistical software (R project, which is an open source statistical
package. littp: //www. i--L)i-oiect.

orý-),
/'), it is relatively straightforward to do
regression analyses, then withdraw those parameters that are not
statistically significant to prediction.
4 Calculate all the coefficients by using R project. Both step 3 and step 4
149
will use 5 years accident data, and the data of the remaining years will
be used to test thesecoefficients.
This section has sketcheda possible researchagendato extend our work on
accident information from purely historical data to include accident
prediction. As mentioned at the beginning of this section, once research
suggeststhe general public are interested in prediction, this fiInction can be
added into our localized road accident information system. Then
experimentsare neededto determine whether possible accident information
about the future would affect people's safety awareness.
Above all, human beings are the most important factor in road accidents.
This thesis is concerned with trying to use localized road accident
information to increase people's road safety awareness. The key
contributions of this thesis are, firstly, from risk perception questionnaire
and focus group discussion, we find that current road safety campaign
websites don't have great impact on road users' safety awareness.People are
not very interestedin the information presentedby those websites. Secondly,

through our road planning experiments,we find that localized road accident
information will affect road users' route planning decisions if they have
access to such information, which means road users will consider local
accident information before they plan their journey and this could
potentially increase their safety awareness. Although we made some
progress,there is a great deal of work to be in this areain the future.
150
Appendix A: A brief Review of road safety research
1 Statistical Models
Statisticians, engineers (road and car), psychologist, economists,
sociologists, politicians all work in this area. The study of occurrence of

is
accidents mainly done by statisticians becauseroad accidentsare regarded
as chance processes.Statistics can be used to find out certain patterns of
accidents. For example the relationship between the number of road

fatalities and a given number of motor vehicles and a given population is
called Smeed'sLaw [45]. This law, in its most general form is describedby
an equation:
DIN = a(N/P)-b
D represents the number of road accident deaths in a country, N is the
number of registered motor vehicles, and P is the population. Smeed fitted
this function to 1938 data for 20 countries, and got a value for a of 0.0003
and a value for b of 0.667. This law has been retested in 1968 against data
from 16 countries for the years 1957-1966 and once more in 1970 against
data from 68 countries for the period 1960-67 and it was found that it still
performed well with the same coefficients. This example shows, although
some people have argued that road accidentsare chanceprocesses,there are
certain statistical rules which may apply to all accident situations.
Statistical models can also be used to support accident prediction. There are
two main topics in prediction. One is to find an ideal prediction model

(algorithm). The other is to identify suitable parameters for the model.
Generalized linear models [47,60] and Negative binomial regression
models [63,1] are the two most popular models for prediction. There are
many others, such as Andrew and Stig's multilevel model [5] and Matthew
151
and Ioannis's hierarchical tree-basedregressionmodel (HTBR) [58]. But all
these models have their own strengths and weakness. For example,
generalized linear models suffered an "overdispersion" phenomenon.

According to the pure Poissonmodel, the variance is equal to its mean. But
in practice, people found actual accident variance is greater than the mean.
Different models will have different sets of parametersand even the same
model may have different sets of parametersfor different areaswhich have
their own characteristics. Expert judgment, statistical analysis tools and
researchexperienceall could help to identify those parameters.
2 Accident Process
Research into the accident process can be divided into three categories:
vehicle, road environment and road users.
Vehicle defects causeaccidents.The layout of seats and controls, tyre
condition, vehicle lighting, braking performance,and steering all affect

driving safety.Therefore,vehiclescanbe madesaferto drive in two ways(1)
by improving their design and (2) by keeping them in better condition
throughout their lives. Through research on individual car accidents,

their
manufacturers, and
engineers designers
can find existing defectsand
thenimprovedesignfeaturesandmanufacturingtechniques.For example,in
December2000,Ford recalledtheir 2001Ford Focusbecauseof a defectin
the seats.Accordingto the report,"the subjectvehicleshavea folding 60/40
secondseatthat becomes
a load floor in the folded position.When the 60%
portion of the seatis folded down and a load is appliedto the front edgeof
the load floor, the outboardhinge pivot could disengagefrom the hinge.If
the seat is then returnedto the upright position without re-engagingthe
hinge pivot, the seatand seatbelts may not provide the intendedlevel of
152
performance in the event of cmsW' (www. automotive.com).
The road environment plays another very important role in road safety
research. The following are features which may influence the risk of
accidents[29]:
a) Road geometry, such as road and lane width, shoulders and medians,
horizontal and vertical curves, vertical alignment, the layout of
junctions, and roundabouts.
b) Road delineation and marking, such as lane markers, pedestrian
crossingmarkers, and the marking of the entrancesto and the exits from
motorways.
c) Traffic signs and traffic signals, such as direction signs, warning signs
and mandatory signs.
d) Road surfaces, such as surface texture, spray problem in wet weather,
surfacesof verges, shouldersand central reservations.
e) Lighting, such as street lighting and anti-dazzle screens.
0 Obstructions on and off the carriageways,such as street furniture (lamp
posts, trees, telegraph poles, traffic signs... ), crash barriers, kerbs,
in
obstructions the road, parked vehicles.
g) Weather,such as rain, snow, wind.
Researchon thesefeaturescan help people to find and build safer roads. For
instance, for two-lane roads, 12 foot (3.7m) lanes have been found to be
safer than 9 or 10 foot (2.7 or 3m) lanes [35]. Researchalso found that roads
153
The Impact of Localized Road Accident Infor-niation on Road Safety Awareness
with long straights and few bends, particularly if they are sharp bends, may
have higher accident rates than roads with more bends [73 1.
According to the research work by Sabey in 1975 [80], road users make the
most important contributions. He attempted to identify the "main
contributory factors" responsible for 2130 accidents which were
investigated in great detail. Evidence was obtained by observations of roads,
vehicles and road users, by interviews, and by assessing errors which were
made by the road users, by defects in vehicles, and by adverse features of
vehicle and road. The answers depended on opinions which are formed by
members of the research team. They categorized those contributory factors

into 3 areas (road environment, road user, vehicle), some accidents being
caused by a single factor, some caused by multiple factors. The result shows
8.5 percent of the causes could be assigned to vehicles, 28 percent to the
road environment, and 95 percent to road users; these add up to more than
100 percent because many accidents were considered to be the result of
more than one contributory factor. Figure A. I shows this result briefly.
Road
environment Road user Vehicle
Single factor 2.5 65 2.5
Double factors 24 4.5
Treble factors 1.5
154
Much recent research focuses on driver behaviours to find potential
dangerousactions or bad habits which may causeaccidents.
The driving task has three main aspects: the intake of information, the
making of decisions and the control of the vehicle. Any error in these
aspects could be dangerous. Drink driving, drug-driving, speeding and

driver distraction are typical problems which could affect drivers' behaviour.
According to the "Think! " website[89], on average 3,000 people are killed
or seriously injured each year in drink drive collisions and nearly one in
seven of all deathson the road involve drivers who are over the legal limit.
Researchhas shown that speedand speedrelated causesare major factors in
about one-third of all road accidents. Surveys show that almost all drivers
and riders exceed speedlimits at some time. Observation of vehicle speeds

in Great Britain in 1998 showed that 69% of cars exceededthe 30mph limit
and 29% exceededthe 40mph limit in free flowing traffic [94].
Education, training, propagandaand enforcement are some of the ways in
which, without altering the roads or the vehicles, an attempt is made to
improve human behaviour on the road [27]. The first three methods can
raise road users' safety awareness so that they understand what kind of
actions would carry risks. Related traffic law and legislation could also
reduce dangerous activities because road users have to consider the
before
consequences they take any improper actions.
3 Accidents Consequences
The researchon accident consequencesplays another important role in road
accidents investigation and research.Injury to road users is one of the most

important consequences from accident. In order to save lives, much
155
attention has been paid to the causesof injuries. Alan formed a hypothesis in
1967 [94] that "many accidents cannot be prevented, but most traffic deaths
and injuries can be". Although I don't agree with the first part of his
hypothesis, from figure in table 1.2, we can see there is a big reduction in
the number of killed or seriously injured from 1993 to 2002.
According to the research [29(pl2l)], there are two most important factors
in injury production. First is the velocity change of the vehicle, the other is
the duration of the impact. Alan believed there are 5 factors which can affect
the severity of the injuries:

1) "the magnitude of the force
2) the area of the body that is struck
3) the rate of onset of the force
4) the duration that the force must be withstood
5) the total areaof the body over which the force is distributed" [94]
The injury researchin 1963 [72(p462-p466)] showed that the head related
injuries accounted for most of the injury cases. In Alan's research [94],
similar results were found. His data were collected in an analysis of 150
accidentswhich were selected from a total of 3,308 accidents that occurred
in 1965 in California. The data showed that there were 32 percent face
injury, 19 percenthead injury and 2 percent neck injury.
After analysing injuries, two kinds of protective measurescan be taken to
minimise the injury when an accident has occurred. The first kind is the
protective devices which can restrain and control the movement of the
occupantand prevent impacts with the interior of the car. The secondkind is
alterations to the construction of the car. The seat belt is a typical device of
the first kind. The most important function of seat belts is to protect heads
and bodies of the wearers in frontal impacts. Table 1.4 compares that the
156
injuries to the occupantsof the front seatsof cars wearing and not wearing
seat belts [38]. The data were based on 1163 unbelted and 490 belted
front-seat occupantsof cars.
Region Unbelted Belted
gesinjured) (percentages injured)

Head 23.7 10.6
Neck 1.2 1.6
Shoulders 2.1 1.4
Chest 5.2 3.9
Spine 1.5 0.6
Abdomen 2.0 1.2
Pelvis 0.7 0.4
Hip joints 0.6 0
Thighs 2.1 0.6
Knees 2.2 1.8
Lower legs 1.5 0.6
Feet/ankles 1.5 2.4
Arms 3.7 2.0
TableA. I Injuries comparisonbetweenunbelted and belted occupants
From the above table, we can see, apart from neck injuries, seat belts do
help to reduce injuries. When all types of accident are included, seat belts
reduce the probability of serious injury by about 42 per cent.
In the secondkind of protective measures,an example is the design of the
steering assemblyso that even when the front of the car is heavily deformed,
the steering wheel is not forced back on the driver. Another example is
safety glass for windscreens. The objective of safety glass is that when
occupantshit the windscreen, fragments are less liable to causeseverecuts.
There are two types of safety glass in Britain, laminated and toughened
glass. After fracture, fragments of toughened glass are less sharp than
laminated glass. But laminated glass has better visibility than toughened
glass.
157
4 Economic Impacts of Road Accidents
Road accidents have a big impact on national and regional economics.
According to WHO's (World Health Organization) "World report on road
traffic injury prevention in 2004", the cost of road crash injuries is estimated
at roughly 1% of gross national product (GNP) in low-income countries,

1.5% in middle-income countries and 2% in high-income countries. The
direct economic costs of global road crashes have been estimated at US$
518 billion in 2004. The estimated annual economic cost of injury in China
is equivalent to US$12.5 billion --- almost four times the total public health
servicesbudget for the county.
Road accidents also affect the economy indirectly. People aged between 15
and 44 years are the most productive age group. But they are also the
majority group in road traffic injuries. The economic impacts of injuries in
this age group are therefore especially damaging. According to the WHO,
people in this group, "tend to affect productivity severely, particularly
among the lowest-income groups whose exposure to risk is greatest and
whose earning capacity is most likely to rely on physical activity".
Although governments are paying more and more attention to road traffic
safety, the WHO believe "current road safety efforts fail to match the
severity of the problem. Road travel brings society benefits, but the price
society is paying for it is very high."
158
Appendix B: Collection of Road Accident
Information
Accurate, adequate and reliable road accident data are crucial to any
systematic, scientifically-based analysis of the road accident situation and

the developmentof rational countermeasures.In this section, I describe how
accident data are collected in Great Britain and what kind of data are
needed.
In the UK, the collection process and accident data collected vary across
local authority and police force areas. They reflect local road safety
requirementsand circumstances.However, each local area is askedto report
the same set of accident records for national purposesand to transmit them
to central government. This is the 'STATS19' collection system, which was
established in 1949. STATS19 is a standardised form used by police to
record road traffic accidents.
Recordedaccidentsonly consistof thosein which someonewas injured.

Damage-onlyaccidentsdo not appearin the collectedaccidentinformation.
Injuriesareclassifiedaccordingto the following criteria [29]:
a) Slight injury. An injury of a minor character such as a sprain,
bruise or a cut or laceration not judged to be severe.
b) Serious injury. An injury for which a person is detained in
hospital as an in-patient, or any of the following injuries
regardless of whether he is detained in hospital: fractures,
concussion, internal injuries, crushings, severe cuts and

lacerations,severegeneral shock requiring medical treatment.
159
C) Death. Death within 30 days.
The STATS19report form consists of an accident record, a vehicle record to
be completed for each vehicle, and a casualty record for each casualty
arising from the accident.
The accidentrecord includesthe severity of the accident,the number of
vehiclesand involved,
casualties time and location,road classand number,
speedlimit, hazards.
weatherandroadconditions,andcarriageway
The vehicle record includes type, location and manoeuvre at time of
accident, and details of the driver (age, sex and breath test results);
The casualty record includes casualty age, sex, injury severity and whether a
driver, passengeror pedestrian.
Figure I is an image of an accident record in a STATS19 form. Figure 2
shows a vehicle record in a STATS19 form. Figure 3 shows a casualty
in
record a STATS19form.
160
DETRtSOMO Accident Ro"rd Attendant Cimurnstances STATSIO(I"9)
RoadType 0 1.202 Pardestrum

CrosskV 1.23 RoadSurloca,
Conchon
1.1 RecordType 1.14
I A... kboul
- HumanControl I Cry
11 N- mood
IS A.. WWmkWt,. -d 2 0. .y wed 0 No' fac" Welh so Rob" 2 war loemp
SDU. lM g )-26. - F11Y a" WIRY " 1 11-
PoliceForm 4 Dd. -Ang.. Y. S. -knes by . 0.1had p~ 4 fd 0ka,
12
6 91ý91.caftp. " - W4. "M read I C., "byddar.. "pow I fbw hurbowwwwraverkm
1.3 Acckkn Rat No 6 SWO -bg--y .2 In- lote fn 2 CoArd by ~ adWbW p~ 4 00 0 dmal
In dW.Cft* 7 Mud
7 Shv. cem-g-w -a bras too Way
1.5 Number of Vehicle P-Ayl I. 2Db Pedesirian Crossing
Records 8 Sno caMagow" -4 ar - tares - Physical Facilrues
Special Conditions at Sit* [3
0- -y-pa" 1.24
9 Uld- p1r) f. My
1.6 Numberal Casualty soý 0 ""
I Z4,400. d. kA.. k antic wend &A
Records 1.15 SpeedLimit(mph)
4 pwý PAFK "- ar Ww 2 Aummirec Ift Wgnd p. "
Y 1.16 JunctionDetail ED ). -a- I NO wo-*v 2 parmarwalow molhwI ki
1.7 Date S Pdadaript-tValk" dW4c*m
arcbKumd
W Notatorw"20mawrodpAxten AMC" 4 A, sp, -
01 9CwwMnAW-waVwc, 6 RoadMOM 60.0re
ON. 02 Mw I FmbrMp ar o-way
I'll Terw d Day 03 Tarsaigg-ilp-don
24 h- 05 sop Pow
06 cm.. da 11.21 LightConditions [3 1.25 CarriagawayHazards E3
07 hlu")-dan
1.10 LocalAuthonty at 1 0.1,191k

Or" 110"1 a
09 Olnerp"M 2 Dayot ne i&w OgIgh I Dhiodgd Wkb MadIn O. Mg... y
S~" ww- 2 Ovw Olga in cwftg"
1.11 Location Juncdort Accidents Only 4 Da*- swaai1gVftp,. -I ord 3
10 dkjK05 GddRefervaceM. MM 6 Dwrlwý %V.M M02%F but 4 call I. -An-. y
13 6 Dwký- no W. 0 NV" I 01har &OW a In
17 JunctionControl
, I ALMhorhadPanwn I Dan,,ý weat
Ewi-9 2 AUk.. k VeRkSignal
3 S"'paig.
Weather 1.26 Place Accident Reported [3
4 Gk... Y. 19.wnd*g, 1,22
1.12 1 st Road Class a Uncw&okd
I Fiv-10ýAhighOrkKAM I AA-
I bk"-y is 2nd Road Class I F. " vrM-0 higharOft 2
11AM I waý" 3 U-" WOW MghWnda
.A 2 A(M) 4 F#-1hhYhaWM
49 3A Ral"ýMh highwkl&
C 46 Sn-" WM" *x1s 1.27 DETR Special PMx[=
: C 7 Fog-M-11hiumd
Ud.. Md a Cw-
1.13 tat Road Number Lki-n
= Road Number
Figure B. 1 STATS19 - Accident Record
DETRISOIWO Vehlcl* Record STATSIS (1"9)
[] 13
Rmrd Type ED 2.8 2.11 Skxkkq and Overtming 2.16 FirstPoirdof bypict
2.1 v-"kl-
Con"ssm-nwft
Iscit 1:1El
F- T.
21 Nowiolikla 0 0 Didm b"d a ORM"
26 A-M I It ss Park.* I SkkMd I F-d 4 t4N"oft
voldd. Wcord Eo
2 NE SSW am Stlait, * SkIddd aid ii-ift-d 2 Sack
M 311 7w 9 Jack4mAvd
2.2 Police Form
4 SE 6 NW a Mrb 4 Ja&-brA%dNWiwaliffimil 2.17 Other Vehicle Hit
2.3 AccidertRot No 6 OINNN'" Rd ft ofcow imbicle
M Pwt(s)DamoqW
2.4 Vehicle Rot No 2.12 HkObpdiliCarriaqewuY 2.18
2.9a Vehicle Locaticirt at rý ol
ACCKIBFt- Road Co m- 06 0 N- 3 CIP.Ift 6 Udw. M.
2.5 Type d voww (" Piwý, Iddorl OT Ildlard I I Fm-t 4 N-wdft 7 Mkl"
02 P. O.. " Of OpM CNION CO 2 8. * 5 Pd
01 P. Myd. 16 00- - I E'Aarkgo. ead. .0 03 P. 1k.11Wicia-Ift (IS Cýal Wails
04 PavitdWftle 2.21 Semof Driiror 0
02 M. Pad 16 RkIftniviriie 3 MIMMIN.A
03 mokir Cycle125cc Sim 4 Qnft~md - Wall 10 Kb
05 D"-wd 11 otwo*d I MIS a Frm 3 Not traced
is T,"Iugm . 11 2.9b Vehicle Location at Tom d C]
04 MoWcyds~lZScC [: ] 1313
is C.. ft A ccKlerS - Restric tedL ane / 2 13 Vehicle UnNin g Carna," ay 2 22 Age of DrKw
09 Tod . .
09 braes -P Amy firm Main Carriageway Y-
10 20 Golm 0 Did t b- Wg-ey
u- d 0 O... lMC. Wq-. y-MI-WkWd I 1.114C.
11419-" 2.23 Breath Tot
pia-9.0-44
11 lb-CoMh(1701' Imeea, -P I Loftca logowayroursiftandrobwadod
- P----V- .. e, zi G om I Tra,,i I ugit amwe& 1 0 Nd Wpkbi. S D%w m
14 OOMý Wkb knowimpanclaimir a Ikele- 4 L. It cariagio" *I" -rA* cww-d I P*Wh* at
&-yMmkoilliggAdodkommoy) "o. 6 Nd
3 NO-q-MMd
26 Toming and Ar"abm 13 4 Cyd. MM On aiiiii
6 cya-y*p-wk-.. b
I Lft -"-" W*b ft w"
4 PAknWlDpmvid.
0 Pkim-arodkWallm 3 Cavoiwi -. Wg... h 6 Let Map-" dhido aridm-d
, Anh#"W at*% 4 Sr4iitrabw 6 C)nby4)ywhm*mAdw cow" Pose"Ob". 2.24 Hit and Run
2 DWAw-ii"Irmar goirer 1- 7 E.*" "W bud ArWillof I L. A.. Wgm". Nift
S L., AVWy-by. hadW. Adm, I US= Mg- -ý off" are Nboundd 0 Caer 2
CD I pm.WR- " hit
27
.
Manomwes 9 2.14 Hil 01ject0" C-. M
w-y
01 R-wrilp 12 Chur" 2.25 DETRSpecialProjec=
oz P~ 13 ovoirts" 2.10 Adim Location of Vatic@: ] w a"
at First kripact 01 Rind Wy, I Tft dpg 2.26 VeNd* Reigistation
03 1-1, g.. hd velid-lisdisift
beh. 1d.p 14 02 LAW pow Mark (vw)cj]: ]=
04 S"P" "- m 0 Noralm-ow w om m -M-4 03 T-waphp-1--dily" SpecialCodo.'
06 Star" Is 0--w" I VhkWqp-d*. gkAd-arpdW , 04 Trae 2 F-0 I Do-oft 4
05 uti'a is k-d- app-h 06 a. am I a. stk. 3 mftq
07 T,. "bt lowd a Vhkbl.. Wftdk.. M. 06 CýW-hbwrdwr
06 vi" w Qumkift 17 Gohq Mead 3 Vdkbd-W). CO-. PXkWA 07 Now.M. . oftift -h harder 2.27 Driver
h. W - Mw.. 00 SuWwgod Inmaikor OMM*ftly) Postcom
09 T-*-911gm
10 vm" 0 ft-dgm is Go" ahftd 4 Did not vvpm 09 ErOmrddkk sp., w C."u a N-UK.. iftt
11 Cf.. " b.. Wkt 10 cotierpi-ittod I UM- 3 P~ and
Figure B. 2 STATS19 -- Vehicle Record
161
VETRISOIWO Casualty Rocord STATSIS(1"9)
RacmdType [jo 3.7 Sexof casusky 0 3.11 Pod.WA. M.. ý* 0 3.13 SdWWPupNCmusky 0
3.1
31 N-CM. My-d I mms 0 Nt a P. Mmem I s&. dpýPNwPW. YqD$rkm
36 A-W. 0 O-fty -d 2 F-W I how
a Q-" 9- dtý% -wwo 0 ClhW
-
byp. b. d. mft., ýWtd.
a C-" k. awmnid.
4 C. " in. ON. % omo
byp. ftdwmWw .. Ndo 13
[10 Age at Cmwky 6 14 Car P"S"W
3.2 Poke Foý 3A W." --, - -M 3.15
I'l-W It "... my Y- I-m-kq w pw"
4b Urftgý" #Aft-l, 0 Nds-
- MW 9
bmr. *q a O. OV. -Wd Fmmýp-qw
. 901. ". dwd. ftwý
7 Vh". bngb, 1-f-bg
AccWwtRol No sr-q of comoky
3.3 3.9 13 S Vh" wftd. 9 I* I- buk
I raw llft
2 5.6- urf- a
3 SW 3.16 BLmorCoschP*ssengw [3
3.4 VeNcleRd No
Mir"
Swn*qpm-W
3.10 Pedestim Locatim 3.12 PainVion Di 4 S.Owp... w
31 CaujokyRd No 00 PAM.
pd. Mm C.. P- Pk* bwd
91 In whpwA aam" m
IN
42 m cwftg.. ". 9 20 2 NE
1-d-" 3E 3.17 DETR SpKW Prqel=
03 ft cwd"mp . 9 wo- zio. 4.
3.6 Casusky Chm 041.

VAN. 10 AN- d pd. 04- 7w
DAý w ddý 05 IN h. A- : mw
. Vhicb. pNmp-w"w 0, Onfoci a.. " 3.18 Comoky
3 On rdugs. cw" kWW a cwftid 0 560" so Postcods
ahýd . Ibg-y. m- SP. " -d-
cw" kwd a C. " I Ud-
09 g 2 H-UK MOM
10 Urdý w olhw
Figure B.3 STATS19 - Casualty Record
162
Appendix C
Questionnaire for Road Accident Prediction
1. What kinds of accidentpredictionmodel areuseful to your job?
a. all accidents
b. all injury accidents
C. certaintypes of accidents (single accidents,rcar-endaccidents,crossingaccidents,
turning accidents... ), or on certaintype of roads(Motorway, A, B, other)
d. other,pleasespecify
1. Normally, every prediction model has algorithm for road links and junctions. In
order to predict your chosenaccidenttype, pleasescalethe following data based
on importanceto the prediction.(from I to 10,10 the most important)
RoadLinks:
a) traffic flow 123456789 10

b) length of road section 123456789 10
C) speedlimit 123456789 10
d) one/two-waytraffic 123456789 10
e) numberof lanes 123456789 10
f) road width 123456789 10
g) speedreducingmeasures 123456789 10
h) cyclist facilities 123456789 10
i) footway 123456789 10
j) centralisland 123456789 10
k) parking facilities 123456789 10
1) bus stop 123456789 10
in
m) otherparameters STAT 19
163
Junctions
a) traffic flows 123456789 10

b) numberof lanes 123456789 10
C) traffic island 123456789 10
d) turning lane 123456789 10
e) bicycle facilities 123456789 10
f) signalised/non-signalised 123456789 10
g) numberof arms 123456789 10
h) otherparametersin STAT 19
164
Appendix D
Questionnaire for gathering experts information and initial system
requirements
The Road Traffic Accident Information Systemhelp people find out about accidents
detail in Glasgow. Currently, the Road Traffic Accident Information Management
System is only prototype. To build a functional system, we need to get more
requirements from potential users. Traffic police and people in the Glasgow city
council who deal with road traffic are selectedto answerthis questionnaire.
I What is your job title?
a) traffic police b) road planning engineer c) road accidentresearcher

a) other,pleasespecify
2 How long have you worked as this job?
a) lessthan a year b) 1-3 years C) 3-5 years d) more than 5

years
3 How do you deal with accidentdata?
a) recordnew data b) classify and submitto higher government

C) analysishistorical data d) other,pleasespecify
4 How often do you dealwith accidentdata?
a) almosteveryday b) oncea week c) few times a month d) several

times a year
5 How often do you usea computer?
a) almosteveryday b) oncea week c) few times a month d) several

times a year
6 How do you rateyour computerknowledge?
Novice (1) 2 3 4
5 Expert
L-
-1 165
7 Are statisticalanalysismethodsinvolved in your daily work?
a) Yes b) No
8 How do you rate your statisticalknowledge?
Novice 1 2 3 4
5 Expert
If you have ever receivedtraining on statisticaltechniques,pleasestatethe level (e.g.

school,college,university... )
9 We have already got accident data for the Glasgow area. What kinds of accident
information do you want to be displayed in the map-based system? Please tick the
top 5 you think are more important.
a) YearofAccident
b) Accident ReferenceNumber
C) Police ForceCode
d) Accident Severity
e) Numberof Vehicles
f) AccidentDay
g) AccidentMonth
h) Day of week
i) Time of Day (Hour, Minute)
j) Location (Easting,Northing)
k) I" RoadClassand I't RoadNumber
1) RoadType
M) SpeedLimit
n) JunctionDetail andJunctionControl
0) 2ndRoad Class nd
and RoadNumber
2
P) PedestrianCrossing-HumanControl andPedestrian
Crossing-PhysicalFacilities
q) Light Condition,WeatherConditions,RoadSurfaceConditions
r) SpecialCondition at Site
S) CarriagewayHazards
166
10 The following are somepotentialways to display accidentdata.Pleasetick the top

5 of your choice.
a) by year
b) by road
C) by road type
d) by time of the day
e) by day of the week
f) by month
g) by light condition
by weathercondition
by road surfacecondition
others,pleasespecify
I Someelementsof the systemdisplay can be changedby the users.Which of the

following featuresdo you think are better if they are editableby users?Tick top 5.
If you areunclearwhat this means,pleaseask.
a) Roadcolor
b) Roadwidth
C) Accident point size
d) Accident point color
e) Information to be displayedon the right handside text area
f) Information to be displayedon the tooltip
9) Backgroundcolor
h) Font ( e.g. style of text)
12 In addition to accidentsdata,we also have casualties'dataand vehicle datawhich

currently are not displayedin the system.Do you think it is necessaryto display
this informationwith the accidentdata?
a) yes b) no
If yes, what kind of casualty information do you want to be displayed in the

system?Pleasetick the top 5.
a) CasualtyClass.(Driver or rider, Passenger,

Pedestrian)
b) Sexof Casualty
C) Age of Casualty
d) Severityof Casualty
PedestrianLocation
PedestrianMovement
PedestrianDirection
167
h) Car Passenger
i) Bus or CoachPassenger
j) CasualtyType
k) SeatBelt Usage
13 On the abovelist, which onesdo you think are necessaryto be madeas prepared
queriesto display accidentdata?Pleasealso tick top 5.
a) by CasualtyClass.(Driver or rider, Passenger,

Pedestrian)
b) by Sexof Casualty
C) by Age of Casualty
d) by Severityof Casualty
e) by PedestrianLocation
f) by PedestrianMovement
g) by PedestrianDirection
by Car Passenger
by Bus or CoachPassenger
by CasualtyType
by SeatBelt Usage
14 For the vehicle data, what kind of information you want to be displayed with
accidentdata?Pleasetick top 5?
a) VehicleType
b) Towing andArticulation
C) VehicleManoeuvre
d) VehicleMovementCompassPoint
e) VehicleLocation
f) JunctionLocation at Impact
g) Skidding/Overturning
Hit ObjectIn Carriageway
VehicleLeaving Carriageway
Hit Object Off Carriageway
I stpoint of Impact
Driver BreathTest
Driver Hit and Run
Damage
VehiclePrefiyJSuffixLetter
15 On the abovelist, which onesdo you think are necessaryto be madeas prepared
queriesto display accidentdata?Pleasealso tick top 5.
a) by VehicleType
b) by Towing andArticulation
168
C) by VehicleManoeuvre
d) by VehicleMovementCompassPoint
e) by VehicleLocation
f) by JunctionLocation at Impact
g) by Skidding/Ovcrturning
by Hit Object In Carriageway
by Vehicle Leaving Carriageway
by Hit Object Off Carriageway
by I Stpoint of Impact
by Driver BreathTest
by Driver Hit andRun
by Damage
by VehiclePrefiyJSuffixLetter
16 Whenyou find all the datayou need,which of the following tools arenecessaryor
useful. Pleasetick all you want.
a) display thesedataby using numericaltable

b) display thesedataby using different kinds of charts,suchas
line chart,bar chart.
C) Extract thesedataout of the system,suchasprint, II
saveas anotherformat (text, graphical)
d) What else
17 If you have already used some

similar system, could you point out some
limitations of that system or the features that
you wish to improved?
18 Any other commentson this work?
169
Appendix E
User Interface Evaluation Form

User InterfaceandNavigation:
I. Wasthe presentationof contentclear?
(no) 12345 (yes)
2. Wasthe user interfaceattractive and appealing?
(Bad) 12345 (Good)
3. Is the sequenceof the screensclear?
(Confiising) 12345 (very clear)
4. Is readingcharacteron the screeneasy?
(hard) 12345 (easy)
5. Wasuser orientation ("Where am IT') within theApplet easyand intuitive?
(Bad) 12345 (Good)
6. Wasnavigation of theApplet ("Can I

go whereI want to go?") easyand intuitive?
(Bad) 12345 (Good)
7. Wasthe layout of theApplet appropriate?
(Bad) 12345 (Good)
8. How well did the use of color, graphics, images and fonts enhance the user's
experienceof the site?
(Bad) 12345 (Good)
9. Canuser control paceand sequenceeasily?
(Difficult) 12345 (Easy)
10. Is systemspeedfast enough?
170
(Too slow) 12345 (fast enough)

Terminologyand SystemInformation
1. Is use of terms throughout system consistent?
(Inconsistent) 12345 (consistent)
2. Is position of messageson screen consistent?
3. Is remembering names and use of commands difficult?
(Difficult) 12345 (Easy)
4. Is the terminology always related to the task?
(never) 12345 (always)
5. Does computer always informs about its progress?
(never) 12345 (always)
6. Is text clear and easy to read?
(Bad) 12345 (Good)
7. Does the highlighting simplify the task?
(not at all) 12345 (very much)
8. Is the organization of information clear?
(confusing) 12345 (very clear)

9. Under nonnal conditions,doesthe programrun consistently?
Overall reactionto the software
How do you think the software?
(terrible) 12345 (wonderful)
171
(frustrating) 12345 (satisfying)
(inadequate power) 12345 (adequate power)
List the most negative aspect(s):
I.
2.
3.
List the most positive aspect(s):
I.
2.
3.
User Interface Evaluation Data
Q1 Was the presentationof

content clear? No. 2345 Yes
13 11 14
Q2 Was the user interface

attractive and appealing? Bad 12345 Good
7 12 10
Q3 Is the sequence of the

screen clear? Confusing 12345 Very Clear
7 14 8
172
Q4 Is reading character on
the screen easy? Hard 12345 Easy
36 14 6
Q5 Was user orientation

within the Applet easy and
intuitive? Bad 12345 Good
37 10 9
Q6 Was navigation of the

Applet easy and intuitive? Bad 12345 Good
135 17 4
Q7 Was the layout of the

Applet appropriate? Bad 12345 Good
1 19 9
Q8 How well did the use of Bad 12345 Good

color, graphics, images and
fonts enhance the use's
experience of the system? 47 12 6
Q9 Can user control pace

and sequence easily? Difficult 12345 Easy
13 16 9
Q10 Is system speed fast Fast

enough? Too slow 12345 enough
123 11 12
Ql 1Is use of terms Inconsistent 12345 Consistent
173
throughout system
consistent?
1 12 16
Q12 Is position of
messages on screen
consistent? Inconsistent 12345 Consistent
11 11 16
Q13 Is rememberingnames
and use of commands
difficult? Difficult 12345 Easy
23 13 11
Q14 Is the terminology

always related to the task? Never 12345 Always
4 17 8
Q 15 Does computer always

informs about its progress? Never 12345 Always
35 12 36
Q16 Is text clear and easy

to read? Bad 12345 Good
54 13 7
Q17 Does the highlighting

simplify the task? Not at all 12345 very much
123 17 6
Q18 Is the organizationof

information clear? Confusing 12345 Very Clear
23 17 7
Q19 Under normal

conditions, does the
program run consistently? Inconsistent 12345 Consistent
174
1 14 12
Q20 How do you think the

software? terrible 12345 wonderful
6 19 4
frustrating 12345 satisfying
114 19 4
inadequate adequate
power 12345 power
186 10 4
175
7lie Impact of Localized RoadAccident Information on Road Safety Awareness
Appendix F
An example of calculating sign test
RI
Subjects Ranking (no acc) Ranking (acc) Direction of difference
sl 73+
S2 770 (omitted)
S3 43+
S4 73+
S5 76+
S6 73+
S7 74+
S8 76+
S9 73+
slo 73+
Lessfrequentsymbolis ", the numberof '-'s is 0.

The critical value for 5 per cent is I (N=9). Therefore,
sign test suggestthe
dependentvariable (accident information) has had an effect on behaviour
(route planning).
176
References:
Abdel-Aty, M. A., Radwan, A. E., 2000. Modeling traffic accident

occurrence and involvement. Accid. Anal. Prev. 32,633-642.
2 Aberg L., Larsen L., Glad A., Beilinsson, L. - Observed Vehicle Speed
and Drivers' Perceived Speed of Others, Applied Psychology: An
International Review, 46(3), p. 287-302,1997
3 Ajzen, 1. (1985). From intentions to actions: A theory of planned

behavior. In I Kuhl & J. Beckmann (Eds.), Action control: From
cognition to behavior (pp. 11-39). Berlin: Springer-Verlag
4 Alcohol and pedestrians, London: Department for Local Government,

Transport and the Regions
5 Andrew P, Jones; Stig H. Jorgensen;The use of multilevel models for

the prediction of road accident outcomes. Accident Analysis
and Prevention 35 (2003) 59-69.
6 Andrews, K. H. and D.B. Kandel 1979, "Attitude and behaviour: a

specification of the contingent consistency hypothesis". Amer. Sociol.
Rev., 44
7 Ajzen, I. And M. Fishbein 1980, Understanding

attitudes and
predicting social behaviour. Prentice Hall, Englewood Cliffs, New
Jersey
8 Ben Shneiderman, The Eyes Have It: A Task by Data Type

Taxonomy for Information Visualisation, IEEE Computer
Society Washington,DC, USA, 1996
9 B. Fildes, G Rumbold & A. Leening, Speed Behaviour and Drivers'

Attitudes to Speeding, Monash University Accident ResearchCentre
-
Report #16 - 1991
10 Bier, E.A., Stone, M. C., Baudel, T., Buxton, W., and Fishkin, K., A
Taxonomy of See-Through Tools, Proceedings of CHI'94, ACM
Conferenceon Human Factors in Computing Systems,Boston, 1994.
I Booch, Grady. Object-oriented analysis and design : with

applications, 2nd ed., Redwood City, Calif ; Wokingham
Benjamin/Cummings Pub. Co., c1994
177
12 Box, George E. P. Statistics for experimenters : an introduction to

design, data analysis, and model building, New York ; Chichester
Wiley, 1978
13 Bradley E. Heinrichs, Boyd D. Allin, James J. Bowler and Gunter P.

Siegmund, Vehicle speed affects both pre-skid braking kinematics
and average tire/roadway friction, Accident Analysis & Prevention,
Volume36, Issue 5, September2004, Pages 829-840
14 Card, S. K., Pirolli, P., and Mackinlay,J.D., The Cost-of-knowledge

Characteristic Function: Display Evaluation for Direct-Walk
Information Visualizations.Proceedings of CHF94,ACM Conference
on Human Factors in Computing Systems, Boston,1994
15 Chambers, J.M., Cleveland, W. S., Kleiner, B., and Tukey, P. A.,

Graphical Methods for Data Analysis, Pacific Grove, CA: Wadsworth
and Brooks/Cole, 1983
16 Charles Goldenbeld, Peter Bernadus M. Levelt and Jelle Heidstra,

Psychological perspectives on changing driver attitude and
behaviour, Recherche Transports Sjcuritj, Volume 67, April-June
- -
2000, Pages 65-81
17 Christie, N. (1990) The

effectiveness of the 1988 police national
motorway safety campaign. In Proceedings of Road Safety and
Trafic Environment in Europe, M, Linkoping, Sweden
18 Cleveland, W.S., The Elements of Graphing Data, Summit, NJ:

Hobart Press,1994
19 Cleveland, W.S., and McGill, R., Dynamic Graphics for Statistics.

Pacific Grove, CA: Wadsworth and Brooks, 1988
20 Corbett Claire, The effects of speed cameras: how drivers respond,

London : Department of the Environment, Transport and the Regions,
Road Safety Division, c 1999
21 Current road and vehicle safety research commissioned by the

Department of Transport, London: Dept. of the Environment,
Transport and the Regions, 1997
22 Dangerous driving and the law, London : Department of the

Environment,Transportandthe Regions,2002
178
23 David D. Clarke, Pat Ward, Craig Bartle and Wendy Truman, An

In-depth Study of Work-related Road Traffic Accidents, School of
Psychology, University of Nottingham, Road Safety Research Report
No. 58, Departmentfor Transport: London, August 2005.
24 Dominique Lord and Bhagwant N. Persaud, Estimating the safety

performance of urban road transportation networks, Accident
Analysis & Prevention,Volume 36, Issue 4, July 2004, Pages609-620
25 Edward R. Tufte, 1983, The visual display of quantitative

information Cheshire, Conn. : Graphics Press,c1983
26 Fazio, R.H. and M. P. Zanna 1981, "Direct experience and

attitude-behaviour consistency". In: Berkowitz, L. ),
(ed. Advances in
experimentalsocial psychology, 14. Academic Press, New York/London
27 Finn Jorgensen and Tore Wentzel-Larsen, Optimal use of warning

signs in traffic, Accident Analysis & Prevention, Volume31, Issue 6,
November 1999, Pages 729-738
28 Freni-Titulauer, L. W. J., and Low, W. C., Comparisons of some

graphical methods for exploratory multivariate data analysis, The
American Statistician, 38,184-188,1984
29 Geoffrey Grime, Handbook of road safety research, London

Butterworths, 6987.
30 Gossain, Sanjiv. Object modeling and design strategies : tips and

techniques. Cambridge : Cambridge University Press ; New York
SIGS Books, 1998
31 Graham, Tony, Road safety education in the Scottish curriculum,

Edinburgh: Stationery Office, 2000
32 Harold BrodskyA. Shalom Hakkert, Risk of a road accident in rainy

weather, Accident Analysis & Prevention, Volume 20, Issue 3, June
1988, Pages 161-176
33 Hauer E. Bias-by-selection:overestimation of the effectivenessof

safety countemreasures caused by the process of selection for
treatment. Accid. Anal. andPrev.12(2):113-117;1980
34 HauerE. On exposureand accident rate. Traffic Engng and Control

36 (3):134-138;1995
179
Tle Impact of Localized Road Accident Information on Road Safety Awareness
35 Hazardous road locations: Identification and countermeasures,

1976, Organisationfor Economic Cooperation and Development,Paris
36 Hendon, K. P., and Meyer, T., 3D Widgets for Exploratory Science

Visualization, Proceedings of UIST'94, ACM Symposium on User
Interface Software and Technology, 1994
37 Hendrickx, L. and Vlek, C. (1991) Effects of risk infotmation on

speed choice in blind curves. In BehaviouraZ Researchin Road Safety.
TRRL, Crowthome, England
38 Hobbs, C.A. 'The effectiveness of seat belts in reducing injuries to

car occupants'. Transport and road research laboratory report, LR
811
39 HVE (HumanVehicleEnvironment)hqp://www.edccoM.com/r)roducts/hve.
html
40 Ian Sommerville, Software Engineering, Fifth Edition, Addison

Wesley, 1995
41 Improving road safety by attitude modification, Road Transport

Research,OECD, 1994,Paris
42 Jacobson, I., Christerson, M., Jonsson, P., and Overgaard, G,

Object-Oriented Software Engineering. Workingham, England:
Addison-Wesley, 1992
43 Jarrett, D.F. Statistical analysis of road accident frequencies. In Proc.

Symposiumon Safety in the Road Environment, LNEC, Lisbon; 1994
44 Jens Rasmussen,Keith Duncan, and JacquesLeplat, New technology

and human error, Chichester: Wiley, 0986
45 John GU. Adams, Risk and Freedom, University College London,

1985
46 Joonhwan Lee, Jodi Forlizzi, and Scott E. Hudson, Studying the

Effectiveness of MOVE: A Contextually Optimized In-Vehicle
Navigation System, Human-Computer Interaction Institutet and
School of Design, Camegie Mellon University, CHI 2005, Portland,
Oregon, USA, 2005
47 Joshua,S. C.; Garber,N. J. Estimating truck accident rate and

involvementsusing linear and Poissonregressionmodels.Transport.
PlanningTechnol.15:41-58; 1990.
180
48 Jovanis, P. P.; Chang, H. L. Modelling the relationship of accidents

to miles travelled. Transport. Res. Record 1068: 42-51; 1986.
49 Keskinen, E, M. Hatakka, A. Katila and S. Laapotti 1992, "Was the

renewal of the driver training successful? The final report of the
follow-up group". Psychological Reports, 94. University of Turku
50 Key publications, Statisticsfor wales,

htti)://www. wales.gov.uk/keypubstatisticsforwales/content/publication/tr
ansport/2003/raw2OO2/raw2OO2-chI 3/raw2OO2-chl 3.htm
51 Kulmala, R., 1995. Safety at rural 3- and 4-arm junctions.

Development and application of accident prediction models. VTT
Publications 233
52 Lawrence K. Saul; SamT. Roweis; An Introduction to Locally Linear

Embedding, http://www. cs.toronto.edu/-roweis/Ile/Papers/Ileintro.pdf
53 Linda Mountain, Bachir Fawaz and David Jarrett. Accident Prediction

Models for Roads with Minor Junctions. Accid. Anal and Prev., Vol
28, No. 6, pp. 695-707,1996
54 L. John Horwood and David M. Fergusson,Drink driving and traffic

accidents in young people, Accident Analysis & Prevention, Volume3Z
Issue 6, November2000, Pages 805-814
55 MacLean A., Young R., Bellotti V, Moran T. Questions, Options, and

Criteria: Elements of Design SpaceAnalysis, Human-Computer
Interaction, Vol.6(3&4), 1991
56 Maher M. J. The evaluation

of road safety schemes: problems with
statistical appraisal. In Proc. County Surveyors' Society/University
Transport StudiesGroup Joint Seminar The Evaluation Road
on of
Safety Schemes.London; 1991
57 MarkusR. Schinidt,Questionnaireon Risk Perceptionregarding

Environment,Agriculture and Health, Instituteof Risk Research,
Universityof Vienna,Austria
58 MatthewG Karlaftis;IoannisGolias;Effects of road geometryand

traffic volumeson rural roadway accidentrates.Accident
AnalysisandPrevention34 (2002)357-365.
59 Mattisse http://www.help2travel.
co.uk/
181
Ile Impact of Localized RoadAccident Information on Road Safety Awareness
60 McCullagh, P.;Nelder, I A. Generalized Linear Models. London:

Chapmanand Hall; 1983.
61 Miaou, S.-P., Hu, P.S., Wright, T., Rathi, A. K., Davis, S.C., 1992.
Relationship between truck accidents and highway geometric
design: a poisson regression approach. Transport ResearchRecord
1376
62 Mike Schneider,SaraKiesler, Calling While Driving: Effects of

Providing Remote Traffic Context, Human-Computer Interaction
Institute, CarnegieMellon University, CHI 2005, Portland, Oregon,
USA, 2005
63 Milton, J., Mannering, 1998. The relationship among highway

geometrics, traffic-related elements and motor-vehicle accident
frequencies. Transportation25,395-413.
64 MunemasaShimamura,Minoru Yamazaki and Goro Fujita, Method to

evaluate the effect of safety belt use by rear seat passengerson the
injury severity of front seat occupants, Accident Analysis &
Prevention,Volume 37, Issue 1, January2005, Pages5-17
65 National Statistics httv://www. statistics.jzov.uk/
66 NIST/SEMATECH e-Handbook of Statistical Methods,

http://www. itl. nist.izov/div898/handbook/
67 OECD (Organisation for Economic Co-operation and Development),

Effectiveness of road safety education programmes, Paris
Organisationfor Economic Co-operation and Development, 1986
68 OECD (1994) Improving road safety by attitude

modification. Paris,
France
69 PAI Ulleberg and Torbjom Rundmo, Personality,

attitudes and risk
perception as predictors of risky driving behaviour among young
drivers, SafetyScience,Volume41, Issue 5, June 2003, Pages 427-443
70 Parker,Dianne,Influencing driver attitudes and behaviour, London:

DETR,2001
71 Plaisant, C., Carr, D.A., and Shneiderman,B., Image Browsers:

Taxonomyand DesignGuidelines,IEEE Software,1995
182
72 Research on road safety, Road Research Laboratory, London

H.M. S.O., 1963.
73 Research on road traffic, Road Research Laboratory, London

H.M. S.O., 1965.
74 Road Accidents Scotland 2003, A Scottish Executive National

Statisticspublication, 2004
75 Roadway Safety Guide, Roadway Safety Foundation, Chapter 2,

html
htt-p://www. roadwaysafety-orR/toc.
76 Robert Spence,Information Visualization, Addison-Wesley,2001
77 Ross, L. (1977) The intuitive psychologist and his shortcomings:

Distortions in the attribution process. In L. Berkowitz (ed.),
Advances in experimental social psychology (Volume 10, pp. 173-240),
Orlando, FL: Academic Press
78 Ross, R. Entity Modeling: Techniques and Application. Boston, MA:

DatabaseResearchGroup, 1987
79 Rothengatter J.A. Psychological behaviour, In

- aspects of road user
Frese M. (ed.): Applied Psychology: An international Review, Special
IssueTmffic Psychology, 221-314,1997
-46, p
80 Sabey, BE and Staughton, GC. Interacting roles of road

environment, vehicle and road user in accidents. Fifth International
Conference of the International Association for
Accident and Traffic
Medicine, and the Third International Conference
on Drug Abuse of the
International Council on Alcohol
and Addiction, London, 1975
81 Schlegel, R-P and D. Ditecco 1982, "Attitudinal

structures and the
attitude-behaviour relation". In: Zanna, M. P., E.T. Higgins and CY
Herman (eds.), Consistency in social behaviour. Erlbaum, Hillsdale,
New Jersey
82 Schuurman, Nadine. GIS: a short introduction, Malden, MA.

Blackwell Pub., 2004
83 Slovic, Paul, 1938-,The perception of risk, London; Sterling, VA:

Earthscan Publications,2000
84 Sneath,P.H., andSOKAL, R. R. Numerical Taxonomy,SanFrancisco:

W. H. Freeman,1973
183
85 Stephanie Blows, Rebecca Q. Ivers, Jennie Connor, Shanthi

Ameratunga and Robyn Norton, Car insurance and the risk of car
crash injury, Accident Analysis & Prevention, Volume 35, Issue 6,
November 2003, Pages987-990
86 Stuart K. Card, Jock D. Mackinlay, Ben Shneiderman,Readings In

Information Visualization - Using Vision To Think, San Franciso,
Morgan Kaufmann Publishers,Inc., 1999
87 System Three. Social ResearchUnit, The effectiveness of leaflets in

road safety, Edinburgh: Scottish Office Central Research,1997
88 The AA, Latest Traffic News httr)://www. theaa.com/travelwatcb/travel news.isp
89 Think Website, www. thinkroadsafely. gov. uk
90 Tillman, W. A. and Hobbs, C. E. (1949) The accident-prone

automotive driver. American Journal of Psychiatry
91 Tomorrow's roads: safer for everyone. The Government'sroad safety

strategy and casualty reduction targets for 2010, Department of the
Environment, Transport and Regions. London: DETR, 2000
92 Torbj om Rundmo, Hilde Iversen, Risk

perception and driving
behaviour among adolescents in two Norwegian
counties before and
after a traffic safety campaign, Safety Science, Volume 42, Issue 1,
January 2004, Pages 1-21
93 Tou, F., Williams, M. D., Fikes, R., A. Henderson,

J., and Malone, T.W.,
RABBIT: An Intelligent Database Assistant. Proceedings
of
AAAI'82,1982
94 Traffic safety, a national problem, a symposium

sponsored by the
National Academy of Engineering; held in Washington, April 28,1966.
Saugatuck : Eno Foundation for Highway Traffic Control, 1967.
p118-pl46
95 TxDOT Exprcssway,road conditionshtt-P:

//www.dot.statc.tx.us/hcr/main.htm
96 UK DataArchives. httv: data-archive.

ac.uk/
97 What Are Focus Groups? Section on Survey ResearchMethods,

American StatisticalAssociation,1429 Duke Street,Alexandria, VA
22314-3415USA, 1997
184
98 Wilde GJS. Risk homeostasis theory and traffic accidents:

propositions, deductions and discussion of dissension in recent
reactions. Ergonomics 1988;
31:441-68
99 William CT.Jacoby, Statistical graphics for visualizing multivariate

data, Sagepublications, London, 1998
100 William S. Cleveland, Visualizing Data, Hobart Press, Summit, New

Jersey,1993
101 W.M. Hirst; L. J. Mountain; M. J. Maher; Sources of error in road

safety scheme evaluation: a method to deal with outdated accident
prediction models. Accident Analysis and Prevention 2003
102 World's flrst road death, RoadPeace,The United Kingdom's national

charity for road crash victims.
103 Yunan Zheng, Chris W Johnson, A Preliminary Evaluation of the

Impact of Local Accident Information on the Public Perception of
Road Safety, Department of Computing Science, University of
Glasgow, RESS(Reliability Engineering & System Safety) special issue
on infrastructure reliability, Accepted and to appear.
iss

2007 Zheng PHD

Uploaded by

Copyright:

Available Formats

2007 Zheng PHD

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2007 Zheng PHD

Uploaded by

Copyright:

Available Formats

The Impact of Localized Road Accident Information

on Road Safety Awareness

A Thesis submitted for the degree of Doctor of Philosophy

The University of Glasgow

The World Health Organization (WHO) estimatethat road traffic accidentsrepresentthe

I would like to thank both my first and secondsupervisorsChris W Johnsonand

1.1 An Introduction to Road Traffic Accident 1

Chapter 2 Identified Problems and Research methodology 12

2.1 Human Error and Road accidents 12

Chapter 3 Evaluation of Existing Road Safety Websites 29

3.1 Existing road accident related information systems 29

4.1 Data Sources for Road Accident Information System 53

Chapter 5 Theories for visualizing multivariate data 65

5.1 Statistical graphics for visualizing multivariate data 66

Chapter 6 System Design and Implementation 80

6.1 Interface Design 81

Chapter 7 System Evaluation 98

Chapter 8 Route Planning Experiment ill

8.1 Motivation and Goals ill

Route Planning Experiment with University Staff and Students 129

9.1 Experiment Design 130

Chapter 10 Conclusions and Future Work 141

10.1 Conclusions 141

Appendix A: A brief Review of road safety research 151

Table 1.1 Accidental deaths: by cause 1971-1998

Figure 1.1 Percentage contributions to road accidents

Figure 1.2 Snapshot of ReCap system

Table 2.1 The major error-shaping factors at each level of performance

Figure 2.1 Dynamics of generic error-modelling system (GEMS)

Figure 2.2 The theory of Planned behaviour

Figure 2.3 Different kinds of road accident information

Figure 2.4: UK Department of Transport Road Safety Site

Table 2.2: Scottish Executive Accident Frequencies for Urban Junctions

Figure 3.1 Screenshotsof HVE

Figure 3.2 Screen shot of TxDOT

Figure 3.4 Screen shot of the AA

Figure 3.5 An example result of KeyAccident system

Figure 3.6 Snapshots of AIS

Table 3.1 Statistical analysis of the evaluation questions

Figure 3.16 Please rate how controllable this risk is.

Figure4.1 Format for Accident Records in UK Road Traffic Accident Data

Figure 4.2 Sample Accident Information from the UK Data Archive

Figure 4.3 Waterfall Model

Figure 4.4 Some results of experts background

Figure 5.1 Multi-code plotting symbols

Figure 5.2 Examples of Histogram Plots

Figure 5.3 Example of Star Plots

Figure 5.4 Example of Scatterplot Matrix for Acedient data

Figure 5.5 Example of conditioning plot for accident information

Figure 5.6 Pre-computer creation of a visualization tool

Figure 5.7 The creation and use of a computer-based visualization tool

Table 6.1 Relationships between task categories, requirements and techniques

Figure 6.1 An example result of KeyAccident system

Figure 6.2: Overview of the first version Map-Based Interface

Figure 6.3: Reviewing Details for Several Selected Incidents

Figure 6.4 Second version of the User Interfaces

Figure 6.5 Filtering control dialog

Figure 6.6 Option panel for more filtering functions

Figure 6.7 Details for selected accident