17911. cl. il.’ 1.“ a
1 31.1: it‘lrvév . ‘01}. \.

 

 

 

\L .Il.lfhl.l.u?.

. ‘I.§ 1.94.. ‘. Ir. .

. _..v...w...2.u "v.95?" prul... . . Al. 9. :1. . .

 

mcmam ST

ll [lllllllll "‘ 5'"

 

lllllllllllllllllllllll

1293 00792 4003

l'.

 

 

 

r \
E LIIMRY
Hichigan State
University
\fi 1'

 

This is to certify that the

dissertation entitled

"Development of a Highway Safety Improvement
Program for the Rural Environs of Pakistan"

presented by

Zubair Ahmad

has been accepted towards fulfillment
of the requirements for

Doctor of Philosophy degree in Civil Engineering

 

 

/ , " _ / //
L446 /: m c; , / Lac/5;. 1

M ajor professor]

 

Date NoveQer 4, 1992

MSU is an Affirmunw Action /Equal Opportunity Institution . 0-12771

 

_4(-_,_‘_,A . -. _ 4

PLACE IN RETURN BOX to tomove this checkout from your record.
TO AVOID FINES return on or betote date due.

DATE DUE DATE DUE DATE DUE

 

 

‘ Ct

' ‘Q ‘
I'm, . f,

 

 

 

 

 

 

 

 

 

 

 

 

 

L____.# l _______

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

MSU Is An Affirmative ActionlEqual Opponunity Institution
ammonia-9.1

//

 

DEVELOPMENT OF A HIGHWAY SAFETY IMPROVENIENT
PROGRAM FOR THE RURAL ENVIRONS OF PAKISTAN

BY

Zubair Ahmad

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Civil and Environmental Engineering

1992

 

BY
Zubair'Ahmad

A Highway Safety Improvement Program (HSIP) was
suggested for alleviating rural trunkline accident problems in
Pakistan. The HSIP is a contemporary term for a sequential
plan of implementing highway related safety improvements.
However, the principal restraining factor anticipated in the
transfer of HSIP technology'was the absence of an.adequate and
accessible accident data base in the country. This research
was, therefore, conducted to develop accident prediction
models using various highway and traffic hazards as the
surrogate measures of safety.

The independent variables developed for this study
represented the hazards in terms of inadequate access control,
deficient pavement and shoulder width, deficient pavement
markings, guardrail deficiencies, potential intersection
conflict points, low pavement serviceability and roadside
obstructions. The ambient hazardousness was quantified using
three types of procedure: use of design standard deficiencies
as a measure of hazard; use of erratic maneuvers and traffic
conflicts as a measure of hazard; and use of an expert team
for subjective rating of hazardousness. Consequently, three
types of data sets were generated: measurements; counts; and

ratings.

A three-year period (January 1988 to December 1990)
accident data were retrieved from police records to be used as
the dependent variable in the study. The experimental site was
comprised of 86 kilometers of rural two-lane, two-way and
four-lane divided sections of the National Highway (N-S) in
the District Rawalpindi, Pakistan.

Multivariate linear regression analyses were performed to
investigate the statistical significance of the hypothesized
relationship between the hazards and accidents. The analyses
indicate existence of a statistical relationship between the
hazards and.the accidents, and.show'that inadequate control of
access and operational friction are significantly correlated
with accidents. These findings are substantiated by the
results of previous studies made in the United States and
other countries.

The results of this research provide a means to implement
a HSIP in Pakistan even when archival accident records may not
be available. The research findings also expose vital issues
for the planners and policy makers that would arise from the
incorporation of preventative safety measures in future
highway transportation facilities. The most significant impact
of implementing these measures would be on the land-use

pattern of the country.

JACIUWONVLIHNGEmflETFTS

I wish to express my earnest feelings of gratitude and
indebtedness for Dr. William C. Taylor, my academic advisor
and the chairman of the doctoral guidance committee, for his
continued guidance in the processing and completion of this
research. I firmly believe that his encouragement and trust in
my capabilities, from the very beginning of my graduate
studies, led to this accomplishment.

I am extremely grateful to Dr. Richard w. Lyles, Dr.
Thomas L. Maleck and Dr. Shlomo Levental for being members of
my committee. Their comments were always a source of
inspiration and. guidance, and. certainly' resulted in. the
improvement of this work.

I would like to express my special appreciation to the
Department of Civil and Environmental Engineering, Michigan
State University; the Chairperson, Dr. William.E. Saul; and my
advisor for the financial support during my Ph.D. program.

I am also thankful to Dr. David Horner and Elda Keaton,
Office for International Studies and Scholars, Michigan State
University for approving a fellowship award.

I would like to thank Gene V. George, Chief, Office of
Engineering, and. David Esch, Chief, Human IResource
Development, of the United States Agency for International
Development, Islamabad, Pakistan. Their help has definitely
resulted in the timely completion of my dissertation.

iv

My thanks are due to Mr. M. Sadiq Swati, Senior Chief,
National Transport Research Centre, Islamabad, for providing
me the opportunity to work at the Centre, and for extending
his auspices in data collection.

I am particularly thankful to my employer, Government of
the Punjab, and specially the former Secretary, Communication
and Works Department, Mr. Manzoor Ahmad, for administrative
actions facilitating completion of my dissertation.

I express my gratitude to my father, Khalid Ahmad, and my
mother, Shaukat Jehan, for their affection and love which
determined my moral directions. Their prayers were the source
of my growth and elevations, and today I owe my every
achievement to them.

My heartfelt thanks are due to my wife, Qaisra, for
deferring her own professional activities during the days of
my graduate school to replenish my negligence from home and
children. With deep feelings of appreciation and love, this
dissertation is dedicated to her.

I would like to appreciate my children Amna, Isma, and
Rehan for demonstrating their excellent adaptability in
switching cultures, peer' and schools during' my' periodic
sojourns between the United States and Pakistan while
completing this research. Their style undoubtedly furnished
carefree time for concentration on my work.

Finally, I submit myself before my God, The Almighty
Allah, with most humble feelings of thanks for bestowing His

Benevolence and enabling me to complete this task.

TABLE OF CONTENTS

page
LIST OF TABLES x
LISTOF FIGURES xii
LISTOFABBREVIATIONS xiv

 

1.1 THE THRESHOLD ............. ........ .................
1.2 HIGHWAY SAFETY’ADHINISTRATION IN’PAKISTAN ..........
1.3 MOTIVATION .........................................
1.4 OBJECTIVE ..........................................
1.5 RESEARCHLBPPROACH ..................................
1.6 ORDEROFPRESENTATIOH.............................. 10

ommhpp

CHAPTER 2
Woso...coo-0000000....000.000.000.000... 11

2.1 HISTORICALPBRBPBCTMooooooooooooooooooooooooooooo 11

2.2 THE CONCEPT OF LIABILITY-ASSIGNMENT IN ACCIDENT
CAUSATION J......................................... 17
2.2.1 Liability Studies in Developing Countries .... 19
2.2.2 The Theology of Single-Liability-Assignment .. 20
2.2.3 The Interactive Role of Accident Causative

Factors ...................................... 22

2.3 “ECONCBPTOFBXPOBURB0.0.0....OOOOOOOOOOOOOOOOOOO 29
2.3.11nducedExposure 0.00.00.00.00...OOOOOOOOOOOO. 31

2.4 THE CONCEPT OF RISK-HOMEOSTASIS AND
RISK-COMPENSATION .................................. 31
2.4.1 The Role of Highway Engineering in
Risk Compensation ............................ 33

vi

2.5 HUMANENGINEERINGAPPROACH......................... 34
2.5.1 Functionalization of Human Engineering
Approach in the Developing Countries
Through HSIP ................................. 35
2.6 THE HIGHWAY SAFETY IMPROVEMENT PROGRAM (HSIP) . . . . . . 37
2.6.1 Framework and Overview of HSIP ............... 38
2.6.2 Implementation of HSIP ....................... 41
SURROGATEMEASURESOPHIGHNAYSAPETY............... 41
2.7.1 Application of Surrogate Measures in
Present Research ............................. 44

2.8 USE OF TRAFFIC CHARACTERISTICS AND HIGHWAY DESIGN
FEATURES FOR ACCIDENT PREDICTION MODELING . . . . . . . . . . 44
2.8.1 Accident Prediction Modeling in

Developing Countries ......................... 51
2.8.2 The Indexation Approach for the
Identification of Hazardous Locations ........ 53

2.9 RELEVANCE OP LITERATURE REVIEW TO CURRENT RESEARCH 58
2.9.1 Transferability of the HSIP Technology ....... 60
2.9.2 Technical Gap in Transferability ............. 62
2.9.3 Suggested Method to Bridge the

Transferability Gap .......................... 63

2.10 FORMULATIONOF HYPOTHESIS .......................... 64

 
 

CA 0.0.0.000... ‘6

3.1 “BMIABLBBOFINTBRBBTsooooooooooooooooooooooooo 66

302 mnnslanornnnxma0000.00.00.000000000000000 71

3.2.1 The Ambient Highway Hazards .................. 72
3.2.2 Development of Measures of Hazard (MOH)....... 73
3.2.3 MOH Specifications and On-Ground Application.. 74
3.2.4 The Applied Procedures for Data Analysis ..... 81

30 3 m Rasmcn DATA . 0 0 . 0 . . . 0 0 0 0 0 . 0 . . . 0 0 0 0 . O 0 . 0 0 0 0 . 0 . . 83
3 0 3 0 1 The Hazard Data 0 0 0 . 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 . . 0 0 0 0 0 0 0 . 0 84
3 0 3 0 2 The ACCident Data . . . . 0 . 0 . . . . O . . . . . 0 . . . 0 . . 0 0 . 0 85
3 0 ‘ Bnnlmn SITE 0 . . . . 0 0 0 0 . . . . . 0 O O . 0 0 0 0 0 0 0 0 0 0 0 0 0 . 0 0 86

3.5 BXBCWIONOPTHBBXPBRIW00000000000000.0..0.0000 89

vii

CHAPTER4
W.00.00....0......00000000.00.000.0000000000 91

4. 1 THE ANALYTICAL APPROACH AND PRESENTATION . . . . . . . . . . . 91

4 .2 DATA ANALYSIS FOR TWO-LANE, TWO-WAY SECTIONS . . . . . . . 99
4.2.1 Descriptive Statistics 99

4 2 2 FrequencyDistribution 101

4 2 3 SimpleCorrelations 107

4 2 4 Eigenvalues and Condition Indices . . . . . . . . . . . . 109

4.2.5 Variable Tolerance & Variance Inflation Factor 111

4 2 6

4 2 7

4 2 8

Regression Analysis and the Crucial Variables 112
Accident Freq. vs. Rate as Dependent Variable 114
Final Accident Prediction Model and Test of

Hypotheses ................................... 117
4.2.8.1 Analysis of Residual & Predicted Values 118

4.3 DATA ANALYSIS FOR FOUR-LANE DIVIDED HIGHWAY SECTIONS 124
4.3.1 Descriptive Statistics 124

4 3 2 FrequencyDistribution 126

4 3 3 Simple Correlations . 132

4 3 4 Eigenvalues and Condition Indices . . . . . . . . . . . . 134

4.3.5 Variable Tolerance & Variance Inflation Factor 136

4 3 6

4 3 7

4 3 8

Regression Analysis and the Crucial Variables 137
Accident Freq. vs. Rate as Dependent Variable 139
Final Accident Prediction Model and Test of

Hypotheses ................................... 140
4.3.8.1 Analysis of Residual & Predicted Values 141

4.4 ANEVALUATIONOFTHE DATATYPE ..................... 146

4.5 HAZARDIHDEXATION.................................. 148
4.5.1 Development of HI for the Present Research . . . 149
4.5.1.1 HI for 2-Lane, 2-Way Sections ....... .. 150

4.5.1.2 HI for 4-Lane Divided Sections . . . . . . . . 153

CHAPTERS
c088 0000.....0...00000000....0000000000 15‘

5.1 PERFORMANCE EVALUATION OF RESEARCH INPUTS
ANDRESULTS........................................ 156
5.1.1 Evaluation of Independent Variables 158

5 1.2 Evaluation of Data Type 160

5.1.3 Evaluation of Predictive models 161

5 1.4 Evaluation of Hazard Index 165

5 1 5Evaluation of Accident Data 165

viii

5.2

5.3

 

ACCIDENT PREDICTION MODELS AND HSIP . . . . . . . . . . . . . . . .

HSIPANDPOLICYDECISIONMAKING ....................

RESEARCHLIMITATIONS....... ..... ...................

PROMINENTFINDINGS.................................
6.2.1The IdentifiedHazards
6.2.2 Implementation of the HSIP
6.2.3 PolicyIssues

BUGGBBTBDRBBmcn0.0.00.0...0000000000000000000000
REMAININGCOMPONENTSOFTHEHSIP ..... ..............

6.4.1 The Implementation Component
6.4.2 The Evaluation Component

W......................................

W

Financial Statement for the Removal of Blackspots.
An Example: Rawalpindi Civil Division, Punjab.
(19.80-1987).0.............0...0.000...000.000.....0

Some Photographs of the Experimental Site Showing
theseleCtedHazards 00.0.00.....0...0.......0.00.00

The Hazard Data and the Quantified Hazardousness
Usingthe DevelopedMOH

TheACCidentData .00....00.0.....000000000000000000

The SPSS Program for DataAnalyses

ix

166

167

169
169
170
171
172
172
173
174

174
175

177

187

188

194

212

225

LIST OF TABLES

2.1 Summary of Socio-Economic Macro Models of
nghwaysafety0.00....OI0......OOOOOOOOOOOOOOOOOOO 13

2 . 2 Standards for Highway Safety Programs . . . . . . . . . . . . . 16

2.3 Causes of Road Accidents in 5 Developing
countrieS[38] ..000000.00.0.00000..0...0..00000.0. 19

2 . 4 Traffic Accident Causes in Turkey [39] . . . . . . . . . . . . 20
2 . 5 Traffic Accident Causes in Pakistan [40] . . . . . . . . . . 20
2 . 6 Traffic Accident Causes in Jordan [47] . . . . . . . . . . . . 25
2 . 7 Computation of Hazardousness Index [101] . . . . . . . . . . 55

2.8 Regression Models of Traffic Safety in
saudiuabia [103] ..00...........0...0..00...0.0.. 56

2 . 9 Procedures of Planning Component [71] . . . . . . . . . . . . . 61
2 . 10 Procedures of Implementation Component [71] . . . . . . . 62
2 . 11 Procedures of Evaluation Component [71] . . . . . . . . . . . 62

3.1 Second Order Elimination - Procedures for
Identifying Hazardous Locations . . . . . . . . . . . . . . . . . . . 68

3.2 Third Order Elimination - Inventory of Hazardous
HighwaYFeatures0.00.0...0000............0.00..... 70

3.3. The Developed MOH and their Use Characteristics . . . 74
3.4 The Hazard Measurement System 81

(2-Lano, 2-Way Sections)
4 . 1 Descriptive Statistics of Variables . . . . . . . . . . . . . . . 100

4.2 TheCorrelationMatrix............................ 108

4.3 Collinearity Diagnostics: Eigenvalues and
ConditionIndices........ ........ .............. 110

4 . 4 Collinearity Diagnostics: TOL and VIP . . . . . . . . . . . . . 111
4.5 The Variable Removal Order by Backward Elimination 112
4.6 The Statistical Results of Regression Analysis . . . . 113
4.7 Indicated Appropriateness of Independent Variables 113

X

4.10
4.11
4.12

4.13
4.14
4.15

4.16
4.17
4.18
4.19
4.20

4.21
4.22
4.23

4.24
4.25
4.26

4.27
4.28

Reported Variation in Traffic Volumes . . . . . . . . . . . . .

Correlation of Indicated Variables with
AccidentFrequencyandRates

Statistics for the Residuals and Predicted Values
Outlier - Standardized Residual . . . . . . . . . . . . . . . . . . .

Spreadsheet Implementation of Accident Predictive

”Odel00000000........0..000...000..0.....0...00.00

(4-Lano Divided Highway Sections)
Descriptive Statistics of Variables . . . . . . . . . . . . . . .

TheCorrelationMatrix............................

Collinearity Diagnostics: Eigenvalues and
conditionIndices.0..........00.00........0....0.0

Collinearity Diagnostics: TOL and VIP . . . . . . . . . . . . .
The Variable Removal Order by Backward Elimination.
The Statistical Results of Regression Analysis . . . .
Indicated Appropriateness of Independent Variables.

Correlation of Indicated Variables with
Accident Frequency andRates

Statistics for the Residuals and Predicted Values. .
Outlier - Standardized Residual . . . . . . . . . . . . . . . . . . .
Spreadsheet Implementation of Accident Predictive

"Odel000000000000000.00.00.0.000.0...0.00000.0000

Regression Analyses Results from One Type of Data. .
Hazard Index: 2-Lane, 2-Way Sections . . . . . . . . . . . . . .

Correlation of Individual Variables and HI with
Acc1dents (2-Lane, 2-Way Sections) . . . . . . . . . . . . . . . .

Hazard Index: 4-Lane Divided Highway Sections . . . . .

Correlation of Individual Variables and HI with
Accidents (4-Lane Divided Highway Sections) . . . . . . .

A Comparison of the Predictive Modeling and HI
Approacn0.0.0000..00..0.00..00000000000000.0000...

Model Limits in Accident Prediction . . . . . . . . . . . . . . .

Examples of non-Accident Based MOEs . . . . . . . . . . . . . . .

xi

114

116
118
119

122

125
133

135
136
137
138
138

139
142
142

145
148
150

151
153

154

155
164
176

LIST OF FIGURES

Figure Page
2.1 Percentage Contributions to Road Accidents as

Obtained in a British and US Accident Study [46]. . . 24
2.2 Percentages of Accidents in which

Human Factors were Identified as Definite or

Probable Causal Factors [34] 24
2.3 Overview of the Highway Safety Improvement

Program(HSIP)[71].......... .......... ...... 38
2.4 Highway Safety Improvement Program at the

Process Level [71] . ....... 39
2.5 Highway Safety Improvement Program at the

Sub-Process Level [71] ................... . . . . . . . . . 40
2.6 Cumulative Adequacy Index vs. Accident Rate [97] . . 50
3.1 [PWIDTH] - A Hyperbolic Function of Pavement

Width Variation from the Standard . . . . . . . . . . . . . . . . . 76
3.2 The Geographical Location of the Experimental Site. 87
3.3 The Magnified Diagram of the Experimental Site . . . . 88

(2-Lans, Z-way Sections)
4 . 1 Frequency Distribution of Variable [RIBBON] . . . . . . . 101
4 . 2 Frequency Distribution of Variable [SPATHS] . . . . . . . 102
4 . 3 Frequency Distribution of Variable [GDRAIL] . . . . . . . 103
4 .4 Frequency Distribution of Variable [PWIDTH] . . . . . . . 103
4 . 5 Frequency Distribution of Variable [SWIDTH] . . . . . . . 104
4 . 6 Frequency Distribution of Variable [PMARKS] . . . . . . . . 104
4 . 7 Frequency Distribution of Variable [INTSEC] . . . . . . . 105
4 . 8 Frequency Distribution of Variable [ISLAND] . . . . . . . 105
4 . 9 Frequency Distribution of Variable [PVCOND] . . . . . . . 106
4 . 10 Frequency Distribution of Variable [SIDEOB] . . . . . . . 106
4 . 11 Frequency Distribution of Variable [ANACFQ] . . . . . . . 107

xii

4.12
4.13
4.14

4.15

4.16
4.17
4.18
4.19
4.20
4.21
4.22
4.23
4.24
4.25

4.26

4.27
4.28
4.29
4.30
4.31

The Compounding Effect of Traffic Volume Variation

Histogram - Standardized Residual . . . . . . . . . . . . . . . . .

Normal Probability (P-P) Plot - Predicted Values . .

Difference Between Actual and Predicted Accidents .

(4-Lane Divided Highway Sections)

Frequency Distribution of Variable [RIBBON]

Frequency Distribution of Variable

Frequency Distribution of Variable

Frequency Distribution of Variable

Frequency Distribution of Variable

Frequency Distribution of Variable

Frequency Distribution of Variable

Frequency Distribution of Variable

Frequency Distribution of Variable

Frequency Distribution of Variable

Frequency Distribution of Variable

Frequency Distribution of Variable

Histogram - Standardized Residual . . . . . . . . . . . . . . . . .

[SPATHS]
[MEDOPN]
[GDRAIL]
[PWIDTH]
[SWIDTH]
[PMARKS]
[INTSEC]
[ISLAND]
[PVCOND]
[SIDEOB]
[ANACFQ]

Normal Probability (P-P) Plot - Predicted Values . .

Difference Between Actual and Predicted Accidents .

A Graphical Representation of Hazard Index

Vs. Accidents (2-Lane, z-Way Sections) . . . . . . . . . . . .

A Graphical Representation of Hazard Index

Vs. Accidents (4-Lane Divided Sections) . . . . . . . . . . .

xiii

115
120
121
123

126
127
127
128
128
129
129
130
130
131
131
132
143
144
146

152

155

AASHO
AASHTO

EOS
FHWA
GNP
HI
HSIP
JICA

Kmph
MOE
MOH

NHSB
NHTSA
NTRC
OECD

PDO
PHD
RRR
TOPICS
TRRL
TRB
USAID
USDOT

LIST OF ABBREVIATIONS

American Association of State Highway Officials
American Association of State Highway and
Transportation Officials

Annual Daily Traffic

End of Shoulder

Federal Highway Administration (USA)

Gross National Product

Hazard Index

Highway Safety Improvement Program

Japan International Cooperation Agency
Kilometer

Kilometer per Hour

Measure of Effectiveness

Measure of Hazard

Million Vehicle Kilometers

Million Vehicle Miles

National Highway Authority (Pakistan)

National Highway Safety Board (USA)

National Highway Traffic Safety Administration (USA)
National Transport Research Centre (Pakistan)
Organization for Economic Cooperation and
Development

Property Damage Only

Punjab Highway Department (Pakistan)
Resurfacing, Restoration and Rehabilitation
Traffic Operations for Increased Capacity and safety
Transport and Road Research Laboratory (UK)
Transportation Research Board (USA)

United States Agency for International Development
United States Department of Transportation
Vehicle Miles of Travel

xiv

CXLKPTFHK l

 

1. 1 THE THRESHOLD

Human knowledge on the subject of highway traffic safety
is beyond infancy and, at present, relatively improved
explanatory axioms are employed for having a refined
comprehension about the crash mechanism. Over the past three
decades the rate of motorization and highway network expansion
has increased worldwide. This growth, particularly in the
developed countries, has acted as a strong stimulus for
attaining the present extent of behavioral and technological
research concerning highway transportation. Over a period of
time, this scholastic enterprise has induced some fundamental
conceptual shifts and contributed many innovative notions to
the transportation knowledge-base.

One such development is perceived in observing that the
word cause has largely disappeared from the technical

literature on highway safety, since the term conveyed the

2

notion of a single responsible factor in the deterministic
sense in which it was used in the physical sciences and
engineering literature [1] . In fact, the results of the state
of the art review indicates that current research no longer
supports the classical single-liability-assignment model of
accident causation based on the typical taxonomy of a
"vehicle-road-user" system. The current research rather looks
at the interactive and multiple level role of these three
basic factors in each traffic crash [2].

A crash is certainly initiated by a set of circumstances
that usually include these three factors. However, it would
seldom result from an unambiguous single cause [3]. Quite
often a single cause is associated with a crash occurrence
because accident reports may ask explicitly for one, and to
probe for the intricate and interactive reasons may not be
easy. In fact, modeling a highway traffic safety system is
very complex because of the high degree of interrelation among
the system variables. This approach requires the modeler to
predict and compare the individual and interactive
effectiveness of changes in various parameters in increasing
overall safety benefits. For these reasons of complexity, even
the highly motorized and developed societies lack a composite
safety system model and their present highway safety practices
reveal a quasi-integrated but simultaneous effort to improve
each prominent component.

One of these practices, with engineering orientation and

proven effectiveness for safety improvement, is the

3

identification and correction of hazardous highway locations.
There is increasing evidence from the developed countries, and
also with particular relevance to the Third World, that
relatively detailed spot investigations combined with low cost
remedial measures can be highly cost effective and impose a
very marked effect on road safety [4,5,38]. The practice of
identifying and correcting hazardous highway locations evolved
in the early forties as the road mileage and use of
automobiles increased dramatically in the United States.
Through persistence, the practice and the techniques have
attained a high degree of sophistication. At present in the
United States, a vast knowledge 'base exists on this
discipline, and the implementation- strategy (usually a
multiple step sequential model) is referred to as a Highway
Safety Improvement Program (HSIP).

It follows (from the details given elsewhere in this
dissertation) that implementing a HSIP would be the ultimate
desirable option fOr Pakistan for alleviating the country's
highway safety problems. However, initiation of a formal HSIP
in a developing country like Pakistan is associated with many
limitations stemming from various financial, technical and
administrative constraints. These impediments, in.conjunction
with a preponderance for correcting the most critical
individual factor first, has resulted in polarized priorities
targeted to enhance highway safety. As such, there is an
absence of a definite national policy on highway safety in the

country. In spite of the public desire, media campaigns and

4
government efforts, no strategy so far has been effective in

alleviating the problem of increasing highway traffic crashes.

1.2 HIGHWAY SAFETY ADMINISTRATION IN PAKISTAN

The public sector highway administration in Pakistan
operates at two levels of government: i) Provincial, and ii)
Federal. Most of the important inter-provincial trunk lines
and major highways of the country are controlled by the
federal government through an administering agency referred to
as the National Highways Authority (NHA) while the primary
(i.e., inter-district) and the secondary (i.e., intra-district
and agricultural roads) networks are administered by the
provincial highway departments and local bodies.

Traditionally these agencies are only regarded as
"highway construction and maintenance organizations” rather
than the potential saviors of trauma and perpetual misery in
human life. Their functional charter does not necessarily
include participation in a formal HSIP. At present these
highway organizations have practically no formal procedure for
identifying hazardous elements, though they are frequently
seen and often reported by the maintenance workers, police and
inspecting officials. As a result, casual attempts at
improving highway safety for isolated locations are practiced.
However, due to the absence of a well structured and

integrated program the selection of suspect sites and

5
countermeasures is not based on sound statistical procedures.

The motorization rate in Pakistan during the past few
years has been increasing. The motor vehicle population of the
country reached 1,220,145 by the end of 1986 as compared to
191,851 in 1970 [6]. This annual growth rate of 13% includes
a higher proportion of trucks and buses resulting from the
deregulation of trucking in 1960 and a partial deregulation of
buses in 1970. The increasing traffic volume and axle-load has
consistently resulted in expansion and rehabilitation of the
highway network. For example, in the province of Punjab‘about
250 kilometers of the National Highway (N-S), the country's
most important strategic and trade route, have been upgraded
to a four-lane divided highway and almost an equal length is
currently undergoing such improvement. Besides, during the
decades of 1970 and 1980, some busy segments of provincial
highways were upgraded as dual carriageway sections and many
by-passes were provided to avoid interaction of urban traffic
with the main-stream flow.

Pakistan is committed to an ambitious new’ highway
construction;program.IMany'prioritized.highway‘rehabilitation
and construction programs have been completed and some are
under implementation with the assistance of ' various
cooperating agencies for international development; like the
World Bank, the Asian Development Bank, USAID and JICA.
However, these programs clearly address the issues of access,
capacity and structural adequacy with little emphasis on
safety, and regretfully reveal that no important lessons have

been learned from the highway loss experience. These

6

improvement programs are appallingly replicating the safety

deficiencies and persistently adding to the size of problem.

1.3 MOTIVATION

Unfortunately Pakistan continues to be among the
developing nations of the third world having a consistently
high road crash and fatality rate. During the period from 1971
to 1989, the number of total road crashes in Pakistan
increased from 5,892 to 11,238 per year, and fatal crashes
from 1,793 to 4,371 per year [6,7]. Although the statistics
for the year 1988 indicate a very slight drop in crash and
fatality rate per ten-thousand vehicles (which may be
attributed to the relatively higher rate of motorization than
in the previous years), the fact remains unchanged that the
increasing trend of road crashes has not declined, and the
status of highway safety in Pakistan remains much lower than
the developed countries of the world. For example, the present
road fatality rate per ten thousand licensed vehicles in
Pakistan.is estimated about 10 to 12 times higher than‘USA.and
UK respectively [38].

In the early 1980's, a program for the removal of highway
blackspots ‘was initiated in 'the Province of Punjab (and
sizeable funds were allocated by the government to meet the
prospective expenditure. Unfortunately this vital program,
instead of gaining significance, tapered off and was

eventually discontinued after 1987 because its effectiveness

7
was neither measurable nor perceived by the public. As an
example, a financial statement for the removal of highway
blackspots in Rawalpindi Division is presented in Appendix A.
The obvious cause of this unimpressive performance was the
execution of safety programs without following a prescribed
and systematic procedure.

It is quite apparent that identification of hazardous
locations is the basic step in order to embark on a formal
process of highway safety improvement. Traffic crashes are
believed to be the most direct measure of safety of a highway
location. However, attempts to estimate the relative safety of
a highway location using this approach are fraught with the
problems of unreliable accident records and the time required
to wait for adequate sample sizes. In Pakistan, three research
studies [9-11] were conducted for' the identification. of
black-spots which used traffic accident data to accomplish the
task. While one of these studies [9] concluded that due to
fragmentary accident data, identification of black-spots at
reasonably exact positions was not possible, the other two
[10, 11] categorically enunciated that these spots could not be
identified due to incomplete police reporting regarding
accident location.

Herein lies the fiber of motivation for the conclusion
that surrogate measures of highway safety need to be
investigated to develop accident prediction models for the
identification of hazardous locations, so that a HSIP could be

formally initiated and practiced in Pakistan.

1.4 OBJECTIVE

The primary objective of this research is to develop a
logical procedure to identify hazardous locations by employing
surrogate measures of highway safety as predictors of
accidents. This would provide the vital missing link required
for initiating a formal HSIP in Pakistan in the absence of a

reliable and accessible accident data base.

1.5 RESEARCH APPROACH

To determine the course of research and to pinpoint the
crucial variables, a backward screening process was employed
at four distinctive levels, constituting the essence of the
research approach. Briefly, this screening process was
employed to achieve the following objectives.

1) To rationalize the road as the principal factor of
interest in the "road-vehicle-user" classification
of accident causation. A review of relevant
literature [2-5,38,39,68,71] provides the necessary
back-up to this rationale.

2) To select a method of identifying hazardous
locations.

3) To screen out ambient highway hazards from an
inventoried template and identify their presence.

This operation resulted in developing the

9
experimental design for determining a relationship
between hazards and accidents.
4) To distinguish significant independent variables
for utilization in predictive modeling.
This screening operation provided an instrument by which
all the prominent safety aspects and technical options were
considered prior to fine-tuning the research approach and

selecting the variables of interest. A detailed‘dte’scripctionggf

 

t3.‘3...£9‘1£.3§!§13._9f.fuming?” #8 Presented...“ Cha.PFer....'°.hres~

The experimental design is based on the outcome of this
screening. The research data is comprised of information on
ambient highway hazards and accident history of highway
sections. These data were collected in Pakistan in early 1991
[12] and correspond to the time period of January 1988 to
December 1990. The experimental site was comprised of 86
kilometers of the National Highway (N-5) passing through the
rural areas of District Rawalpindi. Of these, 52 kilometers
were 2-lane, 2-way, and 34 kilometers were 4-lane divided
highway sections.

Finally, computer routines for multivariate regression
analysiswwflwereh “employed to investigate statistical.

\nww" ' N,

relationships betweeth eflt o entities. Accident prediction
models were then developed using this information with annual
accident frequency per kilometer as the dependent variable.
The feasibility of aggregating the data in terms of a hazard
index to determine the accident potential of a highway section

was also examined.

10

1. 6 ORDER OF PRESENTATION

This dissertation is organized in six main chapters
followed by appendices A to E. Following this chapter, a
literature review based on global research experiences in a
chronological order is presented in Chapter Two. Since
literature from the USA and UK is frequently referred for
prospective application in a developing country, a critical
assessment of specific literary material in terms of its
applicability and transferability is included.

Chapter' three jpresents the details of the research
approach and the specifications of the experimental design.

Chapter four concerns statistical analysis and accident
prediction modeling. The development of a hazard index based
on an adequacy rating of a highway section to determine its
accident potential forms a part of this chapter.

Chapter five offers a thorough discussion on various
inputs, the implementation processes and the results of the
research. This includes an evaluation of the independent
variables; data type; predictive models; hazard index; and
accident data.

Chapter six presents the research limitations, inferences
and conclusions, and the suggested research. The chapter also
covers the remaining steps of the HSIP.

The research data, some selected photographs of the
experimental site, and the SPSS program for data analyses are

presented as the relevant appendices.

CHAPTER 2

2.1 HISTORICAL PERSPECTIVE

Intellectual concerns about the public health problems
inflicted by automobiles and highway-travel started surfacing
in the motorizing societies of the world in the early 1940's.
As the level of motorization gradually increased during the
1960's, accompanied by an upward trend in traffic crash
frequency, the issue of resulting property damage, injury and
pre-retirement of life became a matter of wide public concern.
Although the time frame in which various countries attained a
certain level of motorization differs, they had a common
viewpoint on road safety which included a high degree of
national concern for research and improvement programs.
However, it was not until the late 1960's and early 1970's
that the global importance of highway safety was further
emphasized through legislation and research in many countries.

The United States, Canada, Australia, and various other

11

12
countries of Western Europe (e.g. , Great Britain, France,
Belgium and Sweden), having the highest rates of motorization,
were the pioneers to assign a national importance to the
subject of highway safety.

Besides assignment of national importance, the topic of
highway safety was intensively discussed at an international
level. For example, Organization for Economic Cooperation and
Development (OECD), and Commission of the European Community
(CEC) were particularly active in Western Europe in road
safety research and dissemination of results. The efforts of
these organizations in transportation research mostly
converged to one common objective: to ameliorate road crashes.
It was anticipated that the wisdom of the western European
nations would be able to frame effective international
regulations before automobiles confirmed their increasing
reputation as "the plague of the 20th century". The Overseas
Unit of the Transport and Road Research Laboratory (TRRL),
U.K., performed numerous studies to investigate the highway
safety problems in many developing countries.

Extensive efforts were made in the motorizing countries
to find a plausible explanation of the traffic crash
phenomenon in terms of its rationale and apparent attributes.
These studies were the genesis of many theories and macro
models of highway safety relating accidents, fatalities or
injuries. with a myriad of independent variables. These
variables represented a broad spectrum of technological,

social, economic, demographic, biographic, psychological and

13
even the religious aspects of driver and road system [14-23].
Since a discussion about the macro models is not the main
scape of this work, only the findings of some of the selected

references is summarized in Table 2.1 to maintain continuity.

Table 2.1 Summary of Socio-Econosic Macro Models of Highway Safety.

 

REF RESEARCH VARIABLES OF INTEREST TYPE'

 

 

 

 

 

 

 

 

 

 

 

 

lo. PARTICULARS or
W Independent Variable(s) AMY“
V-idilele)
Significant 80% + level an
Skuifioent
(f) Coeff. (-) Coeff.
1,2 Average speed Alcohol- CS
Driver age consumption (AC)
income Cost/accident
1 WT, GNP Unemloyeent TS
Vehicle population
16 Sivak 1 Young drivers Suicide CS
(1983) Murder rate rate
17 llatoper 5,6,7 VH1 Vehicle length Cost TS
(1984) Average speed [accident
Rural/Urban travel Income
18 hautzinger 2 Vehicle population CS, TS
(1986)
19 Gaudry 2,3,4 Young drivers Cost/accident TS
(1987) Seat belt, Ac Uneaploynent
20 Loeb 3 Rural/Urban travel Length of CS
(1987) Average speed, Ac arterial roads

 

 

 

 

 

 

 

* CS 8 Cross-sectional
TS 8 Tina Series

 
   
   
   

Lhflfllfl!fifl&lflflflun=

1. Fatalities in road accidents per vehicle ailes travelled (VMT).
2. Injury accidents per capita or per VHT.

3. Fatality road accidents per capita or per VHT.

4. Fatalities at night.

5. Total number of road fatalities or injuries.

Number of notor-vehicle-occupants fatalities.

Number of trian fatalities or injuries.

 

 

 

 

The technological aspects of highway safety were further

segregated into two distinct areas of interest: the

l4
automobiles and the highways. In this dissertation, only the
highway-related. safety' aspects. are considered, being ‘the
specific orientation of the work. A chronological overview of
the evolution of this subject in the motorizing countries is
presented in the following pages.

During the decade of the 1950's a large number of studies
were carried out in the USA on the relationship between the
highway geometric design elements and traffic flow
characteristics, and road safety. In 1954, the AASHO published
design guidelines [24] which reflected the major findings of
these studies. The main theme of these guidelines was to
classify a road hierarchy, and to assign different standards
to different. types of road. In 'this hierarchy, access-
controlled and divided highways were categorized as a
relatively safer class of highways. In 1965, AASHO issued the
next edition of these guidelines [25] placing greater emphasis
on the complexity of road accidents and the role of the human
element. However, the prime importance continued on highway
geometrics including topics on crash barriers, road side
obstacles, and specific criteria for climbing lanes. (These
guidelines were further revised.and.expanded,byuAASHTO in 1984
and 1990).

In 1963, the Automotive Safety Foundation and the US
Bureau of Public Roads published a major study [26] on highway
safety considering the relationship of traffic control and
roadway elements with traffic crashes. In this study traffic

volumes, (access control, cross-section, alignment,

15

intersections and interchanges, at-grade railroad crossings,
driveways, speed, pavement surfaces, one-way streets,
illumination and parking were thoroughly investigated for
their relationship with highway safety. In 1966, the AASHO
special Traffic Safety Committee undertook a critical survey
of the safety characteristic of the interstate and other
highway system and recommended improvements [27] applicable to
two distinct areas: roadside design and appurtenances, and
traffic operations.

The increasing frequency of highway fatalities during the
mid 1960’s drew the attention of the Us Congress to the need
for an expanded federal role in Highway Safety. In 1965, the
rigorous lobbying of Ralph Nader and his distinguished
publication [28] raised public consciousness of the issue of
highway safety and stimulated national concerns for reducing
traffic accidents and fatalities. Numerous Congressional
hearings were held and these resulted in enactment of two
important pieces of legislation concerning highway safety,
i.e., The National Traffic and Motor Vehicle Safety Act, and
The National Highway Safety Act, of 1966. These legislative
actions significantly expanded the federal role in highway
safety by creating NHSB (the predecessor of the present
NHTSA), and by bringing new focus to research for the
advancement of the knowledge-base in highway safety [29]. The
Highway Safety Program Manual was accordingly developed by the
USDOT which included the Standards for various specialties, as

shown in Table 2.2.

16

Table 2.2 Standards for highway Safety Programs.

 

These standards covered 18 program areas which are now
administered by NHTSA and FHWA. While efforts in the United
States featured a coordinated approach to improve multiple
aspects of highway safety at all levels of government, two
major European publications in the mid 1960's represented the
state of the art in other knowledge areas.

In Sweden, a strategy based on the relationship between
urban planning and road safety emerged for alleviating
problems of traffic accidents. The Swedish National Board of
Urban Planning issued guidelines in 1968 [30] giving more
attention to hierarchy and strict design standards with more

conservative speeds applicable to urban and rural environments.

17

In Great Britain, the TRRL in 1963 presented the combined
results of studies on pedestrians, drivers, vehicles and road
design [31]. This study presented, for the first time, a
coherent compilation of research findings on the interaction
among road users, vehicle characteristics and highway design
as integrated components of the highway transportation system.

The evolved phrase of "vehicle-highway-user" system
forthwith became a popular entity and generated areas of
interest for further research and development during the
1970's and early 1980's. However, attempts to‘assign liability
and to concentrate efforts on the correction of a specific
component was a major pitfall, soon discovered by the
developed world. Unfortunately this was not realized by most

developing countries.

2 .2 THE CONCEPT OF LIABILITY-ASSIGNMENT IN ACCIDENT CAUBATION

In the era after the :mid 1960's, the three basic
components of the highway transportation system, i.e. , the
vehicle; the roadway; and the user were thoroughly critiqued
for malfunctioning. Their individual responsibility toward
highway safety in terms of perpetrating serious social
problems were assessed. The classical literature indicated a
minor contribution from the roadway and the vehicle as
compared to the driver in accident causation. Following is a
brief component-wise overview of these findings from various

countries of the world.

18

Vehicle: Typically vehicle failures were found not to
have a major (i.e. , not more than 10%) contribution in causing
traffic accidents in the developed countries. In the USA,
Hoback [32] reported that only 9.3% of the accidents on
Oklahoma Turnpikes resulted from adverse causes related to
vehicles. In the UK, Sabey and Staughton [33] found an overall
contribution of 8% due to vehicle failures. In Germany, Bitzl
[34] reported 6.9% of accidents on the.Autobahnen.as being due
to tire failure and defective vehicle mechanism.

Roadway: The roadway, quite similar to vehicles, was
shown to have a relatively small causal relationship to
traffic accident. Michaels [35] reported that highway
characteristics played a significant role in only 5% of the
accidents. In analyzing the role of roadway elements in
Pennsylvania Turnpike accidents, Eckhardt et a1. [36]
concluded that considering the three main components of
driving operation, the roadway design was well ahead of the
driver and his vehicle. Treat [37] reported a contribution
factor of 3% for the USA, while Sabey and Staughton [33]
attributed the road environment to cause only 2% of the
accidents.

Driver: Historically, the driver has been identified as
the most significant single component of accident causation in
context of the vehicle-roadway-driver system. In the USA,
Treat [37] , utilizing data from Indiana, found that the driver
was the exclusive factor in 57% of total accidents. In the UK,
Sabey and Staughton [33] found that driver errors alone were

the causative factor for 65% of the accidents.

19

2.2.1 Liability Studies in Developing Countries

Various studies [38,39,40], made in the developing
countries during the period 1970-85 to analyze the causes of
accidents, revealed similar results and showed a low
contribution of the roadway and the vehicle, as compared to a
very significant role of the user. The Overseas Unit of the
TRRL collected annual police reports from a number of
developing countries which provided a basic summary of the
road accident situation and the major "cause" of the accident.
Though the police had ascribed a "single cause" to each
accident rather than listing the "factors involved", this
information was nevertheless considered to provide an insight
to the police viewpoint of major factors involved in road
accidents. The results of the study based on the police data
of 5 developing countries are reproduced in Table 2.3. Road
user error was identified as the main cause in 71-95% of the

road accidents [38].

Table 2.3 Causes of Road Accidents in 5 Developing Countries [38].

 

 

 

 

 

 

 

 

 

 

 

 

 

 

MALAYSIA Halli-KONG
, _ 1976 1977
» HAIR CAUSE OF ACCIDENT +P ! +P +P +P +1
‘ Road-user Error 95 77 71 87 92
Vehicle Defect 1 16 12 1 * I
Adverse Road Conditions 1 5 2 8 *
Other 3 2 15 4 8 II
I TOTAL ”m“ _ 100 100 100 100

+P I POO included 11’ I POO inclusion not known +1 - injury accident only

20
Ergun [39] found similar ordered figures for Turkey, as

summarized in Table 2.4.

Table 2.4 Traffic Accident Causes in Turkey [39].

 

 

 

 

 

 

 

 

CAUSE PERCENT
CONTRIBUTION*
Road User Error 94
Vehicle Defect 5
Road and Environment 1
Others -

 

 

 

* Total accidents considered.

Swati and Downing [40] also reported the road-user as the
single major cause of accidents in Pakistan. The percentage

contribution of each cause/factor is reproduced in Table 2.5.

Table 2.5 Traffic Accident Causes in Pakistan [40].

 

CAUSE/FACTOR PERCENT
CONTRIBUTION*

Road User 90

Road and Environment 6
vehicle 4 I

* Percentage of accidents in which cause/factor was identified.

 

 

 

2.2.2 The Theology of Single-Liability-Assignment

The above cited factor-contribution studies mostly used

police accident reports. Their results merely indicate that

21
driver errors, often accompanied by law violation, are in the
chain of events leading to 70 to 90% of all highway accidents.
Commenting on this situation, Oglesby [41] observed that
single minded proponents of driver education or strict
enforcement sometimes distort this statement by saying that
driver errors cause 90% of accidents.

In many developing countries, highway safety efforts are
extremely influenced by the single liability approach and
consequently their primary focus is on driver correction. For
example, Al-Isa [42] reported that in Saudi Arabia highway
safety authorities held the traditional "violation-error"
attitude toward accidents. Therefore, the safety programs in
that country were directed toward changing the behavior of
violator drivers by imprisonment or fines.

The recommendations of Somnemitr [43], for traffic
accident prevention in Thailand, indicated the need for road
user education and traffic law enforcement because of
unlicensed drivers, disobedience of traffic laws, and use Of
amphetamine and other stimulants while driving. In Pakistan,
Swati [44] stressed prioritized safety measures oriented
toward improving the road user’s knowledge of traffic rules,
enforcement, and updating laws for alleviating highway safety
problems.

In the United States as well, most early highway safety
initiatives ‘were focused on. the driver being the :major
contributor to motor vehicle crashes. These efforts included

safety campaigns, driver education, training and testing

22

programs, and use of punishment as a deterrent to violations.
However, the jeopardy of isolating the human component for its
prioritized correction was timely detected in the motorized
countries. To them, it became obvious beyond doubt that the
importance of vehicle and highway related safety programs
could not be ignored at the cost of improving human traits.
For example, in the USA, due to a dominant focus on the driver
as the primary cause of crashes, vehicle crashworthiness
research remained a largely undeveloped area until the mid
1960's. Significant advancement and application of knowledge
in this area was noticed after the mid 1960's. Similarly,
highway-related programs, having‘ indirect and direct bearing
on safety, like TOPICS, RRR, and HSIP were launched after the
mid 1960's, which later on demonstrated a definite achievement
of objectives.

It was shown by Koshi [45] that accident reduction in
Japan in the 1970's was largely attributable to improvement.of
the environments of road users rather than improvement of the

road users themselves.

2.2.3 The Interactive Role of Accident Causative Pastors

In the 1970's two major studies [37,33] were carried out
in the USA and the UK to investigate the independent and the
interactive role of factors associated with large samples of

crash data. The US study was performed at the Indiana

23

University by Treat [37] while the British study was performed
at the TRRL by Sabey and Staughton [33]. In both the studies
a multi-disciplinary accident investigation team was employed
to find the accident causes. This approach had been previously
employed with reasonable accuracy to determine accident causes
in.other'man-machine systems such as, aviation, railroads, and
shipping. This approach was supported by other techniques like
experimental analysis and incidence reporting.

Rumar [46] summarized.the results from.both studies in an
interesting pattern as shown in Figure 2.1 with the following
interpretation:

1) The vehicle is identified as the sole factor in 2%

of the crashes;

2) the interaction between vehicle and road user is
identified as a factor in 6% of the crashes;

3) the interaction between vehicle, road user and
environment is identified as a factor in 3% of the
crashes; and

4) the interaction between vehicle and road
environment is identified as a factor in 1% of the
crashes.

The corresponding values for the UK study are 2%; 4%; 1%;

and 1% respectively.

The analysis of the US study [34] were further extended
by classifying the type of human errors involved and are
reproduced in Figure 2.2. It can be seen from Figure 2.2 that

recognition and decision error predominate. These type of

24

l —

 

Vehicle

Road Road
E nvironment“ User

 

 

 

 

 

 

 

 

 

28/34 95/94 8/12

 

Figure 2. 1 Percentage Contributions to Road Accidents as Obtained in a
British and US Accident Study [46] .

 

    

  

 

 
       
     
      

 

60
III DefianCauuiicuus
50 I-
553 DOfll‘lIt. or Probable
.e.e.b.e:e cm a
40 __ ...;.:.;.; Severity-hermim
3.3.3. Factors
96 Of
Accidents

20

1O

 

 

        

   

          

 

0' In- On- In- On- In- On- In- On- In- On-
Depth Site Depth Site Depth Site Depth Site Depth Site

Recognition Decision Performance Critical- Non
Errors Errors Errors Non- Accident

Performance (e.g.
(Blackout Suicide)
Dozing)

Pigure 2.2 Percentages of Accidents in which Susan rectors were
Identified as Definite or Probable Causal Factors [371.

25

errors are obviously caused by "inappropriate information
acquisition and processing." This study further specified
human errors in the decreasing order of frequency of
occurrence revealing the following hierarchy of errors:

1. Improper lookout.

2. Excessive speed.

3. Inattention.

4. False assumption.

5. Improper maneuver.

6. Internal distraction.

Investigating the interactive role of these factors in
Jordan, Balbissi [47] also found results quite similar to the

US and British studies. His results are shown in Table 2.6.

Table 2.6 Traffic Accident Causes in Jordan [47].

 

 

 

 

 

 

 

 

 

 

 

REASONS OP ACCIDENT PERCENT
-__- CONTRIBUTION*

Human Errors 65.00
Combined Human and Road Elements 24.00
Combined Human and Vehicle Elements 04.50
Combined Human, Road and Vehicle Elements 01.25
Road Elements 02.50
Road and Vehicle Elements 00.25
Vehicle Elements 02.50 4'

 

 

* Averaged over 5 years (1979 through 1983).

These three completely separate and large studies [33,
37,47] of several thousand accident records, and corresponding
to different geographical locations, were almost unanimous in

assigning the road user as the dominating cause of highway

26
traffic accidents.

However, many words of caution have been offered by
researchers such as Klien and Waller [48], Shinar [49],
Campbell [50], Blatnik [51], Jacobs and Sayer [38], and Rumar
[46] about drawing instantaneous and raw inference from the
results of such studies and interpreting their findings. The
reasons for citing such cautions.by these researchers are many
and diverse.

Klien and Waller [48] maintained that data collected
through police investigation or through self reporting were:

1) Incomplete (in terms of the relevant observations
that were to be recorded).

2) Unreliable (in terms of the citizen's or the police
interpretation of the observation that were
recorded).

3) Unrepresentative (in terms of crash investigations
that did not represent a cross-section of all
crashes that occurred).:

Overall, they concluded:

-'a number of carefully designed research studies have attempted to
identify causal factors, and most of these have developed
conclusions that differ markedly from those reached from the use of
police data or common sense".

Campbell [50], referring to the results of the State Road
Commission's research on two-lane rural highways in West
Virginia, observed that roadway features were associated with

the driver in accident involvement. He further commented that

27
technological innovations for highway improvement could
substantially help protect the vehicle and occupants after the
accident dynamic was activated.

Blatnik [51], as the chairman of the special sub-
committee on the Federal-Aid Highway Program categorically
emphasized highway-related safety improvements. He rejected
the approach ascribing the majority of accidents to "driver
failure" because of human limitations (e.g., imperfect sight
and hearing, limited intelligence etc.) , and a complex of

emotions that no one fully understood. His deposition stated:

- " when a driver falls victim to an accident despite his best
efforts, it may not be the driver who has failed".

Jacobs and Sayer [38], commenting upon the use of police
accident records to analyze causes of accidents in various
' developing countries, concluded that it could be dangerous to
draw conclusions about variations between the countries as
there were likely to be differences in the types of accidents
reported to the police, and in the way in which the police
analyzed the accidents for causes. Even for a single case,
their observations were:

- ' Also it is likely that the percentages are under estimates of

the true contribution of these factors because in many of these

accidents there are probably several factors involved and not just
one. Thus the percentage of accidents due to adverse road conditions
and environments may in reality be much higher because many of the

road use errors could have been due to inadequate road signing or
marking".

Rumar [46] observed that the weakness of the liability

assignment approach was evident because these studies lacked

28

an explicit theoretical basis, their results were hard to
relate to other types of data, and they tend to use the human
factor as a scrap box. Referring to Haight’s article [52], he
quoted the example of the Japanese White Paper of 1982 which
listed " failure to drive safely" as the major cause of
accidents. He concluded that typical human errors contributing
to accidents were both perpetual and decisional, and were
related with information acquisition and processing.

Evans [1] commented on the high rate of driver’s
involvement in traffic crashes and discredited existence of a
relationship between driver performance and characteristics by

citing young drivers record in the USA. His observations were:

- ”While various aspects of driver performance are related to
safety, there is not a coherent pattern. The findings of no effect
from driver education and knowledge, and that younger drivers, with
the best visual acuity and shortest reaction times, have the highest
crash rates, suggest that driver performance is not the driver
characteristic which has the largest influence on traffic safety".

Numerous studies have attempted to identify human traits
that were common in individuals involved in traffic crashes.
Due to a variety of psychological traits apparent in chronic
traffic violators and accident repeaters, such as
aggressiveness, intolerance, and resentment of authority, it
was concluded by Goldstein [53] that it would be difficult if
not impossible to use human characteristics as reliable
predictors of accident involvement.

The accumulated main theme of these perspectives implies

that the liability'approach.neither explains the traffic crash

29
phenomenon nor assigns a substantial importance to highway-
related safety improvement programs concurrent to efforts

aimed at promoting driver’s performance.

2.3 THE CONCEPT OF EXPOSURE

The concept of exposure takes into account the amount of
opportunity for accidents which the driver of the traffic
system experiences. Quite in contrast to the liability
concept, many studies showed that exposure was the most
convincing explanatory approach to the interpretation of
traffic crash situations. Blunden [54] categorically pointed
out that there had been too much concentration of effort in
the past on the liability factor, and urged that more emphasis
be placed on the study of exposure. Commenting upon the
explanatory potential of the exposure approach, Chapman [55]

observed that:

- ”If the number of accidents is found to be closely related to the
amount of travel, the amount of the traffic can be regarded as a
measure of exposure. If no relation is found, does this invalidate
the use of traffic as such measure? The answer to this is negative;
the variation which has not been explained by travel may be due to
something which has not been measured. This problem is faulty
experimental design, with no control of variables other than those
under study”.

In this study, Chapman [55] has presented a fairly
complete review of the exposure literature describing the
concept and application of exposure, and various terms and

extensions associated with it. In a different comparative

30
study of exposure measures at intersections, Chapman [56]
found that the accident rate at cross roads were significantly
higher than at T- or Y-junctions; twice higher in urban areas;
and five times in rural area, irrespective of many variants of

accident measures.

Erlander et. al [57], showed that significant variation
in the daily accidents in rural areas could be explained by
the amount of traffic. Baker [58] further commented on these
results that correlated exposure with accidents. He

enunciated:

- ' The absolute number of fatalities and injuries has steadily
increased, but so has the population and amount of travel. As the
population increases, the number of travellers and vehicle-miles of
travel will increase for the same level of mobility for individuals.
As a result, the degree of ”exposure” to accidents is increased, and
a greater number of accidents would be expected if no improvements
were made in the highway transportation system“.

In a study by Operation Research Inc. [59] , strong
attention was paid to exposure, viewing it as a systematic
process affecting the crash system, which was an outcome of
the continual interaction of driving behavior with the ever
changing environment. The study considered that three basic
elements of exposure were important:

1) Characteristics of drivers and vehicles;

2) Characteristics of the road system and intensity of

system use; and

3) Environmental conditions (weather, day/night.etc.).

31

2.3.1 Induced Exposure

Literature showed that it was not always possible to
obtain an appropriate estimate of the exposure in terms of the
above mentioned elements. Thrope [60], first of all, proposed
a method which did not require directly measured exposure
(i.e., in terms of traffic and roadway characteristics) but
which induced exposure from the accident data. Subsequently
many other researchers, e.g., Carr [61] and Haight [62], also
followed this approach to develop various mathematical models
pertaining to highway safety. Their fundamental concept
presumed that the population of innocent accident involvement
could be taken as the representative of the entire population
at risk. A comprehensive validation of this concept was made
by Taylor and DeLong [63] employing more sophisticated
asymmetrical models.

An extensive review of the literature on exposure, and
affiliated material, is beyond the scope of this dissertation.
The objective of briefly citing exposure literature was to
demonstrate its potential and applicability toward explaining
the traffic crash phenomenon, which was something beyond the

concept of simple liability assignment.

2.4 THE CONCEPT OF RISK-NONEOBTABIB AND RISK-COMPENSATION

The theory of risk-homeostasis was first presented by

Wilde [64] in the early 1980’s. According to this theory, risk

32

taking behavior involves an attempt to balance perceived risk
and desired risk, and people adjust their behavior in
response to changes in perceived risk. A concurrent theory,
presenting a more acceptable view of the effects of safety
measures on driver behavior, concerns offsetting driver
behavior or risk compensation. According to this theory road
users adapt to conditions and regulations in a way that alter
their level of risk, and in some cases even negate their
original desired intent. The findings by Crandall [65]
partially'discredited.this‘theory'by'showing'that.theiNational
Traffic and Motor Vehicle Safety Act was effective in
significantly reducing the car-occupant fatality rate.

The concept of utility maximization, quite analogous to
these theories, has specific application in many areas of
transportation engineering including' planning and policy
decision making. In the context of safety, this concept would
imply that an individual driver will choose between safety and
his other activity options (i.e., work, recreation etc.), and
then will weigh the benefits and cost of safety features with
the set of anticipated driving conditions. Literature's
[66,67] distinct indication of practical application choice
models in travel demand analysis which were based on the
principle of utility' maximization, seem to substantiate
applicability of the risk compensation theory in highway
safety modeling. From this analogy, it may be presumed that
properly designed HSIP are most likely to modify driver

behavior for safer driving.

33

2.4.1 The Role of Highway Engineering in Risk Compensation

Highway Engineering can play a dominant role in driver's
behavior of risk-compensation if it can be demonstrated that
traffic crashes are most frequent in those circumstances where
relatively higher demand is placed on the drivers ability to
perceive and cope with the situation. A study by Vercase [68]
provided important information on this aspect. He studied
fourteen highway variables in which factor analysis techniques

were applied to roadway and accident data. He concluded:

-"only one single factor emerged from the vast amount of the data in
this analysis which explained where accidents occurred. Although
only highway variables were included in the analysis, this one
factor conveys a psychological meaning: There are more accidents at
those places where situation places greater demands on the momentary
perceptual-decision-motor capacities of the driver. The drivers
basic psychological capacities are heavily exercised when he must
deal with a situation around him that is changing rapidly."

This implies that traffic crashes are most frequent in
those circumstances where traffic friction or conflict is
greater i.e., where one encounters more cars and where there
is traffic flow interference from intersections and driveways.
In.this study it was clearly found that accident frequency'was
proportional to the load or rate of demand placed on the
drivers basic ability to perceive and cope with the situation.

A synthesis of the Vercase findings with the results of
Treat's study [37] (that stratified human recognition and
decision errors, and showed that "inappropriate information
acquisition and.processing" were the obvious causal factors),

imparts the real significance of highway engineering in risk-

34
compensation. The specific discipline which takes an account
of human limitations in the driving task, when considering
improvement in the highway transportation system, is referred

to as the human engineering approach.

2.5 HUMAN ENGINEERING APPROACH

The highway features and the physical changes which
require the driver to make a decision in an extremely short
period of time while driving, may be termed as failure of a
highway system. The driver has to perform three sequential
operations in this short span.of time while driving: 1) detect
hazards or potential dangers; 2) evaluate the overall
situation and decide the ideal action; and 3) take the final
action. It is evident that highway engineering and technology
is directly related to the first two items of the driving
task. The design and signing system alone could avert an
impending accident situation at three levels of technology:

1) Primarily, by providing a relatively hazard free

designed highway; i

2) Auxiliarily, by providing appropriate information

about highway hazards if present; and

3) Over and above, by allowing the highway system to

forgive the driver even if he misjudged the ambient
conditions at certain points.

The vehicle driver is the most important single component

35

of the driving process and the overall highway transportation
and safety system, and is the most difficult to understand.and
control. The human engineering approach attempts to detect
human performance limitations in the complexity of the entire
driving task and uses these findings to redesign or improve
the system to make it compatible with the needs and
capabilities of the road. users. For example, Evans [1]
suggested that, since drivers were poor judges of speed of
oncoming cars, technological innovations providing such
information could increase traffic efficiency and safety in
overtaking maneuvers.

To sum up, it may be concluded that there is a growing
body of literature denoting that highway-related engineering
measures can drastically reduce accident potential of a
highway location or a section. Claes [69] estimated that

proper engineering could reduce the accident rate by 70%.

2.5.1 Punctionalisation of Human Engineering Approach in the
Developing Countries Through HSIP

Some important lessons learned in the area of highway
safety by the motorized societies, through extensive research
and persistent sufferings, could be of significant benefit to
the developing nations of the world. The review of the cited
literature explicitly showed that in the developed countries,
an integrated and simultaneous approach toward the highway

safety problems is the crux of state of the art practices. It

36
may be, therefore, inferred that highway-related safety
improvement strategies in the developing countries can not be
ignored or delayed at the cost of another causal factor deemed
to be improved first.

In this context, Ergun [39] pointed out that without the
provision of a bare minimum in engineering and safety
standards, efforts directed toward improvement of driver
behavior in the developing countries might result in a total
waste of resources. He further suggested that the safety
concepts represented by terminologies like, "design for
safety"; "forgiving highways"; "driver expectancy"; and
"design consistency" should be incorporated in highway design

policies of the developing countries. He enunciated:

- ”Since developing countries are still in their ”infancy” of
motorization and 'their highway' network. expansion, it is very
important for them to incorporate safety concepts in highway design
as it will become more difficult and costlier to correct such design
errors later“.

Jacobs and Sayer [38] asserted that for developing
countries, safety features such as those involving geometry,
signing and delineation, should be introduced at the design
stage rather than added later (almost as an "after-thought")
for the reasons of increased costs and relocation of at-ground
services.

To incorporate safety concepts at a post-design level, a
systematic approach is essentially required on account of
three basic reasons: 1) For identification of hazardous

locations; 2) For selection of appropriate corrective measures

37
to replenish chronic mistakes; and 3)’ For evaluation of
effectiveness of such incorporation.
The pragmatic response to this important technical
requirement is offered by the HSIP as described in the

following section.

2.6 THE HIGHWAY SAFETY IMPROVEMENT PROGRAM (HSIP)

The IHighway Safety' Improvement. Program. (HSIP) is a
contemporary US terminology [71] representing a sequential
plan for highway-related safety improvements structured in
terms of various components, processes, sub-processes, and
procedures. These terms are briefly described as follows:

W: These are the three basic phases of the HSIP,

i.e., Planning; Implementation; and Evaluation (also see

Tables 2.9-2.11, pages 61-62).

zxggggggg: These are the sequential subsets within each

component. For example, there are four processes in the

Planning Component.

fink:pgggggggg: Each process is often divided into

subprocesses, which are the categorized technical

operations.

[Igggggzgg: These are the suggested specific methods to

perfonm the technical operations. For example, in the

process of identifying hazardous location, there are

seven procedures listed to perform this operation.

38

2.6.1 Framework and Overview of HSIP

A flow-chart presentation of the overview of HSIP is
given in Figure 2.3. The magnifications of the chart at the

process level and at subprocess level are presented in Figure

2.4 and Figure 2.5 respectively.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

   

 

 

    

 

r——>l PLANS FOR THE TOTAL HIGHWAY SYSTEM l
_ *6 '6 , 6 V
PLANNING OPERATION
AND CONSTRUCTION AND

DESIGN MAINTENANCE
_» "I-I-I-l-vI-I-I-I- -I-I1
:3 - HICHHAY SAFETY IRPROVERENT i
g ! PROGRAM _
-9 d l
as s ! .
g8 . | I
.§ - i
:2 § ! ,
E35 ! !
52: I I
:8 to I '
~¢ - EVALUATION |
gg'é’ I COMPONENT i
U ! .
LI-I-I-I-I I-I-I-I-l-IJ

 

 

‘1

Figure 2.3 Overview'of the Highway Safety Improvement Prograe.[71].

39

 

 

 

 

PLANNING PROCESS 1
COMPONENT COLLECT AND
MAINTAIN TA
1
PROCESS 2

 
  

IDENTIFY HAZARWUS
OCAT NS AND LEM NTS

 

   

PROCESS 3
CONDUCT ENGINEERING
STUDIES

1

PROCESS 4
EST ABLI SHTPEOJECT

    

.—.-._._.—;—.—.-.-.-.

 
  
  
 

.-.-.-.-3.-.-._.-.-.

LD-I-I-I-I-D-D-I-I-I-D-I-PIR- -IS-I-I-IJ
P.-.-l-I-I-I-I-I-I-IA-I-I-I-D I-I-I-I-

I. IFI’LERENTATIM PROCESS 1 I
'fi MONENT SCNEOULE ANO IMPLEMENT fl"

LI-D-I-I-I-J-I-I-D-I-I-I-I- -I-PI-I-Ij

 

 
  
 

FD-I-I-I-I-I-I-I-I-I-I-I-I- -I-D-I-l‘

 

q I

I EVALUATION PROCESS 1 I

' COMPONENT DETERMINE THE EFFECT ' j
or man“ SAFETY “'4

 

 

   

i i
LI-I-I-l-I-I-I-I-I-I-I-I-I-I-I-I-I-IJ

Figure 2 .4 Highway Safety Iaproveeent Prograe at the Process Level [71] .

40

 

   
 

 
 
       
 
 

 
       
 

 

 

 

 

 

 

 

   
 

   
 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I
: PROCESS I. CdLECT AID MAINTAIN DATA :
| I
I sumac“: I I
“Fl! M NIGNUAT '
I LOCATION REFERIICL
I SYSTEM |
' I
I
| I
| CILCCI AD :sAIleAI ' |
I ACCIsiII DATA Nieuv DA1A I
name
r-.1 cannaewr +-.T
‘ I
I I PROCESS 2. IDENTIFY HAZARDOUS LOCATIONS AID ELEMENTS I I
: PROCESS 3. CONDUCT ENDIIEERIN; STUDIES I
I ' '
| means: 1 m3 2 W” J I
i w: 3:. arm: I
' A. I
I ,
| I
l [ PROCESS 4. ESTABLISI PROJECT PRIORITIES ] I
| 1 gr
I PROCESS I. SCMEDuE AIo IMPL NT SAFETY IIPROVEI'ENT PROJECTS |
L—‘u man-new
:w I I
' I
l
'----_----------—----—--—- -------'
I l _ 1 4
: PROCESS I. DETERMIIE TIE EFFECT OF IIIDNIMY SAFETY IMPROYEIIEIITS :
I : '
I
| I
H F
W
| I
l I
I I
I I

Figure 2.5 Highway Safety Improvement Prograe.at the Sub-Process Level
[71].

41

2.6.2 Implementation of HSIP

Identification of hazardous locations is one of the most
fundamental process, and requires collection and recording of
accident data to perform this task. However, a substitute
strategy will be employed in this dissertation since the
objective of the current research is to develop a HSIP
practicable in conditions where accidents records are not
reliable or readily available.

The following two sections of this chapter precisely
cover the literature on the utility of roadway features and
traffic characteristics as a surrogate measure of highway

safety, and their use in accident prediction.

2.7 SURROGATE MEASURES OP HIGHWAY SAFETY

By’ convention, 'traffic crashes are ‘the :most. direct
determinants of hazardous locations. However, some critiques
seriously opposed this perception. For example, Hauer and
Persaud [72] suggested that there were serious problems of
identifying hazardous locations using accident data, and
showed that in two cases a significant proportion of deviant
accident sites remained unidentified while many sites which
were subjected to countermeasures were not deviant at all.
Hauer [73] showed that there would be a reduction in the
number of accidents at sites identified by a high number of

traffic crashes even if the countermeasures were ineffective.

42

An explanation to this type of phenomenon was offered by
Griffin et al. [74] in terms of a mathematical expression
referred to as "regression to the mean". In simple words, it
means that since sites selected for treatment generally had
much higher than average crash rates, these rates would tend
to be lower in subsequent years regardless of treatment.
Various studies [e.g. , 75-77], therefore, developed algorithms
for the identification of hazardous locations to reduce the
statistical bias caused by accident over-representation from
the average. However, these arguments were presented from a
purely scholastic standpoint, and were case-specific.

Perkins and Harris [78] had a different reservation in
using accident records for identifying hazardous locations,
especially for intersections. They suggested that the accident
potential of a location should be objectively measured without
waiting for an accident history to evolve - an approach which
they referred to as "dynamic evaluation of an intersection".
They used traffic conflict characteristics as measures of
accident potential, and observed that, in three 12-hour
observation sessions, it was possible to completely evaluate
an intersection using the obtained information, which was more
comprehensive than that normally available from accident
records. Their studies showed a high level of association
between traffic conflicts and the reported accident
frequencies.

A comparison of direct and indirect methods for
determining accident potential was made by Pahl [79] . He

concluded that the outstanding problem in using accident

43

records was the moral issue of having to wait a certain number
of accidents before any statistically reliable results could
be obtained. He asserted that, in principle, the correlation
between a direct candidate measure and the accident potential
of a highway site appeared to be feasible. Pahl' s findings
supported the use of indirect measures and pointed out that
the indirect candidate measure needs to be correlated with
accident data.

However, concurrent with these findings, the use of
accident records is still the state of the practice for
diagnosis of deficiencies and application of corrective
countermeasures. At present, practically every highway agency
having access to a comprehensive accident data base, uses some
variant of the rate or number method to identify hazardous
'locations, for discharging its obligation toward highway
safety.

The specific literature on HSIP indicated several methods
for identifying hazardous highway locations utilizing accident
histories (see Table 2.9, Page 61) . However, in circumstances
where accident records are not available, adverse highway
features and geometrics, and operating traffic characteristics
which deviate from the norm, may act as the surrogate
determinants of safety. A few examples of adverse highway
features are: deficient geometric design, roadside obstacles,
slippery pavement surface conditions, and lack of access
control. Likewise, the examples of deviant traffic
characteristics are: traffic conflicts, erratic maneuvers,

short headways, extreme lateral placements, and digressing

44
speed distributions. For a detailed description of the use of
such surrogate measure of highway safety, some selected

studies [80 to 85] are included in the list of references.

2.7.1 Application of Surrogate Measures in Present Research

Based on the above cited examples, various highway
features and traffic characteristics were selected for this
dissertation, as representative surrogate measures. To
evaluate the prospective hazardousness of a highway section,
as represented by these surrogate measures, the following
three types of hazard—quantification procedures were employed.

1) Direct physical measurements,

2) Unobtrusive observations, and

3) Subjective ratings by an expert team.

Accordingly, these procedures resulted in the generation
of three types of corresponding data sets representing ambient
hazardousness of highway locations. A detailed description of

the data collection process is given in Chapter three.

2.8 USE OF TRAFFIC CHARACTERISTICS AND HIGHWAY DESIGN
FEATURES FOR ACCIDENT PREDICTION MODELING

The classical literature on highway safety revealed many
studies on the relationship between various highway features

and accident rates. These studies mostly examined the effect

45
of one or more highway element or design aspect on traffic
accidents. For example, pavement and shoulder width were
traditionally investigated for their effect on accident rates
by various researchers.

Blensly et al. [86] studied the relationship between
accident data and gravel shoulder widths in Oregon and found
insignificant effects at lower volumes. However, for volumes
between 3600 to 5500 ADT, there was a significant relationship
between accidents (total and PDO) and shoulder width. Stohner
[87] , considering the entire system of rural two—lane roads in
New York state, found that there existed a measurable
relationship between shoulder width and accidents rates, which
was especially true for property damage accidents. His finding
showed that the wider shoulder, within reasonable limits, were
associated with a lower accident rate.

Raff [88] studied the effect of a number of design
features on accident rates on rural highways. The factors of
interest included number of lanes, ADT, degree of curvature,
sight distance restrictions and traffic flow characteristics
at intersections. He concluded that traffic volumes and sharp
curves caused accidents and wide pavements and shoulders
increased safety on two lane curves. This finding
substantiated the causal relationship, reported by Blensly et
al. [86] , between personal injury accident frequency and paved
shoulder width for specific volumes.

Belmont [89] investigated the effect of shoulder width on

accidents on two lane tangents, using 1333 accident records

46
for 533 miles of roads in California, and obtained the

following regression equations:

(i=0. 4766+D.2202M

(With no restrain on S) (2.1)

i/Xzo. 1013+o. 01971M+D . 4514,522-

(For S < 6 ft.) (2.2)

JE=D . 1018+0. 005485M+D . 4514,43

(For 3 > 6 ft., and v > 5000) (2.3)

where,

Number of accidents;

Average daily traffic volume;
Length of the road section; and
Shoulder width.

mB<V

Musick [90] investigated the effect of pavement edge
marking on two lane rural state highways in Ohio. His findings
showed that pavement edge markings resulted in a significant
reduction (i.e., a net decrease of 37% at 0.02 level) in
fatality and injury causing accidents. This was specially true
for intersections, alleys and driveways whereas accidents
between access points showed no significant change. In the
context of pavement edge marking, Williston [91] studied
various locations in Connecticut and found that on 2-lane and
4-lane divided highways, the presence of a painted line along
the outer edge of pavement effected the lateral position of
vehicles and the most significant change in position occurred

during darkness.

47

Taylor et a1. [92] analyzed Ohio’s curve delineation
program and tested the effectiveness of roadside delineators
in the presence of pavement edge lines. Their before-and-after
study, using 557 test and 357 control sections, showed that
Ohio's delineation program provided significant reduction in
accidents. However, the use of additional parameters was
suggested to increase program efficiency.

Agent. et .al. [93] studied ‘the :relationship .between
roadway geometrics and accidents for various types of highways
in Kentucky. Information on accident severity, road surface
conditions, light conditions, road characteristics, and type
of traffic control were included in the study. Their findings
showed that four-lane undivided highways had the highest
average accident rate, while parkways (toll roads) had the
lowest rate. Single vehicle accidents, those involving
pedestrians, and accidents which occurred on curves had the
highest severity index.

The effects of gradient and curvature on accidents on the
London-Birmingham motorway were studied by Hillier et a1.
[94]. They reported evidence of a definite trend in accident
rate with gradient on straight three-lane sections. On the
steeper up-gradients, the accident rate was found to be
higher.

Apart from. finding’ the significance of effects and
relationship between the highway features and accidents, the
literature offered a multitude of approaches to develop
mathematical equations that can be used to predict accidents

from such features. In this context, Schoppert [95]

48
represented equations, developed by the Oregon State Highway
Department, that could be used to predict accidents on rural
two-lane highways employing roadway elements such as ADT, lane
width, shoulder width, sight distance restrictions, driveways
and intersections. Based on a sample of 1400 miles of 2-lane,
2-way highways and three years of accident records in Oregon,
the following set of equations for western regions were

generated:

Pg; ADI 3000 - 3992:
A I 7.69 - 0.21 SH + 0.10 CDW + 0.030 SDR - 0.41 LA (2.4)

F9; 5Q: 4909 - 4922:

A I 8.51 - 0.58 SH + 0.23 CDW + 0.004 RDW (2.5)
EQI_ADT_§QQQ_:_§222=

A - 4.84 — 0.12 sa + 0.19 00w + 0.310 RDW (2.5)

:9; 521 ngg 7990:
A 3 3.75 - 0.24 SH + 0.26 RDW + 0.160 ADT (2.7)

For highways in Eastern Oregon and for ADT 3000 and over

the forecasting equation was:

A - 1.04 + 0.23 CDW + 0.11 RDW +0.08 INT + 0.12 SH (2.8)
Where,
A - Total non-intersectional accident experience for a one-mile
section during one-year period;
ADT 8 Average Daily Traffic divided by 100;.
CDW - Number of commercial driveways per mile;
INT - Number of intersections per mile;
LA - Lane width in feet;
RDW - Number of residential driveways per mile;
SDR - Percent of the one-mile section where sight distance was

restricted expressed as a whole number, (i.e., 10% s 10.0); and
Shoulder width in feet.

49

Dart and Mann [96] investigated the relationship of rural
highway geometry to accident rates in Louisiana. They
developed mathematical models for total accidents per 100 MVM
and used traffic volume and mix proportion, pavement and
shoulder width, cross-slope, alignment, and roadside friction
as independent variables. Regression analysis technique
revealed the following relationship with R2= 0.46 and F-ratios

significant at the 0.05 level.

y = 41.32 - 1.2311, - 0.5411, - 0.67x, + 0.03x,x2 + 0.03x,x,

where, - 0.0009x2x, + 0.0261(2):u - 0.1210!“ + 0.009x,x, (2.9)
y - Total Accidents per loo-MVM,

X, - Percentage of trucks,

x, - Traffic volume ratios,

x, - Lane width,

X, - Shoulder width.

x. - Cross slope

X, - Horizontal alignment, and

X" - Traffic conflicts.

Sparks [97] used ten independent variables representing
various highway geometrics and use characteristics and

developed the following equation correlating accident rates.

- 0.1l9x6 + 0.335x7 -O.295x8 (2.10)
where,
AR . Accident Rate
xl - Surface type index
x2 - Curvature index
x3 - Gradient index
x4 - Stopping sight distance index
x5 - Passing sight distance index
x6 - Hazard index
x7 - Surface condition index
x8 - Shoulder condition

The other two variables contributed nothing statistically

and were not included in the final equation. Since the

50
standard error of the equation was in excess of 5.00, it was
considered. practically’ worthless for predicting accident
rates. However, to provide better correlation with accident
rates, another approach was employed by aggregating the data
into a cumulative index. The revised relationship in terms of
the cumulative index yielded an improved significance and
statistical methods showed that 45% of the variance that
occurred in accident rates was attributable to the magnitude
of the adequacy index. This relationship was shown
graphically, as reproduced in Figure 2.6, and was represented

by the following equation:

Y = 1.604 - 0.0483 (X) (2.11)
where,

Y I Cumulative Adequacy Index, and
x I Accident Rate (MVM)

 

YI I504 - .0483 (X)

++++++

4.

CUMULATfiflEADEUWKN'WDEX
+ +

 

AOGDENTIUWECMIanmth‘hI)

 

 

 

Figure 2.6 Cueuletive Adequacy Index vs. Accident Rate [97] .

51
Kihlberg and Tharp [98] investigated the effects of
specific geometric features on accident rates. The studied
geometric features included number of lanes, access control,
median presence, curvature, gradient, ADT, and the presence of
intersections and structures. Accident prediction monographs
relating traffic volumes with accidents for various geometric
conditions were developed. Though the data explicitly
represented Florida, Connecticut and Ohio, the results may be
generalized, based on reasonably acceptable R2 values, for
similar ambient conditions elsewhere. Their principal findings
showed that:
1) Access control had the most significant accident
reducing effect;
2) The multi-vehicle accident rate increased with.ADT,
and single-vehicle accident rate decreased.
3) The presence of gradients, curves, intersections,

and structures increased accident rates.

2.8.1 Accident Prediction Modeling in Developing Countries

The relationship between accident rates and geometric
design standards in two developing countries (Kenya and
Jamaica) was investigated by Jacobs [99]. He developed
mathematical equations predicting the accident rate in terms
of several independent variables representing various highway

and traffic features. Stepwise multiple regression analysis

52

produced the following set of equations.

533;;KEEDEE3
y I 1.45 + 1.02X5 + 0.017X3 (at 5% level). (2.12)
y = 1.09 + 0.62x5 + 0.031x3 + 0.0003x4 + 0.062x2 (2.13)
(at 10% level).
132;;llaiuiisza=
y I 5.77 - 0.755 x1 + 0.275 x5 (at 5% level). (2.14)
No further factor was reported to enter at 10% level.
where,
y I Accident Rate per MVkm,
x1 I Road width (m),
x2 I Vertical curvature (m/km),
x3 I Horizontal curvature (degrees/km),
x4 I Surface irregularity (mm/km), and
x5 I Junctions per km.

In Greece, Frantzeskakis [100] studied the relationship
of traffic accidents to traffic characteristics (e.g., volume
and Level of Service), on two non-controlled access National
Highways. The accident analySis which were carried out per

kilometer resulted in the following equation.

 

I; x 10‘
R. - (2.15)

P; (v/c).

Where,

R. I Accident Rate for a LOS n.

A. I Number of accidents occurring under LOS n.,

l; I Number of hours each section examined operates under LOS n.

(v/c).:-.Average ratio of volume to capacity for the LOS n for all one

kilometer sections examined.

The analysis of accident rates for each kilometer of the
two sections was used to identify hazardous locations. The
study concluded that the operational characteristics created
by a certain volume of traffic significantly affected the

accident rates.

53
2.8.2 The Indexation Approach for the Identification of
Hazardous Locations

The literature indicated several examples [101-104] of
employing indexed values for the identification of hazardous
locations. These indices were, in some manner, related to the
ambient hazardousness and addressed the safety issue both at
micro and macro levels. For example, at the micro level,
Hazard Indices (HI) were developed for the identification and
ranking of hazardous highway locations by Taylor and Thompson
[101] in USA, by Gharaybeh [102] in Jordan, and by Al-Isa et
al. [103] in Saudi Arabia. At a macro level, an Accident
.Hazard Index (AHI) was developed by Shen [104] to measure
relative safety of 46 counties in South Carolina. These index
development studies employed one of the following three types
of hazard indicators: 1) purely accident based factors; 2) a
combination of accident and non- accident based factors; and
3) purely non-accident based safety surrogate factors.

The effectiveness of the Hazard Index (HI) approach is
supported by both theoretical and practical considerations. In
the HI approach, a sufficiency (or deficiency) evaluation
criteria is employed to rate the composite hazardousness of a
highway location. From a theoretical standpoint, this
composite hazardousness rating provides a reasonably accurate
prediction of future accident experience. On the contrary, any
single hazard might not possess such predictive
characteristics. This point is further elaborated in a review
of the study by Taylor and Thompson [101], presented in the
following pages.

54
The practical advantage of the HI approach lies in the
simplicity of computations as compared to the more complex
approach of accident modeling and forecasting. These studies
found that HI approach could be used to measure the relative
highway safety so that proper countermeasures could be

developed accordingly.

Taylor and Thompson [101] developed a hazardousness
rating formula (HRF) which provided a means for arriving at a
hazardous index for any suspect site. The formula incorporated
data inputs for both accident and non-accident based nine

indicators as listed below.

de sed s

1. Number of Accidents per Year
2. Accident Rate
3. Accident Severity

WW
4. Volume/Capacity Ratio

5. Sight Distance

6. Traffic Conflicts

7. Erratic Maneuvers

8. Driver Expectancy
9. Information System Deficiency

The general form of the HRF was as following

U%(IVhl .
HI I (2.16)
Wk
Where,

HI I Hazardousness Index for the site under study

Wl I Weighing factor for indicator i.

IV I Indicator Value (subjectively scaled values from 0 to 100) for
indicator i.

Each indicator was a measure of hazardousness. Some

55
indicators were considered more powerful than others and this
was reflected. in ‘the 1differing’ weights assigned to ‘the
individual indicators. An example of determination of HI
employing their formula is presented in Table 2.7.

Taylor and Thompson established that the composite
hazardousness rating provided a reasonably accurate prediction
of future accident experience, as compared to any single
hazard that might not possess such predictive characteristics.
In this context, they reported that restricted sight distance
was a definite factor in hazardousness at a given location but
analysis of sight distance restrictions in themselves did not

provide good estimates of future accident experience.

Table 2.7 Coeputation of Hazardousness Index [101].

 

Type: Rural Intersection

 

 

 

Indicator Data Units Indicator Weight Part
Value Value 8.1.

1. Number of Accidents 7.67 acc/year 59 .145 8.6
2. Accident Rate 2.47 acc/HEV 49 .199 9.8
3. Accident Severity 12850 dollars 70 .169 11.8
4. Vblume/Capacity Ratio 0.17 unitless 22 .073 1.6
5. Sight Distance Ratio 2.00 wt. avg 0 .066 0
6. Traffic Conflict -- conf/hr -- .053 0
7. Erratic Maneuvers -- E.H./hr -- .061 0
8. Driver Expectancy 2.19 wt. avg 37 .132 4.9
9. Info. System Deficiency 2.79 wt. avg 47 .102 4.8
1.000 41.5

 

Sum of Applicable Weights I 0.886 [*1

* (Do not include weights for indicators not used at this site).

H.I. I Sum of Partial H.I.s/Sum of Applicable Weights I 41.5/0.886 I 47
Relative Strength of Evaluation I 0.886 x 100 I 89%

 

Gharaybeh [102] developed a danger index (DI), by
aggregating four types of hazard indicators, for the

identification of accident-prone locations in Greater Amman,

56
Jordan. These four indicators were: accident frequency;
accident rate; a danger factor (DF), computed as a ratio of
accident rate to critical rate; and an equivalent total

accident number (ETAN), expressed by the following equation.

ETAN = aF + bJ + TAN (2.17)
in which,
E I Number of persons who died in at the site;
J = Number of persons injured at the site;
TAN I Total number of accidents at the site; and

”a" and ”b" are the calibration factors.

Al-Isa et al [103] developed a hazardousness index for
determining accident potential of urban intersections in Saudi
Arabia. Their models expressed various traffic conflicts as a
function of geometric and traffic variables, as represented in

Table 2.8.

Table 2.8 Regression.Nodels of Traffic Safety Study in Saudi.Arabia [103].

Variable Indicator/Transformation of fi i
Manmiuu Rflmtamuui
Side Swipe Conf. Conflicts
I-lnmmuuudanmnna
Intercept -0.243 -0.8323
IPSRD Perception Reaction Sight Distance 1.712 -
PRSD11 1/PRSD - 0.2231
AA1 Approach Volume - 4.531
LEFRIG Left and Right-turns in intersection 0.00231 55.06
DIYPE Type of intersection 0.346 0.959
qudllsujmuflunu
I 0.422 0.451
F(signif) 24.86 20.74
H-iflnsmflnusumz2&Ma
intercept 1.555 -0.4256
IEHCRSD Crossing Sight Distance 2.279 4.2395
AA Approach Volume - 47.679
CC Crossing Volt-e 9.11 10.297
SPEED Average Speed 0.0391 -
APPWIDTH4 Approach Width - 78.98
i§HIUJ£E£flunflfll
I 0.56 .

088
Flsignif) 11.91 47.88

57
Shen [104] developed an Accident Hazard Index (AHI) for
46 counties in South Carolina. The AHI was computed by
determining the arithmetic average of three accident rate

indices, assigning equal weights to each, as expressed by the

following formula.

I PAI + VAI + MAI l
(AHI)i = {l ’i (2.18)
3 i

where,

PAI I Population-based accident rate index;
VAI I Vehicle-based accident rate index; and
MAI I Mileage-based accident rate index.

The validity of the constructed indices was tested by
using a correlation analysis in which per capita economic loss
(PEL) resulting from accidents was used to determine the

adequacy of AHI. The per capita loss was defined as following.

Cnxh‘+ ‘3 xn" ‘3 xn
(PEL), a (2.19)
P:

in which,

(PEL),I Per capita economic loss in County i;
Cl I Estimated average cost per fatal accident;

C, I Estimated average cost per injury accident;
C, I Estimated average cost per PDO accident; and
P, I Population of county 1.

The terms x“ ; xn ; and x3 represented the average number of
corresponding type of accidents per year in the county 1.

58

2.9 RELEVANCE OF LITERATURE REVIEW TO CURRENT RESEARCH

The current research has its primary prospective
applicability in Pakistan, alcountry'placed.in'the category of
infancy of motorization [39]. The nation's highway safety
problems are not necessarily the same as those faced by the
developed countries today. As such, the state-of-the-art in
certain areas of technology may not be applicable to the
conditions in Pakistan. Therefore the material selected for
review, primarily, represented the era of early motorization
of the present developed countries.

Specifically, frequent citations were made to literature
from the USA and the UK reflecting the state of the art of
1960's and 70's. A synthesis was made with the results of
research work pertinent to developing countries, like Saudi
Arabia, Jordan, Turkey, Greece, Thailand, and Pakistan.
Material from the Overseas Unit of the TRRL has been of
particular utility in exposing safety problems in countries
like Jamaica, Ghana, Botswana, Malaysia and Hong-Kong. More
sophisticated safety analysis in terms of accident prediction
modeling were cited for Jamaica and Kenya.

The literature review identified three important
considerations in highway safety that have a direct bearing on
the present research.

1) Assignment of liability on any single causative

factor does not necessarily explain the traffic

crash phenomenon. Hence, resources should not be

59
concentrated on isolated correction of a single
cause, which traditionally has been found to be the
driver.

2) There are many other theories, in contrast to the
single liability approach, which offer a more
refined explanation of accident occurrence, and
theoretical models based on such postulates have
been validated by empirical estimation.

3 ) The importance of highway-related safety
improvement is consistently augmented, whether
inferences are made based on a crude liability
approach, or a refined conceptual framework
including exposure and risk-compensation is
employed.

As such, participation in a systematic program, oriented
toward highway-related safety improvements, was indicated as
a valid strategy for alleviating traffic accident problems in
Pakistan.

The literature review also revealed that a systematic
program, such as the system referred to as HSIP, could be made
functional in Pakistan. Specific references to literature were
made to identify various surrogate measures of highway safety,
I and their use in accident prediction. In addition to accident
forecasting, the development and use of a hazard index in
terms of a composite hazardousness for identification of
hazardous locations was demonstrated. This information
provided important guidelines for the experimental design, and

is extensively used in this research.

60

2.9.1 Transferability of the HSIP Technology

Over the past four decades, the motorized countries have
gradually attained considerable experience in administering
highway-related safety improvement programs. These programs
are the outcome of extensive research spanning several years
and stand tested for their technical soundness and economic
viability. This knowledge-cum-experience base could be of
immense benefit for the developing nations in evaluating their
highway safety needs, provided a compatibility of critical
parameters and transferability of relevant technical
procedures could be established.

To be specific, the US model of HSIP [71] is examined in
this dissertation for application in Pakistan. A detailed
overview of this strategy are described in Sections 2.6 and
2.7. Since the HSIP is purely a technical oriented approach,
a comparison of social, economic or demographic parameters
between US and Pakistan is not required. However, it will be
necessary to look into the technical aspects.

The contents of the three main components of the HSIP
are tabulated in Tables 2.9, 2.10, and 2.11. A cursory glance
at these tables reveals that "procedure" is the basic unit of
the program. Therefore, to qualify for technology transfer, it
is essential that, from all technical and administrative
standpoints, at least one procedure should be accomplishable
in Pakistan. The transferability status of the HSIP procedures

is indicated in column 2 of these tables.

61

Table 2.9 Procedures of Planning Component [71].

 

PLANNMWGCKNWHONENT

' 0 Process 1 ' CDtLECT AID IIIIIAII DATA

     

>
Procedure - Milepost Method ............... ...... ....... ........
Procedure 2 - Reference Point Method ....... ... ........ ... ........
Procedure 3 - Link Mode Method .......... ... ......................
Procedure 4 - Coordinate Method ..................................
Procedure 5 - LORAM-C Method ..... ................................

> Suggrggggg a - Coligg; and Maingiig Aggjgggt Qgtg
Proc re 1 - F1 e of Acci t Reports by Location ...............
Procedure 2 - Spot Maps .......................... ......... . ......

Procedure 3 - Systenwide Computerization of Accident Data ........

> - 0 let Maintain Traffic t
Procedure - Routine Manual Traf IC Counts ......................
Procedure 2 - Use of Mechanical/Electronic Traffic Count Devices ..
Procedure 3 - Permanent Count Stations ...........................
Procedure 4 - Maintenance of Traffic Data on Maps of Files .......
Procedure 5 - Systamwide Computerization of Traffic Data .... .....

. > 4 - o and Maintai Hi hwa D t

Procedure 1 - Systemwide Manual Collect on of highway Data .......
Procedure 2 - Photologging and Videologging . .............. . ......
Procedure 3 - Maintenance of Highway Data on Maps of Files .......
Procedure 4 - Systemwide Computerization of highway Data ... ......

0 Process 2 ' IDEITIFY IIZAIDOUS LDCIJIGIB AID ELEIEITS

Procedure 1 - Frequency Method .................. .................
Procedure 2 - Accident Rate Method ................. ..... .........
Procedure 3 - Frequency Rate Method ..............................

Procedure 4 - Rate Duality Control Method ........................
Procedure 5 - Accident Severity Method ...........................
PM. 6 - ".1.” lmx ".th“ ......OOIOOIOOOIIOO0.00.00.00.00
Procedure 7 - hazardous Roadway Features Inventory ...............

O PfDCUII 3 ' CDIDUCT EISIIEEIIIG STUDIES

> - ll

‘ Procedure 01-05 - Accident Studies ...............................
Procedure 06-14 - Traffic Studies . ........ . ..... .... ......... ....
Procedure 15-20 - Environmental Studies ..........................
Procedure 21-24 - Special Studies ............. ..... ..............

>mmhkfimmmwn

' Procedure - Accident Pattern Tab es ............... ...... .. .....
Procedure 2 - Fault Tree Analysis ................ ............. ...
Procedure 3 - Multi-disciplinary Investigation Team ... ...........

’ 5MEEI2ES1l.Z.:.2£!!i£E_££2£§g£i§2
Procedure 1 - Cost E ect veness method ..........................

Procedure 2 ~ Benefit to Cost Ratio Method ......................

Procedure 3 - Rate-of-Return Method ..............................
Procedure 4 ~ Time-of-Return Method ..............................
Procedure 5 - Met Benefit Method ...... .................... .......

0 Process 4 - ESIADLISI PROJECT PRIORITIES

Procedure 1 - Project Development Ranking ........................
Procedure 2 - Incremental Benefit to Cost Ratio .. ....... ... ......
Procedure 3 - Dynmnic Programing .......... ............ ..........
Procedure 4 - Integer Programming ................................

Transferable.

Mot Transferable.

Transferable with certain limitations.
Mot Transferable without modifications.

*(')
'(*)

 

TRAMIFEBAIIJTY
ITATUO

I0000

0000

I
A

0
v

0

+
4‘? . 4 + o .
V

00000

0000

 

62

Table 2.10 Procedures of Implementation Component [71].

 

IMPLEMENTATION COMPONENT WW

oanmnil-SUEMAEAI>"Pumansnnnvimnmmmmtpmnans ----------------------

 

> - Pro'
Procemre 1 - Cantt Charts ....................................... f
l Procechre 2 - Progru Evaluation and Review Technique (PERT) ..... f
Procedure 3 - Critical Path Method (CPM) ......................... *
Procechre 4 - Multiproject Schemling System ..................... *

 

 

 

 

 

Table 2.11 Procedures of Evaluation Component [71].

 

EVALUATION COMPONENT

 

O m ‘l - DETEIII‘ TE EFFECT N MIGHT SAFETY [WE

Procedare 1 - Perform Accident Based Evaluations ................
Procedue 2 - Perform Mon-Accident Based Evaluations ............
Procedure 3 - Perform Progr- Evaluation ........................
Procemre 4 - Perform Adinistrative Evaluation .................

 

 

+ I Transferable.
- I Mot Transferable.

2.9.2 Technical Gap in Transferability

The information displayed in Tables 2.9 through 2.11
reveal that "identification of hazardous location using
accident data" (i.e., Process 2 of the Planning Component) is
the only process which lacks transferability in the entire
HSIP. The hazardous roadway features inventories could be

used as an alternative to serve the purpose. However, it would

63
be implicitly desirable to validate the technical soundness of

this alternate approach.
2.9.3 Suggested Method to Bridge the Transferability Gap

As mentioned above, the proposed alternative method of
identification of hazardous location needs to be validated
prior to its application. This could be accomplished by
quantifying the hazardousness represented by the hazardous
roadway features inventories, and then correlating it with
accident records. The specifications of the procedures devised
for quantification of ambient hazardousness are specified in
the Experimental Design (Chapter 3).

Since a reliable accident data base is not likely to be
readily available, this may be developed from police records.
In case a significant correlation is indicated between the two
entities (i.e., hazardousness and accidents), the former may
be adopted as a surrogate in identical environmental
situations.

In this dissertation this course is specifically adopted
to bridge the transferability gap and constitutes the basis .of
the research. In the experimental design, the hazardous
roadway features are identified using an inventoried checklist
developed on ‘the basis of information. available in ‘the
literature and in synthesis with indigenous hazard conditions.
The ambient hazards include detrimental highway elements,

adverse geometric design and pavement deficiencies, and

64
inadequate control of access.

In the perspective of technology transfer, it also became
apparent that accident files need to be constructed for
testing the highway hazard data against the accident records.
This was accomplished by retrieving accident reports for
selected highway segments from police records. The retrieved
information was further authenticated using linear plans to
cross check the police narrative or sketched description of

the accident location.

2.10 FORMULATION OP EYPOTEESIB

The mathematical models reviewed in Section 2.8
demonstrated a high degree of correlation between the
accidents and the explanatory variables as indicated by the
various determinants of statistical significance (i.e., RE
F-ratios and probability levels). This significance led to the
practical application of these models for the prediction of
accidents.

For example, Schoppert [95] asserted that the equations
presented by him could be used to predict total accidents on
one-mile sections of rural two lane highways with similar
characteristics in Oregon. Kihlberg and Tharp [98] explicitly
demonstrated the application of their developed.monographs to
predict accidents for various geometric conditions. As a

corollary to this, they also demonstrated how various

65
geometric designs could be evaluated for safety.

As an analogy to these findings, it may be postulated
that the accident potential of a highway location in Pakistan
may' be assessed ‘using’ appropriate surrogate 'measures of
hazardousness. It is, therefore, hypothesized that the
tangible hazardousness of a rural highway section in Pakistan
is representative of its accidents potential. The precise

specification of the null hypothesis is as following:

' There is no relationship between the ambient hazards
and the accident potential of'a.rura1 highway section in

Pakistan.'

To ‘test. this Ihypothesis, an. experimental design is

presented in the following chapter.

CHAPTER 3

11‘! 1b,. US 111 3.0.5.0 0 ‘5” A! '| !_NT‘._,_ I__, i.

3.1 TEE VARIABLES OF INTEREST

The topic of highway safety may be viewed from a variety
of perspectives representing several aspects of individual and
communal interests. In this dissertation, this topic is purely
dealt with from an engineering standpoint, and in specific,
the technological aspect of highway-related safety improve-
ments are addressed.

To rationalize the course of research, and to select the
crucial variables, a four-level screening procedure was
employed. At each level, a group of entities relevant to
highway safety were screened by backward elimination. This
means that all the group-components were considered for their
appropriateness prior to retaining the pivotal ones. A
detailed description of the four levels of elimination is

presented in the following pages.

66

67
W

In the first level of elimination, the three conventional
accident-causal factors (i.e., road, user, and vehicle) were
considered. The objective was to validate the relevancy of
highway-related safety improvement programs in the given
situation.

The improvements aimed at the vehicle and the user were
filtered on three important considerations. First, the
country's indigenous industrial research and production base
was not likely to effect automotive safety improvements in the
near-future. Second, the literature indicated that there was
a low probability of a change in user behavior in a short time
period, even by education or enforcement. Third, any strategy
oriented toward the correction of the highway system would
interactively address the other two factors in achieving

safety.

W

In the second level of elimination, various procedures
for identifying hazardous locations were reviewed, since this
operation is the foremost step in the design and
implementation of a highway related safety improvement
program. The specific literature on implementation of a HSIP
[71] suggested seven methods of identifying hazardous
locations as listed in Table 3.1.

The first six methods require application of accident
data in some form. The last method was considered most

appropriate for use in Pakistan. The choice was based on the

68

Table 3.1 Second Order Elimination - Procedures for Identifying
Hazardous Locations.

 

   
 

STATUS

PROCEDURES SELECTION I

   

 

Frequency Method

 

Accident Rate Method

 

 

 

Rate Quality Control Method

 

5 Accident Severity Method

 

l 0
2 0
3 Frequency Rate Method 0
I 4 0
l o

6 Hazard Index Method 0-1 I

 

 

 

 

Hazardous Roadway Inventory Method

0 I Can not be selected due to technological limitations.
0-1 I Can be selected with certain limitations.
1 I Can be selected without any limitation.

fact that a reliable and accessible highway accident data base
was not likely to be available in the country. However,
preceding any generalization, this approach needs to be

further authenticated with the help of selected accident data.

Winslow
In the third level of elimination, various highway

hazards were examined to select those to be used in this
study. Literature [105] indicated that there was no
universally accepted definition of a hazardous location or an
element. In fact, highway safety personnel recognize various
types of highway locations and features which, if not

corrected, are likely to be associated with high accident

69

frequency or severity. For example, roadway sections with
closely located fixed obstacles and with low skid resistance
properties are considered to have an increased potential for
accidents. Similarlyy a. drop-off‘ of several inches from
pavement-edge to shoulder is considered a potential hazard.
The AASHTO [106] established a generally accepted group of
hazardous road elements identifying specific types of hazards.
The FHWA [107] also developed a similar classification of
hazardous elements. The aggregated hazardous situations, based
on the cited literature and those commonly found on rural
trunk lines in Pakistan, are summarized in Table 3.2.

This table eventually served as a checklist to determine
the number of hazards and to spot their physical location on
the ground. According to this checklist, 22 types of hazards
were found to be present in the experimental site, that are
marked either as "P" or "PNC" in the table. However, only 19
(out of the 22 present) hazards were considered, that are
marked as "P" in the table. The three hazards which were
present but not considered are marked.as "PNC". These included
inconsistent use of signs and traffic control devices, and
poor illumination conditions.

The experimental design is based on the outcome of this
screening process. Twelve Measures of Hazard (MOH) were
developed for the quantification of hazardousness. A.detailed
description of these MOH and the results of their operation is

given in Sections 3.2.2 and 3.2.3.

70

Table 3.2 Third Order Elimination - Inventory of Hazardous
Highway Features.

 

P I Present
A I Absent
PNC I Present (but Not Considered)

71

rde l a o

The fourth, and final, level of elimination constitutes
the principal part of this research. In this process the
independent variables , representing the quantif ied
hazardousness, are tested for their correlation with accident
data extracted from the police records. This operation results
in the development of accident prediction models. The entire
process is presented in Chapter 4 which deals with the

analysis of the research data.

3.2 THE DESIGN OF EXPERIMENTS

Chapter 2 concluded with the formulation of a hypothesis
that there are no identifiable hazards on rural highway
sections in Pakistan which are the determinants of accident
potential. To test this hypothesis, it was intended to examine
the statistical significance of any relationship between the
quantified hazardousness and the accident data. Multivariate
regression analysis were performed to test correlations with
a control on multicollinearity, since the hazard data was
represented by multiple descriptors. The details of the
proposed analysis are covered in Section 3.2.4. The
experimental plan essentially requires information on two
types of data: 1) The Hazard Data, quantified from hazardous
highway features and adverse operating traffic
characteristics; and 2) The Accident Data, extracted from

archival accident records.

72

The primary objective of the experiment was to determine

if accident surrogates could be confidently used as a

substitute for actual accident histories which are unavailable

or deemed unreliable. This substitution is a requirement to

bridge an obvious gap in transferability of HSIP technology

for implementation in Pakistan. The following pages describe

the salient features of the experimental plan.

3.2.1 The Ambient Highway Hazards

As a result of third order elimination (see page 68), the

following 19 types of hazards present on the experimental site

were considered for this study.

1)
2)
3)
4)
5)
5)
7)
8)
9)
10)
11)
12)
13)
14)
15)

Narrow bridges, and bridge approaches.

Guardrail deficiencies.

Fixed objects (roadside obstacles), trees etc.
Sight distance restrictions (e.g., vegetation).
Utility and signal poles.

Animal and pedestrian crossing.

Narrow shoulders and shoulder drop-offs.

Smooth, slippery pavements.

Culverts, headwalls, drainage facilities.

Steep side slopes, high fills, ditches.

Deficient intersections (e.g., blind approaches).
Inadequate or worn pavement markings or delineation.
Substandard geometry.

Deficient bridge rail and connecting guardrail.

Rough pavement surface (e.g., potholes etc.).

73

16) Narrow lane and pavement.

17) Barriers, fences and stone walls.

18) Obsolete geometric design.

19) Buildings.

Some photographs of locations within the experimental
site showing the ambient highway hazards are presented in
Appendix B.

The next logical step in the experimental design was to
develop some appropriate techniques to quantify these hazards.
As such, the following three types of procedures were employed
for the quantification of hazardousness: 1) Use of design
standard deficiencies as a measure of the hazard; 2) Use of
drivers erratic maneuvers and traffic conflicts as a measure
of the hazard; and 3) Use of an expert team for subjective
rating of the hazard.

Application of these quantification procedures resulted
in three type of corresponding data sets: on-ground
measurements; unobtrusive observational counts; and 3) scaler

rating numbers.

3.2.2 Development of Measures of Hazard (MOH)

Twelve MOH were carefully developed to measure the 19
types of hazards utilizing one (of the three) quantifying
techniques. These MOH depicted the typical features of a rural
trunk line highway and operating traffic characteristics in

Pakistan, and ultimately acted as the independent variables

74
for mathematical modeling. These MOH and their intended use

are presented in Table 3.3.

Teble 3.3 The MOH Development Plan and their Use Characteristics.

 

Quantification Location-wise
Technique

 

[ACCESS]
[GDRAIL]
[PWIDTH] Measurement Non-intersection
[SWIDTH]
[PMARKS]

 

[INTENT]
[INTSPD] Intersection
[INTCOL] Observation

 

[SPDCHG]
[LANCHG] Non-intersection

 

[FWXWD]
[SIDEOB]

 

 

3.2.3 MOH Specifications and On-Ground Application

The application of these MOH resulted in generation of
both continuous and discrete numbers, depending upon the
nature of the hazard, and provided a numerical basis for the
descriptive statistics of the composite hazardousness of a
location. The specifications of the developed MOH and their
application method for hazard quantification are described in
the following pages.

1. MOH [ACCESS]: This MOH was employed to evaluate the

hazardousness due to inappropriate access control. This

75

deficiency not only causes intrusion of pedestrian and animals
on the road, but also encourages bus drivers to off-load
passengers upon demanded. Any section of road naturally or
otherwise unprotected against such accessibility was measured.

Three types of lateral access conditions were included in
the study: 1) linear ribbon development [RIBBON]; 2) specific
lateral paths [SPATHS]; and 3) median openings [MEDOPN] in the
case of 4-lane sections. A presence of any of these condition
was assumed a potential hazard and was measured in meters per

kilometer.

2. MOH [GDRAIL]: Installation of guardrail often has a
contradictory effect on overall safety i.e., it reduces the
number of fatalities but increases the number of injuries and
accidents [108, 109]. Though guardrail by itself may not
precisely avert an impending accident situation, the
evaluation of hazardousness due to the absence of guardrail on
warranted highway sections was considered appropriate for this
study. Accordingly, this MOH was used to evaluate the
hazardousness due to the absence of guardrail at bridge
approaches and embankments of three meters or higher. This

hazard was measured in meters per kilometer.

3. MOH [PWIDTH] : For evaluating the prospective
hazardousness caused by deficient pavement width and narrow
bridges, sections having a lateral width below 3.65 meters

(12 ft.) per lane were included in the analysis. It is

76
pertinent to point out here that some sections of two-lane,
two-way highway and either carriageway of a 4-lane divided
highway had pavement widths more than 7.30 meter (24 ft.)
because of stage construction or intermittent road widening
programs. Though not deficient in width, these sections were
considered to pose an increased threat to highway safety by
causing non-channelization and overtaking potential due to the
available extra width. Such hazardous sections were also
covered by this MOH and included in the study. The variation
of width from the specified was taken as a hyperbolic (second
degree) function to incorporate the effect of negative
deviations and to ignore minor deviations in evaluating

hazardousness as shown in Figure 3.1.

4. MOH [SNIDTH]: This MOH was used to determine
prospective hazardousness due to the shoulder deficiency. The
standard shoulder ‘width for ‘various types of roads are
specified by AASHTO [25] depending upon the traffic volumes
and the type of road. For the purpose of this study, a
shoulder width less than 3.0 meter was counted as deficient.
This deficiency, on the average, was considered a measure of

hazard and was recorded in meters per kilometer.

S. MOH [PMARES]: Absence of longitudinal pavement
markings poses a hazard in terms of non channelization,
improper overtaking and off-tracking of vehicles. Evaluation

of this hazard was an important factor because dangerous

77

 

 

 

 

 

 

 

 

 

 

 

 

 

0.5 n‘_\
a" \
A -(
m
L
i o2
E o \\
>
O
o H
-o.2
q
-o.4 \
3 3.2 3.4 3.6 3.1: 4 4.2

 

Povernani Widih (meiars)

-><- 2nd degree * linea'

 

 

 

Figure 3.1 [PWIDTH] - A Hyperbolic Function of Pavement Width variation
tnmithelnmmumd.

overtaking was earlier identified as a principal cause of
fatal accidents in Pakistan [40]. The hazardousness was
measured in kilometers per kilometer both for the longitudinal

sections and intersection areas.

6. MOH [INTENT]: This MOH, along with the next two i.e.,
MOH [INTSPD] and MOH [INTCOL] were simultaneously used to
evaluate hazardousness at an intersection with permanent

obstructed visibility and adverse geometry. The hazardousness

78
was evaluated by unobtrusively recording the number of non
stopping vehicle entries in the intersection per unit time and
in proportion to the operating volume. Thus the devised unit

for the measure of hazard was a ratio of vehicles per hour.

7. MOH [INTSPD]: This MOH was conceived as an indicator
of the hazardousness due to adverse geometry of the
intersection area and approaches. In developing this measure
it was assumed that the vehicle driver who entered an
intersection area without stopping would either try to quickly
stop or clear off the intersection in haste, to avoid an
impending collision. A change in speed exceeding 15 Kmph after
a vehicle's entry in the intersection area was defined as an
"abrupt change" for this study. The number of such speed
changes per unit time, and in proportion to operating traffic
volume, were unobtrusively recorded and used as a measure of

hazard.

8. MOH [INTCOL]: This MOH is an indicator of the degree
of adverse intersection geometry as evidenced by a near
collision. The evidence of having two or more vehicles
reasonably close (defined as one meter or less for this study)
and trying to avoid a collision were unobtrusively recorded
per unit time and used as a measure of hazard.

The MOH [INTENT], MOH [INTSPD] , and MOH [INTCOL] were
specific to intersections. Based on the values of these three

MOH, an aggregated variable [INTSEC] was employed in the

79
analyses to represent the composite intersection
hazardousness. Besides evaluating the independent effect of
these MOH, their interdependent effect were also included in

the analysis.

9. MOH [SPDCHG] : This MOH was a measure of roadway hazard
in terms of spatially located skid-zones, corrugations, humps,
ditches and pot-holes. The number of abrupt speed reduction
events to avoid such roadway hazards (in a distance of 30
meters or less) were recorded by a front-seat passenger using
the following rated weights.

1 I Low hazard (for speed reduction up to 15 kmph).

2 I Medium.hazard (for speed reduction up to 30 kmph).

3 I High hazard (for speed.reduction.more than 30 kmph).

10. MOH [LANCHG]: This MOH was used as a conjugate to the
previous one and was devised with the assumption that, when
confronted with an impending accident situation due to adverse
lane conditions, a driver may suddenly change his lane. Only
road hazard based incidents were recorded. The number of such
actions were recorded by a front-seat passenger who observed
lane changes in.a¢distance of 30 meters or less. The following
weights were used for hazard evaluation.

1 I Low hazard (1/4 lane change).

2 I Medium hazard (1/2 lane change).

3 = High hazard (full lane change).

80
The MOH [SPDCHG] and MOH [LANCHG] were specific to
isolated lane hazards. By aggregating the values of these two
MOH, a variable [ISLAND] was employed in the analyses to
represent the total hazardousness caused by isolated lane

defects like corrugations, humps, and pot-holes.

11. MOH [PVCOND]: This MOH represented the road surface
and riding-quality hazards. The recommended procedures [110]
were adopted for pavement condition rating in terms of surface
cracking, rutting, roughness and edge drop-offs. Each
kilometer of the test section was rated on a scale of 0 (least
hazardous) to 100 by a two member expert team comprised of

experienced highway engineers.

12. MOH [SIDEOB]: Utilizing this MOH, the road-side
obstruction caused by fixed objects which were liable to
impose increased accident severity (e.g. road-side trees,
poles etc.) were measured in number of their occurrence. The
employed MOH provided a relatively simple approach to evaluate
the road-side obstructions by the observers of an expert team.
The distance of the road-side obstacle from the pavement was
not prefixed and the rating-judgement alone was the sole
criterion to determine the hazardousness caused by such
obstructions.

The employed MOH and their measurement attributes are

summarized in Table 3.4.

£11

Table 3.4 The Hazard Measurement System.

a .
iscnnxous

 

 

Ilse of Desim Stmzbrd Deficiency as a Measure of Hazard

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1 IDH [ACCESS] meters/km linear measurement taping

2 IR)" [GDRAIL] meters/km linear measurement taping

3 MOH [PMIDTH] meters/km linear measurement taping

4 MOH [SMIDTH] meters/km linear measurement taping

5 MOH [PMARKS] lune/km linear measurement taping

Ilse of Traffic Conflicts and Erratic Mmeuvers as a Measire of Hazard

6 IBM [IHTEHT] S/hr. comting mobtrusive field observations
I: :23; MOH [INTSPD] I/hr. comting unobtrusive field observations

8 situation 10H [IHTCDL] S/hr. comting mobtrusive field observations

9 NH [SPDCHG] Slkm. comting co-driver's observations

10 IDH [LANCHG] #lkm. comting co-driver's observations

lbeofmiExpertTe-‘sabjectiveRatimasaMaasu'eoflazard

[11 IBM [PVCGIDJ #Ikm. slbjective rating I expert team's judgement

 

   

   

 

NH [SIDEml slbjective GMTIM

 

l expert tema's judgement

 

The printouts of the hazard data and the numerical values
for the quantified hazardousness employing the above described

MOH are presented as Appendix C.

3.2.4 The Applied Procedures for Data Analysis

Multivariate linear regression analyses were performed to
test the null hypothesis of no relationship between the
hazardousness and accidents. This is one of the most versatile

data analysis techniques available for model building, and has

82
a demonstrated applicability [17-18, 96-97, 111-115] in safety
research. The proposed accident prediction model can be

expressed as:
Af . Bo + 8px“ + BZ'XZI 'I' ...... + Bp'xpi + 9;

Where, A, is the accident frequency on a particular
kilometer of the highway section, expressed in #lyear. 14,,
represents the value of the pth independent variable for the
kilometer i. The B coefficients are the unknown parameters and
were determined as the result of regression analysis. The e,
terms are assumed as independent random variables that are
normally distributed with mean 0 and constant variance 0’. It
is also assumed that in the proposed model, the dependent
variable (accident frequency) has a normal distribution for
every combination of the values of the MOH (independent
variables). The following statistic was used to test the
hypotheses.

3:
t-

881

where, B1 is the slope of the regression line, and S is the
standard error of Bl. The distribution of the statistic when
the hypothesis of no relationship is true would be the
Student's-t distribution with N-2 degrees of freedom.

These analyses were made using a commercial software
package, Statistical Package for Social Sciences (SPSS). The
t statistic and their two-tailed observed significance levels

83
are displayed in the standard output results of SPSS
regression routines.

Multicollinearity Diagnostics: Since the hazard data is
represented by a variety of descriptor independent variables
(i.e., MOH), the regression analyses were performed with a
control on multicollinearity. This terminology refers to a
situation in which there is a likelihood of high correlation
between the independent variables. The problem with such a
situation is that the different MOH would provide similar or
interrelated information about the hazardousness of a
location. A check on this phenomenon enables the analyst to
retain the pivotal MOH. Any decision on the retention of the
pivotal MOH was further subjected to practical considerations
of variable acquisition. Two measures of collinearity,
Tolerance of variable (TOL), and variance Inflation Factor
(VIF) were employed as diagnostic tools.

The extended capabilities of the SPSS regression routine
includes computations for the TOL and VIP values. In addition
to these statistics, eigenvalues and condition indexes are
also useful tools to examine collinearity of a data matrix. A
specific application of these proposed statistical procedures

is made in Chapter 4 which deals with the analysis of data.

3.3 TEE RESEARCH DATA

The research data is distinctively divided in two

categories: 1) The hazard data (independent variables); and

84
2) The accident data (dependent variable). In developing the
experimental design, it was planned that the MOH should act as
the independent variables. The accident data were primarily
used in frequency version rather than rate format. The details

of these data are presented in the following subsections.

3.3.1 The Hazard Data

The requisite data on highway hazards were collected
according to the experimental plan as described in Section
3.2.3. The following information, inventories and data were
initially required for the study.

1) Information on district boundary, kilometer posts,

and topographical plans of the sample sites.

2) Inventories of intersections, curves, and bridges

with their geometric details.

3) Information on control of access, ribbon

development, and median openings.

4) Embankment height data.

5) Pavement width data.

6) Shoulder width data.

7) Pavement marking details.

8) Pavement condition rating data.

9) Road-side obstructions data.

10) Traffic volume counts.

These initial data on physical roadway features and the

85
topographical site plans were furnished by the National
Highway Authority, Ministry of Communications, Islamabad; and
their Consultants, M/S Kampsax International A/S. The
furnished information and plans were further checked at site

to incorporate any changes due to construction and land-use

development.

3.3.2 The Accident Data

The accident data were not readily accessible in terms of
a computerized data base. Therefore accident data files were
created by retrieving information from the police records
which were mostly in narrative form, occasionally formatted or
sketched. The following eight police stations (PS) provided
accident data for the three year period January 1988 through

December 1990.

221M121: MW
1) PS Gujar Khan. 1481 - 1496
2) PS Mandra. 1497 - 1510
3) PS Riwat. 1511 1520
4) PS Sehala. 1521 1530
5) PS Rawalpindi (CL). 1531 1546
6) PS Tarnol. 1547 1564
7) PS Taxila. 1565 1570
8) PS Wah. 1571 1583

The accident locations as described in the police records

86
were cross-checked with the help of linear highway plans to
authenticate the narrative report, and to identify their
correct position, The printoutS»of the accident data files are

presented as Appendix D.

3.4 EXPERIMENTAL SITE

The National Highway (N-5) in the Rawalpindi District,
excluding the municipal limits of urbanized areas, was
selected as the experimental site for collecting the required
road hazard data. The sample site constituted a rural stretch
of 86 kilometers, beginning from Kilometer 1481 and ending at
Kilometer 1583. This site excluded the 17 kilometers of the
urbanized sections of highway passing though the cities of
Gujar Khan and.Rawalpindi. The first 52 kilometers of the site
were two-lane, two-way, while the remaining 34 kilometers were
four-lane divided highway. The geographical location of the
experimental site, and its diagrammatic magnification are
shown in Figure 3.2 and Figure 3.3 respectively.

The selected site had the following intrinsic qualities
to the advantage of the experimental plan.

a W. The Rawalpindi District
includes the city of Rawalpindi - a Divisional Headquarter,
and the territories of the Federal Capital, Islamabad. The
national importance of Rawalpindi together with the diplomatic
significance of Islamabad, has resulted in a reasonably

efficient highway patrol system in the District and especially

87

 

  
  

n t
bk" 0 (Nufll‘lioio If

  

c )3
"Psalm
2 6’1."

  

Fmi° “G“ H

film

i

‘ (’7“""06I|zwn"‘fm m
35p

pl-t
medO

' ‘18:"?

;"“
. nor-3‘" -
WEE/i

    

     

COL:
"Manama”

:0

 
  
  

0 mutul 0:”

“fie.

.‘ “a“
(an J 45.5%“... n
:- ..u

  

   
 
   
  

r l “2?
Misha. . u/l
$51.34 ...:

CO"

44: n in!) .-
s'

\_\\\a.5m°fl:‘¢w 0° "8‘1'0

aha. 0""?

.20: hwfl'flfi "Ni-",1!
O

K
gum-",5- mm .
emsrdheo “‘3

Oman—....

 

  

 

India

Subcontinent
and Afghanistan

CONIC PROJECTION
SCALE 0F MILES

International Boundar aria

Provincial and State Boundaries _ . _
Canals ...........................
Scale 1. 14,500,000

to
Rggmgring
o no) no no
Capitals 01 Countries...
Provincial and State Capitals.

an no

......................... R
..e

 

 

 

 

 

 

Figure 3.2 The Geographical Location of the Experimental Site

(Map Source [70]).

NATIONAL HIGHWAY (N-SI
DISTRICT RAWALPINDI. PAKISTAN.

II ‘ ' DISTRICT ASSOTABAD

 

  
  
 

DISTRICT ATTOCK

\3’
“”1492 Gujar khan
Knn488 “

‘Knfl480

DISTRICT JI'IELUM

 

4-Lane Divided Sections

2-Lane, 2-May Sections
Urban Limits

(not included in the study)

 

 

I<__IIII.__>I

 

Figure 3.3 The Magnified Diagram of the Experimental Site.

=93!

89
on N-5. As a result of this proficiency, the accident
reporting rate for the test site was more consistent and
comprehensive than for less significant areas of the country.

0 MMWMUM- As
mentioned above, the experimental section of N-5 is comprised
of both 2-1ane and 4-lane divided highways. This distinctive
situation provided a site to investigate the hazard-accident
relationship characteristics for two basic categories of
highway system.

0 W. The N-5 is the most
important trade and passenger route in the country, and
therefore operates a traffic mix representative of both
private and commercial vehicles. The study was, therefore, not
specific to any single important mode of highway
transportation system.

e W. The terrain of the
experimental site ranged from flat to mild-rolling, and was
therefore considered a reasonable representative of the
topographic features of many other sections of N—5 and other

important arterials in the country.

3.5 EXECUTION OF THE EXPERIMENT-

The highway hazard data were collected on the
experimental site using the twelve MOH developed for this

study. These data were collected during January to March 1990

90
under the supervision of the author by the technical staff of
the National Transport Research Centre (NTRC), Islamabad. The
data collection procedures, applicable to the experimental
design, were explained to the staff and were demonstrated on
site before the start of a particular operation.

A part of the experimental plan also required use of a
front seat passenger for recording observations on driving
maneuvers to collect data concerning MOH [SPDCHG] and MOH
[LANCHG]. A Ford.Transit 1989 van with.necessary equipment was
provided by the NTRC to accomplish the task. The component of
subjective rating in the experimental plan was covered by an
expert team of two civil engineers having their Masters degree
earned in the united States. Both the team-members were the
employees of the Centre and had their professional expertise
in highway engineering.

A technical report was published by the NTRC based on the
data collected on highway hazards [12] . This report only dealt
with the description of the hazards that were physically
observed at the experimental site (i.e., in terms of their
magnitude and frequency of occurrence), and did not address
the location’s relationship with accidents. Subsequently
accident data were extracted from the police records in the
required format and are now being used in this dissertation to

achieve the objective of the present research.

CHAPTER 4

DAT Y

4 . 1 THE ANALYTICAL APPROACH AND PRESENTATION

The foremost objective of this data analysis was to
determine the statistical significance of the presumed
relationship between the ambient hazards and accident history

of a highway section as represented by the following model.

Af = Bo + 13,.xli + 132-X21 + + 135,.xpi + ei (4.1)
Where,
Af = Accident frequency on a particular kilometer i of
the highway section, expressed in #lyear.
Xpi = The value of the pth hazard for the kilometer i.
B = Unknown coefficients (parameters) to be determined

as the result of regression analysis.

e. = Independent random ‘variables that were assumed
normally distributed with. mean 0 and constant
variance 0’.

The proposed model had its genesis in the theoretical

91

92

perception of accidents as a function of adverse highway

features, expressed as following.
Accidents = f(adverse highway features) (4 . 2)

For the jpresent research, a null hypothesis of no
relationship 'was assumed. Primarily, multivariate linear
regression analyses were performed to test the hypothesis.
Since it was assumed that the adverse highway features would
have a direct causal relationship with accidents, the
variables having a negative correlation with accidents were
considered to violate this assumption and were controlled.
Besides negativity, the regression analyses were performed
controlling for multi-collinearity since the hazardousness was
represented by a variety of independent variables.

The regression analyses, performed with these
specifications, resulted in the formulation of mathematical
models predicting the accident potential of a highway section
in 'terms. of crucial. non-accident. based 'variables. These
sections (analytical units) had a uniform length of one
kilometer each and included the intersections. The analytical
unit was not further segregated by the type of location since
accident records did not furnish information in terms of
intersection and non-intersection accidents that was required
for such discrimination. These analyses were primarily made

using the Statistical.Package.for’Social Sciences (SPSS), with

93
other supporting software packages used for graphics and
spreadsheet computations.

An analytical approach based on the following described
eight tasks was developed to accomplish the overall objectives
of data analysis. The results of implementing these tasks are
contained in Sections 4.2 (page 99) and 4.3 (page 124) for the

z-lane, 2-way and the 4-lane sections respectively.

. Task 1. g2nnute_neasriatize_fitatiatisa

The descriptive statistics of the variables was
comprised of the mean, standard deviation, and minimum
and maximum values. This procedure was important in three
ways: First, it provided the basic information on the
limits and variation of the variables. Second, it served
as a confirmatory tool to ascertain flawless transfer of
research data brought from Pakistan on micro computer
disks to the local operating system. Third, it
established the accuracy of the created SPSS system-files

in responding to data files.

0 Task 2- Defermia2.2rssneasz_niafributiga
The frequency distribution histograms were displayed

to examine the nature of occurrence of] hazards and
accidents on the experimental site. Though not an
ultimate measure, the frequency distributions furnished

important logical checks and were one of the criteria

94

used for variable selection in model building.

Task 3. De e e S e Co elations

These analyses were made to examine the type and the
order of any relationship between the dependent and the
independent variables. A significant and non-negative
relationship provided the necessary rationale to proceed
further with the analysis and test the postulate of a
hazard-accident relationship. The insignificant and
inverse relationships were noted as a point of concern to
be considered when selecting the final variables for
model building.

Moreover, the correlations also provided information
on the inter-relationship between the various independent
variables. The prominent interrelated independent
variables were further examined with multicollinearity
diagnostic tests to decide their importance in model
building. The term multicollinearity refers to a
situation in which the independent variables are
correlated and tend to provide similar or interrelated
information. A check on this phenomenon resulted in
identifying the statistical appropriateness of candidate
variables. The two diagnostic techniques employed were:
1) the eigenvalues analysis; and 2) the variance
inflation factor (VIF) method [116] . Their description is

covered in the following two tasks.

95

Task 4. C v e o i e

To examine the collinearity between the independent
variables, the eigenvalues of the scaled uncentered
cross-products matrix, and decomposition of regression
variance corresponding to these eigenvalues were
computed. There is an evidence of near dependency of
variables when there is a high proportion of the variance
of two or more coefficients associated with the same

eigenvalue [116,117] . The condition index is defined as:

Cond. Indexi = [eigenvalue]m / eigenvaluei]°'5 (4.3)

The presence of near-linear dependencies results in
small eigenvalues and, consequently, larger condition
indices. In these computations, the number of large
condition indices is the determinant of the near-

dependent number of cases.

Ink 5- MW
W

The literature [116] indicated that the tolerance of
a variable was another commonly used measure of
collinearity. The tolerance of a variable is defined as:

Tolerance (TOL) = (l-Rf) (4.4)

where Ri is the multiple correlation coefficient of the

96
ith independent variable when it is predicted from other
independent variables. The variance inflation factor
(VIF) is the reciprocal of the tolerance and, therefore,

for the.ith independent variable, it can.be expressed.as:

VIF = [1 / (1-R3IJ (4.5)

The extended capabilities of SPSS regression routine
produced computations for the eigenvalues, condition
indices, TOL and VIP values. These computations displayed
the required statistical characteristics of the variables
and provided a logical check on their appropriateness for

model building.

Task 6. e e a

Variablgg
The process of BACKWARD ELIMINATION in the SPSS

regression analysis routine was used for retaining the
significant variables. This process starts with all the
entered variables of the prospective multivariate model
and sequentially removes them. Two removal criteria are
employed: 1) the minimum F value (FOUT) that a variable
must have in order to remain in the equation; and 2) the
maximum F value (POUT) a variable can have.

All the independent variables were initially entered
into this process for the final selection. The system

default values of 2.71 for (FOUT), and 0.10 for (POUT)

97
respectively, were adopted in the analysis for the two
removal criteria.

Besides the BACKWARD ELIMINATION, the other two
available procedure: the FORWARD and the STEPWISE, were
additionally employed to authenticate the entire process
of regression analysis. The former procedure employed
entry of one variable at a time, while the later examined

each variable for entry or removal at each step.'

'1'“): 7- MW
meanest—Vang};

The accident data were primarily used as
frequencies. However, a noticeable variation in traffic
volume was reported [118] for different sections of the
experimental site. As such, accident rates were taken
into account and the effect of operating volume on the
hazard-accident relationship was investigated in

developing the final predictive models.

Tank 3. W

This task was comprised of constructing two separate
multivariate models for predicting the accident potential
of a highway section, corresponding to the two types of
highway section studied. The choice of variables was
essentially based on the outcome of the first seven

tasks, which. provided. a logical check for ‘variable

98

selection. In this final task, the null hypotheses of no

relationship between the quantified hazardousness and the

accident history of highway sections were tested.

Explicit inferences about the acceptance or rejection of

hypothesis were made employing the statistics of the

final accident prediction models. The analysis of
residuals and comparison of predicted accidents to actual
were covered in this task.

In addition to accident modeling using regression
analysis techniques, a hazard index (HI) was also developed
for the identification and ranking of hazardous locations. The
developed HI, based on an adequacy rating concept, represented
the composite hazardousness of an analytical unit. The
competence of HI approach was shown by the studies [101-104]
indicating that composite hazardousness rating could provide
a reasonably accurate prediction of future accident
experience. A detailed description of the HI approach is
covered in the literature review, and its application in the
development of indices for this study is presented in Section~
4.5. The SPSS program to accomplish the entire analytical
approach is presented in Appendix E.

The data analysis and the results, based on the above
described eight tasks, are separately presented for the two
types of highway sections studied: two-lane, two-way; and
four-lane divided highway sections. In the following pages,
this analytical approach is virtually replicated for the two
highway types.

99

4.2 DATA ANALYSIS FOR TWO-LANE, TWO-WA! SECTIONS
4.2.1 Descriptive Statistics

The descriptive statistics of variables indicated that
data for the entire 52 kilometers were processed and there
were no missing values. A summary of the descriptive
statistics of the variables employed in the study’is presented
in Table 4.1.

A list of 13 variables is exhibited in Table 4.1. The
first ten are the independent variables developed for accident
modeling. The details of development of these variables are
presented in Section 3.2.2 (page 73). These candidate
variables were further screened based on the criteria of
variability; non-negativity; independency; and repressibility
for employment in the model.

Item 11 in Table 4.1, the [ACCFRQ], represents the three
year period.accident.data. The [ANACFQ], i.e., item 12, is the
average annual accident frequency based on [ACCFRQ].
Primarily, the [ANACFQ] was used as the dependent variable in
predictive modeling. Item 13, the [ANACRT], represents the
annual accident rate, computed to account for the reported
variation of traffic volume in different segments of the
experimental site. A detailed description of the effect of
traffic volume variation on accidents, and using [ANACRT] as

the dependent variable, is presented in Section 4.2.7 (page 114) .

 

100

uses essence: unease «m «e.m 66.6 no.” em.” .emoeze. .na

sozueoume ezeauooc amass: an ac.» oo.o he.a on.” “annexe. .nu
sozueouxu azuanooc um oo.fl« oo.o oo.m m«.m .eesooc. .HH
mzoueoamemmo mementos «m cm.oc oo.o no.nu nm.en .monamm_ .6”
causes onaanoo azuzn>em um oo.~o om.n~ ao.o nv.ee .nzoosm. .m
muHozuHoHeue uzea nueeuomn um eo.e oo.o ou.H um.o .nzaama_ .e
muHozuHoHsue zoneoummnezu um eu.o~ oo.o e«.o oo.n .oumezu_ .e
moznxmez azuxu>¢m eanoHsue «m no.n oo.u ea. mo.« .nxmere. .8
menu: «someone azunousun um oo.n oo.o no.o eo.o .maaazm. .m
means azuzu>¢e azuuoueun «m «0.4 60.6 ee.o eu.o .zeauza. .v
mzoneoum uHamamceo azuHoHeun «m oo.umo~ oo.o mm.oo~ oo.oe~ .a~¢mao. .n
mused accuses oHeHoumm «m oo.mn oo.o so.e oe.ma .mzaeem_ .u

aznxaoau>un saunas um oo.u~on oo.o nn.oov me.ome .zonmnx_ .a

 

defied 2 MI! Cd! >On Cum CID: Canedud> .02

 

.ueaa-«uae no noeaueaeau openness-ea «.4 sun-a

101

4.2.2 Frequency Distribution

An investigation of the frequency distribution of the
studied hazards and accidents was considered important to gain
some insight on their characteristics. The frequency
distribution histograms of the hazards and accidents are

presented in Figures 4.1 through 4.11.

 

Comt Midpoint RIBBON DEVELOPMENT (meters per kilometer)

1" ‘3 -=_

’t 139 —=-

3 235 — -

‘ 331 —=

’t 523 —-

’° ‘19 _=

3 rs .... .

2 m1 ... .

2 907 - O

4 1°03 _3—

0 1099 .

0 1195

0 1291

2 1387 .3-

1 1‘83 :-

2 199 a.-
I....+....I....*....I....*....I....+....I....+....I
0 I 8 12 16 20

Histogram freqmncy
Percentile : Value I 15.0 : 0.00, 85.0 : 992.10

Figure 4.1
Frequency Distribution of Variable [RIBBON] - RIBBON DEVELOPMENT.

In these frequency histograms, the "count" column
represents the number of kilometers. Their sum always equals
52, representing the total length of the 2-lane, 2-way test
section. The "midpoint" column represents a scaled axis for
the histograms, and is divided into equidistant intercepts

based on the minimum and the maximum value of the variable.

102
For example, in Figure 4.1 for variable [RIBBON], the
first reading (14) in the "count" column signifies the number
of kilometers having a midpoint value of the hazard as 43
units per kilometer. This vertical axis is divided into
equidistant intercepts of 96 units each to cover the minimum-

maximum range of 0 to 1622 of the variable value.

Comt Michoint SPECIFIC LATERAL PATHS (meters per kilometer)

a e e e e e e a e
MOUOUIOUIOUIOUIOUIOUIOUI
eel

dddd
...
.

garasuss.4.y.e.~.
0| 0

ddONN-INUI‘OUIUIU‘03‘0-D
as. I
as
a
II "II
In
I
e
a

 

O“
a
I
a
O
0
e
e
e
O
N"-
I
D
I
e
0
O
a
e
a
be:
a
a
a
a
...
e
a
e
0
0mm
e
e
O
C
0
I
I
I
ae-e
_a
O

Histogr- fremency

Percentile : Value I 15.0 : 7.00, 85.0 : 25.10

Figure 4.2 '
Frequency Distribution of Variable [SPATHS] - SPECIFIC LATERAL PATHS.

Midpoint

Count

d

flOOONO-IMNUW-D50mum

91
156
221
286
351
616
681
546
611
676
761
806
871
936

1001
1066

Percentile : Value I

Figure 4.3

26

103

DEFICIEHT GUARDRAIL (meters per kilometer)

 

Histogram frequency
15.0 : 0.00, 85.0 : 546.85

Frequency Distribution of Variable [GDRAIL] - DEFICIENT GUARDRAIL.

i

Midpoint

.0
.3

e
0

e e e e e e e
OWNOOWOVJ~HOUI~O

b
HOOOOOOOOOON-INe‘NO

55$HNFUNNNdd-I

Percentile : Value I

Figure 4.4

 

DEFICIEHT PAVEMEHT HIDTH (meters per kilometer)
IIIIIIIIIFIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII
- e
- '
- e
I e
-e
II
1....+....I....+....I....+....I....+....I....+....I
0 8 16 26 32 40

Histogram frequency
15.00 : 0.00, 85.00 : 0.49

Frequency Distribution of Variable [PWIDTH] - DEFICIENT PAVEMENT WIDTH.

104

(meters per kilometer)

Comt Midpoint DEFICIEHT SHOULDER mom
0 '.1 .
17 -1 _=—
0 .3 .
9 -5 —=_
1 .7 _ .
11 -9 _=—
0 1.1 .
1 1.3 _ .
2 1.5 _ I
2 1.7 _ .
7 1-9 —i_
0 2.1 .
0 2.3 .
0 2.5 .
O 2.7 .
2 2.9 :
0 3.1 _
1....*....I....+....I....+....I....+....l....+....l

 

0 4 8 12 16
Histogram frequency

Percentile : Value 8 15.00 : 0.00, 85.00 : 2.00

Figure 4.5

20

Frequency Distribution of Variable [SWIDTH] - DEFICIENT SHOULDER WIDTH.

Micboint DEFICIEHT PAVEMENT HARKINGS

g
f.

HOOONOO-O-IOO-OOOOOS

NNNNNNNNyNNNNNNN-b
umuaauuuw ......
aomatNUOM-DVuOUI-bgax

 

0 10 20 30 (.0
Histogram freqrency

Percentile : Value 8 15.00 : 2.00, 85.00 : 2.00

Figure 4.6

(kilometers per kilometer)

I
1....‘O0.0IOOOC+OOOOI..I.*COOOlIOO.‘OOOOIOIOI+IOOOI

50

Frequency Distribution of Variable [PMARKS] - DEFICIENT PAVEMENT MARKINGS.

105

Count Midpoint INTERSECTION DEFICIENCY (# of traffic conflicts per hour)

'3 .
'1 .

1 _=—
3 .

5 .

7 .

9 .

11 .
13 .
15 .
17

19 :'
21
23 m
25
27 .
29

O-DO-IO-INOOOOOOOVOO

 

0 10 20 30 40 50
Histogram frequency

Percentile : Value I 15.00 : 0.00, 85.00 8 0.00

Figure 4.7
Frequency Distribution of Variable [INTSEC] - imsnsecnou DEFICIENCY.

 

COUNT VALUE ISOLATED LANE DEFICIENCY (# 0f lane change per kilometer)
33 ~00 —=—
3 1-00 — -

3 2.00 _ .

2 3.00 ..

0 6.00

0 5.00

0 6.00

1 7.00 .
1... ...... 1.... ..... I.........l ......... I ......... I
0 3 16 26 32 60

Hi stogr- frequency
Percentile : Value = 15.00 : 0.00, 85.00 : 1.00

Figure 4.8
Frequency Distribution of Variable [ISLAND] - ISOLAT- LANE DEFICIENCY.

Midpoint

22.5
25.0
27.5
30.0
32.5
35.0
37.5
40.0
42.5

Count

N-INUlbUINONUIhUINONO-D

Percentile : Value s

Figure 4.9

106

PAVEMENT CONDITION RATING

 

Histogram frequency
15.00 : 34.950, 85.00 : 54.50

_ -

— -

—=-

—=—

—=

--

-=_
1....+....I....+....l....+....I....+..
0 2 4 6

(rated # per kilometer)

Frequency Distribution of Variable [PVCOND] - PAVIIENT CONDITION RATING.

Count

Midioint
-1

d

dNOO‘to-‘NNNUINN-‘ou-D

11
14
17
20

32
35

41
47

Percentile : Value a

Figure 4.10

ROAD SIDE OBSTRUCTIONS

 

0 4 8
Histogram frequency

15.00 : 4.975, 85.00 : 30.175

12

(counted 0 per kilometer)

l....‘I...‘....*..I.l.CCI+COCIlIIII+OOOOlOO00*....!

16 20

Frequency Distribution of Variable [SIDEOB] - ROAD SIDE DBSTRUCTIDNS.

107

 

Comt Midpoint ANNUAL ACCIDENT FREQUENCY (# of accidents/year per kilometer)

o -.s .

9 -° _=—

10 1.3 —:—

3 . — .

1: g-g —=_

2 2:5 — -

3 3-0 —

3 3-5 — -

1 (0.0 - e

2 4.5 -3.

2 5.0 .3-

1 5.5 :-

0 6.0 .

2 9'3

0 73 "
J..-.+....I....*....lC...+....l....+....l....+....l
0 4 8 12 16 20

Histogrm frequency
Percentile : Value = 15.00 : 0.00, 85.00 : 3.683.

Figure 4.1 1
Frequency Distribution of [ANACFQ] - ANNUAL ACCIDENT FREQUENCY.

The frequency distribution of the independent variables,
[PWIDTH], [PMARKS] and [ISLAND] showed. that. the. hazards
represented by them were neither very frequently occurring nor
had a noticeable variance. This point, though not totally
decisive, was of interest for variable retention in the final
predictive model. The interpretation of the frequency
distribution histogram for variable [INTSEC] required caution
in the sense that the output was only for the five
intersections of the experimental site. These were isolated
spots and were not distributed in 52 kilometers of the test

section like the other hazards.

4.2.3 simple Correlations

The correlation matrix showing the characteristics of the

relationship between the variables is presented in Table 4.2.

108

388.5 coco. 598. SB... 259.. 039.. 053.. 98.... 83. 8Q. gnome. Snug

00080.. 33.- o~5~. ~03. 98. 33. ~03... 553.. 535.. man... "Bun—8
0088.5 «~23 $9.. nn-.. o~n~.. “mm—... 33.. ~n¢~.. 039.. "93?:

0088.5 508.. nm~p.. pump: 33.. n5n~.. 33. 855.. . "gas

:3.— :3. N3... 35. 5~o~. . ~30. . ~mnp . Bumps:

cage; 0%. ~58. n55... ~08. n59. "mug

sag; e33. .62... §~.. 88. 2.5233

:88... :8. . m¢-. . 33.. 2.5235.

59:8... 98—. ~¢o~f adaflu

00089.5 053. 5355.2

388.5 n22:

.Nduulx flOdvld’uuoo Cg N.v CHAIR

109

The correlationumatrix indicated the variables [PWIDTH],
[SWIDTH] , [PMARKS] , [INTSEC] and [ISLAND] had a negative
relationship with accidents. This counter-intuitive effect
was noted as a jpoint of concern ‘while selecting final
variables for model building. The decision about the retention
of these variables was further subjected to various diagnostic
tests for checking multicollinearity, and.multiple procedures

for performing regression analysis.

4.2.4 Eigenvalues and Condition Indices

The computed eigenvalues and the condition indices are
shown in Table 4.3. Each column of the table, after the
condition index, indicates the proportion of the variance of
each of the coefficients associated with each of the
eigenvalues. For example, for the [RIBBON] coefficient, 0.464%
of the variance of the coefficient is attributable to the
first eigenvalue.

In this table, the variables with high proportions of
variance for the smallest eigenvalue are highly dependent. The
number of suspect collinear variables is indicated by the
number of higher-valued condition indices. In other words,
there are as many near dependencies among the variables as
there are large condition indices [116].

Referring to Table 4.3, the last eigenvalue i.e. , 0.00052

(at serial number 11) is the smallest, and accounts for 98.553%

110

. 58. .82. ~88. e88. 388.
. 58. some. .88. .88. .88.
. ammo. 33.. 88. 88. 38.
. .98. 50.8. 0.8. .38. «flat

8mg» g 9.52 um»;— nug 353:» ...—235 4:88 03.53 g; “saga

con—o
'88 n88
NE 0028 0
~02! ~02! 5
8g 0028 o
858. ~88. mg. 3.08. 808. 033. 3N3. 530—. :80. n..-~. 880. 535 035m. 9.
. . 8g 3. c
288 088 n
g 088 ~
~c§ ~88 5

use. 83-»
28282.. 85...; 53.28 ESE teas...

.aeuuvau suave—500 uni aeaueeseudn 333-2535 huwuaendunoo n.v edaea

111
and 69.461% of the variance of [PMARKS] and [INTSEC]. This
implies that these two variables are inter-dependent. Since
other independent variables have small variance proportions
for the 11th eigenvalue, they do not seem to have
multicollinearity. There are two condition indices (at serial
10 and 11) with magnitude (19.484 and 111.186) significantly
higher than the rest. This indicates twosuspect cases of
collinear variables. Both [PMARKS] and [INTSEC] were also
indicated for filtering out of the equation, based on non-

negativity of the simple correlation criteria.

4.2.5 Variable Tolerance (TOL) and Variance Inflation
Factor (VII)

The second diagnostic test.of multicollinearity involved
computing variable tolerance and variance inflation factors.

These statistics are produced in Table 4.4.

Table 4.4 Collinearity Diagnostics: TOL and VIP.

 

 

 

Variable Tolerance VIP
[SIDEOB] .787332 1.270
[PNIDTH] .790045 1.266
[INTSEC] .256323 3.901>>
[RIBBON] .675306 1.481
[PVCOND] .799510 1.251
[GDRAIL] .705304 1.418
[ISLAND] .702357 1.424
[SWIDTH] .627266 1.594
[SPATHS] .613606 1.630
[PMARKS] .234617 4.262))
>> High VIP.

In Table 4.4, high VIF values for [INTSEC] and [PMARKS]

indicate that the two independent variables are collinear, and

112
substantiate the previous findings based on the eigensystem

diagnostic test.

4.2.6 Regression Analysis and the Crucial Variables

All the ten variables were entered in to the process of
BACKWARD ELIMINATION for regression analysis. This process was
employed to observe the removal order of variables. The
observed. elimination sequence (in 'the, order’ of variable
significance based on a POUT = 0.100 criteria) is summarized

in Table 4.5.

Table 4.5 The Variable Removal Order by Backward Elimination.

 

 

Step # Variable Removed T Sig T
1. [SWIDTH] -.842 .4040
2. [ISLAND] .467 .6427
3. [INTSEC] -.540 .5920
4. [PWIDTH] -.858 .3953
5. [SIDEOB] .289 .7741
7. [SPATHS] .461 .6467
8. [PVCOND] 1.283 .2056

 

Consequently, [RIBBON] and [GDRAIL] were retained as the
two most significant variables. The FORWARD and the STEPWISE
methods also indicated [RIBBON] and [GDRAIL] as the two most
significant variables for equation building. The significance
of T-statistics of the filtered out variables indicated that
any additional variable could not be selected even by relaxing
the adopted limit of POUT. The output results of the

regression analysis are presented in Table 4.6.

113

Table 4.6 The Statistical Results of Regression Analysis.

 

Multiple R .59827
R Square .35792
Adjusted R Square .33172
Standard Error 1.36230
mm

D! Sum of Squares Mean Square
Regression 2 50.69269 25.34635
Residual 49 90.93765 1.85587
P - 13.65739 Signif F - .0000

Dependent Variable. [ANACFQ] ANNUAL ACCIDENT FREQUENCY

 

 

Variables in the Equation

Variable B SE B Beta T Sig T
RIBBON .001993 4.266638-04 .557777 4.672 .0000
GDRAIL .002726 7.62760E-04 .426730 3.574 .0008
(Constant) .112287 .370548 .303 .7631

 

The indicated appropriateness of selecting independent
variables, based on the different criteria employed in the

study, is summarized in Table 4.7.

Table 4.7 Indicated Appropriateness of Independent variables.

 

VARIABLE SELECTION CRITERIA

 

No. Variable
FD NN NC RC

 

 

1. [RIBBON] + + + +
2. [SPATHS] + + + -
3. [GDRAIL] + + + +
4. [PWIDTH] - - + -
5. [SWIDTR] + - + -
6. [PMARRS] - - - -
7. [INTSEC] + - - —
8. [ISLAND] - - + -
9. [PVCOND] + + + -
10. [SIDEOB] + + + -
Symbols

+ - Variable indicated for inclusion in the predictive model.
- - Variable not indicated for inclusion in the predictive model.

Abbreviations

ED 8 Frequency distribution.
NN - Non-negativity.

NC Hulticollinearity.

RC 8 Regression.

114
Table 4.7 shows that, based on all four criteria of
variable selection, the most appropriate indicated variables

for employing in the model were [RIBBON] and [GDRAIL].

4.2.7 Accident Frequency Versus Rate as the Dependent
Variable

A noticeable variation in the operating traffic volumes
was reported [118] for the experimental sites. To incorporate
the effect of traffic volume variation, accident rates for the
various sections of the experimental sites were calculated.
The average annual traffic volume for the entry section (i.e.,
Gujar Khan - Mandra section, Kilometer (1480 - 1502) was taken
as the base value to discount the remaining sections. The

reported variation and the calculated discounting factors are

shown in Table 4.8.

Table 4.8 Reported variation in Traffic Volumes.

 

 

No. Section Year Mean Discounting
1988 1989 1990 Factor
]< ----- ADT ----- >1 (mean/8656)
1. Gujar Khan - Nandra 8146 8739 9083 8656 1.00
2. Nandra - Rawalpindi 9470 10332 11982 10595 1.22
3. Rawalpindi - Tarnol 15654 15882 16103 15880 1.83
4. Tarnol - Taxila 12366 15355 15616 14446 1.67
5. Taxila - Wah 10635 11072 11415 11041 1.27

 

The compounding effect of the traffic volume variation on
the experimental site is shown in Figure 4. 12. The peak
represents the increased traffic flow in the suburbs of

Rawalpindi and Islamabad.

115

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Sectional Variaiion in
Traffic Volume
3

26
L.
o
6 2
a
u.
g. /\-\
'5 1.5 ’/V \T
c
a
o
a
E i
o
t)

(15

O
I 2 3 4 5
Highway Seciions

 

 

 

 

 

 

2-Lane, 2-Nay Eighway
Section 1 - Gujar Khan - Nandra.
Section 2 - Nandra - Rawalpindi.

4-Lane Divided Highway

Section 3 - Rawalpindi - Tarnol.
Section 4 I Tarnol - Taxila.
Section 5 - Taxila - Wah.

Figure 4.12
The Compounding Effect of Traffic Volume Variation.

The correlation of the variables with the two versions of
accident data (i.e., frequency and rate), was also examined.
This exercise was considered appropriate to explore the
possibility of having an incremental gain in the predictive

model's statistics by incorporating the reported variation in

116
traffic volume.

For this purpose, Annual Accident Frequency [ANACFQ] and
Annual Accident Rate [ANACRT] were used as the two dependent
variables. The correlation between the two versions of
accident data and the variables indicated a nominal difference

in the coefficients as shown in Table 4.9.

Table 4.9 Correlation of Indicated Variables with
Accident Frequency and Rates.

 

 

Variables [ANACFQ] [ANACRT]
:RIBBON] .4365** .4144*
:spnrasl .2398 .2451
[GDRAIL] .2682 .3151
:pwxara] -.1080 -.O982
:SWIDTI'I] -e0379 ". 0535
:PMARKS] -.0959 -.1086
[ISLAND] -.1501 -.1421
:PVCOND] .0997 .0816
:SIDEOB] .0669 . 0705

 

 

1-tailed Signif: * - .01 ** - .001

 

The comparison presented in Table 4.9 indicates a very
nominal difference between the coefficients of correlation of
the variables. Moreover, model development using [ANACFQ] and
[ANACRT] as two separate dependent variables showed a gain of
only 1.77% in the value of R2 for the "rate" format as
compared to "frequency" (i.e., R2 for [ANACFQ] was 0.35792
versus 0.37564 for [ANACRT]) . Since there were no strong
reasons to switch, the [ANACFQ] was retained as the dependent
variable of the predictive model. This selection also seemed
justified from a practical standpoint considering that
computation of accident rate required additional information

on traffic counts.

117

4.2.8 Final Accident Prediction Model and Test of Hypotheses

The equation which finally emerged from the regression
analysis for predicting accident potential of a 2-1ane, 2-way
section provided the following relationship between the

accident frequency and the hazardousness.

Y“, = 0.112287 + 0.001993 8, + 0.002726 1:, (4.6)
where,
Ym I Accident frequency per year per kilometer.
xl I Length of ribbon development on both sides (meters).
x2 I Length of highway section with deficient guardrail (meters).

The presence of variable [RIBBON] (i.e., x,) in the
equation indicates that deficient control of access is related
to accidents in Pakistan. This finding is supported by the
results of many studies [e.g. ,95,98,99] reporting that control
of access was a significant factor associated with accidents,
both in the motorized and motorizing countries.

Based on the following statistics of the predictive model

R I 0.59827

R? I 0.35792

R’M I 0. 33172

F I 13.65739

you I 0.0000
Variable T Big T

x1 4.672 0.0000

x2 3.574 0.0008

the null hypothesis of no linear relationship between the
ambient hazardousness and accidents was rejected, and it was

concluded that at a probability of F=0.00, 35.79% of the

118
variation in accidents was explained by two types of
identified hazards:
1) Deficient control of access; and

2) Deficient provision of guardrail.

4.2.8.1 Analysis of Residual and Predicted Values

The following four statistics for examining the residuals
and predicted values were calculated.

1. PRED Unstandardized predicted values.

2. RESID Unstandardized residuals.

3. ZPRED Standardized predicted values.

4. ZRESID Standardized residuals.

These statistics are presented in Table 4.10.

Table 4.10 Statistics for the Residuals and Predicted values.

 

 

Min Max Mean Std Dev N
*PRED .1123 3.7719 1.7628 .9970 52
*RESID -2.8610 3.2281 .0000 1.3353 52
*ZPRED -1.6555 2.0152 .0000 1.0000 52
*ZRESID -2.1001 2.3696 .0000 .9802 52

 

Durbin-Watson Test I 1.92816

 

The Durbin-Watson statistic is a measure of auto-
correlation in the residuals. Regression analysis assumes that
the residuals are not auto-correlated, in which case the
Durbin-Watson test should have a value near 2.00 [116]. The
indicated value of 1.92816 for the constructed model signifies

a satisfactory non auto-correlation between the residuals.

119
Regigua; Outliers
The SPSS system default produces the ten worst outlier
cases based on absolute values of the residuals and the
standardized residuals. The information on outliers presented
in Table 4.11 reveal that the model is good enough to predict
accidents by a maximum error of i 2 (taken as a discrete

rounded value for accidents), with two exceptions.

Table 4.11 Outlier - Standardised Residual.

 

 

 

Case # Km *RESID *ZRESID
22 1506 3.22809 2.36958
52 1536 -2.86098 -2.10010

7 1487 -2.39181 -1.75571
25 1509 2.22977 1.63677
26 1510 2.16272 1.58755
20 1504 2.14652 1.57566
51 1535 -2.10017 -1.54163
47 1531 1.98800 1.45929
21 1505 -1.96217 -l.44034
4 1484 -1.91629 -1.4D666

Wit!

The histogram for the standardized residuals is presented
in Figure 4.13. This histogram depicts the observed number of
residuals (labeled "N") , and the number expected (labeled "Exp
N") in each interval. The extreme intervals (labeled "Out")
contain more than 3.16 standard deviations from the mean. The
expected frequencies and the overlap between expected and
observed are indicated by a period and a colon respectively.
In the histogram presented in Figure 4.13, the distribution
seems to be fairly normal except for a mild clustering of

residuals at the center.

120

N Exp N (* = 1 Cases, . : = Normal Curve)
0 .04 Out

0 .08 3.00

O .20 2.67

1 .46 2.33 *

0 .95 2.00 .

3 1.74 1.67 *:*

S 2.85 1.33 **:**

2 4.19 1.00 ** .

3 5.52 .67 *** .

8 6.51 .33 ******:*
4 6.88 .00 **** .
10 6.51 -.33 ******:***
6 5.52 -.67 *****:

5 4.19 -l.00 ***:*

2 2.85 -1.33 **.

2 1.74 -1.67 *z

1 .95 -2.00 :

0 .46 -2.33

0 .20 -2.67

O .08 -3.00

O .04 Out
Figure 4.13

Histogram - Standardised Residual.

Predicted Value;

The normal probability plot in Figure 4.14 presents a

comparison of the probability of the observed and

predicted values .

entities were plotted against each other and examined

The cumulative distributions of the

deviation from the expected straight line.

If the

the
two
for

two

distributions are identical (the zero-error case), a straight

line should result. By observing the scatter of the predicted

121.
points about the expected straight line, it may be inferred
that the probability of having the predicted values from the

developed model will be reasonably close to the actual values.

 

 

 

 

1.0 i i i *
*‘A'
.*
.1”:
*
75 .. .141”. ...
***

o *t***
b *
s **.
e .5 .... ****. «r-
r *‘t
v ** .
e *
d

.25 .. *** ..

*t.
**
.*
v** i l 1 Expected
.25 .5 .75 1 0

Figure 4.14

normal Probability (P-P) Plot - Predicted values.

For a kilometer-wise prediction evaluation, a spreadsheet
implementation of the model is presented in Table 4.12. The
resulting deviations of the predicted accidents from the

actual are shown in Figure 4.15.

122

Table 4.12 Spreadsheet Implementation of Accident Predictive Model
( -Lane, 2-Way Section).

Variables Acc Frequency Constant X{* x5' Predicted Difference

Km.# x, x, 3-Yrs Annual (0.001993) (0.002726) Accidents

1481 901 _ 545 6 2.00 .112287 1.795693 1.48567 3.39 -1.39365
1482 0 180 1 .33 .112287 0 .49068 .60 -.269634
1483 0 500 0 .00 .112287 0 1.363 1.48 -1.47529
1484 0 784 1 .33 .112287 0 2.137184 2.25 -1.91614
1485 0 412 1 .33 .112287 0 1.123112 1.24 -.902066
1486 570 670 13 4.33 .112287 1.13601 1.82642 3.07 1.258616
1487 319 1092 4 1.33 .112287 .635767 2.976792 3.72 -2.39151
1488 980 150 10 3.33 .112287 1.95314 .4089 2.47 .8590063
1493 142 822 13 4.33 .112287 .283006 2.240772 2.64 1.697268
1494 214 602 5 1.67 .112287 .426502 1.641052 2.18 -.513174
1495 0 0 2 .67 .112287 0 0 .11 .5543797
1496 310 50 4 1.33 .112287 .61783 .1363 .87 .4669163
1497 100 328 9 3.00 .112287 .1993 .894128 1.21 1.794285
1498 50 188 D .00 .112287 .09965 .512488 .72 -.724425
1499 50 630 6 2.00 .112287 .09965 1.71738 1.93 .070683
1500 810 0 4 1.33 .112287 1.61433 0 1.73 -.393284
1501 98 0 2 .67 .112287 .195314 0 .31 .3590657
1502 0 399 1 .33 .112287 0 1.087674 1.20 -.866628
1503 305 150 4 1.33 .112287 .607865 .4089 1.13 .2042813
1504 600 200 12 4.00 .112287 1.1958 .5452 1.85 2.146713
1505 1597 0 4 1.33 .112287 3.182821 0 3.30 -1.96177
1506 1040 582 21 . 7.00 .112287 2.07272 1.586532 3.77 3.228461
1507 396 301 5 1.67 .112287 .789228 .820526 1.72 -.055374
1508 1032 284 8 2.67 .112287 2.056776 .774184 2.94 -.276580
1509 718 450 15 5.00 .112287 1.430974 1.2267 2.77 2.230039
1510 660 517 15 5.00 .112287 1.31538 1.409342 2.84 2.162991
1511 990 0 2 .67 .112287 1.97307 0 2.09 -1.41869
1512 629 0 5 1.67 .112287 1.253597 0 1.37 .3007827
1513 610 0 3 1.00 .112287 1.21573 0 1.33 -.328017
1514 821 423 11 3.67 .112287 1.636253 1.153098 2.90 .7650287
1515 534 195 6 2.00 .112287 1.064262 .53157 1.71 .291881
1516 409 50 4 1.33 .112287 .815137 .1363 1.06 .2696093
1517 710 142 10 3.33 .112287 1.41503 .387092 1.91 1.418924
1518 0 200 0 .00 .112287 0 .5452 .66 -.657487
1519 0 200 2 .67 .112287 0 .5452 .66 .0091797
1520 523 488 6 2.00 .112287 1.042339 1.330288 2.48 -.484914
1521 518 0 9 3.00 .112287 1.032374 0 1.14 1.855339
1522 0 88 0 .00 .112287 0 .239888 .35 -.352175
1523 913 0 3 1.00 .112287 1.819609 0 1.93 -.931896
1524 400 0 4 1.33 .112287 .7972 0 .91 .4238463
1525 222 266 0 .00 .112287 .442446 .725116 1.28 -1.27985
1526 0 285 0 .00 .112287 0 .77691 .89 -.889197
1527 0 200 0 .00 .112287 0 .5452 .66 -.657487
1528 755 0 8 2.67 .112287 1.504715 0 1.62 1.049665
1529 0 88 2 .67 .112287 0 .239888 .35 .3144917
1530 170 50 1 .33 .112287 .33881 .1363 .59 -.254064
1531 1622 0 16 5.33 .112287 3.232646 0 3.34 1.988400
1532 356 100 9 3.00 .112287 .709508 .2726 1.09 1.905605
1533 229 212 0 .00 .112287 .456397 .577912 1.15 -1.14660
1534 1340 0 5 1.67 .112287 2.67062 0 2.78 -1.11624
1535 1499 0 3 1.00 .112287 2.987507 0 3.10 -2.09979
1536 1379 0 0 .00 .112287 2.748347 0 2.86 -2.86063

123

 

 

 

 

 

 

 

 

 

 

 

 

 

s
4

’E d A

g 2 + ‘f+ 1

'6

3* 1 7x11 A

3. y 171 N 'K¥ I I x

a:

,_ yvy V Wkly R

0

C

0

3-2

a. X
-4
.6IIUIIUIITWITTIITITTIIIIIIIT‘TIIIIIIDTWTTTIITTTTTI[III
1431 1195 1507 1

KilometerNo.
Figure 4.15

Difference between Actual and Predicted Accidents
2-Lane 2-Nay Section (HIS), Kilometer 1481 - 1536.

124

4.3 DATA ANALYSIS FOR POUR-LANE DIVIDED HIGHWAY SECTIONS

The analyses for the four lane divided highway section
(kilometer 1550 - 1583) are presented in the following pages.
Since the analytical approach, in essence, is the same as that
followed for the two lane analysis, any repetitive description

of the statistical procedures and terminologies is avoided.

4.3.1 Descriptive Statistics

The descriptive statistics of variables indicated that
data for the entire 34 kilometers were processed and there
were no missing values. A summary of the descriptive
statistics of the variables employed in the study is presented
in Table 4.13.

A list of 14 variables is exhibited in Table 4.13. The
first 11, are the independent variables, developed for the
quantification of hazard and use in modeling, as described in
Section 3.2.2 (page 73). As before, the [ACCFRQ] represents
accident data for a period of three years; [ANACFQ] is the
average annual accident frequency; and [ANACRT] represents the
annual accident rate. Primarily, [ANACFQ] was used as the
dependent variable in predictive modeling. However, the effect
of traffic volume variation on accidents, using [ANACRT] as
the dependent variable, was also examined and is presented in

Section 4.3.7 (page 139).

125

 

 

 

seem azuouoor 040222 on 00.0 00.0 on.” 00.0 .amoczc. .en
uuzuooemm azunHooc gross: In 00.5 00.0 H5.A 00.A .osoazc. .na
uozunoumn azunnooe In 00.nn 00.0 «A.m 0n.0 .ommoo¢_ .«A
onaoamenmo unannsos 0n 00.nn 00.0 mn.0 nu.u~ .nou0Hm_ .AA
causes onaanoo aznxu>¢o en ma.nm 00.0“ 00.5 0A.00 .0zoo>o. .0H
nuaozunoabun used absence” on 00.0 00.0 an." up.“ .achmu. .0
muuozuuoumuo zonaounmuazn on an.0d 00.0 0~.n 50.0 .ounazH_ .0
mozmxmrx azurn>¢o azuHoHuuo In 0«.v 00. H0.H 00.A .mmmazo. .5
means mongoose azunoubun 0n 0m.m 00. 0~.A In.“ .meomsm_ .0
means azuxm>¢o azuHoHuun In mn.n 00. 00. mm. .manuzo_ .m
quasnmroo azuuouuun 0n 00.000” 00.0 50.00" «0.0ma .0H4u00. .0
noszuoo zaHan: on 00.0An 00.n. en.vm no.5» _zooouz. .n

are accused ouunouon 0n 00.A0 00.A nn.o an.an .mmacon_ .«
azarooan>un saunas 0n 00.0mm” 00.0 00.n00 no.0me .zoanam. .A

Huang 2 «or as: >00 can 000: eanuaue> .02

.ueuneaueb «0 unevenness eruuneuoeea n~.0 ensue

126

4.3.2 Frequency Distribution

An investigation into the frequency distribution of the
studied hazards and accidents was considered important to gain
some insight into their characteristics. The frequency
distribution histograms of the hazards and accidents are

presented in Figures 4.16 through 4.27.

 

Comt Micboint RIBBGI DEVELOPMENT (meters per kilometer)
1° 44 —=—
2 I“ _=
1 233 —

’° 335 _=—

3 ‘32 —=-

3 529 —i-

1 626 — I

0 723

’° 32° —=—

0 917 .

0 1014

1 1n1 ...

1 1mm ...

1 1305 -3-

1 1402 :

1 11.99 :.=

1 1596 :-
1....4’....I....*....I....*....I....‘0'....I....+....I
0 2 4 6 8 10

Histogrn freqiency
Percentile : Value I 15.00 : 0.00, 85.00 : 1189.50

Figure 4.16
Frequency Distn'bution of Variable [RIBBON] - RIBBON DEVELOPMENT.

127

Count Midpoint SPECIFIC LATERAL PATIiS

a a
OWOUIOUIOUIOUIOUIOUUOU‘O
I
II
II
II
I
I
I
II

Nd‘dd

dO-lNuNuuUl-DUJ‘WNOO-I

 

Histogr- frecpency
Percentile : Value I 15.00 : 12.00, 85.00 : 32.00

Figure 4.17

(meters per kilometer)

I
I
I
I
+
I
I
I
H
I
I
I
+
I
I
I
bl.
I
I
I
I
4'
I
I
I
I
H
I
I
I
I
+
I
I
I
~
I
I
I
+
I
H

Frequency Distribution of Variable [SPATHS] - SPECIFIC LATERAL PATHS.

Comt limoint MEDIAN (PENIIGS

9 —=_
23 —-—

‘7 —
65 —-
85 _ .
104
123 .
1‘2 I
161 .
130
199
218
237
256
275
29’.
313

..a-a
dOOOOOOOOOOONJ‘NOU

 

0 6 B 12 16
Histogram fremency

Percentile : Value = 15.00 : 6.50, 85.00 : 65.75

Figure 4.18

(meters per kilometer)

IIII*IIIIIIIII+IIIIIIIII+IOIIII.II+IIII!IOII+IIIOI

20

Frequency Distribution of Variable [MEDOPN] - MEDIAN OPENINGS.

128

Comt Hicboint DEFICIENT GUARDRAIL

22 ‘1 —:
129 — -
217 —=
305 _
393

‘81 -:
569 .
657 .:
745 .
833
921
1009
1097
1185
1273
1361
11.1.9 _

HOOOOOOOOdONOd§U

 

T
a
a
a
I
a
a
a
e
a-a
a
a
a
a
I
e
a
a
a
a-
*
a-a
I'
e-

Histogram frequency
Percentile : Value e 15.00 : 0.00, 85.00 : 313.00

Figure 4.19

(meters per kilometer)

Frequency Distribution of Variable [GDRAIL] - DEFICIENT GUARDRAIL.

 

(meters per ki lometer)

CM! "1‘30"“ DEFICIENT PAM." UIDTH

‘9 -° _=_

6 -2 _=—

1 .I. _ .

1 .6 _ .

0 .8 .

0 1.0 .

0 1.2 .

1 1.1 _ .

0 1.6 .

0 1.8 .

1 2.0 .3

2 2.2 .3-

1 2.‘ :-

0 2.6 .

1 2.0

o 3.0 -

1 3.2 _
1....+....I....*....I....+....I....+....I....+....I
0 ‘ 8 12 16

Histogr- freqnncy
Percentile Value = 15.00 : 0.00, 85.00 : 2.185

Figure 4.20

20

Frequency Distribution of Variable [PWIDTI-I] - DEFICIIVT PAVEMENT WIDTH.

129

DEF I CIEIIT SIIGJLDER HIDTII

1s

1n

0-

I ...
. I"

d
«3‘
u
U.

I.
e

N
e

-a
O
-a

d

d-‘OONOOI‘NNuNN-DONN
u N
a a a a a
J‘ ‘1
a 0
V
.III

 

Histogram frecpency
Percentile : Value = 15.00 : 1.075, 85.00 : 3.000

Figure 4.21

(meters per kilometer)

Frequency Distribution of Variable [SWIDTH] - DEFICIENT SHOULDER WIDTH.

(kilometers per kilometer)

Comt MIMI"! DEFICIENT PAVEHENT HARKIIIGS
0 '.3 .
2 .0 :-
1 .3 .3
0 .6 .
5 -9 -=_
1 1.2 _ .
0 1.5 .
0 1.0 .
12 2-1 —=—
3 20‘ — O
1 2.7 _ .
6 3-0 —=_
0 3.3 .
0 3.6 .
1 3.9 :
1 1.2 1.
0 4.5
1....‘I‘....I....*....I....‘P....I....+....I....+....I

 

0 I. 8 12 16

Histogram frequency
Percentile : Value -- 15.00 : 1.00, 85.00 : 3.00

Figure 4.22

20

Frequency Distribution of Variable IPMARKS] - DEFICIENT PAVEMENT MARKINGS.

130

 

Count Midpoint INTERSECTION DEFICIENCY (# of traffic conflicts per hour)
29 0—=—
1 _ .

1 2_ .

1 3_.

0 6 .

0 5 .

0 6 .

0 7 .

0 8 .

0 9

1 1o.I

0 11

0 12

0 13

0 14

0 15

1 16-
I....+....I....*....I....+....I....+....I....+....I
0 6 12 18 24 30

Histogram frequency
Percentile : Value 8 15.00 : 0.00, 85.00 : 0.556

Figure 4.23

Frequency Distribution of Variable [INTSEC] - INT SECT ION DEFICIENCY.

i
E

HNOUIUIUIUbO

00

1 00

2 00

3 00

4.00 :
00

6 00

7 00

8 00

 

l.........I.........I... ...... I ......... I ......... I
0 2 6 6 8 10

Histogram frequency
Percentile : Value 8 15.00 : 0.00, 85.00 : 5.00

Figure 4.24

ISOLATED LANE DEFECTS (I of lane change per kilometer)

Frequency Distribution of Variable [ISLAND] - ISOLATED LANE DEFICIIVCY.

131

 

Comt Mimint PVT. CONDITION RATING (rated # per kilometer)

6 29-0 -=—

0 30.5 .

1 32.0 —:-

1 33.5 _:

0 35.0 .

1 36-5 — -

0 38.0 .

0 39.5 .

3 “-0 —=—

2 62-5 _ -

I “-0 —

3 ‘5-5 _=—

5 67-0 —=—

3 63-5 —=_

3 50-0 —=_

2 51-5 —=-

5 53-0 —=—
1....+....I....‘0'....I....‘6....I....+....I....+....I
0 1 2 3 lo 5

Histogram frequency
Percentile : Value 8 15.00 : 32.875, 85.00 : 52.063

Figure 4.25
Frequency Distribution of Variable [PVCOND] - PAVEMENT CONDITION RATING.

Count Midpoint ROADSIDE OBSTRUCTION (comted # per kilometer)

2 .

6 —=—

6 _=_

3 _=—
10 _3_

12 —-

16 —
16 _-
13 _ -

2° - -

22 .
2" I:—
26 .
28 .
30

32—
34

O-POOONO-DNU-DI‘UIVUIWO

 

1...-*OIOIlCI.I+OIIIIIOII+IIIIIIII.*IIIIlIIII+IIIII
0 2 lo 6 8 10
Histogrll freqnncy

Percentile : Value = 15.00 : 5.00, 85.00 : 18.50

Figure 4.26
Frequency Distribution of Variable [SIDEOB] - ROAD SIDE OBSTRUCTIONS.

132

 

Count Midpoint ANNUAL ACCIDENT FREQUENCY (# of accidents/year per kilometer)

0 '.5

7 -° —=—

g 1-3 _=—

- —=—

3 1-5 _ -

0 2.0

3 2-5 _ -

2 3-° — -

3 -5 _=—

o ‘00 .

0 4.5 .

‘1’ 3'3 _‘

0 ah I

1 :1 _

o 73
1....t....I....*....I....+....I....‘f....I....+....I
0 2 6 6 8 10

Histogram frequency
Percentile : Value 8 15.00 : 0.00, 85.00 : 3.50

Figure 4.27
Frequency Distribution of [ANACFQ] - ANNUAL ACCIDENT FREQUENCY.

The frequency distribution of the independent variable
[PWIDTH], much like the 2-lane case, showed that deficient
pavement width neither occurred very frequently nor had a
noticeable variance. The non-variability reflected that
[PWIDTH] was not a robust candidate for modeling. The
interpretation of the frequency distribution histogram for
variable [INTSEC] required caution in the sense that the
output was only for the six intersections in the experimental
site. These were isolated spots and were not distributed in

the 34 kilometers of the test section like the other hazards.
4.3.3 simple Correlations
The correlation matrix showing the nature and the order

of the relationship between the variables is presented in

Table 4.14

133

2.88... 39.. 33. NN—o. canoe. gen. 5.8. 93. e93. «N2... 03... :33. "20:5
6688., 83.. can; :3. 88. Kc... can: 350.. .636. «on. EN. name—8

e088... has; SON. $13.. 933. mp8. 82.. one} 53.. 009. "gen

:88.— p-n.. n58. ~o-.- not... 83.. 009.. 559.. 33. ~35”:

6688... 58¢. «NMN. $3. hemp. 33.. 2.2.. can. 38::

:88.- thp . . 38. 33. meow. . 3-. 32.. august:

:88... gnu . - .68. . hope. . e23. 093. 3338

5089.. on“. n37. 52.. 03.4... 9.5.3.:

2.3; 83.- comp. mono. 328$.

588$ «no. coo—... "28°95

assoc... ~08. sushiaa
66608.. nag—E

.Nduuei gdvedeuuoo DAB 11¢ CHAIR

134
The correlation matrix indicated all variables, except
[MEDOPN] and [SIDEOB], had a positive relationship with
accidents. This effect was noted as a point of concern while
selecting final variables for model building. The decision
about the retention of the variables was further subjected to
various diagnostic tests for checking multicollinearity, and

multiple procedures of performing regression analysis.

4.3.4 Eigenvalues and Condition Indices

The computed eigenvalues and the condition indices are
produced in 'Table 4.15. (A detailed. description. of the
statistical terminologies and interpretation of this type of
table is given in Section 4.2.4, page 109).

The last eigenvalue in Table 4.15 (i.e. , 0.00658 at
serial number 12) is the smallest and accounts for 21.173,
88.776, and 18.922 of the variance of [ISLAND], [PVCOND], and
[SIDEOB] respectively. This implies that these three variables
are susceptible to near dependency. Since other independent
variables have small variance proportions for the 12th
eigenvalue, they do not seem to have multicollinearity.
However, the collinear variables are not sharply distinguished
as there are only two condition indices (at serial 11 and 12)
having their magnitude 12.000, and 33.664 respectively, higher
than the rest. This indicates that there are two suspect cases

of collinear variables.

135

 

 

 

§
E
s
E‘
a
Ԥ
g
E
§
s
10
§
§
0
a
ii
a
s
0
if!
3
§

eNOSc. ~cmec. meaeN. macaw. «amnm. apnea. nnNee. pwmpe. scoop. annFe. creme. oNeee. oe~.e

ppmee. sumac. mQNNe. mauve. epeae. «mace. woowe. oepee. Nmmpe. ance. mmeea. puppe. amp.a nc~—— o
oewoe. moeee. Nepwe. No—ee. mac—N. ameaF. mveme. Neeee. Nooaa. mpwee. soeoe. aaeee. o~a.m emu—N a
canoe. n—eee. emcee. mecca. ~eeee. comma. chesw. chmee. No~me. cam~e. cunnm. ecaee. sen.m ne¢o~ h
-a~e. e~eae. neae~. cnpop. ph—ee. eOOne. ansee. ue~m—. euebp. mac—e. ounce. vaace. ~na.v nae—n c
«ewee. caeee. nepwe. ~mo-e. ~e~ee. Mmeee. vmcmp. Npmem. emcee. ~eeee. N¢~ne. eeeee. mon.n enema m
a-ee. eeeee. amnne. cooea. Peace. skaee. Noeee. «once. me—Nn. m~eee. eepee. eeeee. sp~.n eoawh a
e—cce. e—eee. «came. cmmm~. hecce. unpea. enema. p-ee. vs—me. epaee. nonNe. Npeae. cae.~ ~cnne n
apnea. —eeee. mompe. mhnvw. peace. -ece. manoe. cease. opece. ~eeee. ne—ee. mecca. env.~ eeno~.— ~
capee. o~eee. n-ea. Pm—ee. whpee. «area. en~ee. o~nee. eawcc. ac—ee. n~nee. N—eee. eea.— reamo.s F

came—m 950).. 92.5— uwm»: «nag aha—3m she—3A .258 880* 9:58 See; cringe .
x85 03¢)
363.63.... 3:33) 53:88 .565 .35..

.eeudaau soaudoaoo use aesnesdemwm .euauaoeaewn hvdueedwuuou m~.¢ sands

136

4.3.5 Variable Tolerance (TOL) and Variance Inflation
Factor (VIP)

The second diagnostic test of multicollinearity was
comprised of computing variable tolerance and VIP. These

statistics are produced in Table 4.16.

Table 4.16 Collinearity Diagnostics: TOL and VIP.

 

 

Variable Tolerance VIF
[SIDEOB] .472085 2.118>>
[INTSEC] .580699 1.722
[GDRAIL] ‘.826755 1.210
[SWIDTH] .571155 1.751
[PVCOND] .602800 1.659
[MEDOPN] .676510 1.478
[RIBBON] .599652 1.668
[SPATHS] .624303 1.602
[PWIDTH] .605702 1.651
[PMARKS] .567050 1.764
[ISLAND] .562582 1.778

 

>> Highest VIP.

In Table 4.16, [SIDEOB] is the only variable having a
clearly higher VIF value than the rest indicating one suspect
case of dependency. The VIP analysis did not clearly show
other dependencies, and only partially supported the findings
based on the eigensystem diagnostic test which indicated
[ISLAND], [PVCOND], and [SIDEOB] having interdependency. The
cumulative conclusion drawn from the two diagnostic tests for
mmlticollinearity (i.e., eigenvalues and the VIP tests) is
that only one variable, [SIDEOB] failed both tests whereas
[ISLAND] and [PVCOND] were not clearly indicated for

interdependency.

137

4.3.6 Regression Analysis and the crucial variables

To explore the most significant variables, all the eleven
independent variables were entered to the process of BACKWARD
ELIMINATION for regression analysis. The observed elimination
sequence (in the order of variable significance based on a

POUT = 0.100 criteria) is summarized in Table 4.17.

Table 4.17 The Variable Removal Order by Backward Elimination.

 

 

Step # Variable Removed T Sig T
1. [PWIDTH] -.251 .8035
3. [SPATHS] .220 .8274
s. [ISLAND] .777 .4440

As a result of this operation [RIBBON], [MEDOPN],
[GDRAIL], [PMARKS], [INTSEC] and [SIDEOB] were retained as the
six most significant variables. However, a negative
coefficient was indicated for [SIDEOB] (as‘was expected.due to
negative correlation). Besides, the FORWARD and STEPWISE
methods excluded [MEDOPN] and [PMARKS] based on a PIN = 0.050
criteria. Thus [RIBBON], [GDRAIL] and [INTSEC] were selected
as the three most relevant significant variables for model
building. The results of various iterative runs of regression
analysis indicated these three variables as the most
appropriate regressors. However, it is important to note that
[MEDOPN] would have been the next most significant variable
for modeling if the selection criteria were relaxed.

The output results of the final regression analysis are

presented in Table 4.18.

138

Table 4.18 Statistical Results of Regression Analysis.

 

Multiple R .81211
R Square .65952
Adjusted R Square .62547
Standard Error 1.04503
Analysis of Variance

DP Sum of Squares Mean Square
Regression 3 63.46281 21.15427
Residual 30 32.76268 1.09209
F = 19.37046 Signif F = .0000

Dependent Variable. [ANACFQ] ANNUAL ACCIDENT FREQUENCY
Variables in the Equation

 

 

Variable B SE B Beta T Sig T
RIBBON .001932 3.79586E—04 .557823 5.089 .0000
GDRAIL .002189 6.41767E-04 .366633 3.410 .0019
INTSEC .158703 .057857 .303268 2.743 .0102
(Constant) .012487 .272011 .046 .9637

 

The indicated appropriateness of selecting independent
variables, based on the different criteria employed in the

study, is summarized in Table 4.19.

Table 4.19 Indicated Appropriateness of Independent variables.

 

VARIABLE SELECTION CRITERIA

 

 

No. Variable

FD NN MC RC
1. [RIBBON] + + + +
2. [SPATHS] + + + -
3. [MEDOPN] + - + -
4. [GDRAIL] + + + +
5. [PWIDTH] - + + -
6. [SWIDTH] + + + -
7. [PMARKS] + + + -
8. [INTSEC] + + + +
9. [ISLAND] + + - -
10. [PVCOND] + + - -
11. [SIDEOB] + - - -

 

Symbols
+ 3 Variable indicated for inclusion in the predictive model.
- 8 Variable not indicated for inclusion in the predictive model.

Abbreviations

PD 8 Frequency distribution.
NN 8 Non-negativity.

MC - Multicollinearity.

RC - Regression.

139
Table 4.19 shows that, based on all four criteria of
variable selection, the most appropriate variables indicated

were [RIBBON], [GDRAIL] and [INTSEC].

4.3.7 Accident Frequency Versus Rate as the Dependent
Variable

To incorporate the effect of traffic volume variation,
accident rates for the various sections of the experimental
site were calculated as explained in Section 4.2.7 (page 114).
The correlation of the indicated variables with the two
versions of accident data: 1) Annual Accident Frequency
[ANACFQ]; and 2) Annual Accident Rate [ANACRT] were examined.
A comparison of the two showed a nominal difference in the

coefficients of correlation as shown in Table 4.20.

Table 4.20 Correlation of Indicated Variables with
Accident Frequency and Rates.

 

 

Variables [ANACFQ] [ANACRT]
'RIBBON] .6403** .6969**
spawns; .1486 .1260
HEDOPN] -.o921 —.1226
GDRAIL] .4245* .3537
'pwxnrn] .0743 .0148
SWIDTH] .0813 .0875
PMARKS] .3421 .2873
INTSEC] .4838* .3750
ISLAND .0122 .1065
pvcounl .2648 .2377
sxusos] -.1079 -.1203

 

 

1-tailed Significance: * - .01 ** - .001

 

The model development using [ANACFQ] and [ANACRT] as two
separate dependent variables showed better statistical results

for the "frequency" format as compared to "rate" (e.g., szor

140
[ANACFQ] was 0.65952 versus 0.62698 for [ANACRT]). Therefore,
the [ANACFQ] was adopted as the dependent variable of the
predictive model. Much like the 2-lane case, this selection
was justified from a practical standpoint as well, considering
that the computation of accident rate required additional

information on traffic counts.

4.3.8 Final Accident PredictionMNodel and Test of Hypotheses

It was concluded in Section 4.3.6 that [RIBBON], [GDRAIL]
and [INTSEC] were the most appropriate independent variables
for model building. The equation which finally emerged for
predicting the accident potential of a 4-1ane divided.highway
section provided the following relationship between the

accident frequency and the hazardousness.

Y“, = 0.012487 + 0.001932 81 + 0.002189 32 4' 0.150703 83

(4.7)
where,

Ym - Accident frequency per year per kilometer.

x, - Length of ribbon development on both carriageways (meters).

x2 - Length of highway section with deficient guardrail (meters).

x, - The aggregated ratio of traffic conflicts to the operating volume
in an intersection expressed as percent.

The presence of [RIBBON] and [INTSEC] (i.e., x1 and x3)
reveal that deficient control of access and lateral entrance
conditions were predominantly associated with accidents in
Pakistan. These findings are upheld by many studies made in
the developed countries, and by the TRRL, 0.x. , for some third
world countries [e.g., 95,98,99].

141.

Based on the following statistics of the predictive model

R - 0.81211

R2 8 0.65952

3’”, - 0. 62547

F s 19.37046

Fm” - 0.0000
Variable T Sig T

xl 5.089 0.0000

x2 3.410 0.0019

x, 2.743 0.0102

the null hypothesis of no linear relationship between the
ambient hazardousness and accidents was rejected, and it was
concluded that at probability of F=0.000, 62.55% of the
variation in accidents was explained by three types of
identified hazards.

l) Deficient control of access;

2) Deficient provision of guardrail on warranted

sections; and

3) Intersection deficiencies.

4.3.8.1 Analysis of Residual and Predicted values

The following four statistics for examining the residuals
and predicted values were calculated.

1. PRED Unstandardized predicted values.

2. RESID Unstandardized residuals.

3. ZPRED Standardized predicted values.

4. ZRESID Standardized residuals.

These statistics are presented in Table 4.21.

142

Table 4.21 Statistics for the Residuals and Predicted Values.

 

 

Min Max Mean Std Dev N
*PRED .0125 5.5138 1.4608 1.3868 34
*RESID -2.0442 3.7555 .0000 .9964 34
*ZPRED -1.0444 2.9226 .0000 1.0000 34

 

Durbin-Watson Test 8 2.09919

 

The Durbin-Watson statistic is a measure of auto-
correlation in the residuals. Regression analysis assumes that
the residuals are not auto-correlated, in which case the
Durbin-Watson test should have a value near 2.00 [116]. The
indicated value of 2.09919 for the constructed model signifies

satisfactory non auto-correlation between the residuals.

e du t e
The SPSS system default produced the ten worst outlier
cases based on absolute values of standardized residuals. The
information furnished on outliers is presented in Table 4.22
which reveals that the model was good enough to predict
accidents by a maximum error of _-I_-2 (taken as a discrete

rounded value for # of accidents), with one exception.

Table 4.22 Outliers - Standardized Residual.

 

 

Case # Km *RESID *ZRESID
23 1572 3.75548 3.59366
30 1579 -2.04417 -1.95609

5 1554 -1.87951 -1.79852
22 1571 -1.56187 -1.49456
34 1583 -1.30388 -1.24769
11 1560 1.17303 1.12248

8 1557 .87809 .84025

6 1555 -.87478 -.83708
15 1564 .84745 .81093

12 1561 .83011 .79434

 

d No

1

143

The histogram for the standardized residuals is presented

in Figure 4.28 (the interpretation of the histogram is given

at page 119).

Here the distribution does not seems to be

normal due to clustering of residuals at the center and a

steep tail toward negative values. The statistical literature

[116] indicated the following supporting remarks:

"it would unreasonable to expect the observed residuals to be
exactly normal - some deviation is expected because of sampling error.
Even if the errors are normally distributed in the population, sample
residuals are only approximately normal".

N Exp N
1 .03 Out
0 .05 3.00
0 .13 2.67
0 .30 2.33
0 .62 2.00
0 1.14 1.67
0 1.87 1.33
2 2.74 .00
3 3.51 .67
5 4.25 .33
12 4.50 .00
5 4e26 -e33
1 3e61 -e67
1 2e74 '1.00
2 1.87 -1e33
1 1e14 -1e67
1 .62 '2.00
0 .30 '2.33
0 .13 '2.67
0 .05 '3.00
0 .03 Out

Figure 4.28

(* = 1 Cases, . : a Normal Curve)
*

O
O
0
es,
are.
***:*
eeeegeeeeeee
stage

II“

Histogram - Standardized Residual .

Erasmus;
The comparison of the probability of the observed and the

predicted values is presented by the normal probability plot

in Figure 4.29.

The cumulative distributions of the two

144
entities were plotted against each other and examined for
deviation from the expected straight line. By observing the
scatter of the predicted points about the expected straight
line, it was inferred that with the exception of a few data
points, the probability of having the predicted values from
the developed model will be reasonably close to the actual

values.

 

**

.**
sees
.75 0 we .p
sees
as .
* 0
*******°
.
e
**

00<nouco
G

N

U"
1
I
fi
1
I

“sets: I

 

 

 

1 Expected

‘—
_

Figure 4.29
Normal Probability (P-P) Plot - Predicted values.

A spreadsheet implementation of the predictive model is
presented in Table 4.23, and the resulting deviations of the

predicted accidents from the actual are shown in Figure 4.30.

145

Table 4.23 Spreadsheet Implementation of Accident Predictive Nedel
(4-Lane Divided Section).

Constant

.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487
.012487

x.*

(.001932)

.382536
.63756
2.812992
2.751168
1.093512
.31878
1.624812

x,*

x5*

Predicted

(.002189) (.158703) Accidents

.726748 1.598139

.06567
.10945
.054725
.13134
.236412
0
.10945
.168553
0
1.041964
0
.096316
.098505
.140096
3.26161
1.521355
.135718
.280192
.560384
0

0
.06567
.50347
1.024452
0

0
.10945
.04378
.2189
.08756
.496903

0
0
0
.4888052
0
0
0
0
0
0
.0603071
0
0
0
0
.3777131
0
0
2.58IMUU6
0
0
0
0
0
0
.0968088
0
0
0
0
0
O
0

urrr.rrrwu
ESEKREEBS

Diff-
erence

.3208463
1.172749
.8306059
-.108803
-.110992

.847417'
-.187030
°.157822
.0522003
.4969863
.1521737
.4726983
-1.56195
3.755295
.0108297
.0331197
-.311820
.0464202
-.504473
-.693827
-2.04438

.148785
-.269569
-.514538
-1.30397

--- Variables --- Accident Freq
Kn.# x, x, x, . 3Yrs Annual
1550 517 332 10.07 11 3.67
1551 121 30 .00 1 .33
1552 0 50 .00 0 .00
1553 0 25 .00 2 .67
1554 818 60 3.08 1 .33
1555 324 108 .00 0 .00
1556 0 0 .00 1 .33
1557 0 50 .00 3 1.00
1558 0 77 .00 1 .33
1559 0 0 .00 1 .33
1560 400 476 .00 9 3.00
1561 50 0 .38 3 1.00
1562 0 44 .00 0 .00
1563 0 45 .00 0 .00
1564 0 64 .00 3 1.00
1565 300 1490 .00 11 3.67
1566 300 695 2.38 7 2.33
1567 1104 62 .00 7 2.33
1568 799 128 .00 7 2.33
1569 1218 256 16.31 17 5.67
1570 439 0 .00 4 1.33
1571 802 0 .00 0 .00
1572 1639 30 .00 21 7.00
1573 590 230 .00 5 1.67
1574 1344 468 .00 11 3.67
1575 500 0 .00 2 .67
1576 437 0 .61 3 1.00
1577 198 50 .00 0 .00
1578 330 20 .00 0 .00
1579 1456 100 .00 3 1.00
1580 1424 40 .00 9 3.00
1581 566 227 .00 4 1.33
1582 165 236 .00 1 .33
1583 841 0 .00 1 .33
Total 149 49.67

146

 

 

 

l

 

N

l

 

O

l

wa MHAWN / \1 1
\/ V

 

Difference (# of occldenis)
A
.1

i

 

-4

 

 

 

T T T T T T T llllll I I T I T T T I T l T TT T I T

f I T I T T
1550 1561 1572 1583
Kilometer N0. ‘

 

 

 

Figure 4.30
Difference between Actual and Predicted Accidents

4-Lane Divided Section (N-S), Kilometer 1550-1583.

4 . 4 AN EVALUATION OF THE DATA TYPE

As the result of modeling the hazard-accidents
relationship by regression analyses, equations 4.6 and 4.7
(page 117, 140) were developed for 2-lane, and 4-lane sections
respectively. The significant variables common to both the

equations, were [RIBBON] AND [GDRAIL], both of which are

147

measures of design deficiencies. Additionally, in the case of
the 4-1ane model, the variable [INTSEC] was also significant.
This measure is comprised of the conflict (observational) data
set. However, neither of the models included a measure from
the expert team data set among the explanatory variables.

One of the objectives of this study was to test
association of each data set separately with the dependent
variable. This operation was important to determine which type
of data set would be the most appropriate for measuring the
ambient hazardousness so that the findings could be applied to
an experimental design for optimizing resources in the data
collection.

To achieve this objective, the three type of data:
geometric data; observational data; and subjective data were
independently used in regression analyses. The categorization

of the variables according to type of data is indicated as

following.
Type of Data variables
1) GEOMETRIC: [RIBBON] [SPATHS] [MEDOPNJ‘
[GDRAIL] [PWIDTH] [SWIDTH]
[PMARKS]

2) OBSERVATIONAL: [INTSEC] [ISLAND]

3) SUBJECTIVE: [PVCOND] [SIDEOB]

 

‘ Applicable to 4-lane sections only.

 

 

148
As before, the three methods of model building BACKWARD,
FORWARD and STEPWISE were employed. The results of this
investigation for both type of highway sections are presented

in Table 4.24.

Table 4.24 Regression Analysis Results from One Type of Data.

 

 

 

DATA TYPE STATISTICAL SIGNIFICANCE
R2 F Fsignif
2-LANE SECTIONS .
GEOMETRIC 0.37809 4.55971 0.0011
OBSERVATIONAL 0.02988 0.75449 0.4756
SUBJECTIVE 0.01505 0.37431 0.6897

4-LANE SECTIONS

GEOMETRIC 0.63591 6.48740 0.0002
OBSERVATIONAL 0.26555 5.60420 0.0084
SUBJECTIVE 0.07181 1.19922 0.3150

 

The determinants of statistical significance in Table
4.24 clearly indicate a hierarchy of data type and show that
geometric data are much better predictors of accidents in the

traffic and environmental conditions in Pakistan.

4.5 HAZARD INDEXATION

Several studies [101-104] employing the hazard index (HI)
approach for the identification and ranking of hazardous
locations were reviewed in Chapter 2. In these studies, the HI
was developed using any one of three type of indicators: 1)

purely accident based factors; 2) a combination of accident

149
and non-accident based factors; and 3) purely non-accident
based safety surrogate factors.

It was shown in these studies [i.e., 101-104] that the
composite hazardousness rating of a highway location provided
a more accurate prediction of future accident experience than
any single type of indicator. For example, Taylor and.Thompson
[101] established that restricted sight distance was a
definite factor in hazardousness at a given location but
analysis of sight distance restrictions in themselves did not
provide good estimates of future accident experience.

The practical advantage of the HI approach lies in the
simplicity of computations as compared to the more complex
approach of accident modeling and forecasting. The above
cited, and many other, studies have found that the HI approach
can be used for measuring relative highway safety so that

proper countermeasures could be developed accordingly.

4.5.1 Development of HI for the Present Research

The feasibility of developing hazard indices from the
present data.was examined.and.it was found.that an aggregation
of the three types of data (as described on pages 73 and 159)
would produce better indices than employing a single type of
data set. A detailed description of the development of hazard
indices for the two types of highway sections is presented in

the following pages.

150

4.5.1.1 HI for 2-Lane, z-Iay Sections

The HI were developed for each kilometer of the test
section by computing the arithmetic sum of the hazard data

values and are presented in Table 4.25.

Table 4.25 Hasard Index: 2-Lane 2-Nay Sections.

1482 .00 5.00 180.00 .00 .00 2.00 .00 1.00 34.00 12.00 234.00 .

1483 .00 17.00 500.00 .36 .00 2.00 .00 .00 35.00 7.50 561.86 .00
1484 .00 15.00 784.00 .00 .00 2.00 .00 .00 37.50 4.50 843.00 .33
1485 .00 7.00 412.00 .00 .70 2.00 .00 .00 40.00 45.00 506.70 .33
1486 570.00 16.00 670.00 .00 .00 2.00 .00 .00 42.00 23.00 1323.00 4.33
1487 319.00 24.00 1092.00 .00 .00 2.00 .00 .00 45.00 9.00 1491.00 1.33
1488 980.00 21.00 150.00 .01 .00 2.00 .00 .00 46.50 12.50 1212.01 3.33
1493 142.00 10.00 822.00 1.44 2.00 2.00 .00 .00 38.00 11.00 1028.44 4.33
1494 214.00 23.00 602.00 .00 1.00 2.00 .00 .00 42.00 7.00 891.00 1.67
1495 .00 10.00 .00 .00 1.00 2.00 .00 2.00 54.50 7.00 76.50 .67
1496 310.00 19.00 50.00 .49 .00 2.00 .00 7.00 50.00 46.50 484.99 1.33
1497 100.00 14.00 328.00 .09 .00 2.00 .00 1.00 45.50 44.50 535.09 3.00
1498 50.00 6.00 188.00 .00 1.00 2.00 .00 .00 47.50 4.50 299.00 .00
1499 50.00 7.00 630.00 .49 2.00 2.00 .00 .00 40.00 5.00 736.49 2.00
1500 810.00 23.00 .00 .04 1.00 2.00 .00 1.00 50.50 9.00 896.54 1.33
1501 98.00 9.00 .00 .49 .50 2.00 .00 .00 53.00 8.00 170.99 .67
1502 .00 31.00 399.00 .09 .00 2.00 00 1.00 58.00 5.00 496.09 .33
1503 305.00 .00 150.00 .00 .00 2.00 . .00 61.00 6.50 524.50 1.33
1504 600.00 11.00 200.00 .00 1.00 2.45 17.69 .00 57.00 15.50 904.64 4.00
1505 1597.00 29.00 .00 .04 .50 2.00 .00 .00 43.00 21.50 1693.04 1.33
1506 1040.00 30.00 582.00 .01 1.00 2.00 . .00 49.00 5.00 1709.01 7.00
1507 396.00 13.00 301.00 .01 1.00 2.00 .00 1.00 54.50 10.00 778.51 1.67
1508 1032.00 9.00 284.00 .09 2.00 2.00 . 1.00 62.00 24.50 1416.59 2.67
1509 718.00 16.00 450.00 .01 1.40 2.00 .00 .00 54.50 19.00 1260.91 5.00
1510 660.00 19.00 517.00 .09 .00 2.00 . .00 55.00 33.50 1286.59 5.00
1511 990.00 5.00 .00 .09 .00 2.00 .00 .00 61.50 30.00 1088.59 .67
1512 629.00 17.00 .00 .09 1.00 2.00 .00 .00 54.00 17.00 720.09 1.67
1513 610.00 8.00 .00 .00 .50 2.00 .00 2.00 51.50 8.00 682.00 1.00
1514 821.00 15.00 423.00 .09 . 2.00 .00 .00 39. 7.00 1307.09 3.67
1515 534.00 17.00 195.00 .09 .50 2.00 .00 3.00 45.50 2.00 799.09 2.00
1516 409.00 12.00 50.00 .09 .00 2.00 .00 .00 45.50 16.50 535.09 1.33
1517 710.00 13.00 142.00 .09 1.70 2.00 .00 .00 45.50 6.00 920.29 3.33
1518 .00 5.00 200.00 .00 1.60 2.00 .00 .00 45.50 33.50 287.60 .00
1519 .00 23.00 200.00 .00 1.00 2.32 17.07 .00 42.50 7.50 293.39 .67
1520 523.00 32.00 488.00 .00 1.50 2 00 . .00 42.50 14.50 1103.50 2.00
1521 518.00 21.00 .00 .49 .50 2.00 .00 1.00 36.00 19.00 597.99 3.00
1522 .00 8.00 88.00 .16 1.50 2.00 .00 3.00 36.50 25.50 164.66 .00
1523 913.00 19.00 .00 .00 .00 2.35 23.32 .00 36.00 .50 994.17 1.00
1524 400.00 14.00 .00 .04 1.00 2.00 . .00 43.00 8.5 468.54 1.33
1525 222.00 4.00 266.00 4.84 2.00 2.00 .00 .00 51.50 .00 552.34 .00
1526 . 16.00 285.00 .00 .00 2.00 .00 .00 51.50 1.00 355.50 .

1527 . 7.00 200.00 .04 1.00 2.00 .00 .00 48.50 5.50 . .

1528 755.00 13.00 .00 .00 .50 2.00 .00 .00 43.50 4.50 818.50 2.67
1529 .00 14.00 88.00 .00 .50 2.20 19.55 1.00 40.50 17.50 183.25 .67
1530 170.00 38.00 50.00 .04 .50 2.00 . 2.00 33.50 5.50 301.54 .33
1531 1622.00 25.00 .00 .01 .50 2.00 .00 .00 38.50 9.50 1697.51 5.33
1532 356.00 19.00 100.00 .81 2.00 2 00 .00 .00 27.50 11.50 518.81 3.00
1533 229.00 27.00 212.00 .81 2.00 2.46 .00 .00 26.50 17.50 517.27 .00
1534 1340.00 10.00 .00 1.44 2.00 2.00 .00 .00 23.50 36.00 1414.94 1.67
1535 1499.00 8.00 .00 1.21 3.00 2.62 26.14 .00 33.00 .00 1606.97 1.00
1536 1379.00 8.00 .00 .00 3.00 2.00 .00 .00 39.00 9.50 1440.50 .00

151.
A simple correlation of the individual "hazard" value of
the independent variables; and the composite hazardousness
[HZINDX] (evaluated by aggregating the data) with annual

accident frequency [ANACFQ] is presented in Table 4.26.

Table 4.26 Correlation of Individual Variables and
HI with Accidents (2-Lane Sections).

 

 

Variables [ANACFQ]
[RIBBON] .4365**
[SPATHS] .2398
[GDRAIL] .2682
(PWIDTH) -.1080
[swxora] -.0379
[PMARKS] -.0959
[INTSEC] -.0718
[ISLAND] - 1501
[PVCOND] .0997
[SIDEOB] .0669
[HZINDX] .5907**

 

N of cases: 52
l-tailed Signif: * - .01 ** - .001

 

The results presented in Table 4.26 are the same as those
displayed in the correlation matrix (Table 4.2, page 108)
except that the new variable [HZINDX] is introduced.here. The
magnitude and the sign of the correlation coefficient of
[HZINDX] clearly indicates a substantial improvement in the
relationship by combining all types of data rather than using
them in a segregated form. This finding is distinctively
supported by the results reported by Spark [97] who found that
aggregating the data into a cumulative index significantly
increased (to the 'extent of 45%) the correlation of
explanatory variables with accidents. A detailed review of

this study is included in the literature review (Chapter 2).

152

However, the vital issue that needs to be addressed here is to
assign appropriate weights to the individual hazard indicators
(independent variables). In most of the above quoted studies
of hazard index development, the variables were assigned
weights by subjective rating of a hazard according to its
importance based on engineering and professional judgement. In
the present research, this question is partially answered in
the sense that significant hazard indicators were identified
in accident modeling by regression analysis.

Figure 4.31 is a graphical representation of the

developed HI with the annual accident frequency.

HAZARD INDEX 2-LANE
l l l l l l 1

 

 

 

 

A 7.5~ -
n 1
n

u

a 1
l 5‘ 11 -

1 1

A 1

c l 1 1 R
c 21

. 2.5- 1 1

1 1 1 1

F 112 11 2 1 1 1
r 1 l 1

e 1.21 2 2 1 1

g on 1 211 12 1 -

225 I 675 ' 1125 ' 1575
450 900 1350
Hazard Index
N = 52, Correlation - 0.59074, R2 8 0.34898, Sig = 0.0000

Figure 4.31

A Graphical Representation of hazard Index Vs. Accidents.

4.5.1.2 HI for 4-Lane Divided Sections

153

The 34 indices for each kilometer were developed as

before by computing the arithmetic sum of the hazard data, and

are presented in Table 4.27.

Table 4.27 Hazard Index: 4-Lane Divided Highway Sections.

.00

.00

.00
300.00
300.00
1104.00
799.00
1218.00
439.00
802.00
1639.00
590.00
1344.00
500.00
437.00
198.00
330.00
1456.00
1424.00
566.00
165.00
841.00

4900
sum
«up
1am:
mum
1am:
54»
4mm:
3am
and)
32m:
69m:
34»
8pc
ELM
nun
3mo
memo
9&0
n4»
6J0
34»
sec
emu
1am:
3mm)
smo
1mm:
rim
mum
31m:

.‘3

1.48
2.26

2.34
3.25
1.96

2.29
.26
.08
.05

2.74
.04
.04
.00
.01
.10
.09
.18
.34
.18
.13

3e”
3.00
2.”

2.38
.00
.00

16.31

10.50
23.50
5.00
9.50
31.00
15.00
7.00
5.50
4.50
7.50
12.50
7.50
4.50
4.50
9.50
12.50
5.00
11.00
18.50
9.50
18.50
19.00
7.00
11.50
7.50
5.00
9.00
16.50
8.00
13.00
15.00
6.00
8.50
23.00

1024.15
332.76
124.76
129.00

1059.41
577.13

95.51
139.79
162.98
111.51
990.58
177.32
143.25
152.91
208.51

1903.69

1082.09

1275.33

1102.50

1586.54
846.84
904.59

1769.00
909.51

1911.65
588.99
536.79
357.41
473.43

1660.38

1562.25
900.25
496.50
949.00

3.67
.33
.00
.67
.33
.00
.33

1.00
.33
.33

3.00

1.00
.00
.00

1.00

3.67

2.33

2.33

2.33

5.67

1.33
.00

7.00

1.67

3.67
.67

1.00

154
Table 4.28 presents a simple correlation of the
individual "hazard" value of the independent variables; and
the composite hazardousness [HZINDX] with annual accident

frequency [ANACFQ].

Table 4.28 Correlation of Individual Variables and
HI with Accidents (4-Lane Sections).

 

 

Variables [ANACFQ]
[RIBBON] .6403**
[SPATHS] .1486
[HEDOPN] -.0921
[GDRAIL] .4245*
[PWIDTH] .0743
[SWIDTH] .0813
[PHARKS] .3421
[INTSEC] .4838*
[ISLAND] .0122
[PVCOND] .2648
[SIDEOB] -.1079
[HZINDX] .7570**

 

N of cases: 34

 

Much like the 2-lane case for HI development, it is
observed from the contents of Table 4.28 that for the 4-lane
sections [HZINDX] has a substantially improved relationship
with [ANACFQ] as compared to each of the independent variable
individually. However, the question of assigning appropriate
weights to the individual hazard indicators (i.e, the
independent variables) still remains to be addressed. The
graphical representation of the developed HI with the annual
accident frequency for the 4-lane sections is presented in

Figure 4.32.

155

HAZARD INDEX 4-LANE

 

 

 

 

1 l l l 1 J 1
A 7.54 ~
n 1
n
u
a 1
l 5- -
A R
c 1 2
c l 1
. 2.5+ 2 1 ~
F 12
r 12 1 1
e 22 1 l 1 1 1
q 0-12 1 1 1 1 ~
'3 I I I
250 750 1250 1750
500 1000 1500
Hazard Index
N s 34, Correlation - 0.75703, R2 = 0.57310,
Figure 4.32

Sig - 0.0000

A Graphical Representation of hazard Index Vs. Accidents.

A comparison of the various determinants of statistical
significance for the predictive models and hazard indices is
presented in Table 4.29 which shows the ascendancy of the

predictive modeling using regression analysis over the HI

approach.

Table 4.29 A Comparison of the Predictive Hodeling and HI Approach.

8 a” su
2-Lane 0.59827 0.35792 0.000
4-Lane 0.81211 0.65952 0.000

Ill Approach
R 309
0.59074 0.34898 0.000
0.75703 0.57310 0.000

C1LAPTIH!.5

5.1 PERFORMANCE EVALUATION OF RESEARCH INPUTS AND RESULTS

To embark on a research program concerning highway safety
in Pakistan was challenging in many ways. First and foremost,
there was a lack of an acceptable theoretical foundation on
which the design of a research program could be based. Efforts
were made to bridge the theoretical gaps with proven
postulates in highway safety as evidenced by the literature,
especially from the United States and United Kingdom. In doing
so, several variables evolved as candidates for explaining and
predicting traffic accidents in Pakistan.

Second, there‘was a lack of coherent empirical studies on
the subject of accident prediction modeling. Therefore, the
experimental design was framed ab-initio with some insight on
the topic furnished by Taylor and Thompson's work [101] .

Subsequently, Schoppert's study [95] on accident prediction

156

157

from roadway elements, and the TRRL studies [99] relating
accidents with roadway characteristics in various developing
countries, provided basic guidelines for designing the
analytical framework. As a result, multivariate regressions
were chosen as the predictive modeling technique.

Third, variable acquisition according to an ideal
scholastic design was not always possible. For example, in the
case of independent variables, the intersection hazardousness
was evaluated in terms.of conflicts rather than sight.distance
deficiencies. In the case of the dependent variable, accident
records did not furnish information on accident type that was
required for certain in-depth analysis.

Fourth, the assumption of design deficiency as an
indicator of hazard was seriously violated in some cases due
to absence of an uniform standard. This was particularly true
in the case of traffic signs. Their absence (or presence) made
no meaningful difference since there were no standards to
quantify the deficiency. This was the basic reason that signs
were included in the "PNC" category in the experimental design
and were disregarded in the present study (see pages 69-70).

Since this study was an innovative effort for accident
prediction modeling in Pakistan, and there was no defined
entry point, it was considered appropriate to expose several
procedure for variable acquisition. The intent here was to
acquire data on a number of potential determinants of

accidents to begin with, and use a stochastic approach rather

158

than following a deterministic approach for variable
selection. Later, the influential variables were identified
employing a series of screening processes.

Prior to drawing any conclusion from the analysis, an
evaluation of the performance of the following research inputs
and results is presented in the subsequent pages.

- Evaluation of Independent Variables

- Evaluation of Data Type

- Evaluation of Prediction Models

- Evaluation of Hazard Index

- Evaluation of Accident Data

5.1.1 Evaluation of Independent Variables

The literature [116,117] cautioned that including a large
number of independent variables in a regression model was
never a good strategy unless there were strong, previous
reasons to suggest that they all should be involved.
Accordingly, it was never intended that all the postulated
determinants of accidents would be used as the independent
variables in :model building. It ‘was expected, that. only
appropriate variables (in terms of independency, randomness
and variability) would be retained.

The following three types of procedures were employed for

the quantification of ambient hazardousness:

159

1) Use of design standard deficiencies as a measure of
hazard;

2) Use of drivers erratic maneuvers and traffic
conflicts to evaluate hazard; and

3) Use of an expert team for subjective rating of
ambient hazardousness.

Correspondingly, the following three types of data sets
were generated by employing the above mentioned procedures of
hazardousness quantification:

1) Measurements: evaluating' a potential hazard in
linear dimensions (e.g., lack of access control was
measured in meters/kilometer).

2) Counts: evaluating a potential hazard in number of
events signifying its presence (e.g. , number of
conflicts in an intersection in proportion to
traffic volume, were counted in #lunit time).

3) Numbers: evaluating a potential hazard by a
subjective rating using a scale of 0 to 100 (e.g.,
each kilometer ‘was rated on this scale, zero
signifying the least hazardous).

The analysis showed that the first two procedures (i.e.,
use of design standard deficiencies; and use of drivers
erratic maneuvers and traffic conflicts), produced the best
predictive variables. The use of subjective rating did not

yield appropriate variables for modeling.

160

5.1.2 Evaluation of Data Type

Three type of research data: geometric; observational;
and subjective data, were used independently in regression
analyses. The results of this investigation, for both type of
highway sections, were presented in Table 4.24 (page 148)
which clearly indicated that geometric data were much better
predictors of accidents in the traffic and environmental
conditions in Pakistan.

In the experimental design, the rationale of using a
particular measuring system to quantify a hazard was based on
the applicability of the techniques and available resources.
For example, traffic conflicts were employed to evaluate
intersections.hazardousness because frequent illustrations of
such cases were found in the literature [78,119]. As another
example, pavement condition was evaluated using subjective
rating techniques because the determination of Pavement
Serviceability Rating (PSR) by this approach was found an
established engineering practice [110].

As a result of this study, it is suggested that the
hazardousness evaluated in terms of observational or
subjective data should be measured in the form of geometrical
dimensions. For example, the intersection.deficiencies should
be measured using sight distance data rather than employing
the traffic conflict approach as incorporated in this study.

As another example, roadside obstruction should be measured in

161

terms of lateral placement from the EOS line rather than

subjectively rating their prospective hazardousness.

5.1.3 Evaluation of Predictive models

The predictive models, for both 2-lane, 2-way, and 4-lane
divided sections, indicated control of access as a significant
factor explaining the accident variance. This finding is quite
understandable intuitively, and supported by the results of
, many studies.

Referring back to the equations 2.4 - 2.8 (page 48)
developed by Schoppert [95], it may be seen that the following
variables (highway elements) were significantly associated

with accidents:

1) "CDW", number of commercial driveways per mile;
2) "INT, number of intersections per mile; and

3) "RDW", number of residential driveways per mile;

Similarly, equations 2.12 - 2.14 (page 52) were developed
by Jacobs [99] representing the relationship between accident
rates and geometric design standards in two developing
countries, Kenya and Jamaica. It may be seen that in both the
equations, the accident rate per MVKm was a function of

junctions per km. Moreover, the findings of Kihlberg and Tharp

162
[98] also revealed that access control had the most
significant accident reducing effect.

However, contrary to expectations, it was surprising to
note that the absence of guardrail on warranted sections
emerged as an important factor in the predictive models. This
association, prima-facie, was not supported by a simple
theoretical basis because guardrail by itself may not cause,
or avert, an impending* accident situation. Its role in
reducing severity was quite understandable and substantiated
by the literature [108,109].

Accident data were, therefore, examined to find if the
association of accidents with the absence of guardrail could
be explained by "run-off-the-road” type accidents on high
embankments. It was found that the accident data did not
provide explicit information on the type of accidents. The
only relevant information in this context was in a narrative
form (i.e., "the vehicle fell into ditch"), and revealed that
these locations were different than the one indicated by the
model. The cause of the accidents was mostly described as
"due to driver's negligence". It was, however, verified that
these accident sites were mostly the high embankment
locations. Another possible explanation is that vehicles leave
the road at random locations but the only reportable accident
were for high embankments, while at other locations the
accidents were not reported being less serious.

These findings indicated that driver behavior was, in

163

some manner, related to the absence of guardrail on high
embankment sections. It was, therefore, assumed that, on high
embankments, drivers preferred to keep the transverse position
of their vehicles away from the outer edge of the pavement for
safety reasons and got involved in accidents due to improper
lateral placement conditions. A plausible support to this
assumption was furnished by Taragin’s study [120] which
reported that transverse positions of vehicles were effected
by the presence of objects and barricades on highway
shoulders. Asla corollary, it was inferred.that.the.drivers on
high embankment sections kept the lateral position of their
vehicles away from the outer edge of the pavement. However,
this premise warrants further research. If a linkage between
lateral placement of vehicles and embankment height is
evidenced by empirical studies then operational friction may
be generalized as the principal factor associated ‘with
accidents on high embankments. The analyses also indicated
association of intersection deficiencies with accidents for
the 4-lane sections. For this study, the intersection
deficiencies were quantified in terms of operational hazards
(i.e., conflicts). This finding is intuitively understandable
and supported by the literature [78,105,119].

The detailed analysis of residual and predicted values
for the developed models were presented in Chapter 4. Here,
the performance of the models is evaluated from a different

angle. The limits of the two models are worked out by plugging

164

in the minimum and the maximum values for the variables (i.e. ,

the values a hazard can possibly have on-ground). The results

of this exercise are summarized in Table 5.1.

The annual accident frequency based on the average of

three years of accident data (January 1988-December 1990)

showed 0 and 7 as the minimum and maximum number of accidents

per referenced kilometer.

The results shown in Table 5.1

indicate that the predicted accident range practically covers

the observed limits.

Table 5.1 Model Limits in Accident Prediction.

 

 

2-1m: 2101;590:100
2a, - 0.112287 + 0.001993 x. + 0.002726 2:,

 

 

 

 

 

 

 

 

 

 

VARIABLE Constant Variable # of Accidents
VALUE
x1 x2 Predicted Actual
Minimum 0.112287 0 0 0.11 0
Maximum 0.112287 1500 1500 7.21 5.33
- v i hw

Y“, = 0.012487 + 0.001932 x. + 0.002189 2., + 0.158703 3:,

 

 

 

 

 

 

 

 

 

 

VARIABLE Constant Variable # of Accidents
VALUE
x1 x2 x3 Predicted Actual

; Minimum 0.012487 0 0 0 0.01 0
1
;W ,°-°1_43'L 130° 3 __11-1_2 “’0 ,
2-Lane, 2-Way Section.
Iw I Accident frequency per year per kilometer.
x1 I Length of ribbon development on both sides (meters).
x, I Length of highway section with deficient guardrail (meters).
4-Lane Divided Highway Section.
Ym I Accident frequency per year per kilometer.
x, I Length of ribbon development on both carriageways (meters).
x2 I Length of highway section with deficient guardrail (meters).
x3 I The aggregated ratio of traffic conflicts to the operating volume

in an intersection expressed in percent.

165

5.1.4 Evaluation of Hazard Index

Concurrent to accident modeling by regression analyses,
the feasibility of an aggregated use of the three type of data
was also examined. These analyses were aimed at developing
hazard indices (HI) for identifying and ranking of hazardous
highway sections. The details of this task were presented in
Section 4.5.

The development of a hazard index to examine the
relationship of composite hazardousness rating of a highway
section with accidents indicated a logical feasibility of
employing this approach for the identification and ranking
purposes for the highway and traffic conditions in Pakistan.
The finding that an aggregation of various types of data
produce the best indices rather than employing a single type
of data is substantiated by the literature [97,102]. However,
the shortcomings of the HI approach lie in the arbitrariness
in assigning the weights to the individual indicators of

hazardousness.

5.1.5 Evaluation of Accident Data

As mentioned in Chapter 3, the accident data were
not readily accessible in terms of a computerized data base.
The accident files were created by retrieving information from

the police records for the three-year period: January 1988

166
through December 1990.

The three years period was considered adequate because
longer period could involve changes in physical, environmental
and demographic characteristics, invalidating the results.‘ The
adequacy of this time-frame was supported by the studies by
Renshaw et al. [122], and by Shen [104] which demonstrated
that a three-year accident.data.base'was an appropriate sample
for accident analysis. Schoppert [95] also used a three-year
accident data in developing predictive models for two-lane
highways in the State of Oregon.

However, the details of the accident data were not
adequate to conduct certain in-depth analyses. For example,
accident data could not be segregated by the type of location
(i.e., intersection and non-intersection) and were integrated
for use on a one-kilometer section that included both type of
locations.

The other example of the accident data inadequacy
concerns inability to segregate by the degree of severity
(i..e., fatal, injury, and P.D.O). This limitation ‘was
primarily due to inadequate sample size for each severity

level and did not necessarily concern lack of details.

5.2 ACCIDENT PREDICTION MODELS AND HSIP

The predictive model for 2-lane sections only explained

36% of the variation in accidents. A comparison with the

167
results of the 4-lane analysis, which explained 66% of the
variation in accidents, leads to a higher degree of confidence
in claiming these models as satisfactory prediction tools for
employing in HSIP.

The primary objective of accident prediction modeling was
to provide a basis for the transferability of HSIP technology
to Pakistan. The results of the present research show that
this transfer is accomplishable, and the formal program of
applying accident countermeasures may be initiated even in the

absence of a reliable accident data base.

5.3 HSIP AND POLICY DECISION HAHINC

An important feature of the prediction models is the
consistency in indicating deficient control of access as a
principal factor associated with accidents for both 2-lane and
4-lane sections of the experimental site. This finding is
supported by many highway safety studies for both developed
and underdeveloped countries as discussed thoroughly in
Section 5.1.3 in the preceding pages.

This is a remarkable point to note in the sense that the
HSIP has emerged as an analytical tool for policy making
rather than a simple means for identifying hazardous locations
on the rural highways in Pakistan. Unlike correcting other
hazards, the remedial measures required to improve highway

access control involve social implications in addition to

168
technical and financial feasibility.

For example, the issues of the social cost of relocating
people, and their properties and businesses to improve the
existing network may become a leading policy matter.
Similarly, incorporation of preventative safety measures in
the future design and construction of new facilities may
constitute vital policy issues. This is particularly true
because safety measures oriented toward the improvement of
highway access control would make.a very significant.impact on

the land-use pattern of the country.

CILKPTEHK 6

lull. I,| , .k . ,,.L iJJl'l .a a I_1 1,,3 4i i

6.1 RESEARCH LIMITATIONS

A detailed discussion of the data inputs and the attained
results was presented in the preceding chapter. As is the case
with any other empirical research, the available data placed
limitations on the resulting analysis. The following factors
should be recognized in extrapolating the results of this
study.

1. The three-year accident data base employed in the
research was quite sparse. It furnished an average of 5.29
(=275/52) and 4.38 (=149/34) accidents per km for the 2-lane
and 4-lane sections respectively (see pages 122 and 145 for
details). Out of 86 kms of the experimental site, 16 kms had
no accidents; 12 kms had one; and 7 kms had two accidents.

2. The accident data base was comprised of only
reportable and retrievable accidents. As such, a bias in terms
of accident severity and accident type prevailed that debarred

in-depth analysis based on these aspects.

169

170

3. The developed hazard measurement system (see page 81) ,
employed a predetermined definition of hazard in certain
cases. For example, the absence of guardrail on an embankment
height of 10 ft. or'more was defined to constitute a hazardous
situation. This limitation prevented an analysis of the
relationship between the embankment height and accidents.

4. Hazardousness evaluation based on sufficiency criteria
provided a sound rationale. However, the consistency in the
occurrence of some hazards did not yield good variables for
regression analysis. Absence of pavement markings is the most
relevant example in this context. This hazard was so
consistently present that, in terms of variability, it did.not
produce an appropriate regressor for accident modeling.

These limitations should be considered in perusing the

following prominent findings.

6.2 PRONINENT FINDINGS

The results of the research indicate existence of a
statistical relationship between the hazards and accidents,
and furnish the necessary basis for initiating a formal HSIP
in Pakistan. Based on these results, the null hypothesis
negating a relationship between the identifiable hazards on
rural highway sections in Pakistan and its accident potential
was rejected.

The prominent findings pertain to the following three
categories of interest: 1) The identified hazards; 2)

Implementation of the HSIP; and 3) Policy issues. These

findings,

171

relevant to each category, are summarized in the

following pages.

6.2.1 The Identified Hazards

Inadequate control of access was the most
significant identified hazard associated with
accidents.

Both for 2-lane and 4-lane sections, the 'most
adverse condition of control of access was the
ribbon development.

The lateral openings in the form of specific paths
did not indicate correlate with accidents.

Median openings would have been an additional
potential indicator of accidents for 4-lane
sections, if the criterion of variable entry in
model building were to be relaxed.

Deficient intersection control was a significant
factor for 4-lane sections.

High embankment sections with deficient guardrail
system were associated with accidents.

The absence of pavement marking and roadside
obstructions were the two frequently occurring
hazards that were not associated with accidents.
The deficiencies of pavement and shoulder width,
and inadequate pavement serviceability were not

frequently occurring hazards.

172

Implementation of the HSIP

The transferability of HSIP technology to Pakistan
is accomplishable, and a formal program of applying
accident countermeasures can be initiated on the
country's rural trunklines in the absence of a
reliable accident data base.

Countermeasures oriented toward improving the
control of highway access should receive priority

implementation.

6.2.3 Policy issues

Inadequate control of access on a rural trunklines,
operating at a volume of 10,000-16,000 ADT, warrant
prioritized treatment, preferably in the form of
full control of access.

Providing access control would require substantial
resources to meet the direct costs and social costs
of relocations.

The general land holding pattern of the country,
having integrated interests on both sides of the
existing highway, could inflict serious problems in
providing an access controlled facility.

New highway projects / facilities may be appraised
based on these findings.

173

6.3 SUGGESTED RESEARCH

While presenting a discussion of the performance of
variables and the predictive models in Section 5.1, it was
pointed out that variable acquisition according to an ideal
scholastic design was not always possible for the present
research. For example, accident, records 1did. not furnish
information on accident type that was required for in-depth
analysis.

It is, therefore, the natural and desirable outcome of
this research that accidents may be segregated by the type of
location in terms of intersection and non-intersection
accidents. This is particularly true because the literature
[e.g., 78,101,105] generally describes them as different
entities.

In this research, the assumption of one or more design
deficiencies as an indicator of hazard was seriously violated
in some cases due to the absence of an uniform standard. For
example, the absence (or presence) of traffic signs made no
meaningful difference since there were no standards to
quantify the deficiency. It is suggested that a method to
quantify such cases using some appropriate procedures be
developed for future research.

Use of different techniques and deployment of different
measuring units to quantify ambient hazard is strongly
suggested. For example, intersection deficiencies may be
quantified in terms of sight distance data rather than in

terms of operational maneuvers or conflicts. This strategy

174
would not only provide means to review the present findings
but may also suggest a different analytical approach for
examining the relationship between hazards and accidents. For
example, discriminant analysis might be effectively used for
determining the membership of the suspect sites to the most

appropriate hazard groups in developing a hazardousness Index.

6.4 REMAINING COMPONENTS OF THE HSIP

Besides the Planning Component which constitutes the
major part of the HSIP, the Implementation and Evaluation
components are the two remaining parts for completing a HSIP
(see Tables 2.9-2.11, pages 61-62). A discussion, specific to
the present findings, is presented in the following pages for

the accomplishment of these two remaining components.

6.4.1 Implementation Component

The purpose of the Implementation Component is to design,
schedule, construct and make necessary final adjustments to
the highway safety improvements which were selected in the
Planning Component [71]. For the present research, the
necessity of the following three types of remedial measures
was indicated.

1) Highway access control improvements.

2) Intersection operation improvements.

175

3) Delineation and channelization improvements on high

embankment locations.

The implementation of relevant improvement projects can
be carried out in Pakistan by the national or provincial
highway agencies. The agencies' managerial capabilities and
the knowledge of administrating and monitoring of engineering

projects is assumed.

6.4.2 Evaluation Component

The Evaluation Component is the final step of the HSIP
which must not be ignored. The purpose of the Evaluation
Component is to assess the value of ongoing and completed
projects which result from the Planning and Implementation
Components [71] . Since surrogate measures of safety were
employed in the present research to identify and rank
hazardous sections, a discussion on non-accident based project
evaluation techniques is warranted.

Non-accident based evaluations refer to methods employing
other than frequency or rate of accidents measures to analyze
the HSIP effectiveness. Such analyses are conventionally’made
in terms of various measure of effectiveness (MOE)'. The
possible examples of non-accident based MOEs for the
evaluation of the countermeasures indicated by the present

research are displayed in Table 6.1.

* An HOE is a measurable unit or set of units assigned to each evaluation
objective. The 'data collected in the units of the NOE will allow the
analyst to determine the degree of achievement for that objective [71].

176

Table 6.1 Examples of Non-accident Based MOEs.

 

 

 

 

 

 

 

 

 

No. Type of Countermeasures MOEs I
I i
1. Highway Access Control Traffic conflicts, auto-
Improvements. pedestrian conflicts,
stopping, loading/offloading
maneuvers, traffic control
violations.
2. Intersection Operation Traffic conflicts, erratic
Improvements. vehicle maneuvers, vehicle
speeds. L
3. Delineation and Channelization Traffic conflicts, erratic
Improvements on High Embankment vehicle maneuvers, lateral
Locations. vehicle placement, vehicle
speeds.

 

 

 

 

The non-accident based evaluations are quasi-robust
analyses and should be replaced by accident based evaluations
once reliable accident data becomes available. To maintain
continuity, it is added that HSIP literature [71] describes
four selected plans for evaluating highway safety projects.

1) Before and After Study with Control Sites;

2) Before and After Study;

3) Comparative Parallel Study; and

4) Before, During and After Study.

The cited reference [71] includes a detailed description
of each evaluation plan. Additionally, Laughland et al. [123] ,
OECD [124], and Tarrants et al. [125] offer very wide ranging

perspectives on safety program evaluations.

10.

11.

REFERENCES

Evans, Leonard. Traffic Safety’ and. The Driver. Van
Nostrand Rienhold, New York, 1991.

Haddon, W. Jr. A Logical Framework for Categorizing
Highway Safety Phenomena and Activity. Journal of Trauma,
12 (193-207), 1972.

Goeller, Bruce F. Medeling the Traffic Safety System.
Accident Analysis and Prevention. Volume 1, (167-204),
1969.

Duff, J.T. The Effect of Small Road Improvements on
Accidents. Traffic Engineering and Control, Vol. 13, No.
6, (244-245), October 1971.

IHT. Accident Reduction and Prevention. Highway safety
Guidelines: International Edition. The Institution of
Highways and Transportation. London, U.K. 1990.

Hundal, Sajjad H. Analytical Review of Road and Transport
Statistics 1947-85. Report NTRC-101. National Transport
Research Centre, Planning Commission, Government of
Pakistan, Islamabad. December, 1987.

UNO. Statistical Yearbook for Asia and Pacific-1990.
Economic and Social Commission for Asia and the Pacific,
United.Nations Organization. Bangkok, Thailand. February
1991.

PHD. Annual Expenditure Record. Budget Section, Office of
the Superintending Engineer, Punjab Highway Department.
Rawalpindi, Pakistan. 1991.

Swati, Mohammad S. , and Idris, Mohammad K. Accident Study
for Punjab. Report NTRC-27. National Transport Research
Centre, Planning Commission, Government of Pakistan,
Islamabad. December, 1978.

Swati, Mohammad S., and Khan, Shamim A. Islamabad Road
Accident Analysis. Report NTRC-130. National Transport
Research Centre, Planning Commission, Government of
Pakistan, Islamabad. January, 1990.

Khan, Shamim A. Road Accident Analysis. Report NTRC-137.

National Transport.Research Centre, Planning Commission,
Government of Pakistan, Islamabad. May, 1990.

177

12.

13.

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

178

Ahmad, Zubair. Highway Hazards: Identification and
Ranking. Report NTRC-145. National Transport Research
Centre, Planning Commission, Government of Pakistan,
Islamabad. March, 1991.

SAE. Accident Reconstruction: Human, vehicle and
Environmental Factors. Report SP-814 (a compilation of 15
papers on crashworthiness and accident reconstruction).
Society of Automotive Engineers, Inc. February, 1990.

Peltzman, S. The effects of Automobile Safety Regulation.
Journal of Political Economy 83(4), 1975.

Eshler, J. M. Filtering of Fatal Accident Rates.
Transportation Research Record 643. Transportation
Research Board. Washington, D.C. 1977.

Sivak, M. Society's Aggression Level as a Predictor of
Traffic Fatality Rate. Journal of Safety Research. Volume
14 (93-99), 1983.

Zlatoper, T.J. Regression Analysis of Time Series Data on
Motor vehicle Deaths in the United States. Journal of
Transport Economics and Policy 18. September, 1984.

Hautzinger, H. Regression Analysis of Aggregate.Accident
Data: Some Methodological Considerations and Practical
Experiences. Accident Analysis and Prevention 18, 1986.

Gaudry, M. Responsibility for Accidents: Relevant Results
from the DRAG Model. Report 544, Transport ' Research
Centre. Montreal, Canada. 1987.

Loeb, P.D. The Determinants of Automotive Fatalities.
Journal of Transport Economics and Policy 21 (279-287).
September, 1987.

Cooper, P.J. The Nature and Causes of the 1982 Accident
Casualty Reductions in British Columbia. Transportation
Research Record 1047, (1-9) . Transportation Research
Board. Washington, D.C. 1986.

Schuster, 0.11., and Guilford, J .P. The Psychometric
Prediction of Problem Drivers. Traffic Safety Research
Review 6 (16-20), 1962.

Kuzma, J.W., Daysinger, P.W., Strutz P., and Abbey D.
Non-fatal Traffic Accidents in Relation to Biographical,
Psychological and Religious Factors. Accident analysis
and Prevention 5 (55-64), 1973.

AASHO. A Policy on Geometric Design of Rural Highways.
American Association of State Highway Officials,
Washington, D.C. 1954

25.

26.

27.

28.

29.

30.

31.

32.

33.

34.

35.

36.

37.

38.

179

AASHO. A Policy on Geometric Design of Rural Highways.
American Association of State Highway Officials,
Washington, D.C. 1965.

Automotive Safety Foundation, and 0.8. Bureau of Public
Roads. Traffic Control and Roadway Elements in their
Relationship to Highway Safety. Washington, D.C. 1963.

AASHO. Highway Design and Operational Practices Related
to Highway Safety. American Association of State Highway
Officials, Washington, D.C. 1967.

Nader, Ralph. Unsafe at Any Speed. New York. N.Y. Pocket
Books, 1966.

TRB. Safety'Research for a Changing Highway Environment.
National Research Council, Transportation Research Board.
Special Report 229, (134-160). Washington, D.C. 1990.

Swedish National Board of Urban Planning. The Scaft
Guidelines 1968 - Principles for Urban Planning with
Respect to Road Safety. Stockholm, Sweden. 1968.

Transport and.Road Research Laboratory, Crowthrone. U.K.
Research on Road Safety. Her Majesty's Stationary Office.
U.K. 1963.

Hoback, W.D. Accidents on Oklahoma Turnpikes. Traffic
Quarterly, Volume 13, No. 4 (546-556). October, 1959.

Sabey, B.E, and Staughton,G.C. Interacting Roles of Road
Environment, Vehicle and Road User in Accidents. 5th
International Conference of the International Association
for Accidents and Traffic Medicine. London, U.K. 1975.

Bitzl, F. Traffic Accidents on the Autobahnen.
International Course in Traffic Engineering (Theme III),
Reports 1954.

Michaels, R.M. Human Factors in Highway Safety. Traffic
Quarterly, Volume 15, No. 4 (586-599). October, 1961.

Eckhardt, Paul K., and Flanagan, John C. Role of Roadway
Elements in Pennsylvania Turnpike Accidents. Bulletin
120, Highway Research Board. Washington, D.C. 1956.

Treat, J .R. A study of Precrash Factors Involved in
Traffic Accidents. The HSRI Research Review Vol. 10 (#6) ,
Vol. 11 (#1). Highway Safety Research Institute, Ann
Arbor, Michigan. USA. 1980.

Jacobs, 0.0., and Sayer, I.A. Road Accidents in
Developing Countries. Transport and Road Research
Laboratory, Overseas Unit. TRRL Supplementary Report 807 .
Crowthrone, Berkshire, U.K. 1983.

39.

40.

41.

42.

43.

44.

45.

46.

47.

48.

49.

50.

180

Ergun, G. Effectiveness of Highway Safety Programs in
Developing and Developed Countries. Sabbatical Research
Report, Texas Transportation Institute, Texas A & M
University, College Station, Texas, 1985.

Swati, Mohammad S., and Downing, A. J. Road Accidents in
Pakistan. Report NTRC-64. National Transport Research
Centre, Planning Commission, Government of Pakistan,
Islamabad. January, 1983.

Oglesby, Clarkson H. Highway Engineering. 3rd Edition,
(253-254). John Wiley & Sons, New York. 1975.

Al-Isa, Muhammad S. Developing a Methodology to Predict
High Accident Locations on Rural Highways in Saudi Arabia
Using Speed Distribution Characteristics. Doctoral
Dissertation, Michigan State University, 1984.

Somnemitr, T. Road Traffic Accident Prevention in
Thailand. Report No. 103. Department of Highways,
Ministry of Communications. Bangkok, Thailand. April,
1987.

Swati, Mohammad S. Real Problem of Highway Safety in
Pakistan. Report NTRC-43. National Transport Research
Centre. Islamabad, Pakistan. 1980.

Koshi, M. Road Safety Measure in Japan. Paper Presented
at a Symposium on Human Behavior and Traffic Safety at
the General Motors Research Laboratory, Warren, Michigan.
September, 1984.

Rumar, K. The Role of Perpetual and Cognitive Filters in
Observed Behavior. Human Behavior and Traffic Safety,
(151-165). Plenum Press, New York. 1985.

Balbissi, Adli H. Role of Road User and Roadway
Geometrics in Road Accidents in Jordan. Transportation
Research Record 1270, (42-45). Transportation Research
Board. Washington, D.C. 1990.

Klien, David, and Waller, J.A. Causation, Culpability and
Deterrence in Highway Crashes. US Department of
Transportation. Washington, D.C. 1970.

Shinar, David, and McKnight James A. The Effects of
Enforcement and Public Information on Compliance. Human
Behavior and Traffic Safety, (385-415). Plenum Press,
New York. 1985.

Campbell, M. Earl. A Countdown for Highway Safety.
Traffic Quarterly, Columbia University Press, New York.
Volume XXIV, No.1 (147-159), January, 1970.

51.

52.

53.

54.

55.

56.

57.

58.

59.

60.

61.

62.

63.

64.

181

Blatnik, John A. (Rep). The Need for Highway Safety
Consciousness. Roadside Hazards, The ENO Foundation for
Highway Traffic Control, Inc. Connecticut, 1968.

Haight, F. A. Road Safety: A Perspective and New
Strategy. The Pennsylvania Transportation Institute, USA.
Working Paper 29, 1983.

Goldstein, L.G. Whither Accident Research? Traffic Safety
Research Review, Volume 7, No. 1 (2-4), March, 1963.

Blunden, W. R. Planning the Land Use Transport Complex to
Reduce Exposure to Accidents. Paper Presented to National
Road Safety Committee, Canberra. March, 1972.

Chapman, Roger A. The Concept of Exposure. Accident

4 Analysis and Prevention 5, (95-110), 1973.

Chapman, Roger A. Some Relationships Between Road
Accident Frequencies and Measures of Exposure to Risk.
Doctoral Thesis, University of London. 1971.

Erlander, S., Gustavsson, J ., and Larusson, E. Some
Investigations on the Relationship Between Road
Accidents and Estimated Traffic. Accident Analysis and
Prevention 1(1),(17-65), 1969.

Baker, R. F. The Highway Risk Problem - Policy Issues on
Highway Safety. Wiley, New York. 1971.

Operational Research Inc. Development of Highway Safety
Indicators. Prepared for the U.S. Department of
Transportation (DOT HS-800 472) . Operational Research
Inc. Silver Spring, MD. 1971.

Thrope, J. D. Calculating Relative Involvement Rates in
Accidents Without Determining Exposure. Australian Road
Research Board, 2 (1), 1964.

Carr, B. R. A Statistical Analysis of Rural Ontario
Traffic Accidents Using Induced Exposure Data. Accident
Analysis and Prevention 1(4), (343-357), 1969.

Haight, Frank A. A Crude Frame-Work for By-Passing
Exposure..Journal.of’Safety'Research.2(1), (26-29), 1970.

Taylor, William C., and DeLong, Matthew W. Validation of
the "Innocent Victim" Concept. Report prepared in
cooperation with Michigan Department of Transportation,
MI-OHSP, NHTSA and USDOT. Michigan State University.
March, 1986.

Wilde, J.G.S. The Theory of Risk-Homeostatic:
Implications for Safety and Health. Risk Analysis 2,
(209-255), 1982.

65.

66.

67.

68.

69.

70.

71.

72.

73.

74.

75.

76.

77.

182

Crandall, R.W., Gruenspecht, H.K., Keeler, T.E., and
Lave, L.B. Regulating the Automobile. The Brooking
Institute. Washington, D.C. 1986.

Kanafani, Adib K. University of California, Berkeley.
Transportation Demand Analysis. McGraw-Hill Book Company.
(87-150, 192), 1983.

Bloomquist G. C. A Utility Maximization Model of Driver
Traffic Safety Behavior. Accident Analysis and Prevention
18, (371-375), 1986.

Vercase, John. Factor Analysis of Roadway and Accident
Data. Bulletin 240 (24-32) , Highway Research Board.
Washington, D.C. 1960.

Claes, M.G. A Study of Accident Rates in Belgium.
International Course in Traffic Engineering (Theme III),
Reports 1954.

Hammond Inc. Hammond Discovery World Atlas. New Jersey
07040, 1989.

Zegeer Charles, V. Highway Safety Improvement Program
(HSIP) : User’ s Manual. Goodell-Grivas Inc. Report. FHWA-
TS-81-218. December, 1981.

Hauer, B., and Persaud, B. The Problem of Identifying
Hazardous Locations Using Accident Data. Department of
Civil Engineering, University of Toronto. 1983.

Hauer, E. Selection for Treatment as a Source of Bias in
Before-and-After Studies. Traffic Engineering and
Control. Vol. 20, (418-421), 1980.

Griffin, Lindsay 1., Powers, B., and Mullen, C.
Impediments to the Evaluation of Highway Safety Programs.
University of North Carolina, Highway Safety Research
Center. June, 1975.

Higle, Julia L., and Witkowski, James M. Baysian
Identification of Hazardous Locations. Transportation
Research Record 1185 (24-36) . Transportation Research
Board. Washington, D.C. 1988.

Chira-Chavala T., and Mak, King K. Identification of
Accident Factors on Highway Segments: A Method and
Applications. Transportation Research Record 1068, (52-
58) . Transportation Research Board. Washington, D.C.
1986.

Maher, M.J., and Mountain, L.J. The Identification of
Accident Blackspots: A Comparison of Current Methods.
Accident Analysis and Prevention. Volume 20, No. 2

(143-151), 1988.

78.

79.

80.

81.

82.

83.

84.

85.

86.

87.

88.

89.

183

Perkins, Stuart R., and Harris, Joseph I. Traffic
Conflict Characteristics - Accident Potential at
Intersections. Highway Research Record 225. Highway
Research Board. Washington, D.C. 1968.

Pahl, Juergen. A Comparison of Direct and Indirect
Methods for Determining Accident Potential. Accident
Analysis and Prevention. Vol 2, (201-207), 1970.

Datta, Tapan K., Perkin, D.D., Taylor, J.I., and
Thompson, H.T. Accident Surrogates for Use in Analyzing
Highway Safety Hazards. Report No. FHWA-RD-82-104,
Federal Highway Administration, U.S. Department of
Transportation. 1983.

Glauz, William D., and Migletz, Donald J. Application of
Traffic Conflict Analysis at Intersections. National
Cooperative Highway Research Program Report 219.
Transportation Research Board. Washington, D.C. February,
1980.

Taylor, William C. Speed Zoning: A Theory and its Proof.
Traffic Engineering 35 (4), January, 1965.

Caples, G.B., and. Vanstrum, R.C. The Price of ‘Not
Walking. Minnesota Mining and manufacturing Company .
September, 1969.

Heimbach, C.L., Allred, P.W., Atkins, J.R., and Hooper,
J .W. Developing Traffic Flow Indices for the Detection of
High Accident Potential Highways in North Carolina. North
Carolina State University, October, 1968.

Rockwell, T.H., and Treiterer J. Sensing and
Communication Between Vehicles. NCHRP Report 51. National
Cooperative Highway’ Research Program. Transportation
Research Board. Washington, D.C. 1868.

Blensly, R.C., and Head, J.A. Shoulders and Accident
Experience on Two-Lane Rural Highways: A Summary.
Bulletin 266, Highway Research Board. Washington, D.C.
1960.

Stohner, W.A. Relation of Highway Accidents to Shoulder
Width on Two-Lane Highways in New York State. Highway
Research Board. Proceedings, Vol. 35, 1956.

Raff, Morton S. Interstate Highway - Accident Study.

Bulletin 74, Highway Research Board. Washington, D.C.

1953.

Belmont, D.M. Accidents Versus Width of Paved Shoulders
on California Two-Lane Tangents - 1951 and 1952. Bulletin
117, Highway Research Board. Washington, D.C. 1956.

90.

91.

92.

93.

94.

95.

96.

97.

98.

99.

100.

101.

102.

184

Musick, James V. Effect of Pavement Edge Marking on Two-
Lane Rural State Highways in Ohio. Bulletin 266, Highway
Research Board. Washington, D.C. 1960.

Williston, Robert M. Effect of Pavement Edge Marking on
Operator Behavior. Bulletin 266, Highway Research Board.
Washington, D.C. 1960.

Taylor, William C., and Foody, Thomas J. Ohio's Curve
Delineation Program - An Analysis. Traffic Engineering.
Vol. 36, No. 9, (41-45), June, 1966.

Agent, Kenneth R. , and Dean, Robert C. Relationship
Between Roadway Geometrics andeccidents. Transportation
Research Record 541, (1-11). Transportation Research
Board. Washington, D.C. 1975.

Hillier, J.A., and Wardro , J.G.‘Effect of Gradient and
Curvature on London-B rmingham. Motorway. Traffic
Engineering and Control Vol. 7, No.10, (617-621),
February, 1966.

Schoppert, David W. Predicting Traffic Accidents from
Roadway Elements of Rural Two-Lane Highways With Gravel
Shoulders. Bulletin 158, Highway Research Board.
Washington, D.C. 1957.

Dart, Olin K Jr., and Mann, Lawrence Jr. Relationship of
Rural Highway Geometry to Accident Rates in Louisiana.
Highway Research Record 312. Highway Research Board,
Washington, D.C. 1970.

Sparks James W. The Influence of Highway Characteristics
on Accident Rates. Public Works, Vol. 99, No. 3, March,
1968.

Kihlberg, J. K., and Tharp, K. J. Accident Rates as
Related to Design Elements of Rural Highways. NCHRP
Report 47, 1968.

Jacobs, G. D. A Study of Accident Rates on Rural Roads in
Developing Countries. Report 732. Overseas Unit,
Transport and Road Research Laboratory. Crowthrone,
Berkshire, U.K. 1978.

Frantzeskakis, John M. Accident Analysis on Two Non-
Controlled Access National Highways in Greece. ITE
Journal, Vol.53, No. 2 (26-32), February, 1983.

Taylor, J.I., and Thompson, H.T. Identification of
Hazardous Locations - Final Report. Report No.
FHWA-77-82, Federal Highway Administration, U.S.
Department of Transportation. December, 1977.

Ghara beh, Fouad, A. Identification of Accident Prone
Locat ons in Greater Amman. Transportation Research
Record 1318, (70-74). Transportation Research Board.
Washington, D.C. 1991.

103 O

104.

105.

106.

107.

108.

109.

110.

111.

112.

113.

185

Al-Isa, M.S., Ergun, G., Al-Senan, S., and Al-Zahrani A.
Development of a Hazardousness Index for Urban
Intersections. Presented at the 3rd IRF Middle East
Regional Meeting, Riyadh, Saudi Arabia, 1988.

Shen, Lon-Li David. Development of Highway Accident
Hazard Index. Journal of Transportation Engineering.
American Society of Civil Engineers. Volume 112, No. 5,
(447-464), September, 1986.

Zegeer, Charles V. Methods for Identifying Hazardous
Locations. National Cooperative Highway Research Program,
Synthesis of Highway Practice 128. Transportation
Research Board. Washington, D.C. November, 1986.

AASHTO. Highway Design and Operational Practices Related
to Highway Safety. 2nd. Edition. Association of American
State Highway and Transportation Officials (AASHTO) .
Washington, D.C. 1974.

FHWA. Highway Safety Standard 12 on Highway Design,
Construction and Maintenance . Federa 1 Highway
Administration, U.S. Department of Transportation.
Washington, D.C. 1977.

AASHTO. Roadside Design Guide. Association of American
State Highway and Transportation Officials (AASHTO) .
Washington, D.C. 1988.

Bronstad, M.E., Michie, J.D., and Mayer, J.D. JR.
Performance of Longitudinal Traffic Barriers. NCHRP
Report 289. National Cooperative Highway Research
Program. Transportatioanesearch Board. Washington, D.C.
November, 1987.

Carey, W.N. , and Irick, P.E. The Pavement Serviceability-
Performance Concept. Highway Research Record 250. Highway
Research Board. Washington, D.C. 1960.

Allsop, R.E. Variance of Regression Estimates of Expected
Accident Frequencies. Proceedings: Symposium on the Use
of Statistical Methods in the Analysis of Road Accidents,
Road Research Laboratory Crowthrone, U.K. Published by
OECD, 1970.

Gustavsson J. On the Use of Regression Models in the
Study of Road Accidents. Proceedings: Symposium on the
Use of Statistical Methods in the Analysis of Road
Accidents, Road Research Laboratory Crowthrone, U.K.
Published by OECD, 1970.

Morlat, G. , and Ropars, A. Multiple Regression Applied to
the Growth of Road Accidents. Proceedings: Symposium on
the Use of Statistical Methods in the Analysis of Road
Accidents, Road Research Laboratory Crowthrone, U.K.
Published by OECD, 1970.

114.

115.

116.

117.

118.

119.

120.

121.

122.

123.

124.

125.

186

Fernandez-Giro, J. S-G. Accidents Involving Casualties:
Simple Correlations and Regression Planes. Proceedings:
Symposium on the Use of Statistical Methods in the
Analysis of Road Accidents, Road Research Laboratory
Crowthrone, U.K. Published by OECD, 1970.

Cleveland, Donald E., and Kitamura, Ryuichi. Macroscopic
Modeling of 2-Lane Rural Roadside Accidents.
Transportation Research Record 681. Transportation
Research Board. Washington, D.C. 1978.

Norusis, Marija J. SPSS/PC+ Statistics 4.0. Statistics
Guide and Program Manual for the IBM PC/XT/AT and PS/2.
SPSS Inc. Chicago, Illinois 60611. 1990.

Draper, N.R., and Smith H. Applied Regression Analysis.
John Wiley & Sons. New York, 1981.

Bakht, M. Traffic Count Data (data communicated by mail) .
National Transport Research Centre, Planning Commission,
Government of Pakistan, Islamabad. May, 1990.

Campbell, R.E., and King, L.B. The Traffic Conflicts
Technique Applied to Rural Intersections. Accident
Analysis and Prevention (209-221), Vol. 2, 1970.

Taragin, A. Driver Behavior as Affected by Objects on
Highway Shoulders. Proceedings 34; (453-472). Highway
Research Board. Washington, D.C. 1955.

Jorol, Nils H. Lateral Vehicle Placement as Affected by
Shoulder Design on.Rural Idaho Highways. Proceedings 41;
(415-432). Highway Research Board. Washington, D.C. 1962.

Renshaw, D.L., and Carter, E.C. Highway Hazard Location
Identification in the Baltimore County Road Rating
Project. Paper Presented at the 59th Transportation
Research Board Meeting. Washington, D.C. 1980.

Laughland, John. C., Haefner, L.B., Hall, J.W., and
Clough, D.R. Methods for Evaluating Highway Safety
Improvements. NCHRP Report 162. National Cooperative
Highway Research Program. Transportation Research Board.
Washington, D.C. November, 1975.

OECD. Methods for Evaluating Road Safety .Measures.
Organization for Economic Co-Operation and Development.
Road Research Group, Paris. June, 1981.

Tarrants, W.E., and Veigel, C.H. The Evaluation of
Highway Traffic Safety Programs. Report DOT HS 802-525.
US Department of Transportation. Washington, D.C.
February, 1978.

APPENDICES

Appendix A

Financial Statement for the Removal of Blackspots
An Example: Rawalpindi Civil Division, Punjab.
(1980-1987)

Financial Statement for the Removal of Blackspots in

187

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Rawalpindi Division. Source: [PHD].

Year Allocation (Rs) Expenditure
1981-82 332,660 315,306
1982-83 343,400 283,739
1983-84 356,870 330,679
1984-85 50,000 49,876
1985-86 536,000 549,030
1986-87 Discontinued --

000000

500000
‘2
0
C

3 400000
L
3
O
B

5 300000
6
5

2 200000
0
9
0:

100000

0

1981-82 1983-84 1985-86
1982-83 1984-85 1988-87
FMawMiYmr

 

[m ms... Expendihn

 

 

 

 

A Graphical Representation of the above Financial Statement.

Appendix B

Some Photographs of the Experimental Site
Showing the Selected Hazards

188

 

 

 

 

 

Deficient Access Control - Ribbon Development: Hazardousness
Evaluated by non [RIBBON].

 

 

Deficient Access Control - Specific Paths: Hazardousness
Evaluated by won [spares].

 

Deficient Access Control - Median Openings: Hazardousness
Evaluated by MOH [MEDOPN].

 

 

 

Deficient Guardrail: Hazardousness Evaluated by MOH [GDRAIL].

190

 

 

 

Deficient Pavement Width: Hazardousness Evaluated by MOH
[PWIDTH] .

 

 

 

Deficient Shoulder Width: Hazardousness Evaluated by MOH
[swrm] .

191

 

 

 

Deficient Pavement Markings: Hazardousness Evaluated by MOH
[PMARKS

 

 

 

Intersection Deficiencies: Hazardousness Evaluated by MOH
INTS

).

 

 

 

Isolated Lane Deficiencies: Hazardousness Evaluated by MOH
[ISLAND

 

 

 

Pavement Condition Rating: Hazardousness Evaluated by MOH

193

 

 

 

Roadside Obstructions: Hazardousness Evaluated by MOH [SIDEOB].

 

 

A Highway Section Having no Hazards that were Included in this
Study.

Appendix C

The Hazard Data and the Quantified Hazardousness
Using the Developed MOH

194

I. [ACCESS]: DEFICIET W N AGES!

Unit of Measurement: meters per kilometer

2-Lane, Z-Way Sections

[RIBBON] I Sum (Ribbon Development)

[SPATHS] = Sum (Specific Paths)

Specific Paths

Ribbon Development

m.

(meters)

(meters)

[SIRIUS]

Right

Left

Left

Right

mwmmmwmmmowwwwmwwwwmwmwmmmwmmwmmmwmmwwmmmmwmmwmmwmmm

3 122ml 111 121

mammmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmm

1536 956660 4629 6 9663 7 539796 3 z 66 6 669162656
2 al slal slal al al slel sl el slel el

mmmmmmmmmmmo mmmmmmmwmmmmmmmmwmmmmmmmmmmmmmmmmmmmmmmm

8 67779527 9. .5“9.e-Da 102566252560355029050607709655656
al elalal. alalalel

mmmmmmmmommmmommmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmm

m m:m mmmmmn mam. .mmw m:mm mmmm mm mam m memmmmm
mmmmmmmmmmmmmmmmmmmmmmmwmmmmmmmmmmmmmmmmmmmmmmmmwmmm
m mmmmm mm m m...-mmmmmmmmmm mm 3mm m -mmmmmm
mammmmmmmmmmmmmmmmmmmmmmmmmmmmmmwmmmmmmmmmmmmmmmmmam
m mmmmm m mmm mmmmzmmmmmmmmmm mm m m m mmmmmmm
mmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmm

195

II mm: KFICIEIT mu. SEC?!“
Unit of Measurement: meters per kilometer

2-Lane, 2-Uay Sections

[ORML] I Sill Deficient Guardrail Sections]

Km. Deficient Guardrail
Sections
(meters)
Left Right mum
1681 212.00 333.00 565.00
1682 100.00 80.00 180.00
1683 250.00 250.00 500.00
1686 367.00 617.00 786.00
1685 216.00 198.00 612.00
1686 350.00 320.00 670.00
1687 510.00 582.00 1092.00
1688 110.00 60.00 150.00
1693 396.00 626.00 822.00
1696 328.00 276.00 602.00
1695 .00 .00 .00
1696 .00 50.00 50.00
1697 162.00 186.00 328.00
1698 .00 188.00 188.00
1699 330.00 300.00 630.00
1500 .DD .00 .00
1501 .00 .00 .00
1502 205.00 196.00 399.00
1503 100.00 50.00 150.00
1506 100.00 100.00 200.00
1505 .00 .00 .00
1506 362.00 260.00 582.00
1507 207.00 96.00 301.00
1508 188.00 96.00 286.00
1509 200.00 250.00 650.00
1510 322.00 195.00 517.00
1511 .00 .00 .00
1512 .00 .00 .00
1513 .00 .00 .00
1516 217.00 206.00 623.00
1515 100.00 95.00 195.00
1516 .00 50.00 50.00
1517 50.00 92.00 162.00
1518 100.00 100.00 200.00
1519 100.00 100.00 200.00
1520 280.00 208.00 688.00
1521 .00 .00 .00
1522 .00 88.00 88.00
1523 .00 .00 .00
1526 .00 .00 .00
1525 70.00 196.00 266.00
1526 100.00 185.00 285.00
1527 50.00 150.00 200.00
1528 .00 .00 .00
1529 .00 88.00 88.00
1530 .00 50.00 50.00
1531 .00 .00 .00
1532 100.00 .00 100.00
1533 66.00 168.00 212.00
1536 .00 .ll) .00
1535 .00 .00 .00
1536 .00 .00 .00

 

196

In (Pinon): DEFICIEII’ PAVBEIT mote
Unit of Measuruent: meters per kilometer
2-Lane, 2-Uay Sections
[PUIDTH] . (Variationf.

Variation I (3.65 x Dimer of lanes - Pvt. Uidth).
A negative variation signifies pavement Hider than specified.

Km. Pvt. Uidth Variation 0|!le
(meters) (meters)
1681 7.30 .00 .00
1682 7.30 .00 .00
1683 7.90 - .60 .36
1686 7.30 .00 .00
1685 7.30 .00 .00
1686 7.30 .00 .00
1687 7.30 .00 .DD
1688 7.20 .10 .01
1693 8.50 -1.20 1.66
1696 7.30 .00 .00
1695 7.30 .00 .00
1696 8.00 - .70 .69
1697 7.00 .30 .09
1698 7.30 .00 .00
1699 8.00 .70 .69
1500 7.50 - .20 .06
1501 8.00 - .70 .69
1502 7.00 .30 .09
1503 7.30 .00 .00
1506 7.30 .00 .00
1505 7.10 .20 .06
1506 7.20 .10 .01
1507 7.60 - .10 .01
1508 7.00 .30 .09
1509 7.20 .10 .01
1510 7.00 .30 .09
1511 7.00 .30 .09
1512 7.00 .30 .09
1513 7.30 .00 .00
1516 7.00 .30 .09
1515 7.00 .30 .09
1516 7.00 .30 .09
1517 7.00 .30 .09
1518 7.30 .00 .00
1519 7.30 .00 .00
1520 7.30 .00 .00
1521 8.00 -.70 .69
1522 7.70 - .60 .16
1523 7.30 .00 .00
1526 7.50 - .20 .06
1525 9.50 -2.20 6.86
1526 7.30 .00 .00
1527 7.10 .20 .06
1528 7.30 .00 .00
1529 7.30 .00 .00
1530 7.10 .20 .06
1531 7.20 .10 .01
1532 8.20 - .90 .81
1533 8.20 - .90 .81
1536 8.50 -1.20 1.66
1535 13.50 1.10 1.21
1536 11 .00 - .05 .00

 

197

ll MDTII]: DEFICIEII’ sauna UlDI’I
Unit of Measurement: meters per kilometer
2-Lane, 2-Uay Sections
[SUIDTH] = Slli Deficiency (+ values only”.

Deficiency - (3.00 - Shoulder Uidth).
A negative deficiency signifies shoulder wider than specified.

Km. Shoulder Uidth Deficiency
(meters) (meters)
Left Right Left Right [SUIDTIIJ
1681 3.00 3.00 .00 .00 .00
1682 3.00 3.00 .00 .00 .00
1683 3.20 3.20 -.20 -.20 .00
1686 3.00 3.00 .00 .00 .00
1685 2.30 3.20 .70 -.20 .70
1686 3.60 3.60 -.60 -.60 .00
1687 3.00 3.00 .00 .00 .00
1688 3.10 3.00 -.10 .00 .00
1693 2.00 2.00 1.00 1.00 2.00
1696 2.50 2.50 .50 .50 1.00
1695 2.50 2.50 .50 .50 1.00
1696 3.00 3.00 .00 .00 .00
1697 3.00 3.00 .00 .00 .00
1698 2.00 3.00 1.00 .00 1.00
1699 1.50 2.50 1.50 .50 2.00
1500 2.50 2.50 .50 .50 1.00
1501 2.50 3.00 .50 .00 .50
1502 3.00 3.00 .00 .00 .00
1503 3.00 3.00 .00 .00 .00
1506 2.50 2.50 .50 .50 1.00
1505 3.50 2.50 ~.50 .50 .50
1506 2.50 2.50 .50 .50 1.00
1507 2.50 2.50 .50 .50 1.00
1508 2.00 2.00 1.00 1.00 2.00
1509 2.30 2.30 .70 .70 1.60
1510 3.00 3.00 .00 .00 .00
1511 3.00 3.00 .00 .00 .00
1512 3.00 2.00 .00 1.00 1.00
1513 2.50 6.00 .50 -1.00 .50
1516 3.00 3.00 .00 .00 .00
1515 3.00 2.50 .00 .50 .50
1516 3.00 3.20 .00 -.20 .00
1517 2.00 2.30 1.00 .70 1.70
1518 2.20 2.20 .80 .80 1.60
1519 3.00 2.00 .00 1.00 1.00
1520 2.50 2.00 .50 1.00 1.50
1521 3.00 2.50 .00 .50 .50
1522 2.50 2.00 .50 1.00 1.50
1523 3.00 3.00 .00 .00 .00
1526 3.00 2.00 .00 1.00 1.00
1525 2.00 2.00 1.00 1.00 2.00
1526 3.00 3.50 .00 -.50 .00
1527 3.00 2.00 .00 1.00 1.00
1528 3.00 2.50 .00 .50 .50
1529 3.00 2.50 .00 .50 .50
1530 3.00 2.50 .00 .50 .50
1531 ' 2.50 3.00 .50 .00 .50
1532 2.00 2.00 1.00 1.00 2.00
1533 2.00 2.00 1.00 1.00 2.00
1536 2.00 2.00 1.00 1.00 2.00
1535 1.50 1.50 1.50 1.50 3.00
1536 1.50 1.50 1.50 1.50 3.00

 

198

mm m: DEFICIEIT PAVBEM! MRI!“
Unit of Measurement: Kilometers per kilometer

2-Lane, Z-Uay Sections

[PMARKS] I Slll Deficient Length x Height] x 1/1000
Heights: 1 I for missing edgeline on one side;
2 I for missing etbeline on both sides: and
3 I for missing centerline and intersection areas.
Km. Deficient Length km. Deficient Length
(Longitudinal Sections) (Intersection Areas)
~1-side 2-sides c-line Sue DMARKS]
1681 .00 1,000.00 1,000.00 1681 5000.00 5.00
1682 .00 1,000.00 1,000.00 1682 5000.00 5.00
1683 .00 1,000.00 1,000.00 1683 5000.00 5.00
1686 .00 1,000.00 1,000.00 1686 5000.00 5.00
1685 .00 1,000.00 1,000.00 1685 5000.00 5.00
1686 .00 1,000.00 1,000.00 1686 5000.00 5.00
1687 .00 1,000.00 1,000.00 1687 5000.00 5.00
1688 .00 1,000.00 1,000.00 1688 5000.00 5.00
1693 .00 1,000.00 1,000.00 1693 5000.00 5.00
1696 .00 1,000.00 1,000.00 1696 5000.00 5.00
1695 .00 1,000.00 1,000.00 1695 5000.00 5.00
1696 .00 1,000.00 1,000.00 1696 5000.00 5.00
1697 .00 1,000.00 1,000.00 1697 5000.00 5.00
1698 .00 1,000.00 1,000.00 1698 5000.00 5.00
1699 .00 1,000.00 1,000.00 1699 5000.00 5.00
1500 .00 1,000.00 1,000.00 1500 5000.00 5.00
1501 .00 1,000.00 1,000.00 1501 5000.00 5.00
1502 .00 1,000.00 1,000.00 1502 5000.00 5.00
1503 .00 1,000.00 1,000.00 1503 5000.00 5.00
1506 .00 1,000.00 1,000.00 1506 Chakual Y-Jmction 650.00 6350.00 6.35
1505 .00 1,000.00 1,000.00 1505 5000.00 5.00
1506 .00 1,000.00 1,000.00 1506 5000.00 5.00
1507 .00 1,000.00 1,000.00 1507 5000.00 5.00
1508 .00 1,000.00 1,000.00 1508 5000.00 5.00
1509 .00 1,000.00 1,000.00 1509 5000.00 5.00
1510 .00 1,000.00 1,000.00 1510 5000.00 5.00
1511 .00 1,000.00 1,000.00 1511 5000.00 5.00
1512 .00 1,000.00 1,000.00 1512 5000.00 5.00
1513 .00 1,000.00 1,000.00 1513 5000.00 5.00
1516 .00 1,000.00 1,000.00 1516 5000.00 5.00
1515 .00 1,000.00 1,000.00 1515 5000.00 5.00
1516 .00 1,000.00 1,000.00 1516 5000.00 5.00
1517 .00 1,000.00 1,000.00 1517 5000.00 5.00
1518 .00 1,000.00 1,000.00 1518 5000.00 5.00
1519 .00 1,000.00 1,000.00 1519 Check Deli Y-Jmction 320.00 5960.00 5.96
1520 .00 1,000.00 1,000.00 1520 5000.00 5.00
1521 .00 1,000.00 1,000.00 1521 5000.00 5.00
1522 .00 1,000.00 1,000.00 1522 5000.00 5.00
1523 .00 1,000.00 1,000.00 1523 lslmbed Y-Jmction 350.00 6050.00 6.05
1526 .00 1,000.00 1,000.00 1526 5000.00 5.00
1525 .00 1,000.00 1,000.00 1525 5000.00 5.00
1526 .00 1,000.00 1,000.00 1526 5000.00 5.00
1527 .00 1,000.00 1,000.00 1527 5000.00 5.00
1528 .00 1,000.00 1,000.00 1528 5000.00 5.00
1529 .00 1,000.00 1,000.00 1529 Sihala Y-Jmction 200.00 5600.00 5.60
1530 .00 1,000.00 1,000.00 1530 5000.00 5.00
1531 .00 1,000.00 1,000.00 1531 5000.00 5.00
1532 .00 1,000.00 1,000.00 1532 5000.00 5.00
1533 .00 1,000.00 1,000.00 1533 Pauli fds. V-Jmction 660.00 6380.00 6.38
1536 .00 1,000.00 1,000.00 1536 5000.00 5.00
1535 .00 1,000.00 1,000.00 1535 Ayib Park T-Jmction 270.00 5810.00 5.81
1535 1535 Attock Dil T-Jmction 350.00 1050.00 1.05
1536 .00 1,000.00 1,000.00 1536 5000.00 5.00

 

159$)

52. .... .392... ._ is

co.._..ou .... . u:

8:2 .283 . «u

oc_ano.a .2 a a:

scucm 3.... . me

~..- nsoo. noun. none. none. :o...:=..h ..o ...... c~ ~n ms ~nu nun.
.o.. mace. Fnoo. ~ooo. s.~o. co...c=..c a... aax< . o a~ uo~. an“.
an... memo. ~o.o. «one. .ooo. an.aucaa.» ...:.a 63 um gs can o~m—
~n.n~ npuo. o.no. no... nose. co..oc:..» >.:gu.= u........ on on on ~n.. nnm.
~o.~. no... na~o. mo~o. ~nho. c...u:=s-» .... ...;o o~ on us «no opm.
oo.~. puma. onwo. acne. ~roo. c....==s.» ..ag.gu an ac ~s .s_. «on.

c....o.3 .gn». gag». .gn’. .gn,.
”numb... .~u<»a_. ndoopz_. noun»... "—.m»a__ ca........c. u. «a a: up ..2

.2952: + 38;: e "camps: 6 pumps: e 88;:
Jose‘s 533...»... .6 83.326 .3. 7

4323363.»): ... 8:25.966): a 2932:

.333 a 38.—...:

.mp3 a 38.—z:

622. . Camp-.8

.c...o.m >.:-~ ..c...~
.52. ..ea m 3:89:83: .6 :5

3:3»; Bum—.38 a: "fight: :2
820332.:— ot: >823 aubu< 325.0 cum..." pg: "Samba: :9.
02:85 25:53 3303135 0»: >55 «Camps: :2
a—s—u—ug Bung;— ufi-wuhl: I!

200

u um): locum I.“ MFICIEICII
ID" [SPDCHG]: ABRUPT SPEED CHANGE
Mil [LANCHG]: ABRWT LANE CHANGE

Unit of Measurement: 8 par kilometer (weighted)
Z-Lane, Z-Uay Sections

USLAMD] ' [SPDCHG] + [LANCHG].
[SPDCHG] I (Lou ASCX‘I) * (Medit- ASCXZ) * (High A8616”.
[LANCHG] I (LOH ALCX‘I) '1’ (Mediu- ALCXZ) 1' (High ALDEN.

km. Abrupt Speed Change Abrupt Lane Change
LOH Medi In M i gh [SPDCHG] LOH Medi in Ii i d! [LANCHG] [ISLAD]

1681 .00 .00 .00 .00 1681 .00 .00 .00 .00 .00
1682 .00 .00 .00 .00 1682 1.00 .00 .00 1.00 1.00
1683 .00 .00 .00 .00 1683 .00 .00 .00 .00 .00
1686 .00 .00 .00 .00 1686 .00 .00 .00 .00 .00
1685 .00 .00 .00 .00 1685 .00 .00 .00 .00 .00
1686 .00 .00 .00 .00 1686 .00 .00 .00 .00 .00
1687 .00 .00 .00 .00 1687 .00 .00 .00 .00 .00
1688 .00 .00 .00 .00 1688 .00 .00 .00 .00 .00
1693 .00 .00 .00 .00 1693 .00 .00 .00 .00 .00
1696 .00 .00 .00 .00 1696 .00 .00 .00 .00 .00
1695 1.00 .00 .00 1.00 1695 1.00 .00 .00 1.00 2.00
1696 1.00 1.00 .00 3.00 1696 2.00 1.00 .00 6.00 7.00
1697 .00 .00 .00 .00 1697 1.00 .00 .00 1.00 1.00
1698 .00 .00 .00 .00 1698 .00 .00 .00 .00 .00
1699 .00 .00 .00 .00 1699 .00 .00 .00 .00 .00
1500 1.00 .00 .00 1.00 1500 .00 .00 .00 .00 1.00
1501 .00 .00 .00 .00 1501 .00 .00 .00 .00 .00
1502 1.00 .00 .00 1.00 1502 .00 .00 .00 .00 1.00
1503 .00 .00 .00 .00 1503 .00 .00 .00 .00 .00
1506 .00 .00 .00 .00 1506 .00 .00 .00 .00 .00
1505 .00 .00 .00 .00 1505 .00 .00 .00 .00 .00
1506 .00 .00 .00 .00 1506 .00 .00 .00 .00 .00
1507 .00 .00 .00 .00 1507 1.00 .00 .00 1.00 1.00
1508 .00 .00 .00 .00 1508 1.00 .00 .00 1.00 1.00
1509 .00 .00 .00 .00 1509 .00 .00 .00 .00 .00
1510 .00 .00 .00 .00 1510 .00 .00 .00 .00 .00
1511 .00 .00 .00 .00 1511 .00 .00 .00 .00 .00
1512 .00 .00 .00 .00 1512 .00 .00 .00 .00 .00
1513 1.00 .00 .00 1.00 1513 1.00 .00 .00 1.00 2.00
1516 .00 .00 .00 .00 1516 .00 .00 .00 .00 .00
1515 .00 1.00 .00 2.00 1515 1.00 .00 .00 1.00 3.00
1516 .00 .00 .00 .00 1516 .00 .00 .00 .00 .00
1517 .00 .00 .00 .00 1517 .00 .00 .00 .00 .00
1518 .00 .00 .00 .00 1518 .00 .00 .00 .00 .00
1519 .00 .00 .00 .00 1519 .00 .00 .00 .00 .00
1520 .00 .00 .00 .00 1520 .00 .00 .00 .00 .00
1521 1.00 .00 .00 1.00 1521 .00 .00 .00 .00 1.00
1522 1.00 .00 .00 1.00 1522 .00 1.00 .00 2.00 3.00
1523 .00 .00 .00 .00 1523 .00 .00 .00 .00 .00
1526 .00 .00 .00 .00 1526 .00 .00 .00 .00 .00
1525 .00 .00 .00 .00 1525 .00 .00 .00 .00 .00
1526 .00 .00 .00 .00 1526 .00 .00 .00 .00 .00
1527 .00 .00 .00 .00 1527 .00 .00 .00 .00 .00
1528 .00 .00 .00 .00 1528 .00 .00 .00 .00 .00
1529 .00 .00 .00 .00 1529 1.00 .00 .00 1.00 1.00
1530 1.00 .00 .00 1.00 1530 1.00 .00 .00 1.00 2.00
1531 .00 .00 .00 .00 1531 .00 .00 .00 .00 .00
1532 .00 .00 .00 .00 1532 .00 .00 .00 .00 .00
1533 .00 .00 .00 .00 1533 .00 .00 .00 .00 .00
1536 .00 .00 .00 .00 1536 .00 .00 .00 .00 .00
1535 .00 .00 .00 .00 1535 .00 .00 .00 .00 .00

201

I. M]: Mm mama IAN“

Unit of Measurement: tuber (rated) per kilometer

2-Lane, 2-Uay Sections

Subjective Rating Muer

Cwlative Rating I A commits rating based on above specified distress.

Sue itemized I Sill [Cracking + Rutting + Roughness + Dropoff].

[PVCGID] I ((81. itemized) + (Cwlative Rating)i/Z.00

Cumalative

itemized Rating

Rating

Cracking Rutting Roughness Dropoff

itami zed

m.
nus 762 6:6n66m .5561 569fiufis 1%.

666066606606 .omwmmmwmmomo

mmmmwmmwmmwmwwwwmmmwwwwmmwwmmmmwmwmmwwwwwmmmmmwmm666
666666666 66 6666665m5666665m66666 .666660 6:666663666

o66666666666666666666666666mmmmwmwwmmmwmwmmmwmmmm666
86636umun9666666663666766:un6666636uu661nnfimwum

666666666mmwwmwmmmmwmmmwwwwmmwmwmwmwwmmmwmmmwmmmm666

554.7749‘658'?’90‘0187‘5020895"798776085600756095‘567

1111 11‘

mmwmmwwmwmmwmmm666666666666mmmwmwmmwwwmwmmmmwmmwmmmm

77667878689180788900'91‘1oz1°99776789877208708056655

“11111

o666666666666666666666666666666666666666666666666666

mmmmmmmmmmommmmmmmmmmmmwmmmmmmmmmmmmmmmmmmmmmmmwmmmm
o1350532565‘76556m875681676232578" 0.415635273535302
alelelelelelelelelalelelelelslelel elelalalelq‘elelelelalelelelel elelelalelelelel elel

202

Unit of Measurement: tuber per kilometer

.

.

.

.

.

a

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

.

m .
.

.

..I .
t .
.

u .
s .
.

v. u
m .
. .
2 u
e .
.

.

.

I— .
. .
2 m. "
Zea.
’12.
ml...-
arr.

.

.

t .

.

m .

.

.

cum.
.3.

.

el .
i.el.
‘tt.
65M“.

.

m .
Ccc"

.l .
an.

660;.

ll.

RR"
8...."

111 .
man"
Boat.

S .
(cc.

(mmtn)

(Count #1)

 

6666666066666666666666mo6066666666£flmmflfl666666666666

12765a92177““6598565150fi9BM78726637uwa 8.Lihmilmmkwm

666666666666mmmm6666666666666666wwmmmmmwmmwmmwwwwmmw

e0l.27.‘..2 8.eeH.eeH.7.6. 9.‘.~.7.8.:Q.6..O. Seal. .mmnmsmhmul 5‘6Lo.el.6.,u ..I.

666666666666wmmmwwwwwmmwwwmmmwmwmwmmwwmwmmwmmmmmmmmm

11155

L 7 7 12 7 0123656789 1 7 1 5
ummmmmammmmmnmmmmmmmmmmmm...any?”nunmammmmmnmmmnmmmnm
elalelelelalelelelalelelelelelalelelelelelelelelelelelelelelelelelelelelelelelelelelelelelelelelelelelel

203

lh3’\

I. mm: KHCIEIT mum. N

Unit of iieeeurenent: utere per kilueter

(b-Lene Divided Sectione)

[RIM] . ” (Ribbon Develop-mt).

[SPAIHSJ 8 Su (Specific Pethe).

new») I Iiedien Opening.

Specific Pethe Median mooning

Ribbon Developent

hetero)

(netere)

teeters)

SOC

mwmmmmwmmwmmmmmmwmmwwmmwmmwwwmmwmm
wmz “muufiwsmufisanm339165366msann3

wmmmwmwwwmmmwmwwmmwmwmwmmmmwwmwwmm
u6unumnufiwvv£esu oauuma1zuaauneuo1

e8l.e6l.3.Q.B.27.9. .518.7.o§.5“5 ‘4Q3808227782 .3..el.

3

mmwwmmmwwmmmmmmmmmwmwwmmmwmmmwmwmw
7

mm m. mo mmem Mmmeme mammmmm
mmmmmmmmmmmmmmmmmmmmmmmmmmmmmmwmmm
m mm mm mmmmm mmumwmmmmmmuaw
mmmmmmmmmmmmmmmmmmmmmmmmmmmmmmwmmm
mm mm m mmmmmmmmmmmmummmmmm
mmmmmmmmmmmmmmmmmwmmmmmmmmmmmmmmmm

NBC 8 North Bound Corriegeuey
sac - South om Corriegmy

204

ll mum: DEFICIENT mu. SEC"!!!
Unit of Neaeurauent: netere per kilo-eter

(6-Lane Divided Sections)

[GDRAIL] I Slli [Deficient Guardrail Sectional.

Kn. Deficient Guardrail Sactione (Deters)
N00 500

Left Right Left Right [011611.]
1550 222.00 110.00 .00 .00 332.00
1551 30.00 .00 .00 .00 30.00
1552 .00 .00 50.00 .00 50.00
1553 .00 .00 25.00 .00 25.00
1556 60.00 .00 .00 .00 60.00
1555 .00 .00 100.00 .00 100.00
1556 .00 .00 .00 .00 .00
1557 .00 .00 50.00 .00 50.00
1550 .00 .00 77.00 .00 77.00
1559 .00 .00 .00 .00 .00
1560 232.00 266.00 .00 .00 676.00
1561 .00 .00 .00 .00 .00
1562 .00 .00 66.00 .00 66.00
1563 .00 .00 65.00 .00 65.00
1566 .00 .00 66.00 .00 66.00
1565 570.00 660.00 230.00 230.00 1690.00
1566 530.00 .00 165.00 .00 695.00
1567 .00 .00 62.00 .00 62.00
1560 50.00 .00 70.00 .00 120.00
1569 136.00 .00 122.00 .00 256.00
1570 .00 .00 .00 .00 .00
1571 .00 .00 .oo .00 .00
1572 .00 .00 30.00 .00 30.00
1573 112.00 110.00 .00 .00 230.00
1576 250.00 210.00 .00 .00 660.00
1575 .00 .00 .00 .00 .00
1576 .00 .00 .00 .00 .00
1577 50.00 .00 .00 .00 50.00
1570 .00 .00 20.00 .00 20.00
1579 50.00 .00 50.00 .00 100.00
1500 60.00 .00 .00 .00 60.00
1501 05.00 .00 162.00 .00 227.00
1502 110.00 .00 110.00 .00 236.00
1503 .00 .00 .00 .00 .00

NBC I North Sound Carriaoeaay
SOC 8 South Domd Carriageuay

205

II WIDTH]: DEFICIEI‘I’ PAM 9107'
Unit of Measurement: netera per kilometer

(6-Lane Divided Sections)

[PNlD‘l’li] . sun [(Variation)’].
Variation I (3.65 x Nulber of lanea - Pvt. Nidth).
A negative variation signifies pave-ant wider than apecified.

Kn. Pavenent Nidth Variation m
(aetera) (eaters)
N00 300 N00 500

1550 7.10 7.50 .20 -.20 .00
1551 7.60 7.30 -.10 .00 .01
1552 7.60 7.30 -.10 .00 .01
1553 7.30 7.30 .00 .00 .00
1556 7.30 7.30 .00 .00 .00
1555 7.50 7.00 -.20 .30 .13
1556 7.20 7.30 .10 .00 .01
1557 7.50 7.30 -.20 .00 .06
1550 7.50 0.50 -.20 -1.20 1.60
1559 7.60 0.00 -.10 -1.50 2.26
1560 7.00 0.00 .30 -.70 .50
1561 0.00 7.00 -1.50 .30 2.36
1562 9.00 7.90 -1.70 -.60 3.25
1563 0.70 7.30 -1.60 .00 1.96
1566 7.20 7.30 .10 .00 .01
1565 7.50 0.00 -.20 .1.50 2.29
1566 7.00 7.60 ~.50 -.10 .26
1567 7.50 7.50 -.20 -.20 .00
1560 7.60 7.50 -.10 -.20 .05
1569 0.00 0.00 -1.50 -.70 2.76
1570 7.30 7.50 .00 -.20 .
1571 7.50 7.30 -.20 .00 .06
1572 7.30 7.30 .00 .00 .00
1573 7.30 7.20 .00 .10 .01
1576 7.60 7.00 -.10 .30 .10
1575 7.30 7.00 .00 .30 .09
1576 7.00 7.00 .30 .30 .10
1577 6.00 7.00 .50 .30 .36
1570 7.00 7.00 .30 .30 .10
1579 7.50 7.00 -.20 .30 .13
1500 7.30 7.30 .00 .00 .00
1501 7.30 7.30 .00 .00 .00
1502 7.30 7.30 .00 .00 .00
1503 7.30 7.30 .00 .00 .00

NBC I North Bound Carriageuay
SBC . South Bound Carriageuay

206

I. mom: fiFICIEIT mm mm

Unit of Measureaent: eaters per kilo-eter

(6-Lane Divided Sections)

[SUIDTN] I Sill [Deficiency (+ values only”.
Deficiency I (3.00 - Shoulder Nidth).

A negative deficiency signifies ahoulder uider than specified.

mom

Deficiency (asters)

Shoulder Nidth (eaters)

NBC

Left Right

Left “flit

Rid“ Left Ridlt

Left

mmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmm
”mama m samumuwwwsaanummuummwsmumm

mmwmmwwwm mwmm wwwwmmwmmwwwwm wwww
I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I
elal el al 2
mmmwwmmmwwmmm ”an wwmwmwmmwmw
I I I I I I I I I I I I I I I I I I I I I I I I
11111 11.1 elalaleleloeh LLLZLehalel.
mmammmmmm....m.m.mmmxwwmwmwwmmmwmm
al elal..al .alelalel 2.1 e1 1111 12211111
mammmwmmwmwmmmwmmwmmmmwmwwwwmwmmw
I I I I I I I I I I I I I I I I I I I I
2 .a... . .a.l .ahel.
mmmmmmwmommmwomommm ommmmmmmmwmmmm

2233333322333222333223333333313333

mmmmmmmwmmmwommmowmmmmmwmommmmwmwm

22222333333322222222‘2222222112222

mo mmmmmommmmmmsmmmmmmmwwomwmmwmmm

2222232332222213222321221221112222

mmmmmowmwommmmmwmwwmmommmmmwmmmmwm

122333233232233‘332232322‘33322333

mausunuyummmmmwmmmmmmnnnnnnnnnmmum
5555555555 55555555555 5
111111111111111111‘111‘1111111‘111

NBC I North Bound Carriageuay
SBC I South Bound Carriageuay

.2()7

8303.28 ...-0. .333 09.
8300228 353 5.5.. n um:

... 8” n8. 8 8. 8. 8. . .
8. 8. 8m. 8 8. 8. 8. W. %. Mm"
8 ~ 8 8.... .8. 8 8. 8. 8. 8.8... 8. .8.
8.... 8.8.... 8.. 8 8. 8. 8. 8.8... 8. 8....
8... 8.8.... km. 8 8. 8. 8.8... 8...... 8. 2...
”mum fiugm mm" % 8” 8” 88... 88.”. 8. R...
“an.“ $.88... 8.8m 83.5.... 2. ...... 2... .... 8 a”. fi. %. ”man." a”. Rm"
e e I e e I
.... 8.8.. mm... 1.” 3. %. um. %. avg." mm"
8.... 8.8.... Rm. 8 8. 8. 8. 8...... 8. n...
8.. 8..... km. 8 8. 8. 8.8... 8.8... 8. fl
fin» $8... Km" ”u 838”. 8” 88.”. 8.8.”. 8. Km"
8” 8.... 8.8... 8.05...» is .2 Sn. 8 8. . 8 8 8. . 8 ... . 8 ”sow"
8... 8.8... 8.3... 83.5.... .... .3. 8.8... 8.8... 8. 8.8... 8.8.. 8. ....
8 .. 8 8... 8a. 868.. 8.8... 8. 8.8... 868.. 8. 8m.
8”. 8.8.. ...... 8.8a 8.8. 8. 8.8... 868.. 8. ..m.
8 m 8.8.. 8.8m 83.5.... ...... 3:... 3m. 8. 8. 8. 8.8... 8.8.. 8. 3n.
8.. 868. no... 8. 8. 8. 8.8... 8.8... 8.
8. 8. 3n. 8. 8. 8. 8. 8. 8. fl.
8.. 868. n... 8. 868.. 8. 8. 8...... 8. no."
8”. 858. 3a. 8. 868.. 8. 8. 868.. 8. 3..
fl." fiufiu 8.8. 83.5.... ...... 8.. Wm" %. $8.". 8. 8. 868.. 8.. .8.
. . . .8. . 8. 8. 868.. 8. 8m.
8 . 8 .8.. .8. 8. 8.8... 8. 8. ...8... 8.
8e. gogoc e e I e e e I e a“.
.... ... .... ...... ...... a..." a. .. ...... .. ....
“an" ”mam“... “mm" 8” 88.". 8. 8. 868”. 8. mm"
”mum “mug“ 868 8.85.... E... ...... an. n“ %.%m . fl. «“8... mafia." «a. “mm"
a e e I e I e
flu» ““88 mm m $8.“. a” gnaw; «van»; %. “mm"
......... ...mHm.....m..8m. 8.8. 8.85.... .5. ...-...... 8m. 8 $8. . a. fig." gnaw." «u. ...mm"
83...... S... .................................... ”fin“... m0...~ 0E“ ..... Emmnm..mflmmum:flmhm .............
.............................................. m8 8..
anoo.‘ co «octave: “rummmmmnmuuuam.m.m.o.w uuuuuuuuuuuuuuuuuu
................................ WWWWW.....w...m.w...wmmwwm::..H......-.......- . .23.... 5853......9..8 .5.
.IIILI smuwouccuc— v. IC-IIuCOMommmmmhnmmMoham nnnnnnnnn
ace 30.3. .33 .0 2:02! 2...... so» a n
.1... aco .0 2:02.. 9:3? .8. a _. 3.3.0:
.8.: x .22.: x 583 .53.... ...u a 83.3....

.18—woo” 3.); 0.373
3.3.0:... can 23.023 3:33.: ..o «...:
gnu! PES— E—u—ug «g I!

208‘

.52. ..x. 83...... . ......

8...:8 .8.. - u..

530‘ ’30-)” I (m

2.88» 2. - ...

>3... .3... a m.

... 88. 88. 88. ...8. 83.5.... E .8.... .8.. . . e 8.. 2....
...... .8.. S... S... a»... 83.5.... 8.. 3.. o ~. 8 8. 3...
.... 9.8. no... 3... .8.. 83.5.... at. a ~. on .3 ......
SJ 88. 8.8. 88. 2... 83.5.... .38. :5... e e o. 3.... 8n.
8.. .8.. 88. 88. 88. 83.5.... .8.. a . . e 3... ......
8..” ...... 38. £8. .8.. 83.5.... .8.. ...... o o .. 3.. on...
...... fl»... 3... .-.. .8.. 83.5.... .5. ...-...... o. .... 8 N... 8....

83.8.. 22.. 2...... 21>. 21>.
88...... .85.... 38...: 88...: 32...... 838:8... u.. «u v... u. ......

.2“:th + fight: 9 83.—z: ... nbxwhz: I Swabs:
Joe... 2.3083... ... cot-325 .3.

43:88.37“): ... 3:25.979: I 958—3:

6.202 I 38;:

.mp5 I 38.5:

.mhxaz I Camps:

3.8308 .83)... 8.3....
.52. .8.. . Sign... ... 2.5
3:3...» 3.2.38 3.. "38:... ...!
20:323.... 2.: >55 52¢ 3.3.“. cum... ...-.33 «83:... :9.

2.38.5 #3323 szuwmamhz— oh: rcpxu «Camps: =2
ans—9:8 B—hug: «E: I!

209

>503....8 ..Son .338 a com
8303.939 .353 5.3.. a on:

8. 8. 8. 8. 8. 8. 8. 8 n8. 8 8. 8. 8. 8. 8. 8. n8.
8. 8. 8. 8. 8. 8. 8. 8 3m. 8. 8. 8. 8. 8. 8. 8. 8m.
8... 8.... 8.. 8. 8. 8. 8. 8 ...... 8... 8. 8. 8. 8.. 8. 8.. ......
8... 8... 8. 8.. 8. 8. 8.. 8 8m. 8. 8. 8. 8. 8. 8. 8. 8m.
8.. 8. 8. 8. 8. 8. 8. 8 km. 8.» 8. 8. 8. 8. 8.. 8.. km.
8.n 8.. 8. 8. 8.. 8. 8. 8 8.. 8.~ 8. 8.. 8. 8. 8. 8. E...
8. 8. 8. 8. 8. 8. 8. 8 E... 8. 8. 8. 8. 8. 8. 8. km.
8.... 8.... 8. 8. 8.. 8. 8. 8. ...... 8.. 8. 8. 8. 8. 8. 8.. a...
8.. 8.... 8. 8. 8. 8. 8.. 8. E... 8.. 8. 8. 8. 8. 8. 8.. a...
8.. 8... 8. 8.. 8. 8. .... 8 in. 8. 8. 8. 8. 8. 8. 8. .3.
8... 8.~ 8. 8. 8. 8. 8.. 8 n... 8.. 8. 8. 8. 8. 8. 8.. n...
8.. 8... 8. .... 8. 8. 8.. 8 fl... 8.... 8. 8. 8. 8. 8.. 8.. fl...
8. 8. 8. 8. 8. 8. 8. 8 ...... 8. 8. 8. 8. 8. 8. 8. .3.
8. ... 8. 8. 8. 8. 8. 8 8... 8. 8. 8. 8. 8. 8. 8. 8..
8. 8. 8. 8. 8. 8. 8. 8 Sn. 8. 8. 8. 8. 8. 8. 8. 8m.
8... 8. 8. 8. 8. 8. 8. , 8 8n. 8..... 8.. 8.. 8. 8. 8. 8. 8m.
8.. 8. 8. 8. 8. 8. 8. 8 .8. 8.. 8. 8. 8. 8. 8. 8.. so...
8.. 8. 8. 8. 8. 8. 8. 8 8n. 8.. 8. 8. 8. 8. 8. 8.. 8...
8. 8. 8. 8. 8. 8. 8. 8 mom. 8. 8. 8. 8. 8. 8. 8. 3m.
8... 8. 8. 8. 8. 8. 8. 8 3m. 8... 8.. 8. 8. 8. 8. 8.. 3m.
8... 8.. 8. 8. 8.. 8. 8. 8 no... 8.... 8. 8.. 8.. 8. 8. 8. no...
8... 8.~ 8. 8. 8. 8. 8.. 8 8... 8.» 8. 8. 8.. 8. 8. 8. 8..
8.~ 8. 8. 8. 8. ... 8. 8 .3. 8.... 8. 8.. 8. 8. 8. 8. .8.
8.. 8. 8. 8. 8. 8. 8. 8 ...... 8.. 8. 8. 8.. 8. 8. 8. 8...
8. 8. 8. 8. 8. 8. 8. 8 on... 8. 8. 8. 8. 8. 8. 8. on...
8.... 8.. 8. 8. 8. 8. 8. 8. 8m. 8.~ 8. 8. 8. 8. 8. 8.~ on...
8.. 8.. 8. 8. 8. 8. 8. 8. ...... 8.. 8. 8.. 8. 8. 8. 8.~ ......
8... 8. 8. 8. 8. 8. 8. 8 o... 8... 8. 8.. 8. 8. 8.. 8.. on...
8... 8.... 8. 8. 8.. 8. 8. 8. ...... 8.. 8. 8.. 8. 8. 8.. 8.~ an...
8.. 8. 8. 8. 8. 8. 8. 8 an. 8.. 8. 8. 8. 8. 8. 8.. 3....
8.. 8.. 8. 8. 8. 8. 8. 8 . ...... 8... 8. 8. 8. 8. 8.... 8. man.
8.~ 8. 8. 8. 8. 8. 8. 8 «an. 8.... 8. 8. 8. 8. 8.. 8. ......
8.~ 8. 8. 8. 8. 8. 8. 8 .mm. 8.... 8. 8. 8. 8. 8.. 8. .8.
8. 8. 8. 8. 8. 8. 8. 8 8... 8. 8. 8. 8. 8. 8. 8. ......
8.6.3. ...... 5.8.. 8.. ...... 5.8.. .8.. .28.... ...... 5.8.. .... ...... 5.8.. .8..
.8 u... .8 u...
.852. .8.... .8. .93. .... .8.... 8.... it... ....

.23... 8.... + 3.6.... 5.8... + 3.8.... :3. u 8.8......
4.0.8.. 8.... + $6... 5.8... + 3.6.... so... u 8.6....
. .883. + .88.... a 8.3...

283... 8...... .8....
8.8.2.. .....8... .... . 38888... 8 ......
8.2... ...... .58.. . 8.3.8. ...:

9.25.5 oumcm #332 u 8.583 :9.
mung-uric g 91.-Nu "—3: g

210

903099.209 0.89 ...-.99 a 999
90:099....09 ...-.99 5...... u 99..

8.9 8.99 8.99 8.90 8.9 8.9 8.~. 8.9. 8 93 8.9 8.0 8.9. 89. 909.
8.93 8.99 8.3 8.90 8.9 8.9. 8.0. 8.9. 8.93 8.0 8.9 8H3. 8... ~99.
9~.93 8.93 99.99 8.90 8.9 8... 8.3. 8.0. 8.99 8.9 8.0 8 9 8.0 .09.
9.93 8.93 99.93 8.99 8.3. 8.9. 8.0. 8.9. 8.99 8.0 8.9 8.0 8.3 999.
9..~3 8.93 99.93 8.9. 8.3. 8.~. 8.9. 8.0. 8.9~ 8.9 8.9 8H3 8.9 9.9.
9.9 8.99 99.93 8.9. 8.~. 8.9. 8.~. 8.9. 8.9 8.9 8.9 8 s 8.0 99.
9~.93 8.99 99.93 8.99 8.~. 8.9. 8.9. 8.9. 8.9 8.9 8.9 8.9 8.9 9.9.
8.99 8.99 8.9 8.99 8.9. 8.9. 8.9. 8.9. 8.99 8.0 8.9 8.9. 8 9. 099.
99.~9 99..9 99..3 8.99 8.3. 8.~. 8... 8.9. 8.99 8.9. 8.9 8.9 8.9 999.
9~.~3 8.9 99.99 8.90 8.9. 8.~. 8.9. 8.9. 8.99 8.0 8.. 8.0 8.9 399.
8.53 8.99 8.3 8.99 8.9. 8... 8.~. 8.9. 8.93 8.. 8.9 8.9. 8.9. 999.
8.93 8.99 8.3 8.99 8.9 8.9. 8.9. 8.0. 8.9 8.0 8.9. 8..... 8.9. ~99.
99.03 99.99 8.93 8.2 8.9 8.9. 8... 8.9. 8.93 8.3 8.9. 8.9. 8.~. ...9.
8.93 8.99 8.99 8.90 8.9 8.9 8.9. 8.9. 8.93 8.9 8.9 8... 8.3. 9.9.
9~.99 8.90 99.93 8.90 8.0 8.9 8.9 8.3. 8.99 8.9 8.9 8.9. 8.9. 909.
99.99 8.90 99..3 8.90 8.9 8.9 8.9 8.9. 8.99 8.9. 8.0 8..... 8.9. 009.
99.99 8.90 99..3 8.90 8.9 8.9 8.~. 8.3. 8.99 8.9 8.. 8.9 8.9. 909.
9.9 8.90 99.03 8.90 8... 8.~. 8... 8.9. 8.99 8.9. 8.. 8... 8.0. 009.
8.99 8.99 8.93 8.99 8.9 8.9 8.9. 8.9. 8.90 8... 8.9 8.9. 8.9. 909.
8.93 8.99 8.03 8.93 8.9 8.9 8.9. 8..... 8.90 8.9. 8.9. 8.~. 8.0. 309.
9.9 8.99 99..9 8.93 8.9. 8.9 8.9. 8.9. 8.90 8.~. 8.3. 8.9. 8.9... 909.
99.93 8.99 8.3 8.9 8.0 8.9 8.9 8... 8.90 8.~. 8.9. 8..... 8.0. ~09.
99.93 8.99 8.09 8.90 8.9 8.9 8.9 8.9 8.99 8.9 8.9 8... 8.9. .09.
99.09 8.93 8.99 8.99 8.9 8.9 8.~ 8.0.. 8.99 8.9 8.0 8.9 8.0. 909.
9.9 8.99 99.99 8.90 8.9 8.9 8.~ 8.9. 8.99 8.9 8.0 8.9. 8.9. 999.
99.93 99.~9 99.39 8.99 8.9 8.9 8.~ 8.9. 8.99 8.9 8.3 8.~. 8.0. 099.
993 99.93 8.99 8.99 8.0 8.9. 8.9 8... 8.93 8.9 8.9 8.9 8.9. 999.
99.~9 8.93 8.9.. 8.9 8.0 8.9 8.. 8.~. 8.99 8.9 8.9 8.0 8.9 099.
8.39 8.93 8.9... 8.99 8.9 8.9 8.. 8.9. 8.99 8.0 8.3 8.9 8.9 999.
99.9~ 8.99 8.39 8.99 8.0 8.9 8.~ 8.9 8.99 8.0 8.0 8.3 8.9 399.
99.9~ 8.99 8.5 8.99 8.. 8.9 8.0 8.9. 8.99 8.9. 8.3 8.... 8.0 999.
99.8 99.... 8.99 8.9.4. 8.9 8.9. 8... 8..... 8.99 8.9 8.~ 8.3 8.9 ~99.
9~.9~ 8.9.. 99.99 8.9. 8.9 8... 8.9. 8.9. 8.99 8.9 8.3 8.9 8.3 .99.
8.99 99.99 99.03 8.99 8.9. 8.9. 8.~. 8.9 8.90 8. 8.9. 8.9.3. 8.- 999.
2.33.59 03.5..
co... :3: _ _
.9895 .85....- 8259 8.4.9.. . 9.2.9.9 3.1.8.8. 95.59 99.3.9.9
9...... 2.»...
2.32.5. 9...... 3...... 2.33.5. 8.... 03......
999 _ 99..
.89... 835.. 2.2.. 2.38.9. .8.

.............................................................. :3..qu 9.9.2.... .-....i......a-m.m.mmuumsi........-n._.....-.a.umm§....
. $82.. + 8858.. .. 2.38 . 2:8... ..B .. .5. 32......
8.2.2.325. :8... + 83...... So... .. 3.82..

.8....» 08.)... 83-3.
.85... ._ 68...: .85: 38.358... .3 ......

03:: 8:93 E19 nag I!

211

Unit of Honours-nu Wr per kilo-etc:-

(lo-Lono Dividod Sections)

[SIDEOB] 8 SUN [Count '1 + Count l2]!2.00.
Comt £1 8 Rated Comtina by ml- 1.
Count #2 I Rated Counting by nolbor 2.

ablative Rating (WP!)

SBC

NBC
(Comt #1) (Cunt '2)

(Count I1) (Count I2)

mommwmmmmwmwmmmmmmflmm ommomwwmmm

03591575‘7276 .& 9L17596835683
cl 4|

wmwmwmmmmmwmmmmmmmmwmmmmwmmwwwwmmm

Sol—1.617231“‘13673‘8552153o‘oahohoho68‘26-

wwmmwmmmmmwmmwmmmmmwmwmmwmmwwmwmmm

612779251‘65236.3‘0675363 .S‘SLLL‘HB.

mmmmwmmmmmmwwwmmmmmmmmwmwmmwwwmmmw

52336651227331351$8415555551377255

mmommmmmmmmmmmmwwmmmwwwmwmmwwmmmmm

53338852558332‘5381"657‘5“‘10567

use I North mu Col-rim
NWIWMMhMMflCuflumwy

Appendix D

The Accident Data

23142

-...025358 ...........................................................................................................................

993.
003.
093.
003.
003.
003.
003.
003.
003.
003.

99.9.9... 9. 00.9

00.. 990.9 I... .9. .9. soougo>9 9.999
9.9.9 .9. u.9 9.9.9

00.. 990.9 I... .9. .9. ..9 9.9.9
99:9. .9. ..9 .0.

00.. 990.9 .9. .9. accuse». .99

.9. .9. u.9 .99

99.....uon o9. v.9 9.9.9

385.. .933

0.9.9. gas. .9.

9......o. 99 no.9

99.....009 .9. u.9 9.999
5....898. 2.. .... .8...
99.5.9900. .9. «.9 .9.

no.9...o. 9. 09.9

00.9. a...

too-I90 .....999 9.99

. too-:90 9.99

.0.. 999.9 I... 9.9.. .9. 9.9 9.9-:
0... .9. co .9... o. .90 0.9.9.5.). 9.9..
a... .8.. .2. 5...... 2.. .... .8...
9......o. c. 09.9

.8. 2.. .... 5..

90u.0 9.9. .... .99

Stu!!! 2.. .... .55.
9......009 .9. ..9 999...

0.9.1.0 .9.

0.999. .99. .....9.) 00.9. ...
009.990 9uo9

9......9. co 0.09

9......o. so 0909

90.. 999.9 I... 9.999 .9. ..9 9.9..
«0.. 9:9.9 I... .9. .9. «.9 9.9..
0.99990 .9.

a... .9. ..9 .99

5.358.. 2.. .... :33

'-OOOOOOOOOFPOOOOOOOOOOFPPOOOPOOOOOOO

.35. :

.8..
9.9.9
9.9.9
.09
.99..
.00
.09

.9999.
9.9.9
.99999

5......

i

.8
.8...
8.....

9......

...:§

.00
.0999. ..o
.99.:9

0:0

.99

can.)
.8..
9.9..
o...

9.9..
.9.
9.9.9
9.999
.99
.99
.99
9.9.9
.99
.99
9.9.9
9.9.9
9.9..
.9.
9.9..
9.9..
9.9.:
9.9..
9.9..
9.9..
9.9..
9.9..
9.09

:38
9......
:33

.8..

9.9..
9.9..
9.9..
.9.
.00

.93999

N 9 9 930 90x0.\9 003. 09
. . 9 999. 99\N.\. 003. 99
. N o 9N9. 90\0N\.. 003. 39
. N . 999. 90x9.\.. 003. 99
. . 9 990. 90x3 \. 003. N9
. 9 9 939. 99\9 \0 003. .9
. . 9 9.9. 99x9.\N 003. 99
9 . 9 99.N 90x9.\9. 003. 9N
9 9 9 990. 00\9.\N 003. 0N
9 N o 999. 00x0.\. 003. 9N
N . . 993. 00\.9\. n03. 0N
9 9 . 93.. 90\3 \9 903. 9N
9 o . 990. 90\N.\0 903. 3N
9 9 . 990 99\.N\0 903. 9N
N N . 93NN 00\N.\N 003. NN
N 9 3 090. 09\9N\3 003. .N
N 9 N 999. 00x0.\9 003. ON
N . . 990. 00\. \N. 003. 9.
. N . onN. 99\.N\3 003. 0.
. N . 99.. 90\9N\9 003. N.
. 9 . 999 99\NN\9 003. 0.
. 9 3 99NN 00\NN\N 003. 9.
. 9 . 999. 00\..\9 003. 3.
. 9 o 999 00x3.\0 003. n.
9 o . 999N 00x0 \0 003. N.
9 . 9 999 90\0 \9 003. .—
9 . 9 999. 00\9N\. 003. 9.
9 N 9 9.9. 90x9 \3 903. 9
N 9 9 99.. 00\9N\3 303. 0
N N . 990. 00\N.\N. N03. 9
N 9 . 990. 99x9 \0 .03. 0
. 9 9 9.9 99\9.\0 .03. 9
. N . 99.. 90x0.\. .03. 3
. 0 o 9.0 90x9.\9 .03. 9
. N 0 999. 90\3.\0 .03. N
9 . 9 999. 99x3 \9. .03. .
gang
999 9:. 9.9 93.. 0.99 .12 .0009

.999. son-0.09.009. 9.99:99

909.-.03. .ouqlu..u “90.0.09 9:00...< ......—

.i.u.i._ ...-.9958 8.3.... 5..-... >55... ...-......

23143

-.-..9099..co.. ........................ .-................. .................. ........--.....--..-.-.........--..........-...-......-.....-

.3. 8.3:... ... .82. o . 8:8. ...... ~ .. o a... 859. .3. 8
.3. 8.2.... ... ...2. o 8..... ...... ... . o 8. 8.33. .3. ...
.3. 8.5.8 8 .82. .. ... ... ~ 2 o a... 33 ... .3. ..
.3. .... .2. 3.2.8 ..8 o ..8 8..... ~ 9 a m... 8...... .3. k
.3. 8.2.8 53 o .8.... ... 8...... ... ... o o 3.. S3 \. .3. o.
.3. .3.... 2.. ...... .... 8...... o .8.-... 88.. . . o .8. 8..... .3. m.
.3. 8:3 2.. .... 8.2. a 8.... .3.... . ~ .. 8.... 8.2.... .3. e.
.3. .98... 2.. .... 8...... o 3...... 8.2. . o o 3... 33...... .3. n.
.3. ... 2.. .... 8...... .. ... 83.. . ... a on... 8.. 3 .3. fl
.3. 2... .8.... .8 2.. .... 8...... o ..8 8.2. . ~ .. 8. SR 9. .3. ..
.3. 8 2.. .... 8.2. e .- 88.. . . o 8.. 8...... .3. 9.
.3. .82.... 8.2. o 8..... 8.. ...... . o .. m... 89.2... .3. 3
.3. 8... 2.. .... ...... o 8... 8..... . o a on... 3...... .3. 3
.3. .8.... 2.. .... ...... o 3...... 8..... . ~ 9 a... 83.... .3. ..
.3. ... 2.. .... 8..... .. ... 9...... . . o 8.. 8...... .3. 3
.3. ...». 2.. .... 8..... o .8... u. . o . ...... 83...... .3. no
.3. 8.2.8 ... o ... 8.2. . o a 8m. 89.. R .3. ...
.3. 8.2:... 8.. 8.2.8 ...... ..8 o ..8 8..... . .. ... 3.... 83.9. .3. no
.3. 8.38.8. 2.. 8...... c8... . . 8.2. o . e on. 83...... .3. ....
.3. 5.388.. 2.. 8:... 8..... . . 8..... o o . 8. 8.. 3 .3. ..
.3. ......882. 2.. .... 8..... . . 8..... o . .. 8... 39...... .3. 8
.3. 8.88.8... 2.. .... .....3. . . ......5. o o . 3.... 85:. .3. 3
.3. 5.88.... 2.. .... ..8 . . ..8 o o . on. 83 k .3. ...
.3. 8.888.. 2.. .... ..8 . . ..8 o . a 8.... 8..... .3. ...
.3. 5.88.... 2.. .28... ..8 . . .8 o o . 8.. 8...... .3. on
.3. $.88... 2.. .... .3 . . .... o . o 83 83.... .3. an
.3. 8.889... 2.. .... .3 . . 3. a o . 8.. 8.3.... .3. 2..
.3. 5.888. 2.. .... .3 . . .... o o . a... 8..... .3. ....
.3. 5.88.2. 2.. .... .3 . . .... o . e on... 83.3 .3. ....
8.. 8.2:... ... .82. o .3.... .3.... ~ o a 8.. 8..... 8.. .n
8.. 8...:8 8 .32. o 3...... ...... ~ ... ... 8.... 8...... 8.. on
8: ... 2.. .... 8..... o ... 88.. . o . .8. .2. s. 8.. 3
8... ...... a 3.8.88 98...... .8.. o .... ...... o . o 8.... 83 3. 8: 3
.3. 8.5.8 8.. o ...... 8..... ~ . o a... 33 k .3. 3
3... 8.2.... 8 .82. o ..8 ...... ... . o 8.. 83 3. .3. 3
.3. 2.2.8 o ..8 9...... ~ .. o 8.. 85.. .3. 3
3... .3 2.. .... .3 o 8. 8.. ... o a 8.. 33»... .3. 3
.3. .. 2.. .... 83.. a 88.. u. . . a on. 83.9.. .8. 3
3... 8.. ...... ...... ...... 2.. .... 8..... o 8..... 8..... . o e on... 3.. k .3. 3
3... ..8 .. .8.. .... .. ..8 2.. .... 8..... a ..8 8..... . o . ...... S9. 3. .3. S
3... 38...... ..8 o ...... .8 . o o 8. 83.... .3. 3
.8. 5.388.. 2.. 92.8.. 8.2. . - 8.2. a . o 8.. 8...... .3. 3
.3. 5.8.8... 2.. .... 3......» . . 3...... o o . 8... 3: x. .3. on
3... 5.88.... 2.. .... ...... . . .8... o a . on. 8..... .8. ...

...... 9:8... ...... .... 92:8... 88...... .. .2. ...v 8.. ...: ... ...... 88 .5. .82.

214

-...Auo::_ucouv ................... ....-..... ......... . ................... ............-.... ...... .. ..... ...........................-.....

nova
yac—
«6"
«ac?
«Ger
«6"
now-
nocw
nae.
noc—
nuc—
nova
nut?
nuep
nae.
noc—
nocp

ca_guoouon ogu u.g nsxo_n
co.o_..oo co too;
co.u_..oo co v.0;
cu.guuouan ogu u_: coon:
c-_guouuon ozu u.g Lo..ogu
:o_uuuouoa ogu u_: neon
co.o...oo co and:

too-Inn coco:

.xauaa ozu u_: suagu

gnu «cu u.: Juana

goo osu u.g snag»

guagu uc.u:-uu as» u.: use
so._-Lu oc.ucauu 0;“ a“: can
goo ecu u.: nan

...: oz» gu.x toga-go use
c...u..uoa as» u.g o.o.g.,
co.a_..ou ot.o

sod—agu 0:» u_g can
co_gu¢uuon ogu v.5 can

u: 0;» «.5 sou

_u:~:u accuuoso coo-3
coco» as» u.g cue-a

".8.: 8.23. .5 a... :8...
9. .5 a... :8!

8 .5 a... :8!

goo use u.g en)

us as“ u.: Juana

ou_u soon In.» coo-a ecu u.g can
:8 .5 3.. .3

n! 2.» at. can

:-_gunouon ozu u.; coo-1
ca.gunouoa ogu u_g coo-3
:-_Luaouon use «.3 coo-3
:...u..uoa as» u.g ..o.g.»
co—Lauouoa one u.: ca)
cu_guaouon ogu u_: xuagu
.3 to... £80 :3 39.33..
.ogucoo we «no noses» ..o
oc.u-o_uc_ an: on can
co.guoouon on» u.: Lao
c...u.ouoa ogu u.g nan

oo—wron—PPw—Pwoooooeoooee—eo—oooooocooFFPQOF
I

coon:
.xauam

god—cg»
n:-
...:

I:-
so..osp

Leno-g—
ODCOh

9%:
..8
8.

coco:
go__ogp

.8:
:38

NOD000°066°OOOPPPP'PFPPNPOOOOPPPPPPPFNOOONNO

coco-o'-v-oooo—c-aonuu-o—nc-o-o-a—chomv-onuo-OG-OPNQC-

OGPPOONOOFPPOOFOOOOVOOOPOPONOOOOOOOOOOOPOPOO

can oaxo \s
mwop oo\0n\o
mpow noxn’xo
comp ooxppxw
one. noxopxo
m—o— ooxonxm
onnp ooxo~\e
mph. cox» \m
can oo\o \0
arm ooxonxs
ooo~ unxw—xo
uro- on\o~\o
can ooxopxo
ooh nox-xc
menu obxn \Nw
on“ ooxo~xm
ao~p ooxo \e
000— 00\o*\n
cop. oox~ \m
o¢o~ ooxmpxn
menu ooxm~xm
oqa ocxc \c
coo cosh \NF
mwo— oosnmxh
mpp— oa\~p\s
new. ooxo~xo
coop ooxnuxo
coop oa\m~\o
anNN ooxh—xo
ooo~ ao\ow\op
onnp onxowxo
wasp coxo \n
000 ooxn~xh
onbp ooxs-\n
oowp ooxnpxo
n—n oaxopxow
anon onxo~xs
one. oo\0n\op
oo- noxm~xw
nepm cosh-\o
men. ooxpn\m
o—np ooxo \cw
on—— coxo-\~
can ease \~p

mocp «Np
99¢" nNF
«01' -F
coca ww—
cocp om—
oovp o——
mic. n—p
nova sw—
nucp op—
nocp nap
nsep opp
noe— nap
noe- ~pp
nae. w—p
n§cp op-
nfic- oo—
nuep oa—
nwcp sop
nova oa—
Nocp mop
soc— «ca
soc. no.
soc. no.
«on. pop
«09’ new
Noc—
seep
Noe—
Noc—
Noep
Nocp
«oo—
«0"
«new
Noe,
Noe—
«new
Noe—
woe—
Nocp
Noe—
Nov-
woe.
pocp «a

883883888833838288

.-30s A... ui>dos:_ no.0.go> mo .02 ...v can wzu ahm 05.» Gala

co.guuonon ogu toga-go nan

too-luv cyan

non-Ian .xauau and cocoa
nag-sun

.19 .uuoc

22115

.....uoac.ucou~ ................................................................. . ........................................................

3m. 53:8 5 ...... o . .8... .8... ... o . ...... 83.). .8. no.
3.. 23...... .8 8 .3 o . 2.3:. ... ~ . a 8.... 83 3 3m. 3.
3.. ... .... .... .8.. o . u. .8... . ~ .. 8.. 8:... 3m. 8.
3m. .... .... .... .8.. o - .... .8... . ~ . 88 8.8... 3.. 8.
3.... 5.1.8.5.... . . .3.... .8... . o a 8.. 83...; 3... ...
3m. 8 .... .... .3.... o . o. 9.... . . o 8.. 8a.? 3a. 3.
3.. 8..... .... .... ..5 ..8. .... .... o - 5.... .... . . a 8.. .8.. x. .8. on.
3.. .8.. .... .... .... o . .8... .... . . a $9.. 85:. 3... 8.
3m. .3... .... .... .3 o . .3... .... . . ... 9. 83.: 3.. ....
3.. 53...... .... .... .... . . . .... o o . 8... 8...: 3.... 8.
...... ...... 8... .8... .... .... .8... a . .8... .8... ~ . a on. .8. a. no... a.
3m. .... .... .... 5.... a . .... So... . n o .8. 83.: no... r...
3.. .... .... .... .8... o . .... .3.. . . o 8... .8. .... no... ....
n... .8... .... .... .8... o . .8... .8... o . o a... 88.2.. 8.. 8.
...... .8... .... .... .833! o . 85.3.... .8... . ~ . a... 8x. 3 3n. ...
.8. 53...... .... .85.... .3 . . . .... ~ . . a 8B ... .8. on.
.8. .85.... .... o . .... .8... . a o .8... 83...: ...... ....
8.. .... .... .... 5.... a . .... :8... ~ . a 8.. 8x. a 8m. 8.
8.. 5.5.8.. .... .... .8... . . . .8... o o . 8. .8»... 8m. ....
8.. 5......8. .... .... :33. . . . .53. o . o 8.. 8:2... 8.. 3.
8m. .5388. .... .... .3 ~ . . .... o . ... a... 8...... 8m. 9.
8.. 5.5.8 8 ...... o . .... 8.... ~ . a a... .8.... 8.. 3.
.... a. .... .... 3...... o . ......» u. . . a 8.. 85: .3. ......
8... .3 .... .... .33. o . :33 .3 . . o 8.. 83.3 8.. 8.
.... 5.5.88 .... .... .8.. ~ . . .8.. . ... o n... .88.: 8.. ...
.3. 5.5.8.. .... .... .... . . . .... o e . 8.. 8k... 8.. 3.
8.. .... .... .... .3 o . .... .8 o . a a... .8.. 3 .3. on.
.3. .2. ...... ...... 5.... .... .... 5.... o . .8.... 5.... ~ .. . 8. 832.. .3. 8.
.3. 5.5.8 8 5... o . .... .3.... ... o . on. 25...... ...... ....
...... 5.....8 8 .8... o . .8... .8.. ~ .. . 88 8a.... ...... 8.
.3. ...... ...... ...... 8..... .... .... .3 o . :8... .3 ~ ~ .. on. 83.3. ...... an.
.... v85... 8......» ...... o . .3 .8... ~ .. a 88 8:2... .8. 8..
.... ...2. .... .... 5.... o . :8... .8... . e . 8. 832... ...... 8.
.... .8... .... .... ....5. ... a . .8... .....5. ... . a a 8.. .8... a. ...... 8.
.3. .... .... c. .... .3 a . . .2. . o o 9. 83.2. .3. ....
.3. 5.5.8.. .... .... .... . . . .8.. o . o 8.. 8:. k .3. ......
g
8.. .85.... .... a . .... .35... . o o 8.. 8.8: 8.. R.
o... 55...... ... o . .... .3 . a a on... 832. 8... 8.
8.. 53...... .... .... u. . . . ... o . o 8.. 8?. 3. 8.. .~.
8.. .... .... .... .3 a . 8. .... o o o 8.. 8.83 8.. 8.
a... v.53 ...». ....5. ..5 .... o .8.. .....5. .3 ... ~ ... . 8.. 835... no... 3.
.5. .....l... ...... .... 532.... 8.25.. .o .2. ..-. .8. ...: .... .... .... ...... .8..

22115

--.-Avlt.ucouv ................................................................. . ........... . ...............

comp
oom—
oomp
comp
oom—
comp
oom-
comp
oom—
nor...
comp
8a..
soap
son.
now?
son.
hon.
oom—
comp
com.
com.
com-
comp
con—
83
com.
comp
comp
comp
33
8a..
oom—
oom-
comp
comp
comp
comp
oom—
mom—
momp
nom—
moan
comp

co..unonoa 0;. u.g .xauan
ca..uonnon ogu u.g ..u
..o ecu u.g ..u
5.5.8.. .5 .... .3
co.o...oo no use:

.no oz» u.g .ouu-.u

....» .... .... .3

u. one u.: can

co..uuouoa use toga-.0 .xauao
ca..unouon as» u.; aco—
5.5.8.. .... .... ....
ca..uoouoa as» u.: can

.oo 0;. u.g 393..
:o.u...ou co too:

.3.. soon 30.. .au 0:. u.g can
co..uoouon as» u.s 0.0.x.»
exot noc.:u can

.xauaa ogu a.g coo-a
:o.....oo no you:
co.u...oo no and;

sun.“ soou.o>o can

8 .... .... 8.....

on as. a.: .aauaa

u. .... .... .3....
:o.....oo so ago;
co.u...ou so too;
cu..uoonoa as. u.: coo-3
cu..uuuuon as. u.: coo-a
c...u¢onon one u.g coo-a
co..unonoa oz. toga-.0 coon:
ca..uoouon as. u.g 0.9.5.)
an..uoouon on. u.: xu:.u
cn_.u.ouon as» a.g .uanai
5.5.8.. 2.. .... 9.
ca..anouon as. u.: .09
ca..uoonon 0:. u.: ..o
5.5.8.. .... .... .3
5.5.8.. .... .... .3

u: uc.ucauu ogu u.g coo-a
.uauaa 0:. u.: xu:.u

on as. u.: .xzuaa
ca..uouuon as. was...» man
cocoa as» soou ..>o nus.
co.u...oo co nae:

OOFOOOFPPFFPFPPPPFOQOOOOOOOOPQOOFFPPOOODPOPF

LCD

.30

82.

on
.xauaa

...u

...:
....
.8
.3
.8....
.oaoo..
.3
....
:33.
......
...u
.3
.8...
.8...
....
39......
.3

can.)
.8...
.8...

....

u.

.8....
.3...
.....5. ...

80.3
8.....
80.3
s;
......»
.8....
:33.
u:

...u
.8
....
.3
8.....
.3...
:33.
3.
......
.3

NNOPPPOOOOOOOOOGOOPPGPPNNNNOOPPPOOOOPPPNOOOO

P

Inc-OPI-Ov-POPPC-OOOPOP.QPMNPONPMOMOPOv-POI-hv-Nc-PPP

OOPOOPOOPOOOVPFCPOP'OOOOOPOPPOFDFOOPOOOOOOOO

new. ooxn~xw
«pop ooxo \«
one. noxn \.
onh ooxo—xo
onup oo\~ xv
ooo~ coxnusn
new ooxnpxe
ooo- ooxopxm
onop ooxonxh
ooh. oox~pxwp
ooo oo\o~\o
m—o noxb~xn
oop ooxo xv
onu— oo\m~\s
ne~p ooxe \n
ooeN ooxnpxpp
on». ooxo~\p
ono ooxnpxu
ono~ ooxomxw—
onN— noxo \N
one ooxu—xs
nNNP ooxonxs
ooop noxo \n
o noxonxs
onow ooxo x.
onNF ooxoaxs
oovp ooxo~\o—
mpnp ooxo—xo
ooo— ooxonxm
owe. oo\—~\o
o ooxpnsh
nun. oosmpxo
ooN— ooxon\pp
npo ooxppxwp
ooNP aux» \n
op~p noxn \m
ooo ooxn—xp.
onop ooxova
onop oosswss
onop ooxmnxo
npop noxo \pw
onnp oosowxp
ono. noxopxo
onop oosonxo

com—
com.
com.
comp
oom—
con.
oom—
comp
com.
com—
comp
oom—
comp
comp
«on.
non—
mom,
«on.
mom.
«on.
«on.

585

35858

so—
00.

8......»

.o .o: ...v

oos sun A». Il_p Ouoo

217

3:33:35... ...................... ........... ...........

....m. .853 .26 .8 o . - ..au _. p a 80— ooxppxn 3m— om~
.... 5.5.8.. .... .... 5.... . - . 5.... o o . 8.... 8.8... ...... .8
...... 5.5.8.. .... .... ......» . . - ......2. o o . 8.. 8..... ...... n8
...... 5.5.8.. .... .... .....5 . . . .8... o . o 8... 8.. .. ...... .8
.... 8.. ...... .3 .... .... .3 o . .... .... o . o n... 8...... .... 3...
n... 5...:8 5 5... o . 5...: 5.... ~ . a 8... 8...... ...... .8
n... 5.8.8 0...... o - 0...... .....5 . o o 8. 8...... n... .8
...... 555.. 0...... o . 0...... .....5. ... . o o 8... 8... ... n... .8
2... .....5 2.5... .... .... .....5 o . .....5 .....5 ... . . 8.. 8.8.. ...... ~8
2m. ... .... .... 5...... o - ... 5...... . o . 8m. 8.. .... ...... .8
3m. .... .... .... .3 o . 5. .... . ~ o 8.. 8.8.... ....m. 8..
~..... .3 .... ...... 58 .... 5.. o . . .... o o . .8 8...... ...... .8
.... 5.5.8.. .... .... .3 . . . .... a . a 8... 8.8... ...... .8
.... .3 .... ..8. ...». 5...... .. . 5.... .... ~ . n .8 8..... .... .8
...... ......3. .... .... .3 o - .8.... .... . . . .8. 8...... .... 8N

Ala-flag
a... 5...... ... 2.5... .... .... .....5 a . .....5 .....5. .3 ~ . a on 8.. ... a... an...
...... 5.....8 5 5... o . 55.. .... ... n o 8.. 8.. .. ...... ..8
...... 5...:8 5 5... o . .....5 .... ~ N o “.8 8..... .... n8
...... .. .... .... 3...... a . 8 ......8 . . .. .... 8..... ...... «8
...... .... .8.... .....5. ... .. . . .....5. .... . . . 8.. 8...... a... .8
...... 8.. ...... .8... .....5 .... .... 5. o . .... .....5 . ~ .. 8.. 8...... o... .8
a... .....5 .... .... .... a . .... .8... . . a 88 8...... ...... 8...
a... 5.5. ... .... .... 5. o . 5. .....5. .... . . o 88 8.. ... ...... 8.
a... 5.585. .... .... 5.... . . . 55.. o . .. ...... 8.....~. ...... .8
.... 5.5.8.. .5 .... 5.... . . - 55.. o . . ...... 8.8.. ...... 8.
...... 5.5.8.. ...... .... .... .... . . . 5... o n o 8... 8.. .~ ...... .8
o... 0...... .... .... .... o . 0...... .... o . .. a 8.8... a... 8..
...... 5.5.8.. .... .... 5. . . . .... o o . 8o. 8..... ...... 8...
a... 5.5.8.. .5 55.... .... . . . 5. a . a 8.. 8.8.... ...... 8m
...... 5.5.8.. .... .... u. . . . ... a o . 8... 8...... ...... .8
8.... .3 55.5 .... 5...... a 55.5 .... .... .3 . .... .... 5...... ... n .. 8. 8...... .8. 8...
.8. 5...:8 5 5... o . .....5 .8... ~ ~ . . 8... 8..... S... ....
.8. 5.2.... 5 5... o . .8.... :8... N o a 8.. 8.. ... .8. .8
.8. 5...:8 5 5... o . 5...: 5.... ~ o ~ 88 8.8... .8. ..~
.8. 555.. .5. ...... o . .8 5. ~ . o .8 8...... .8. ..~
.8. .... .5 .... .....5 o . .....5 .... . ~ .. m8. 8..... .8. n...
8... 58 55... .3 .. . - .... . . . 8. 8... .o .8. ......
.8. 8.. ...... 5.... .... .... .3 o . 5.... .... . . a 8.. 8.8... .8. ......
.8. 8 .... .... .3 o . 8 .... . . a a... 8..... .8. ......
.8. 5.5.8.. .... .... ......» . . . ......» a . a 8... 8.8... 8.... ....
.8. .....5 .... .... .....5 o . .....5 .....5 o . ... a... 8.. .n .8. .8

in «ft: .805 A... 3.0.5: 00.0.50). yo .0: -..v 8.. ..a— a: a: zoo .18 ..008

221E}

----.uoac.ucoo. ...................... ..---.........-... .................. ...-..--..--...-.--.-.-...-.--..-.-.............-.-....-...-...-

«Nap
Pmnp
FNmp
—~mp
p~mp
p~np
«Na-

on—
o~m—
o~mp
ommp
onp
o~mp
0pm—
own.
sump
swap
h—mp
5pm.
hump
swa—
spar
spu—
s—m—
upm—
own-
own.
own-
owns
npmp
apn-
man.
man.
m—np
mpmp
e¢np
e—mp
e—mp
erm—
epnp
epmp

.nnon A--. untdosc— oo.o.g0> .0 .o: ...v cos ou— 4». cl.» Ouao

... .... .... 0......

...-...... 2.. .... 3..

co...oouon 0:. ..g coo-a
co..u.ouoa as. ..g ..u .x-.
5.5.8.. 2.. .... .8.
co...oouoa 0:. ..z ..u
5.5.8.. 8.3... .... .... .3

co.u...ou co one:
8.5.8 ......

ado-zit u.-u sacs

wanna. .920 .33.:-
co....uuon on. ..g coo-a
ca...novon 0:. ..g .ouoo.u
8.. ...... 55 .... .... .... 5....
5.5.8.. .... .... .3
co.....oo no you:

coon: 0:. ..s coon:

... .... .... 0......

.xanaa 0:. ..g .oxca. ..0
... .... .... ....

gong. 0:. ..s ...

..o 0:. ..s man

.au 0:. ..g can
ca..uoouoa oz. «.5 coo-3
co.u...ou co v.0;
co.u...oo co not;
account 3.03

..o as. ..s «an
co....ouon as. nus-... 0.9.5.)
......5 2.. .... ......5
no.0...oo co one:
co.o...oo co was;
5.5.8.. 2.. .... .3
8..-...... ......

co...oonaa on. ..s coon:
nap on. ..g 4.3..
co.u...oo co and;
co.u...oo co too:

.oxca. ..o 0:. ..g .:n
......a. .... .... 5...:
co.o...oo co too;

...-in.

OOOOOOP0°0°OPOOOOFOOOOOOOOPOPPOOOO

u:
......

.00

...:I» ..o
coon:
u:
.aauau
..u
..u
..u
..u
sun.»
..o
..u
.09
Jun.»
coon:
a:-
sun.»
39:.»

can
coo-3
..u

o:-
.xaana
coo-3

0......
....
5......
....
.8..
LCD
....

any.»

.09

coo-a

an.
o.u.=o>
3.3.»

.3.

gang:

can
some.» ..o
coo-a
sun.»

can

xu:.p
.oxcn. ..o
coon:

......»

OOOOOPF

'PNNNNONNNNNOWNNOOPPFWPPPNOPOOPNNN

OOOPPPN
PFC-000°

OOOOPOGPMMOOOPP'PPOOOPOOPOOOPOOOP
F

NPBMNPOOOMOPNOONOPMPPMNPQPPPOMOPO

C
v
'

new. ooxppxn pun. —o~
oo—— ooxmwx~p FNMP oo~
oonN ooxnwxo mep can
ooep ooxsmx~ —~m— non
nee ooxe \n «Nap soN
upo~ ooxn \~ pwnp can
ooo— ooxo~\~— pwnp no~
4%

meo ooxepxn own. eo~
ooop oo\e~\~ omn- now
own. ooshwxu oflnp Noa
mos ooshpxm ommp po~
npo ooxn \m o~mp oow
onmp ooxupxo omnp gnu
onop ooxopxpp opm— oh~
men. ooxswxop orm— NNN
on- ooxpp\n h—n— nun
ooop ooxn \m swap nNN
ooo— ooxouxo s—np es~
own— ooxonxo swap nNN
oo~p oo\on\n hump NLN
on. ooxo \e spa. psN
one. ooxo—xn upm— os~
on oo\e~\o hpnp QQN
nee oo\n~\o sump ooN
oo~ ooxo~xn 5pm. son
o ooxppxw crap cow
oohp ooxo \n can. now
one. ooxo \p opn— ecu
onsp ooxowxpp orn— non
oan ooxswxo npmp Now
one. ooxouxw. mama wow
oop— ooxo \pp mam. ooa
mnN ooxo ... awn. onN
oan ooxspxmp npnp Gnu
ooow ooxonxs npn— saw
oee ooxomxo epmp onw
new. oo\— .5 epnp nnN
ono- ooxpmxo— ewmw en~
ooop ooso~\o— epn- nnN
onpp ooxe \p- e—np NnN
ne- ooxon\~p epnw pan

...... ......

219

3:32:53 ..................................................................................... . ......
gout-h ..o

Nam—
Nnmp
Nnmp
Nnnp
Nnmp
«nap
pnnp
pnm—
pnmp
pnmp
wnmw
pnm-
pump
pnm-
pnnw
pnmp
pump
pump
pump
pnmw
pnmp

enm—
on—
onp
onp
uNmp
onF
DNmp
cwa
QNm-
ONm-
ONmF
«Na—
«Nap
«Nap
«Nor
nNm—
nNmp
nNmp
me—
—Nmp

av.- xol. .8.; .323 ..o 05 u... ...o
co..unouoa ogu tuna-.0 0.0.x.»
5.883.. a... a... P23 .53.
5.882... a... a... 05.93
5.38.8: 2... u... ...u

9. 2: a... 9.9...

5.333 co to...

coo-a 2: a... 3n

suagu as» u.: cool:

“.3.: 2.» u... con...

9. .5 a... :8!

u: .5 a... coo-...

on.- ocog Cool. 23 u... x33
ou.o econ: no.» u: ogu u.; gown-Lu
533.3 co to...

...—38.8.. 05 a... cool.
c.7533... 2: a... 0.9....»

.xauaa as» log. crow ..o. Lance-con
Ix. 2: In... Son :3 .3983...
8.3.88 .5 a... a...
5.383.. .5 u... ..3

5.3.32. 2.» a... .33-
5.....oo co to...

8.2..3 ca .82.

uc.xou nose mecca

coo-3 as» u.g coo-a
5.238 no .82.

cone: 05 a... Ix.

accuse Lo>o cog-3

.53. 2.» u.... too...
5.5.32. 05 a... cool.
stun-ton .5 .8520 cool.
guns log; .09 ogu u.g coo-3
u: cgu u.s coo-a

00.5 2.» a... It

5.383.. 2.» a... bio...
.xanaa osu u.g coo-3

gun can u.g coca:

05.. .95... mo ...: 2.» u... ..3
5.....oo co .82.

..oo .5 a... ...u

P’OOPFOOOOOOOOOOPPFPO

OOOOOFOOOPPOOOOOODOP

LID

fifiii.....

88
hi-

g
......2§§£2

...:
89:.»
sun

33...,
:33
9.9...

..8
3.3..
..8

an.
89':
coal:
.80.:
cone:
*5...—
g
.599

i

:23.
:33.

no...
..

a 3

ESE§§§85§§§§§§§§§§§

.Iu

Sting

.32. A... .3326. 3.9.5... .9 .2. ...v

p p 0 nm.— ooxpN\¢ Nnmp Nnn
o o p on oo\QN\m Nnmp pan
0 p o moNN ooxs \N— Nnmp onn
o p o onNN o¢\0n\o Nnnp ON»
a o p o oo\0N\n Nnm— uNn
N p a on.. ooxq \c pump «Nu
N p a can. ooxm \op pan. ONn
.N n o awn ooxp \— pump mNn
p p a cow noanxo pnmp eNn
p o p 090— noprxo pnmp nNn
p w o mNN- ooprxo pump NNM
p o p can ooxopxpp pnnp pNn
p p o oan ooxsprp pnmp oNn
p o 9 amp. oo\mp\p pump own
p p o uppN ooonxe pnm— own
a p a one. ooxopxw ammo spn
o o — m-FN aoxo \N pnnp own
a o p “we ooansN pnmp awn
o o p 699 noxoanp pump «pm
a o p emu cesspxn pnn. npn
o o - cop ooxpNxo pump N-n
gunman-g
o o — ooh. noxNNxop onn— ppm
N a 0 one— ooxopsz Oan own
p N o 990 oQ\N \n ONn— oon
N n — coo ob\m \p uwa can
N n 9 mph. ooxo~pr uNmp hon
N N 9 ans. oo\NN\o oan can
N a p can ocvaxo uNn- man
w o .N onwp ooxcwxo onw «on
o N o onpN obxpNxo QNmp man
a p a coup ouxsva NNm— Non
a o — onop ooxo \0 ON“. won
N p 9 cc. oosonxv «Nu. can
p N 9 ans cox: \o cNm— ooN
- c a on» ooxowa— «Nap ooN
o o w onu— oovaxp «Nap hoN
N N o cn—F oosopxs nNmp oaN
w r o ace. ooxpn\Np nNmp noN
p p p can unsoprp nNmp eoN
N N o nee ¢o~opxp wan nnN
N p o onnp ooxo \a— pan NON
09. .... at an: a»... in in:

220

..--.uo::.»couo ........................... ........ ....... . .................... .......-. ....... ....... ...................................

oww»
oww»
oww»
oww»
oww»
oww»
oww»
oww»
oww»
oww»
oew»
oew»
oew»
oew»
oew»
oew»
oew»
oew»
oew»
new»
new»
new»
new»
new»
new»
new»
new»
new»
New»

03. so... to.» cool. 2.. »... 3..
v8! 2.3..» 52.

..8 2.» »... .3..»

on... :9.» 2... 2.» ».... .3..»
on 2.» 2.. ..8

noon-an son

on as» ».g »:n

cu..».ouoa oz» ».g nauo.n
cl..».uuon on» ».g can
:n..»oouon 0:» ».g can
8.398 98....

.8.-3.3 ca .82.

u: 2.» »... .3..»

.38... 2.» .... ......

0.2. 2.» c. :8 ...:- cn...».ouon 2.» »... I...
5.5.8.. 2.. .... .3

no... no... 9. 2.» »... I...

..oo 2.» »... 3..

5.3.8.... 2.» »... .3..»
2.398 98....

c8... .5 2.. .8..»
5.3.30 5 v.2.

c8... 2.» .... .8.»

2.3:. 2.» »... .3.»

00.» a co..».onoa 0:» ».s can
ca..»oouon 0:» ».g gun.»

u. 2.» .... So...

I... 2.» »... u:

co..».onon 0:» ».s can

no: 52.

u: 2.» »... cool.

8.... a .3 3 v2.8... 8
u: 0:» u»: u:

5.233 co v.2.
.8.-.23 co ‘2.

...-null: 53

cu..».onon 0:» ».g can.)
.3..» 2.» »... .3..»

on.» 98.... I...» .3..» 2.» »... 3..
:35. 2.» »... an...»
.8.... .o :8 2.. .... ....

FOOC’O’OOOOOPOOPPOOOOPFFOOOOOOO

OOOOFOOOOOOO

coop:
:93.»
..u

on

nos.»

3....

9....

89:.»

....
.8:
.8...
.8.:

.8
.8...
....
.3....
.2.
3.
.8....
coop:
.8...
.2500...—
3.
....

.3
.3
893.»
..8
c8...
:8...
8o...
80:.»
can
.8...
c8...
.3
n:-

.._8. .... .8335 .33....» .. .2. .-..

N w o me»» oosowxn oww» nun
N o o ooh» ooxn \e oww» Nhn
» w » own» o0\e \o oww» »hw
» N o oo»» nn\»»\w oww» ohm
» o » owo on\»»\n oww» on»
» o o oon» nn\»N\e oww» new
» o » weo» ononx» oww» New
o » o w»o» nnxn»\»» oww» new
o » o wew» nn\nN\»» oww» wow
o o » w»n nnxe \N oww» eon
N o o w»n ooxeNx» oew» new
N o » one» ooxNNxe oew» New
» o » wew» nnxn \w oew» »nn
» » o oon ooxn»\N» oew» onn
» n w o onxo \N oew» own
» o » ow»» nnxn»\n oew» nwn
» N o oww» onanxo» oew» swn
» » o ono nn\» \0 oew» own
o o » w»N» oo\o»\» oew» wwn
N o o oon oo\»N\» new» ewn
N n » w»n» oo\nN\» new» wwn
N » o ooN» ooxw \» new» Nww
» e o ooh» onon\»» new» »wn
» o » oon nnxs \» new» own
» w o one» on\» \0 new» oew
o o » oo» cox» \0 new» new
» o » own» onxn \w new» New
» N o oo»» ooxnnxw new» new
o o » oo»» onwaxo» new» wen
1%
N o o weo» nnxn»\N» wnw» een
» » o oo»» ooxw»\n wnw» new
» o » ooo onxo»\» wnw» New
N N o oow»,nn\0»\e enw» »ew
N N o ooO» on\»»\n enw» oen
N n o oww» nanN\e enw» own
N o o onN» on\0»\»» enw» nwn
o » o oeoN ooxh»\s enw» swn
N » o oo»» nnxn \o» an» own
N » o w»n oo\w»\N an» wnn
» N » oNn» ooxo»\e an» ewn
» o o wn ooxN»\m wa» nun
com on. 4». cl.» o»oo. .19 taco

221

--...vflt..couo......--......-................ .......... . ............... ....... ............. .......... ..... ......................-.......

new» 2.3. 0...... .89. 2.. ... $8.. :3 .35.. o . .. 3.33m » o » oo»» onxn 3 new» e»e
8m. .3 .... .... .8 o . 8. .8 . a . .8. 83:. 8... ....
8.. 5.3.8... .... .... 8.... . - - 8.... o . a 8.. 8.8.. 8.. 2..
8.. 5.3.8... .... .... .3 . . - 8. a o . m... 8... ... 8m. ...
8.. 8...... .5... o . .... .8 ~ ~ . .8. 8..... no... 2.
a... 8...... 8..... o . .8... .... N . o .8. 8...... 8m. 8.
8m. ... .... .... 8.... o . 8..: .... ~ . o o... 8...... 8.. 8..
8m. 8...... .8.... o. 8.. o . .8... .... ~ ~ 9 98. 8.8.. «on. .3
8.. .83 .... .... .... o . .8 .8... ~ . o 8.. 8..... 8... 8..
8m. ..3 .5 .... coo... o . - :8... . o o 8 8.8:” no... no.
no... 83. 8.. .8.. .5 ... .... .83 o . - .8... . .. ~ .8 8.8.. 8.. 3..
a... o. .... .... .3 o . 8 8. . . o 8.. 8..... 8.. no.
a... 5.3.8.. .... .... 8.... . . - :8... a .. o .8. 8.8.. 8.. «3
a... 5.3.8... .... .... ......3 . . - 3...... a . . .8. 8..... 8m. ....
8.. 5.3.8... .... .... .88. . . . . .88. o . o 8.. 8.83 8.. 8.

qua-H.333
3.. .... .... .... .3 a . .8 8. .8.. ~ . o 3... 8.. a. .8. 8.
8.. 53:... ... ...... o . .... .8... . . a 8. 83... 8.. 8n
8.. 5.3.88. .... .... .83 . - . .8... a a . 8.. 8.. .. 8.. ....
.... .... .... .... .83 o . ... .8... . u o 88 8..... .8. 8.
.8. 5.3.8.. .... .... 8.... . . . 8.... o a . .8. 8..... ...... no.
.8. 58 .... cl . .83 .... .8. .8. .3 o . .8... 8. o o . ...... 8..... .8. .3..
8.. 8.. .... ...... .83 .... .... 8.... o . .8... 5.... ~ ~ 9 9.. 8...... 8.. no.
8.. .... ...... .883 .... .... .83 o . 3...... .8... ~ ~ . 88 8..... 8.. ....
8.. .83 2.85.. .... .... .83 o . .8... .8... ~ . . 8.. 8.. k 8.. .8
8.. u. .... .... 8.... o . .- 8..... . . . 82 8.8.. 8.. 8.
8.. 8.. .... ...... ... .... .... ......3 o - .... 3...... . . a 8.. 8...... 8m. .8
8.. .... ...... 8..... .... .... .3 a . 8..... 8. . . a 8. 8...... 8m. 8.
8.. 8..... ....6 8 .8 .3... .... ... .... .3 o . . 8. . 8 o 8.. 8...... 8.. .8
8.. 5.3.8... .... .... 8.... . . . 8.... o . o 8.. 8..... 3.... 8.
8.. 5.3.8... .... .... .83 ~ . . .8... o ~ 9 ...... 8...... 8m. 8..
a... ..3 .... .... .... o . .... .... . o . a... 8.8.. ...... .8
8.. 5.3.8.. .... .... 8..... . . . 8.... o . o 8.. 8..... 8.. 8n
.8. 3.8.2. .85... .85... .83 o . .3.... .8... . o o .8. 8.. x. .8. 8n
.... .8 .5 .... 3:... a . .... ...:E. . n . 8.. 8.8.. .9... .8
.... 5.3.8... .... .... .83 . . . .8... o a . .8. a... a .8. 8.
3.. u. .... .... .88. o . u. .88. . . o 8. 8:2... 8.. ...
.3.. 5.3.8... .... .... .8 o . u. ... . ~ g 8.. 8...... .8. ...
...... 8.2:... 8 ...... o .... .8... .... N . o 8... 8... 8.. n... ...
.... 5.3.8... 8.85.. ...... .... .... a... . . . .8.. . ... a 8. 8...... n... ...
.... .3... 8.. .82 .... ... S8 .... 8. o . - 8. . ~ . 8.. 83 x. .9... ....
a... .83 . 8. .... .... 8.... .8.... .... .... .3 o .8... 8.... a. ...... n . . ... 8..... ...... ....

.lu 3.258.185 A.-. 3.9)... :.......> »o .02 .-.v cos ..x. a». ': Ouoo .lx .Quon

222

1:328:53 ..........................................................................................................................

2.. 833.88 .... .... 8.. N . . 8. o . . 8.. 8....N 2.. ...
2.. 5.3.8.. .... .... 8.. . . . 8. o o . .... .... .... 2.. ....
2.. 5.3.8.. .... .... 8.. . . . 8. . . . 8.. 8..... 2.. ...

«gala—flu
2.. 5.3.... ... .8.. . . 8. ...... N . . SN. .8.. ... 2.. N...
2.. 5.3.8.. .... .... .83 . . . .8.. o . . .... .... .. 2.. ...
2.. 5.3.8.. .... .... 8.... . . . 8..... . . . 8.. 8..... 2.. ...
2.. 5.3.8.. .... .... .... . . . ...... o o . 8.. 8.2.. 2.. ...
.... 8...... 8.... o . ...... .8.. N . . .... 8..... .... 8..
.... 8.3.... ... .8.. o . 8. .8.. N o . .2. .... ... .... ....
.... u. .... .... ...... . . .. ...... . o . .... .... ... .... ....
.... .88. .... .... ...5. ... . . .8.... ...... ... . . . .... .... .. .... ....
.... 5.3.8.. .... .... ...... . . . ...... o . o 8.. .... .. .... ....
.... 5.3.8.. .... .... ...... . . . 8.... o . a .8 8..N.~. .... ......
.... 5.3.8.. .... 8...... ...... . . . ...... o . . .2. 8.2.. .... N3
.... 8..... .... .... 8...... 3...... . . . 3...... o . ~ ...N 8.8.. .... ....
.... 5.3.8.. .... 8...... 3...... . . . 3...... o . . .8 8...... .... ....
.... 5.3.8.. .... .... .83 . . . .8.. o o . 8.. 8.. .. .... ...
.... 5.3.8.. .... .... .88. . . - .88. . . o .8. 8..... .... 8.
.... 5.3.8.. .... .... u. . . .. u. o . a on. 8..... .... ....
.... 5.3.8.. .... .... ... . . . .... o a . .... 8.8.. .... 8.
.... 5.3.8.. .... .... 8.. . . . 8. . o . .... .... .. .... ...
.... 5.3.8.. .... .... 8. . . . 8.. . .. . 8.. 8..... .... ...
.... 5.3.8.. .... .... 8.. . . . 8. o o . .8 8.8.. .... 8..
.... 5.3.8.. .... .... 8.. . . . 8.. o . . ...... 8.. .. .... ......
8.. .. .... .... 5.... . - .. ...... . .. ... 8.. ......N. 8.. ...
8.. ... .... .... .83 . . ... .8.. . N . 8.. 8.. .. 8.. ...
8.. .8.. 8...... .... .... .88. . . .8.. .88. . . . .... 8.~ ... 8.. .N.
8.. ... .... .... ...... . . .88. u. . N a ...N 8......N 8.. 8..
8.. 5.3.8.. .... 8...... 3...... . . . 3...... o o . .3 8...... 8.. a.
8.. 5.3.8.. .... .... ...... . . . ...... o . o 8.. 8..... 8.. 8.
8.. 5.3.8.. .... .... ... . . . ... o . . .... ...N~.N 8.. ....
.... .... ...... .83 .... .... .83 . . .8.. .8.. ~ . ~ . ...N 8...... .... ...
.... .83 8...... .... .... ...... . . ...... .8.. . ~ . .2. 8.8.. .... 8.
.... ...... 8...... 8.. . . . 8. . n o 8... .... .. .... NN.
.... 5.3.8.. .... .... .83 . . . .8.. . . . 8N ...~... .... a.
.... 5.3.8.. .... .... .83 . . . .8.. o . . 8. 8..... .... .N.
.... 5.3.8.. .... .... ... . . . ... o . . .... 8..... .... ....
.... 5.3.8.. .... 8...... 8.. . . . 8.. o . . .... 8....N .... ...
8.. 8...... 8.... . . 8. .8.. N ~ .. 8.. 8..... 8.. ...
8.. 8...... 8.... . . ... ...... N N . .... .... .N 8.. ...
8.. 8.3.... ... 8.... o . ... .... ~ . . .... 8..... 8.. ....

.... 3...... .8.. .... .23.»... .33.... .. .... .... ... ...: ... ...... ....

a: 0‘8“

2233

..uo¢ .... u.>.o)c_ no.o_go> .0 .oa ...v

.12

c-.Luoouoa on» u.: Loo
co_guoouoa 0;“ u.g can
ca_guaonon use u_: coo-a
ca.guuouon ecu u.g snag»
ca_gu¢ovon any a.; Leo
no.5unouon on» u_; coo-3
c-pguoouon as» u.: gnu
c-_.u.ooon cg“ u.g .an
:-_gua~von as» u_: saunas»
cant nongau can
uc_noogo aces:

soda sag» nan oz» u.g coo-3
gnu venue. as» u.: Lou
ug-u ..au 0:» u_z coo-3
soon cog» nan u_g coo-a
non-inn a...

xuagu ogu u.g goo
ca_guuonoa Ogu u_g coca:
on any «.2 .xauaa
c..g..auon as» u.g a...
c-.Luuouon ogu u.g Lou
Loan-nu as» u.g u:

u: 0;» u.g can
co.guaonon ogu u.g snag»
co.guuouon 0;» u.; gonna
c...~..uoa «cu u.; .an
oc.ooogo econ:

noon «ago: ac. 09cc» ogu u.z .xnuau
mucus-u a...

.xanau ac.ucuuu 0:» v.5 goo
can» use u_: coo-a

on as» u.: coo-3

on 0;“ u.g coon:

on o:« a.: sauna

u: as“ ».g can.

on ogu u.= you
ca_guuonon osu u.; coo-z
cowguuouoa ogu u_: coo-1
ca..u.ouoa «g» u_; 89:.“
al.5uuouoa use a.: use.
co.guuouon use v.5 Lou
couguuonoa osa u.: sou
co.guaoton one u.g Loo
co.guoouun as» u_; gnu

PFC-PPFPFOOOOOOOOOOFPPOOPFOFOOOOOOOOPPI—OPFFFI—

us-u ads.

8035»

gang»
coo-3
.xauao
gang
gnu
beau-g»
as.
«935p
snag»
a:-
con-S
..auaa
.3.
gnu
coca:
coo-a
coo-3
89:."
noon
5.9
coo-a
coon:
song»
goes
gnu
sou
sou
..u

GOOGOQuad-Pv-PPFNNNNOOOPPOOOOv-I-PPNNNOQOOI-OOCOO
Ov-v-I-OOOOPOOv-v-Ovv-v-~t°Gov-FPOPGNOQONOMOPPOPOOOPP
POGO!-PPPOPPOOFOOv-OPI-PPOOPOPOOOOOOOOOOPOI—PFOO

onb aoxonxm
“pow noxo \c
ono— onxewxe
ono noxo~xpp
ons— ooxm-xh
ono~ ooxopxnp
ecu oaxn~xm
coup ooxopxh
ono cox—nxn
one onxn \o
anew ooxn \n
onnp ocxon\n
anew oo\p \s
enn- ooxv~xpp
m'o ooxo \pp
9 noxo \u
onn— ooxowxpp
onp— noxo’xm
comp ooxwwxv
mpnw oo\~ \N
coop ocxo~x~
comp ooxo \N
ono— noxo \o
on~p oosowxo
anew ooxn~\Np
mph ooxo \n
80— 83.4.3
oop~ oox—wxn
ooo— anxnuxw
coop noxc \e
095- oaxvpxmp
oon— onxo \N
000— noxouxb
nee noxn~xo
onmp ooxo—xo
nvp~ naxn \0
mos- ooxo \p
upon ocxp—\~
o~n ooxn—xwp
agar ooxswxn
onbp ooxo~xo
095— ooxo—xo
oop— ooxop\b
coop ooxnpx~

at.» tuna

osm—
osm—
bump
95m—
9km.
num—
mun—
nnmw
sump
sump
sump
yum—
yum-
sump
sump
sump
sump
sump
sump
nun.
num—
nun.
nun.
nhmp
Num—
mum—
nun.
Nump
Nsnp
num—
mun—
Nump
who.
thp
msmp
mun—
thp
Num—
~hm—
Nun.
thp
thp

IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII .-...-II.-.IOIIII'I...I|I-.

woe
00v
one
One
sac
One

224

Be 5.338.. .5 a... .3 - - .3 8.2 8: R Ba. «.3
83 Bull. .88 - .88 ..8 82 3:5. ~93 :n
85 .38: .5 a... “.8.; . 38...: .8: 9m 82:“ 3.... 2m
8m. 83:8 8 .82. . :85 3. one Sat: 8.: Sn
.8. 9. 2: a... 3.2.. . 9. 3.9: on. 853. .8. 8n
.8. 5:282. .5 a... ".8.: . . ..8: on 853 .8. Ba
83 is... a a. a... 8.! :33» :82. a. on: 835 8m, 8m
85 83:8 8 .8... . ..8 88: $2 8: R 8m. 3m
8m. ..8 8.23. 2: a... ..25 . ..8 .8.: 8m 83 > Sm. an
89 38...... ..8 . .3 ..8 82 8: a. 8m. 3m
89 u. .5 2.. .3 - 8 5. 32 835 8“. 8m
83 5.3.88 .5 a... “.8.: . . .8.: at 83 > 8m. .8
83 5.3.88. .5 a... ..8 . . ..8 8: 83:“ 8.: 8n

.Iu 8.8.5.. .3: A... .8302: .302; .0 .02 ...v 8.. .... a: ‘3 33 .5 .3...

“FOOOOOFOOOOP
O
OOPPPNHOPNNPO
QmeOOmOWOO—v

Appendix B

The SPSS Program for Data Analyses

 

225
SPSS COMMAND AND PROGRAM FILES FOR DATA ANALYSIS

SET MORE = OFF.

************************************************************

* 2-LANE ANALYSIS.
************************************************************

TITLE ’2-LANE ANALYSIS’.

DATA LIST FREE FILE = ’2L2W.DAT’/ KM (A) RIBBON SPATHS
GDRAIL PWIDTH SWIDTH PMARKS INTSEC ISLAND PVCOND SIDEOB
HZINDX ACCFRQ ACCRAT.

VARIABLE LABELS KM ’KILOMETER NO.’
RIBBON ’RIBBON DEVELOPMENT’ /
SPATHS ’SPECIFIC LATERAL PATHS’/
GDRAIL ’DEFICIENT GUARDRAIL’
PWIDTH ’DEF. PAVEMENT WIDTH’ /
SWIDTH ’DEF. SHOULDER WIDTH’ /
PMARKS ’DEF. PVT. MARKINGS’
INTSEC ’INTERSECTION DEFCY.’
ISLAND ’ISOLATED LANE DEFICIENCIES’ /
PVCOND ’PVT. CONDITION RATG.’/
SIDEOB ’ROADSIDE OBSTRUCTION’ /
HZINDX ’HAZARD INDEX’]

ACCFRQ ’ACCIDENT FREQUENCY’I
ACCRAT ’ACCIDENT RATE’.
SAVE OUTFILE = ’2L2W.SYS’.

GET / FILE ’2L2W.SYS’.

COMPUTE ANACFQ=(ACCFRQ/3.0).

COMPUTE ANACRT=(ACCRAT/3.0).

VARIABLE LABELS ANACFQ ’ANNUAL ACCIDENT FREQUENCY’
ANACRT 'ANNUAL ACCIDENT RATE'.

DESCRIPTIVES / VARIABLES ALL.

* CORRELATIONS.

CORRELATIONS I VARIABLES RIBBON SPATHS GDRAIL PWIDTH SWIDTH
PMARKS INTSEC ISLAND PVCOND SIDEOB ANACFQ WITH RIBBON SPATHS
GDRAIL PWIDTH SWIDTH PMARKS INTSEC ISLAND PVCOND SIDEOB
ANACFQ.

* FREQUENCY DISTRIBUTIONS.

FREQUENCY / VARIABLES RIBBON SPATHS GDRAIL PWIDTH SWIDTH
PMARKS INTSEC ISLAND PVCOND SIDEOB ANACFQ HZINDX / FORMAT
NOTABLE / HISTOGRAM NORMAL / PERCENTILE 15 85 / STATISTICS
DEFAULT.

226
* TEST FOR MULTICOLLINEARITY.
REGRESSION / VARIABLES RIBBON SPATHS GDRAIL PWIDTH SWIDTH
PMARKS INTSEC ISLAND PVCOND SIDEOB ANACFQ / STATISTICS
COLLIN / DEPENDENT ANACFQ / METHOD ENTER.
REGRESSION / VARIABLES RIBBON SPATHS GDRAIL PWIDTH SWIDTH

PMARKS INTSEC ISLAND PVCOND SIDEOB ANACFQ / STATISTICS TOL
/ DEPENDENT ANACFQ / METHOD ENTER.

* REGRESSION ANALYSIS.

* BACKWARD ELIMINATION METHOD.

REGRESSION / VARIABLES RIBBON SPATHS GDRAIL PWIDTH SWIDTH
PMARKS INTSEC ISLAND PVCOND SIDEOB ANACFQ / DESCRIPTIVES /
DEPENDENT ANACFQ / METHOD BACKWARD.

* FORWARD SELECTION.

REGRESSION / VARIABLES RIBBON SPATHS GDRAIL PWIDTH SWIDTH
PMARKS INTSEC ISLAND PVCOND SIDEOB ANACFQ / DESCRIPTIVES /
DEPENDENT ANACFQ / METHOD FORWARD.

* STEPWISE SELECTION.
REGRESSION / VARIABLES RIBBON SPATHS GDRAIL PWIDTH SWIDTH

PMARKS INTSEC ISLAND PVCOND SIDEOB ANACFQ / DESCRIPTIVES /
DEPENDENT ANACFQ / METHOD STEPWISE.

* EXAMINE TRAFFIC VARIABILITY EFFECT.

CORRELATIONS / VARIABLES RIBBON SPATHS GDRAIL PWIDTH SWIDTH
PMARKS INTSEC ISLAND PVCOND SIDEOB WITH ANACFQ ANACRT.

REGRESSION / VARIABLES RIBBON GDRAIL ANACRT / DEPENDENT
ANACRT / METHOD BACKWARD.

* EXAMINE RESIDUALS.

REGRESSION / VARIABLES RIBBON GDRAIL ANACFQ / DEPENDENT
ANACFQ / METHOD BACKWARD / RESIDUALS DEFAULT HISTOGRAM
(RESID BRED) OUTLIERS NORMPROB (RESID FRED).

* REGRESSION ANALYSIS WITH ONE DATA SET.

* GEOMETRIC DATA ONLY.

* BACKWARD ELIMINATION METHOD.

REGRESSION I VARIABLES RIBBON SPATHS GDRAIL PWIDTH SWIDTH
PMARKS ANACFQ / DEPENDENT ANACFQ / METHOD BACKWARD.

 

227

* FORWARD SELECTION.
REGRESSION I VARIABLES RIBBON SPATHS GDRAIL PWIDTH SWIDTH
PMARKS ANACFQ / DEPENDENT ANACFQ I METHOD FORWARD.

* STEPWISE SELECTION.
REGRESSION / VARIABLES RIBBON SPATHS GDRAIL PWIDTH SWIDTH

PMARKS ANACFQ / DEPENDENT ANACFQ / METHOD STEPWISE.

* CONFLICT DATA ONLY.

* BACKWARD ELIMINATION METHOD.
REGRESSION I VARIABLES INTSEC ISLAND ANACFQ / DEPENDENT
ANACFQ I METHOD BACKWARD.

* FORWARD SELECTION.
REGRESSION / VARIABLES INTSEC ISLAND ANACFQ / DEPENDENT
ANACFQ / METHOD FORWARD.

* STEPWISE SELECTION.
REGRESSION I VARIABLES INTSEC ISLAND ANACFQ I DEPENDENT
ANACFQ / METHOD STEPWISE.

* SUBJECTIVE RATING DATA ONLY.

* BACKWARD ELIMINATION METHOD.

REGRESSION I VARIABLES PVCOND SIDEOB ANACFQ I DEPENDENT
ANACFQ I METHOD BACKWARD.

* FORWARD SELECTION.

REGRESSION / VARIABLES PVCOND SIDEOB ANACFQ I DEPENDENT
ANACFQ / METHOD FORWARD.

* STEPWISE SELECTION.

REGRESSION / VARIABLES PVCOND SIDEOB ANACFQ / DEPENDENT
ANACFQ / METHOD STEPWISE.

* HAZARD INDEX DEVELOPMENT.

COMPUTE HI = (RIBBON + SPATHS + GDRAIL + PWIDTH + SWIDTH +
PMARKS + INTSEC + ISLAND + PVCOND + SIDEOB).

DESCRIPTIVES I VARIABLES ALL.

CORRELATIONS I VARIABLES RIBBON SPATHS GDRAIL PWIDTH SWIDTH
PMARKS INTSEC ISLAND PVCOND SIDEOB HZINDX HI WITH ANACFQ.

* GRAPH BETWEEN HZINDX AND ANACFQ.

PLOT I FORMAT REGRESSION I TITLE ’HAZARD INDEX 2-LANE’I

VERTICAL 'Annual Acc. Freq'] HORIZONTAL 'Hazard Index' [PLOT
ANACFQ WITH HZINDX.

 

228

************************************************************

* 4-LANE ANALYSIS.
************************************************************

TITLE ’4-LANE ANALYSIS’.

DATA LIST FREE FILE = ’4LDH.DAT’/ KM (A) RIBBON SPATHS
MEDOPN GDRAIL PWIDTH SWIDTH PMARKS INTSEC ISLAND PVCOND
SIDEOB HZINDX ACCFRQ ACCRAT.

VARIABLE LABELS KM ’KILOMETER NO.’
RIBBON ’RIBBON DEVELOPMENT’ /
SPATHS ’SPECIFIC LATERAL PATHS’I
MEDOPN ’MEDIAN OPENINGS’I
GDRAIL ’DEFICIENT GUARDRAIL’
PWIDTH ’DEF. PAVEMENT WIDTH’ I
SWIDTH ’DEF. SHOULDER WIDTH’ I
PMARKS ’DEF. PVT. MARKINGS’ I
INTSEC ’INTERSECTION DEFCY.’
ISLAND ’ISOLATED LANE DEFICIENCIES’ I
PVCOND ’PVT. CONDITION RATG.’I
SIDEOB ’ROADSIDE OBSTRUCTION’ I
HZINDX ’HAZARD INDEX’I
ACCFRQ ’ACCIDENT FREQUENCY’I
ACCRAT ’ACCIDENT RATE’.

SAVE OUTFILE = ’4LDH.SYS’.

GET I FILE ’4LDH.SYS’.

COMPUTE ANACFO=(ACCFRO/3.O).

COMPUTE ANACRT=(ACCRAT/3.0).

VARIABLE LABELS ANACFQ 'ANNUAL ACCIDENT FREQUENCY'
ANACRT 'ANNUAL ACCIDENT RATE’.

DESCRIPTIVES / VARIABLES ALL.

* CORRELATIONS.

CORRELATIONS I VARIABLES RIBBON SPATHS MEDOPN GDRAIL PWIDTH
SWIDTH PMARKS INTSEC ISLAND PVCOND SIDEOB ANACFQ WITH RIBBON
SPATHS MEDOPN GDRAIL PWIDTH SWIDTH PMARKS INTSEC ISLAND
PVCOND SIDEOB ANACFQ.

* FREQUENCY DISTRIBUTIONS.

FREQUENCY I VARIABLES RIBBON SPATHS MEDOPN GDRAIL PWIDTH
SWIDTH PMARKS INTSEC ISLAND PVCOND SIDEOB ANACFQ HZINDX I
FORMAT NOTABLE I HISTOGRAM NORMAL / PERCENTILE 15 85 I
STATISTICS DEFAULT.

* TEST FOR MULTICOLLINEARITY.
REGRESSION I VARIABLES RIBBON SPATHS MEDOPN GDRAIL PWIDTH

SWIDTH PMARKS INTSEC ISLAND PVCOND SIDEOB ANACFQ I
STATISTICS COLLIN I DEPENDENT ANACFQ I METHOD ENTER.

229

REGRESSION I VARIABLES RIBBON SPATHS MEDOPN GDRAIL PWIDTH
SWIDTH PMARKS INTSEC ISLAND PVCOND SIDEOB ANACFQ I
STATISTICS TOL I DEPENDENT ANACFQ I METHOD ENTER.

* REGRESSION ANALYSIS.
* BACKWARD ELIMINATION METHOD.

REGRESSION I VARIABLES RIBBON SPATHS MEDOPN GDRAIL PWIDTH
SWIDTH PMARKS INTSEC ISLAND PVCOND SIDEOB ANACFQ I
DESCRIPTIVES I DEPENDENT ANACFQ I METHOD BACKWARD.

* FORWARD SELECTION.

REGRESSION I VARIABLES RIBBON SPATHS MEDOPN GDRAIL PWIDTH
SWIDTH PMARKS INTSEC ISLAND PVCOND SIDEOB ANACFQ I
DESCRIPTIVES I DEPENDENT ANACFQ I METHOD FORWARD.

* STEPWISE SELECTION.

REGRESSION I VARIABLES RIBBON SPATHS MEDOPN GDRAIL PWIDTH
SWIDTH PMARKS INTSEC ISLAND PVCOND SIDEOB ANACFQ I
DESCRIPTIVES I DEPENDENT ANACFQ I METHOD STEPWISE.

* EXAMINE TRAFFIC VARIABILITY EFFECT.

CORRELATIONS I VARIABLES RIBBON SPATHS MEDOPN GDRAIL PWIDTH
SWIDTH PMARKS INTSEC ISLAND PVCOND SIDEOB WITH ANACFQ
ANACRT.

REGRESSION I VARIABLES RIBBON GDRAIL INTSEC ANACRT I
DEPENDENT ANACRT I METHOD BACKWARD.

* EXAMINE RESIDUALS.

REGRESSION I VARIABLES RIBBON GDRAIL INTSEC ANACFQ I
DEPENDENT ANACFQ I METHOD BACKWARD I RESIDUALS DEFAULT
HISTOGRAM (RESID PRED) OUTLIERS NORMPROB (RESID PRED).

* REGRESSION ANALYSIS WITH ONE DATA SET.
* GEOMETRIC DATA ONLY.

* BACKWARD ELIMINATION METHOD.
REGRESSION I VARIABLES RIBBON SPATHS MEDOPN GDRAIL PWIDTH
SWIDTH PMARKS ANACFQ I DEPENDENT ANACFQ I METHOD BACKWARD.

* FORWARD SELECTION.
REGRESSION I VARIABLES RIBBON SPATHS MEDOPN GDRAIL PWIDTH
SWIDTH PMARKS ANACFQ I DEPENDENT ANACFQ I METHOD FORWARD.

* STEPWISE SELECTION. '
REGRESSION I VARIABLES RIBBON SPATHS MEDOPN GDRAIL PWIDTH
SWIDTH PMARKS ANACFQ I DEPENDENT ANACFQ I METHOD STEPWISE.

230
* CONFLICT DATA ONLY.

* BACKWARD ELIMINATION METHOD.
REGRESSION I VARIABLES INTSEC ISLAND ANACFQ I DEPENDENT
ANACFQ I METHOD BACKWARD.

* FORWARD SELECTION.
REGRESSION I VARIABLES INTSEC ISLAND ANACFQ I DEPENDENT

ANACFQ I METHOD FORWARD.

* STEPWISE SELECTION.
REGRESSION I VARIABLES INTSEC ISLAND ANACFQ I DEPENDENT
ANACFQ I METHOD STEPWISE.

* SUBJECTIVE RATING DATA ONLY.

* BACKWARD ELIMINATION METHOD.

REGRESSION I VARIABLES PVCOND SIDEOB ANACFQ I DEPENDENT
ANACFQ I METHOD BACKWARD.

* FORWARD SELECTION.

REGRESSION I VARIABLES PVCOND SIDEOB ANACFQ / DEPENDENT
ANACFQ I METHOD FORWARD.

* STEPWISE SELECTION.

REGRESSION I VARIABLES PVCOND SIDEOB ANACFQ I DEPENDENT
ANACFQ I METHOD STEPWISE.

* HAZARD INDEX DEVELOPMENT.

COMPUTE HI 8 (RIBBON + SPATHS + MEDOPN + GDRAIL + PWIDTH +
SWIDTH + PMARKS + INTSEC + ISLAND + PVCOND + SIDEOB).

DESCRIPTIVES I VARIABLES ALL.

CORRELATIONS I VARIABLES RIBBON SPATHS MEDOPN GDRAIL PWIDTH
SWIDTH PMARKS INTSEC ISLAND PVCOND SIDEOB HZINDX HI WITH
ANACFQ.

* GRAPH BETWEEN HZINDX AND ANACFQ.

PLOT I FORMAT REGRESSION I TITLE ’HAZARD INDEX 4-LANE’

/VERTICAL 'Annual Acc. Freq'lHORIZONTAL 'Hazard Index' [PLOT
ANACFQ WITH HZINDX.

FINISH.

 

INTERr