. . :
333‘N ..
. 57:3. :1
. I .2 .

(Ah
.. r5.“

I.— .u
twp.
h .
. . ‘
.133)... u.
l‘f

gags?
24a,

91
. -
bauxiéwfi
3.3.9.5? .
.. Java”. 3 ‘ "hm!
.uueflwuw 2. . J:
‘4‘

$ .

tufidfiz.

a?)

 

£3 . a“ cum
<.F..r.,cnii .Jfiwfi : .M.“
$§a§§fifl$fiflhfiv . _
inn». . h 5&2
in; a» a»... (114*. fig
.. awn a}
3.

.3 .1
1:31.. «1.2!!!! . A.
. .21. 155.7%
3:... > II
V
.

'1‘. on
:5 .WM

.

1.71 .1.

a: . .wr‘.
2.. :3...

g m
.23...
.Mu.‘ .17.? ’5 1‘ Jmhmfii

of x 1.21:. 3
‘u 3:3 ..

 

 

.1...
.-
5%1
2!
,4). \n
. .Kuflfiffw.
A I.
3! IN 3::
3L” 1.
\ 54 x!
.123. 11...
.a (3.4:: .

 

t9. .
s f . a... .:
:15. I Mr. .
, , lama: gm”? 1»!
.53. .tgasv 2... £JK§QIW
a. 3.3.2.9.) . 1
.. £12...

.3. r
viruu‘ a
.27runmvf. nah-1.
mmngiifinm

:\ I .
5:: 3.. u)“;
.Oafittgtfiiltaaux. as.
) . ... Ii .1 . .
Stranxcnilfi mu; .II. .v u? o

55: V

is”

E $2..

 

 

. vr‘_:.m.ak.:bqq4o

 

looé

This is to certify that the
dissertation entitled

An Analysis of CMS Impact on Incident-Based Congestion

presented by

ln-Kyu Lim

has been accepted towards fulfillment
of the requirements for the

Ph.D. degree in Civil and Environmental
Engineering

 

 

51/40ng «Cm/flaw/éi

Major Professor’s‘téignature

7/22/2005

 

Date

MSU is an Affirmative Action/Equal Opportunity Institution

LIBRARY
Michigan $tate
l UniversfiY J

 

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE

DATE DUE

DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

2/05 chIfiEfDateDmJndd-pjs

 

AN ANALYSIS OF CMS IMPACT
ON INCIDENT-BASED CONGESTION
By

In-Kyu Lim

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

2005

ABSTRACT
AN ANALYSIS OF CMS IMPACT ON INCIDENT-BASED CONGESTION
By

In-Kyu Lim

Changeable Message Signs (CMS) are the most visible traffic control devices that
provide real-time traffic information about downstream congestion or potential delays to
drivers. Their use is intended to modify roadway travel choices through en-route
diversion. A high route diversion rate can reduce severe congestion, improve safety, and
network performance. However, given the same information, each of the CMS
deployment locations are not likely to produce the same level of effectiveness. Under
various conditions, based on human, traffic, and geographic characteristics, the CMS
vary in their effectiveness. Evaluation of CMS system performance using observed field
data was not strenuously researched in previous studies, and no studies have evaluated
their effectiveness under various conditions.

To analyze the effectiveness of CMS as it pertains to drivers’ route diversion
behavior, this study measures the percentage of traffic that diverts to an alternate route
during the time the CMS displays a message based on empirical field traffic data under
various conditions. A method to estimate travel time from upstream to downstream
location using the discrete Inductive Loop Detector (ILD) traffic data was developed an
implemented. The sensitivity of the diversion to four different factors: visual observation,
familiarity and time constraints, historical or existing traffic conditions, and geographic

location were also tested.

Dedicated to my parents,
Won-Uong Lim and Hyun-Sook Han
and

Kyung-Koo Lee and Hyo-Soon Ra

iii

ACKNOWLEDGMENTS

I would like to thank all the people who supported, guided and helped to
accomplish my dissertation. I would like to express my deepest gratitude to my esteemed
advisor, Dr. William C. Taylor for his remarkable guidance, inspiration, patience and
providing me with the opportunity and support to accomplish this research. His infectious
enthusiasm and unlimited passion have been motivated me through my graduate career at
Michigan State University. I would also like to express my appreciation to Dr. Thomas L.
Maleck, Dr. Ghassan Abu-Lebdeh, and Dr. Yimin Xiao for their helpful suggestions and
comments serving my dissertation committee.

I am grateful to Hyung-Suk Lee and George Howell, who as good friends, were
always willing to lend assistance to my research and writing efforts over the years. My
appreciation also is extended to my colleagues in Civil and Environmental Engineering.
Especially, I need to express my gratitude and deep appreciation to Linda Steinman,
Laura Taylor and Mary Wiseman for their friendship, hospitality, and administrative
assistance.

Finally, my accomplishments would not have been possible without the love,
patience and support of all my family. I am indebted to my mother, father and elder
brother for granting me their unconditional love and support. I am especially deepest
appreciation to my wife, So—Jung and my daughter Soo-Min for their patience, love and

support.

iv

TABLE OF CONTENTS

LIST OF TABLES ................................................................................................ viii
LIST OF FIGURES .................................................................................................. x
CHAPTER 1
INTRODUCTION ............................................................................................. l
1.1 Introduction ............................................................................................................. 1
1.2 Advanced Traveler Information Systems (ATIS) ................................................... 2
1.3 ATIS Characteristics ............................................................................................... 3
1.3.1 ATIS Benefits .................................................................................................. 5
1.3.2 ATIS Performance Measures ........................................................................... 5
1.4 Description of Research Area ................................................................................. 7
1.5 Description of Problem ........................................................................................... 9
1.6 Objective and Scope ............................................................................................. 10
1.7 Research Approach ............................................................................................... 10
CHAPTER 2
LITERATURE REVIEW ................................................................................. 12
2.1 Introduction ........................................................................................................... 12
2.2 Characteristics of Changeable Message Signs ...................................................... 13
2.2.1 Brief History of CMS Use ............................................................................. 13
2.2.2 CMS Types .................................................................................................... 14
2.2.3 CMS Message Contents ................................................................................. 14
2.3 Data Approach Methods ....................................................................................... 15
2.3.1 Stated Preference (SP) Approach ................................................................... 15
2. 3. 2 Revealed Preference (RP) Approach ............................................................. 16
2. 3. 3 Field Study Approach .................................................................................... 17
2. 4 Drivers’ Driving Characteristics ....................................................................... 18
2.4.1 Route Choice Behavior .................................................................................. 18
2.4.2 Knowledge of Alternate Routes ..................................................................... 21
2.4.3 Route Switching Behavior ............................................................................. 22
2.5 Factors Affecting a Drivers’ Route Diversion Decision ....................................... 24
2.5.1 Factors Related to Human and Socioeconomic Characteristics .................... 24
2.5.2 Factors Related to Traffic Conditions ............................................................ 25
2.5.3 Factors Related to Traffic Information .......................................................... 28
2.6 Prior CMS System Evaluation Studies ................................................................. 29
2.6.1 CMS Performance Evaluation ....................................................................... 29
2.6.2 User Impacts Base on Different Designs and Features of CMS .................... 35
2.7 Summary ............................................................................................................... 37

CHAPTER 3

DATA COLLECTION ..................................................................................... 40
3.1 Introduction ........................................................................................................... 40
3.2 Description of Inductive Loop Detector Data ....................................................... 40
3.3 Description of CMS Message Log Data ............................................................... 41
3.4 Study Site Selection .............................................................................................. 42
3.5 Data Collection Method ........................................................................................ 45

3.5.1 ILD Traffic Data Collection ........................................................................... 45
3.5.2 ILD Traffic Data Conversion ......................................................................... 46
3.5.3 Traffic Data Classification ............................................................................. 48
3.5.3.1 Normal Condition Data Collection ......................................................... 49
3.5.3.2 Normal Condition Data Screening and Filtering .................................... 49
3.5.3.3 Accident Condition Data Collection ....................................................... 53
3.5.3.4 Accident Condition Data Screening and Filtering .................................. 53

3.6 Summary ............................................................................................................... 54

CHAPTER 4

MEASUREMENT METHOD .......................................................................... 55
4.1 Introduction ........................................................................................................... 55
4.2 ILD Traffic Data Collection Method .................................................................... 56
4.3 Travel Time Measurement Method ...................................................................... 60
4.4 Queue Distance (QD) Estimation Method ............................................................ 63
4.5 Summary ............................................................................................................... 70

CHAPTER 5

DATA ANALYSIS AND RESULTS ............................................................... 71
5.1 Introduction ........................................................................................................... 71
5.2 Sensitivity Analysis Factors .................................................................................. 71

5.2.1 Familiarity and Time Constraint Sensitivity .................................................. 71
5.2.2 Visual Sensitivity ........................................................................................... 72
5.2.3 Traffic Condition Sensitivity ......................................................................... 72
5.2.4 Geographical Location Sensitivity ................................................................. 73
5.3 Analysis of the CMS Effect at Site 1 .................................................................... 73
5.3.1 Description of Site 1 ...................................................................................... 73
5.3.2 Diversion Ratio Analysis ............................................................................... 84
5.3.3 Sensitivity Analysis at Site 1 ......................................................................... 86
5.3.3.1 Familiarity and Time Constraint Sensitivity Analysis ............................ 86
5.3.3.2 Visual Sensitivity Analysis ..................................................................... 88
5.3.3.3 Traffic Condition Sensitivity Analysis ................................................... 89

5.4 Analysis of the CMS Effect at Site 2 .................................................................... 93
5.4.1 Description of Site 2 ...................................................................................... 93
5.4.2 Diversion Ratio Analysis ............................................................................... 97
5.4.3 Sensitivity Analysis at Site 2 ......................................................................... 98
5.4.3.1 Familiarity and Time Constraint Sensitivity Analysis ............................ 98

5.5 Analysis of the CMS Effect at Site 3 .................................................................. 100
5.5.1 Description of Site 3 .................................................................................... 100

vi

5.5.2 Diversion Ratio Analysis ............................................................................. 104

5.5.3 Sensitivity Analysis at Site 3 ....................................................................... 106
5.5.3.1 Familiarity and Time Constraint Sensitivity Analysis .......................... 106
5.5.3.2 Visual Sensitivity Analysis ................................................................... 107
5.5.3.3 Traffic Condition Sensitivity Analysis ................................................. 108

5.6 Analysis of the CMS Effect at Site 4 .................................................................. 108

5.6.1 Description of Site 4 .................................................................................... 108

5.6.2 Diversion Ratio Analysis ............................................................................. 112

5.6.3 Sensitivity Analysis at Site 4 ....................................................................... 113
5.6.3.1 Familiarity and Time Constraint Sensitivity Analysis .......................... 113

5.7 Analysis of the CMS Effect at Site 5 .................................................................. 114

5.7.1 Description of Site 5 .................................................................................... 114

5.7.2 Diversion Ratio Analysis ............................................................................. 117

5.7.3 Sensitivity Analysis at Site 5 ....................................................................... 118
5.7.3.1 Familiarity and Time Constraint Sensitivity Analysis .......................... 118
5.7.3.2 Visual Sensitivity Analysis ................................................................... 119

5.8 Geographic Location Sensitivity Analysis .......................................................... 120
CHAPTER 6
CONCLUSIONS AND RECOMMENDATIONS ............................................ 122
6.1 Conclusions ......................................................................................................... 122
6.2 Recommendation ................................................................................................ 125
APPENDIX .................................................................................................... 127
REFERENCES ............................................................................................... 144

vii

LIST OF TABLES

Table 2. 1 Characteristics of Stated and Revealed Preference Data Approaches ............ 17
Table 2. 2 Reasons for Taking Primary Routes (Huchingson et al., 1977) ..................... 18
Table 2. 3 Importance of Factors Affecting Route Choice Behavior (Polydoropoulou et
al., 1994) ................................................................................................................... 20
Table 2. 4 Number of Names and Landmarks in Route Descriptions (Wenger et al., 1990)
................................................................................................................................... 22
Table 2. 5 Case Study Results (Turner, Dudek and Carvell, 1978) ................................. 30
Table 3. 1 Format of Inductive Loop Detector Data ........................................................ 41
Table 3. 2 Format of CMS Message Log Database ......................................................... 42
Table 3. 3 Selected Study Site Information ..................................................................... 44
Table 3. 4 Normal Condition Samples at the Study Site ................................................. 50
Table 3. 5 Accident Condition Samples at the Study Site ............................................... 54
Table 5. 1 Descriptions of Collected Accident Conditions with CMS Message at Site 176
Table 5. 2 Accident and Related Normal Condition Traffic Volume at Site 1 ................ 85
Table 5. 3 One Sample t-Test (p < .05) for Downstream Volume Reduction ................. 86
Table 5. 4 Independent Sample t-Test for Familiarity and Time Constraint (p < .05) 88
Table 5. 5 Independent Sample t-Test for Visual Sensitivity (p < .05) ........................... 89
Table 5. 6 Independent Sample t-Test for Traffic Condition (p < .05) ............................ 90
Table 5. 7 Independent Sample t-Test for Traffic Condition Sensitivity (p < .05) .......... 91
Table 5. 8 Independent Sample t-Test for Route Traffic Condition Sensitivity (p < .05)92
Table 5. 9 Description of Collected Accident Conditions with CMS Message at Site 2 96
Table 5. 10 Accident and Related Normal Condition Traffic Volume at Site 2 .............. 97

viii

Table 5. 11 One Sample t-Test (p < .05) for Downstream Volume Reduction ............... 98
Table 5. 12 Independent Sample t-Test for Familiarity and Time Constraint (p < .05).. 99

Table 5. 13 Descriptions of Collected Accident Conditions with CMS Message at Site 3

................................................................................................................................. 102
Table 5. 14 Accident and Related Normal Condition Traffic Volume at Site 3 ............ 103
Table 5. 15 One Sample t-Test (p < .05) for Downstream Volume Reduction ............. 104

Table 5. 16 One Sample t-Test (p < .05) for Downstream Volume Reduction without
Cases 1, 2 and 3 ...................................................................................................... 106

Table 5. 17 Independent Sample t-Test for Familiarity and Time Constraint (p < .05) 107

Table 5. 18 Independent Sample t-Test for Traffic Condition (p < .05) ........................ 108
Table 5. 19 Description of Collected Accident Conditions with CMS Message at Site 4
................................................................................................................................. 1 1 1
Table 5. 20 Accident and Related Normal Condition Traffic Volume at Site 4 ............ 112
Table 5. 21 One Sample t-Test (p < .05) for Downstream Volume Reduction ............. 113

Table 5. 22 Independent Sample t-Test for Familiarity and Time Constraint (p <.05). 113

Table 5. 23 Description of Collected Accident Conditions with CMS Message at Site 5

................................................................................................................................. 1 16
Table 5. 24 Accident and Related Normal Condition Traffic Volume at Site 5 ........... 117
Table 5. 25 One Sample t-Test (p < .05) for Downstream Volume Reduction ............. 118

Table 5. 26 Independent Sample t-Test for Familiarity and Time Constraint (p < .05) 118
Table 5. 27 Independent Sample t-Test for Visual Sensitivity (p < .05) ....................... 119

Table 5. 28 Independent Sample t-Test for Geographic Location (p < .05) .................. 121

ix

LIST OF FIGURES

Figure l. 1 Generic ATIS System (Schiesel and Demetsky, 2000) ................................... 4
Figure 1. 2 CMS System in Southeast Michigan ............................................................... 8
Figure 2. 1 CMS Configuration ....................................................................................... 15
Figure 2. 2 Route and Time Switching for Home-to-work and Work-to-home Trips
(Mahamassani, Caplice, and Walton, 1990) ............................................................. 24
Figure 2. 3 Effect of Delay on Drivers Diversion (Huchingson etal., 1979) .................. 27
Figure 3. 1 Selected Study Sites ...................................................................................... 43
Figure 3. 2 ILD Data Collect Locations ........................................................................... 45
Figure 3. 3 Normal Condition Profiles Before Filtering at Site 1 .................................... 51
Figure 3. 4 Normal Condition Profiles After Filtering at Site 1 ...................................... 52
Figure 4. 1 Vehicle Trajectory Time-Space Diagram ...................................................... 57
Figure 4. 2 ILD Location Diagram .................................................................................. 60
Figure 4. 3 Upstream and Downstream Volume Profile at Site 1 ................................... 65

Figure 4. 4 Volume and Density Profiles at Site 1 Downstream Location (I-696 EB) 66

Figure 4. 5 Shockwave Speed Diagram ........................................................................... 68
Figure 5. 1 Site 1 Location ............................................................................................... 74
Figure 5. 2 Upstream Volume Distribution at Site 1 ....................................................... 77
Figure 5. 3 Downstream Volume Ratio Comparisons ..................................................... 79
Figure 5. 4 Average Through Traffic Percent Reduction at Site 1 .................................. 87
Figure 5. 5 Average Through Traffic Percent Reduction Based on Visual Sensitivity at
Site 1 ......................................................................................................................... 91
Figure 5. 6 Site 2 Location ............................................................................................... 94

Figure 5. 7 Average Through Traffic Percent Reduction at Site 2 .................................. 99

Figure 5. 8 Site 3 Location ............................................................................................. 101
Figure 5. 9 Downstream Speed Profile between Accident and Normal Conditions (Case 3:
Accident on I-96 EB Express after Greenfield Road 07:10 — 08:10, 03-18-02) ..... 105
Figure 5. 10 Site 4 Location ........................................................................................... 109
Figure 5. 11 Site 5 Location ........................................................................................... 115
Figure 5. 12 Average Through Traffic Percent Reduction Based on Geographic
Locations ................................................................................................................. 120
Figure A. 1 Upstream Volume Distribution at Site 2 .................................................... 128
Figure A. 2 Downstream Volume Ratio Comparison at Site 2 ...................................... 129
Figure A. 3 Upstream Volume Distribution at Site 3 .................................................... 131
Figure A. 4 Downstream Volume Ratio Comparison at Site 3 ...................................... 132
Figure A. 5 Upstream Volume Distribution at Site 4 .................................................... 136
Figure A. 6 Downstream Volume Ratio Comparison at Site 4 ...................................... 137
Figure A. 7 Upstream Volume Distribution at Site 5 .................................................... 139
Figure A. 8 Downstream Volume Ratio Comparison at Site 5 ...................................... 140

xi

CHAPTER 1

INTRODUCTION

1.1 Introduction

As population increases, so too does the demand for greater highway capacity.
This is particularly true in metropolitan areas, and construction of new highways to
accommodate this demand has not increased proportionally to population growth.
Between 1980 and 1999, route miles of highways in the United States increased only 1.5
percent while vehicle miles of travel increased 76 percent (FHW A, 1998). The Texas
Transportation Institute (TTI) reports that the average annual delay per person in the 75
largest urban areas increased from 7 hours to 26 hours between 1982 and 2001. The
measurable, primary costs of congestion in 2001 totaled $69.5 billion: the monetary
impact of 3.5 billion hours of delay and 5.7 billion gallons of excess fuel consumed.
These combined costs have contributed to traffic congestion being recognized as one of
the most significant problems in urban areas.

Traffic congestion is classified into two groups: recurring and non-recurring,
based on the primary cause(s) at any given period of time when occupancy exceeds
capacity. Recurring congestion can be defined as that which routinely appears during
certain peak-hour periods of excessive traffic demand. The most distinctive feature of
recurring congestion is that it occurs virtually every week day, and at the same location.
Therefore, this congestion is sometimes referred as “expected congestion”. “Non-
recurring congestion” is caused by unexpected events including traffic incidents (for

purposes of this paper, the term “traffic incidents” refers to any incident which is sudden,

occurs randomly, and affects the normal flow of traffic, e.g., accidents, stalled vehicles,
etc.), construction on or adjacent to the roadway, severe weather and sudden volume
increases from special events. A number of recent studies show that less than half of the
congestion experienced by drivers in the US is caused by recurring congestion. Slightly
more than half is caused by non-recurring congestion. Non-recurring congestion
dramatically reduces the available capacity and reliability of the transportation system,
and travelers are especially sensitive to these unanticipated or unexpected disruptions in
their personal activity.

The Federal Highway Administration (FHWA) is focusing their efforts to mitigate
traffic congestion problems through several congestion improvement programs. As part
of these programs, they are helping state and local transportation partners develop
regional frameworks for the integrated development of Intelligent Transportation
Systems (ITS) technology, computerized traffic control systems, traveler information
systems, and public transit information management systems. The Advanced Traveler
Information System (ATIS) is one of the most-widely used components of Intelligent

Transportation Systems (ITS).

1.2 Advanced Traveler Information Systems (ATIS)

The ATIS includes Changeable Message Signs (CMS), route guidance, telephone
information, and commercial radio systems, Highway Advisory Radio (HAR), and
personal communication devices such as pagers, the Internet, and designated telephone
numbers (e.g., 511). The ATIS assist motorists in making more-informed decisions on

congestion avoidance by their pre-route and en-route path selection. This system plays a

pivotal role in reducing traffic congestion, improving safety, enhancing mobility, and

improving energy efficiency, which, in turn, reduces environmental pollution.

1.3 ATIS Characteristics

The ATIS provide static and/or real-time traffic information. Static information
includes planned road construction and maintenance, special events, tolls and payment
options, and transit schedules and fares. Static systems provide information on long term
events such as construction activities or road closures. Real-time systems provide minute-
by-minute information on roadway conditions including congestion and incidents. It may
also convey information on available alternate routes, travel time to a destination based
on time of day, transit bus schedules and the availability of spaces in parking lots. This
real-time information is frequently updated in response to current conditions and is useful
in pre-trip and/or en-route traveler decisions. Travelers have repeatedly affirmed the
efficacy of real-time information, stating that it is the most helpful in providing the
information they need to make decisions about their route choice.

Pre-trip traveler information may include road and weather conditions, a business
directory (i.e., tourist attractions, hotels, and restaurants), various routes to a chosen
destination, and typical travel time. En-route traveler information provides traffic
information including congestion, incidents, construction zones, weather conditions and
recommended safe speeds. Figure 1.1 presents the generic ATIS system showing how
information on current conditions is gathered and dispersed through different control

devices (Schiesel and Demetsky 2000).

82a 3325: 2a assume 53% 9: E28 _ a 2:3..—

 

 

 

 

 

 

mSomem HM. Eben. I

 

 

 

 

 

 

 

 

 

 

 

       

   

 

  
      

  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

>955
x3385 8 Bo .. _ U .. 22.30..
mam senses _ .
a » ,
629w 6.29
06mm 2832 85 3835. an": E .m.$ 669:. flowed $2
.8329: 9239.20 2.28.8. .80
r
_ _ _ _ _ _
mSom 925?. 8:89:00 9m“. Senegcfii
oonMmmMfimd mmmfiwwe nae can use Edged new 89:. 856.25
. . 99% .96; $8202 9.628 :39... cozoaficoo
s s s s a s

 

 

   

c2555.... oefaam

 

 

5:952:— £35

 

 

 

1.3.1

ATIS Benefits

Advanced Traveler Information Systems (ATIS) are intended to provide traffic

information that is timely, accurate and reliable to help people make more informed

travel decision. The following list shows the benefits from using ATIS as reported by the

US. Department of Transportation (Mitretek, 1997).

1.3.2

Reduced travel time (4-20%, more in severe congestion)

Decreased traveler stress

Decreased crash risk (4-10%) and fatalities (e. g., reduced driver distractions on
unfamiliar routes)

Enhanced ability to avoid unexpected congestion

Decreased energy consumption and air pollution (decreased HC emissions by 16-
25% and CO emission by 7-33%)

Promotes other travel modes

Reduced inter-modal travel times

ATIS Performance Measures

An ATIS performance evaluation aids in understanding how well the system is

performing with regard to its intended objectives, and how it is likely to perform in the

future given anticipated circumstances. This can help determine whether or not an

applied system was appropriate, identify potential problems, and provide guidance for

solutions. Several Measures of Effectiveness (MOES) have been used to evaluate the

performance of ATIS, and different MOEs are used depending on the type of data

collected and the purpose of the evaluation. The following describes some of the

performance measures used in prior studies classified by purpose.

Ogrational Effectiveness Measures

- Increased volume to capacity (We) ratio
- Decreased average congestion delay
- Increased average traffic speed or decreased average travel time

- Decreased number or percentage of stops

Environmental Effectiveness Measures
- Vehicle emission reduction
NOx reduction

- CO and (CO)x reduction

Economic Effectiveness Measures
- Travel or delay time savings
- Fuel savings

- Reduction of monetary costs associated with vehicle accidents

System Effectiveness Measures

- Decreased average travel time
- Increased Vehicle Miles of Travel (VMT) or Person Miles of Travel (PMT) per

unit time

- Improved level of service (LOS)

- Reduced lost time or delay

1.4 Description of Research Area

The Interrnodal Surface Transportation Efficiency Act of 1991 (ISTEA)
established the national Intelligent Vehicle Highway System (IVHS) program (now
known as the Intelligent Transportation Systems (IT 8)). The program was designed to
promote the use of advanced transportation technologies in the United States. In 1995,
the Michigan Department of Transportation (MDOT) initiated the process of designing
and building the ITS infrastructure in southeast Michigan. It was the largest IT S
deployment and traffic monitoring system in the world at that time. The system consists
of 180 total freeway miles including selected segments of I-96, I-94, I-75, M-39, M-lO
and the I-696/I-275 circumferential freeway, in the metropolitan Detroit area. The system
included 59 CMS locations, 156 closed-circuit television (CCTV) cameras, 61 ramp
meters, 2260 Inductive Loop Detectors (H.D), and 11 Highway Advisory Radio (HAR)
transmitters.

The Michigan Intelligent Transportation Systems Center, known as the “MITS
Center,” in cooperation with the Michigan Department of Transportation (MDOT),
provides motorists with real-time traffic information via the Changeable Message Signs
(CMS). The CMS conveys highway traffic information to drivers and alerts them to
sudden or unexpected changes in traffic conditions. This information includes accidents,
disabled vehicles, construction, road maintenance activities, and severe weather. All

messages inform drivers of what has caused or will cause the upcoming change in travel

conditions (reason message posted), the location of the change, and what effect it will
have on traffic conditions. Figure 1.2 shows the CMS system in the southeast Michigan

area.

 

 

 

 

 

 

Figure 1. 2 CMS System in Southeast Michigan

1.5 Description of the Problem

Many previous studies have measured the performance of Changeable Message
Sign systems through Stated Preference (SP) or Revealed Preference (RP) methods. Very
few studies used the field study method. These SP or RP approach methods collect data
based on survey questionnaires (field, mail, telephone, etc.), focus group interviews, and
on-screen or full-scale driving simulators. However, these methods have a number of key
limitations. Firstly, the data is collected under controlled hypothetical scenarios created
by the researcher so the results are always under a scaled response. Secondly, the
responses directly indicate the whole impact based on a sequence of prior results;
therefore, some valuable considerations may not be interpreted correctly. Finally, the
most critical shortcoming is that respondents may vary their answers at different times
and they often over-state their actual behavior. Therefore, attitudinal surveys or
simulations that simply ask people how they will respond in a given situation are not
generally viewed as reliable (although participant responses can provide some indication
of relative behavior).

Due, at least in part, to these reasons, the reports on the effectiveness of CMS
were not consistent across previous studies. Some of the previous research concluded that
the CMS system influences drivers’ en-route travel choices, diverting them to less-
congested routes thereby alleviating downstream congestion and improving a wide level
of hi ghway-network performance measures. Conversely, other studies concluded that
CMS only minimally influence driver diversion behavior, thus reporting that they do not
provide cost-effective benefits to either the transportation system or the drivers (Schiesel

and Demetsky, 2000).

1.6 Objective and Scope

The purpose of a CMS system is to reduce congestion, and improve safety and
network performance by providing real-time traffic information. However, given the
same information, each of the CMS deployment locations are not likely to produce the
same level of effectiveness. Under various conditions, based on human, traffic, and
geographic characteristics, the CMS likely vary in their effectiveness. No evaluation of
CMS system performance using observed field data were found in the literature, and no
studies have evaluated their effectiveness under various conditions.

Broadly stated, the objectives of this research are to measure the performance of
CMS and to compare their effectiveness as a function of the human, traffic and

geographic conditions. The scope of research includes:

1. Identifying Measures of Effectiveness (MOE) to quantify the impact of CMS
2. Measuring the impact of CMS on the selected MOE under various conditions, and

3. Developing recommendations for the future placement of CMS.

1.7 Research Approach

This research is based upon the analysis of empirical field traffic data collected in
the metropolitan area of Detroit, Michigan. The traffic data, such as volume, speed, and
occupancy, were collected by Inductive Loop Detectors (ILDs) which store minute-by-
rrrinute traffic data, 24 hours a day and 365 days a year. The effectiveness of a CMS
message is measured by the percent of traffic that diverts to an alternate route during the

time the CMS displays a message. The traffic is measured by comparing the downstream

10

traffic volume with the upstream traffic volume when the CMS was displaying a message
with similar measurements under normal conditions (Non-message). Therefore, two
different conditions (an accident condition with a CMS message and normal conditions
without a CMS message) are compared by collecting traffic volume data at different

locations (one upstream and the other downstream) of an interchange.

This study developed a method to estimate travel time between the upstream and
downstream locations using the discrete ILD traffic data. The sensitivity of the diversion
to four different factors; visual literacy, familiarity and time constraints, historical or

existing traffic conditions, and geographic location were tested.

11

CHAPTER 2

LITERATURE REVIEW

2.1 Introduction

Changeable Message Signs (CMS), also known as Variable Message Signs
(VMS), or Dynamic Message Signs (DMS), are the most commonly used Advanced
Traveler Information System (ATIS) devices in the United States. A CMS provides non-
personalized but real—time information on traffic congestion or potential delays to drivers.
This information is intended to influence drivers’ en-route travel choices, diverting
vehicles to less congested routes such as an alternate freeway or arterial surface road.
This diversion reduces the duration of the congestion, improves the level of service and
network performance, and enhances traffic safety. Different Measures of Effectiveness
(MOE) can be used to measure the performance of traffic system control devices,
depending on the data collection approaches and research purpose.

A driver’s decision to divert from their present route to an alternate route is
affected not only by traffic information, but by many other factors. Therefore, an
understanding of how these other factors influence a drivers’ route diversion behavior is
very important in determining where to place CMS to achieve the desired diversion.

This chapter provides a review of previous research related to the objectives of
this study. The chapter is broken down into five different sections. Firstly, characteristics
of CMS are described; secondly, data approach methods to identify and quantify human

behavior are indicated; thirdly, the literature on drivers’ driving characteristics is

12

reviewed; and fourthly, factors affecting a drivers’ route diversion decision is explored.

Finally, prior CMS performance and functional evaluation studies are reviewed.

2.2 Characteristics of Changeable Message Signs

Changeable Message Signs (CMS) are intended to provide en-route real-time
information to drivers and alert them to sudden or unexpected changes in traffic
conditions. The message displayed may be in the form of either simply an information
message, or as an advisory message. In addition to the CMS, drivers may receive
information from other sources, such as commercial radio traffic reports. This
information is used, along with their own travel experience, to make en-route travel

choices, such as diverting to alternate less congested routes.

2.2.1 Brief History of CMS Use

CMS have been used in highway applications in the United States for over 30
years. The first CMS were operated by sliding appropriate messages into a CMS board.
Fold-out, blank-out (including neon), rotating drum, and rotating tape (scroll) signs then
came into being and provided the capability to display information in “real-time”. Even
though these signs were innovative at that time, they were only capable of displaying a
small number of messages. With computer technology, CMS now have the capacity to
display a nearly unlimited number of messages.

In the early 19705, the bulb matrix became the most popular technology for

motorist information systems. However, new technologies such as fiber optics, light-

13

emitting diode (LED) and liquid crystal displays (LCD), which have lower operating

costs and improved visibility are used today.

2.2.2 CMS Types

There are two different types of CMS used on the roadways: a portable CMS
(PCMS) and a permanent CMS. Typically, permanent CMS are used on high density
roadways to advise the driver of both non-recurring congestion and recurring congestion.
PCMS are typically used only for non-recurrin g congestion caused by temporary capacity

reductions from construction, maintenance or severe weather conditions.

2.2.3 CMS Message Contents
Message Elements

CMS can be used as any of the three sign categories; advisory signs, guide signs,
or regulatory Signs. Messages can contain words, numbers or symbols and are used to
provide traveler information, to warn of accident or incident conditions, or display speed

limits or lane restrictions.

Message Format

A consistent format reduces the time required to understand the meaning of the
message. If the information is presented in a non-standard format, it may confuse drivers,
and will increase the time required to understand the message. Guidelines on the design
of CMS, contained in the Manual on Uniform Traffic Control Devices (MUTCD),

suggest the following message elements in the sequence shown:

14

e What is the problem;
e Where is the problem; and
e What is the effect (or suggested action)

The recommended configuration of a CMS boards is shown in Figure 2.1.

easel. neeell Lido
IV V! I? '0
o-

o. .-
eeb-oe u”.
’1 r!
e e

 

Figure 2. 1 CMS Configuration

2.3 Data Approach Method
Three different approaches or methods have been used to identify and quantify
human behavior; Stated Preference (SP), Survey based Revealed Preference (RP) and

field studies.

2.3.1 Stated Preference (SP) Approach

The Stated Preference (SP) approach relies on respondents making their choices
when presented with hypothetical scenarios. The respondents are asked to indicate their
preferences among a set of hypothetical alternative choices such as “If under certain
specified conditions, you were presented with each of these diflerent alternatives, which

one would you most prefer? Which is next? " Respondents rank, rate or choose the

alternative from among the set of hypothetical scenarios, which are described by a set of
attributes generated from an experimental design. The highest ranked attributes (from
those included in the questionnaire) are assumed to represent the respondent’s behavior,
and are then used to estimate that driver’s behavior when presented with real choices.
This is a useful approach when attempting to extend our understanding into areas that
cannot be tested under real conditions due to cost or safety considerations. Two

techniques used to extract data are surveys and simulators.

2.3.2 Revealed Preference (RP) Approach

The Revealed Preference (RP) approach is attributed to Samuelson (Economist,
1938). He hypothesized that individual behavior could be described as a series of choices,
so respondents’ actual choices reveal the individuals’ behavior when presented with
available alternatives. Revealed preference data is gathered based on surveys asking
about previous actions, or direct observations in real-life situations (field study approach).
The advantage of the RP Approach is the reliance on actual choices, avoiding the
potential problems associated with hypothetical responses such as strategic responses or a
failure to properly consider behavioral constraints.

However, there are several limitations to the use of the RP approach. First, it is
very difficult to observe the effect of large variations in the variables. Second, there are
often strong correlations between the variables in revealed preference data (e. g., travel
time and travel cost) so it is often difficult to separate the effect of different variables.
Third, it is difficult to estimate utility levels attributed to secondary variables, as opposed

to primary variables. Therefore, the utility weight assigned to secondary variables is

16

usually low. Finally, RP data is based on choices from actual alternatives, and it is
difficult to forecast the responses to new alternatives. Table 2.1 shows the different

characteristics of the SP and the RP methods.

Table 2. 1 Characteristics of Stated and Revealed Preference Data Approaches

 

 

 

 

 

 

Stated Preference (SP) Revealed Preference (RP)
Based on hypothetical scenarios Based on actual behavior
Attribute framing errors Attribute measurement errors
Extended attribute range Limited attribute range
Attributes uncorrelated by design Attributes correlated
Intangibles can be incorporated Hard to measure intangibles

 

Cannot directly predict response to new

Can elicit preferences for new alternatives .
altematrves

 

Preference indicators can be rank, rating,

. Preference indicator is choice
or chorce

 

Cognitively congruent with choice

May be cognitively non-congruent behavior

 

 

 

 

2.3.3 Field Study Approach

The field study approach analyzes human behavior or attitude through real-time
field data observation. The actual behavior in response to stimuli is measured by
observation. The field study approach has all of the limitations of the Revealed

Preference (RP) approach, and in fact is a subset of this more general approach.

17

2.4 Drivers’ Driving Characteristics
2.4.1 Route Choice Behavior

Several previous research studies investigated drivers’ route choice behavior.
Huchingson, McNees, and Dudek (1977) studied commuter route choice behavior using
interviews and mail-in surveys in Dallas and Houston, Texas. This study collected survey
data at two different locations: one in the Central Business District (CBD) of Dallas and
the other at a rest stop along an Interstate highway leading into Houston. Drivers were
asked to describe the routes they regularly took to work and home, and the reasons why
they had chosen the present route. The most frequent reasons for taking the priority route
were that it was more convenient, direct or faster, and the alternate routes took longer and
were less direct. The survey results at the two different locations are given on Table 2.2.
There is clearly a different sequence of priorities that commuters and intercity drivers
used to select their priority route. A faster and shorter route had a higher priority for

commuters while convenience and accessibility were more important to intercity drivers.

Table 2. 2 Reasons for Taking Primary Routes (Huchingson et al. 1977)

 

 

 

 

 

 

 

 

 

 

Percentage of Drivers
Reason Commuters Intercity
Home—to-Work Work-to-Home Drivers

Fastest route 23 24 20
Fewest stops l4 8 3
5:22:53?“ 12 . ..
Shortest, most direct 22 14 20
Less traffic 8 19 5
Good traffic flow 5 10
Other 16 19

 

 

 

 

 

 

l8

Depending on when drivers choose their driving route, two different route choice
behaviors have been described: one is pre-route choice (deciding the route before
departure) and the other is en-route choice (deciding on a route while driving on the road).
Research conducted in Chicago, Illinois by Deniels, Levin and McDermott (1976)
indicated that 69 percent of home-to-work and 64 percent of work-to-home drivers
always choose their route before departure and only 7 and 11 percent of home-to-work
and work-to-home drivers modify their route while on the road.

Khattak, Schofer and Koppelman (1991), also in Chicago, Illinois found similar
results. Seventy-four percent of the drivers in that survey answered they choose the route
before getting in the car for the home-to-work trip. When questioned about how often
respondents modified their route, slightly more than 80 percent of the respondents stated
they had used the same route for more than 1 year, although they had made minor
diversions occasionally.

Polydoropoulou, Ben-Akiva, and Kaysi (1994) attempted to define the preference
factors in route choice using a Likert-style survey. The Likert-style survey uses
questionnaires based on a rating scale (generally five point scales such as strongly
disagree to strongly agree) to determine respondents feelings or attitude. The respondents
indicate their feelings based on various statements providing a series of reliability
statements for each person’s attitude. The data were obtained from Massachusetts
Institute of Technology (MIT) commuters. Table 2.3 shows the results from this study.
Category 1 corresponds to the response “Not important at all”, whereas category 5
corresponds to “Very important”. Time of day (61.2 percent), commute time (76.1

percent), and time spent stopped in traffic (79.1 percent) were reported to be very

l9

important factors to the drivers when they choose their route. Traffic reports (18.5) and
weather (38.0 percent) were relatively less important to the drivers in this survey. It is
interesting to note that habit is rated “very important” nearly four times as often as traffic

reports. Traffic reports have the highest frequency of “not important” responses.

Table 2. 3 Importance of Factors Affecting Route Choice Behavior
(Polydoropoulou et al, 1994)

 

 

 

 

 

 

 

 

 

 

Not Very
Important Important
Number Attribute
1 2 3 4 5
(%) (%) (%) (%) (%)
1 Time of day 15.7 7.1 11.1 16.3 44.9
2 Commute time 9.2 4.6 10.0 20.6 55.5
3 Habit 12.3 8.6 29.8 26.1 23.3
4 Time spent stopped in traffic 5.3 5.3 10.3 29.0 50.1
5 Number of traffic lights 10.9 11.5 24.2 25.4 27.9
6 Traffic reports 30.7 22.7 28.1 12.4 6.1
7 Risk of delay 8.6 8.5 26.1 30.6 26.2
8 Weather 25.9 16.7 19.3 15.9 22.1

 

 

 

 

 

 

 

 

 

Wenger, Spyridakis, Haselkorn, Barfield, and Conquest (1990) studied the
motorist behavior and decision making using personal interviews in Seattle, Washington.
Seventy-three percent of Seattle commuters reported that they received some traffic
information before their departure to work and half of them received traffic information
pertaining to their regular route almost immediately after awakening. However, the

majority of commuters answered that they rarely decide to use an alternate route (65.7

20

percent), rarely decide to use an alternate mode (90.0 percent), and rarely decide to
change their departure time (64.3 percent) on the basis of traffic information received
before departure. Therefore, this study indicated that while commuters may receive pre—
departure traffic information, only about a third of the drivers stated that they use this

information in their route choice.

2.4.2 Knowledge of Alternate Routes

Shirazi, Anderson, and Stesney (1988) conducted a telephone survey concerning
commuter attitude and characteristics in Los Angeles, California. More than 70 percent of
commute drivers reported they have knowledge of alternate routes while 27 percent
stated they do not. This study categorized the respondents who know of an alternate route
by age, gender, travel time and delay. No significant relationship was found between
knowledge of alternate routes and age, gender, and delay. However, a significant
relationship was found between knowledge of an alternate route and travel time. The
drivers who commute for less than 45 rrrinutes are more likely to know of an alternate
route than those with a travel time greater than 45 minutes.

A study from Wenger, Spyridakis, Haselkom, Barfield, and Conquest (1990)
measured the knowledge of a drivers’ primary route and alternate routes based on how
well they could name landmarks and street names on each route. Table 2.4 shows the
means and standard deviations for the number of street names and landmarks named on
each route. The drivers were much more familiar with the street names and landmarks on
the primary route than on the alternate routes. The drivers reported their decision to use

an alternate route was based first on en-route traffic information and secondly on

21

observed traffic condition. When drivers did use an alternate route for any reason, 77.8
percent of commuters answered that they had experienced an increased stress level on the

trip.

Table 2. 4 Number of Names and Landmarks in Route Descriptions

 

 

 

 

 

(Wenger et al, 1990)
Primary Route Alternate Route 1 Alternate Route 2
Mean 8.45 5.02 4.26
Street Names
SD 6.23 3.70 4.01
Mean 1.67 1.03 0.79
Landmarks
SD 1.89 1.48 0.90

 

 

 

 

 

 

 

SD: Standard Deviation

2.4.3 Route Switching Behavior

Research conducted by Shirazi, Anderson, and Stesney (1988) investigated the
route changing or switching behavior for home-to—work commuters in Los Angeles,
California. Forty percent of respondents indicated they had changed their route during the
home-to-work trip at least once, while thirty-one percent never did and twenty-nine
percent gave an invalid response. Only fourteen percent of the respondents who had
changed their route reported that they change their route “very often or often ” and
approximately twenty-five percent of the respondents answered “rarely or sometimes”.
The majority of route changes occurred when drivers observed traffic congestion. Similar
results were reported by Polydoropoulou, Ben-Akiva, and Kaysi (1994) who found 62

percent of the drivers who switched routes changed their route based on their own

22

observation, 12 percent diverted based on radio information and 26 percent for other
reasons.

Mahamassani, Caplice, and Walton (1990) conducted a study focused on time-
and route-switching behavior of commuters in Austin, Texas. Nearly 3000 randomly
selected households in a mostly suburban area of northwest Austin close to a major
technology—based manufacturing and research area were interviewed. Thus, the sample
data was not exclusively CBD oriented commuters, but included a large proportion of
suburban workers. Figure 2.2 shows the results of route switching behavior based on two
different trip purposes. The results indicated that a higher proportion of commuters adjust
their departure time for the home-to-work trip than for the return trip. A slightly larger
proportion of home-to-work commuters adjust their departure time than switch routes. On
the work-to-home commute, a larger proportion of drivers adjusted their routes compared
with the home-to—work commute. Also, a significantly larger proportion of commuters
responded that they were more likely to adjust their travel route than departure time for
the work-to-home trip. These results indicate the switching behavior may be based on

different considerations between the home-to-work and work-to-home commute.

23

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

50 —
41 39
a 40 _ 34 r. ”U...
0’ 28
g 30 _ . 24
E ,
8 20 ~ , . 12 : _ 15
o\° 10 ~ 7 ,
P‘] i
O I r ‘ A T 3‘ I I 3 fl
Neither Route only Time only Both
Type of Switching
l3 home-to-work I work-to-home

 

 

 

 

 

 

Figure 2. 2 Route and Time Switching for Home-to-work and Work-to-home Trips
(Mahamassani, Caplice, and Walton 1990)

2.5 Factors Affecting a Drivers’ Route Diversion Decision
2.5.1 Factors Related to Human and Socioeconomic Characteristics

Khattak, Schofer and Koppelman (1991) investigated factors influencing en-route
diversion behavior through a survey in Chicago, Illinois. The results of this research
indicated that some human characteristics, such as gender and self-assessment statements
about risk behavior, significantly affect the diversion behavior. For example, among the
people who have knowledge of alternate routes, men (62 percent) were more likely to
change travel routes than woman (38 percent) and a person who states they are more
willing to take risks and have an interest in discovery and exploration, has a more
aggressive diversion behavior. However, Mahmassani, Caplice, and Walton (1990) found
different results in their research in Austin, Texas. In particular, their research indicated

that gender, as well as age or work place rules do not significantly effect drivers’ route

24

diversion. This study concluded that the home-to-work route diversion is primarily
motivated by geographic considerations and network considerations rather than by socio-
demographic characteristics. However, departure time switching for this trip was clearly
influenced by lateness tolerance, job position, and other individual characteristics.

Polydoropoulou, Ben-Akiva, and Kaysi (1994) developed a route diversion model
based on revealed preference data. From their analysis, when drivers are under time
pressure, and drivers who often change their regular routes while driving, have the
highest probability of route diversion.

Ratim Pal (1998) attempted to identify drivers’ diversion decisions using an
integrated framework which involved observable socioeconomic and situational factors,
and unobservable latent factors (such as risk acceptance, trust in traffic information
systems, and the expected level and quantity of the information). From their observation,
commuters who have a high-risk-taking tendency and have more trust in the traffic
information system were more likely to divert compared with commuters who were

looking for more detailed information (a low-risk-taking tendency).

2.5.2 Factors Related to Traffic Conditions

Several previous studies investigated the relationship between route diversion and
traffic congestion. Heathington, Worrall, and Hoff (1971) used revealed preference (RP)
to evaluate driver behavior toward route diversion to avoid unexpected delay in Chicago,
Illinois. The results of this research indicated that most drivers are more willing to divert
when they encounter non-recurring congestion than when they encounter recurring

congestion. Also, typically, home-to-work trip drivers are more willing to divert than

25

work-to-home trip drivers. From the diversion percentage results based on the highway
functional classification, expressway drivers were slightly more willing to divert than the
non-expressway drivers when they were faced with the same amount of delay. However,
this difference was not statistically significant.

Huchingson and Dudek (1979) conducted a study to find freeway motorists’
preference and behavior with an imaginary field test. The test was conducted with various
groups from different locations in the United States. The first test measured the
percentage of drivers who stated they would divert for various levels of delay time from
incident conditions. Fifty percent of drivers stated they would divert for a delay
exceeding 15 to 20 minutes. Longer delays naturally induced more drivers to state that
they were willing to divert. However, in severe weather conditions such as rain or ice, the
stated willingness to divert was low for up to an hours’ delay. More drivers indicated they
were willing to divert when the displayed information contained the duration of delay
than when only the type of incident or level of congestion was displayed. There was no
significant difference in the response to the delay from different types of incidents such
as roadwork, an accident, a truck over-turned, or weather conditions such as snow, ice,
and rain. The overall conclusions from this study indicated that as delay increased on the
regular or preferred route, an increasing number of drivers state that they would divert to
an alternate route, and the relationship between length of delay and percentage of
diversion resembles an S-shaped curve. Figure 2.3 shows the shape of the effect of delay

on the percentage of driver diversion.

26

 

100~

90~

80-

70~

50~

% Diversion

40a

304

20«

10—

 

 

0 1 0 20 30 40 50 60
Delay ‘n Nitutes

 

 

 

Figure 2. 3 Effect of Delay on Drivers Diversion
(Huchingson et al. 1979)

This study also investigated the motorist perception of the delay duration when
the adjectives “MAJOR” and “MINOR” are used to modify the word “ACCIDENT”. The
median value of the expected delay for “Minor Accident” meant “a delay of 12 minutes
or less”, where as “Major Accident” meant “a delay of 22 minutes or more A similar
study was conducted by Huchingson, Whaley, and Huddleston (1984). The results
supported Huchingson and Dudek’s (1979) finding that major accident means “a delay of
22.7 minutes or more ” and a minor accident as “a delay of 7.9 minutes or less

Khattak, Schofer and Koppelman (1991) investigated the influence on

commuters’ route diversion in response to delay with a stated preference (SP) survey in

27

 

Chicago, Illinois. About sixty-two percent of the surveyed drivers had experienced en-
route delay during the past 6 months and 84.6 percent of drivers expected 10 - 30
rrrinutes of delay from recurring congestion on their work trips. Thirty nine percent of
drivers stated that they would divert to an alternate route in response to an expected 10 -
20 rrrinutes of additional delay. As expected, more drivers (49 percent) stated they would
divert in response to an estimated increase in delay of 21 — 30 rrrinutes. However, the
percent of drivers who would divert for an expected extra delay of more than 30 minutes
did not increase, and the percentage actually decreased when more than 40 rrrinutes of
extra delay was expected.

These results were based on a small sample, and this may have contributed to the
counter-intuitive results. When asked about their past experience with diverting to an
alternate route, more than seventy percent of respondents who diverted believed that they
saved travel time, and over fifty percent of respondents who did not divert believe that

they would have saved time by taking their best alternate route.

2.5.3 Factors Related to Traffic Information

Daniels, Levin and McDermott (1976) conducted a home-interview survey in
Chicago to investigate diversion based on radio traffic reports. The survey responses
were divided into two groups of drivers, expressway and non-expressway. Over 70
percent of both expressway and non-expressway drivers listen to radio traffic reports on
the way to work. Among them, one out of three expressway drivers and one out of four

non-expressway drivers reported they would divert their route on the way to work based

28

on radio information. Reported accidents induced a higher proportion of diversion than
simple congestion information among both the expressway and non-expressway drivers.

Huchingson, McNees, and Dudek (1977) also conducted a study to measure the
reaction from radio advisory incident information. Two different destination groups
(within city and beyond city) were surveyed. Seventy seven percent of respondents with a
destination in the city and 65 percent of respondents with a destination beyond the city
answered that they would divert their route based on traffic information. The major
reason for not diverting was unfamiliarity (66 percent) of the area, and the main reasons
for diverting were to avoid congestion (48 percent), save time (27 percent), and avoid
delay (20 percent).

These results illustrate the problem with using Stated Preference (SP) techniques
to study traffic diversion. The Revealed Preference (RP) studies all found diversion rates
of approximately thirty percent, while in this study between 65 and 77 percent said they

would divert.

2.6 Prior CMS System Evaluation Studies
There are several previous research studies that evaluated the performance of

CMS, and the impact of different CMS displays and designs.

2.6.1 CMS Performance Evaluation
Dudek, Weaver, Hatcher and Richards (1978) measured the impact of real-time
special events; the annual Fourth of July fire works, the annual football game between the

University of Texas and University of Oklahoma, opening day of the annual Texas State

29

Fair, and the Cotton Bowl football game, by field observation in Dallas, Texas. This
study evaluated 14 real-time messages displayed on matrix signs located on the freeway.
The diversion of freeway traffic to an arterial alternate route was used to measure the
driver response. From the analysis results, around 40% of drivers diverted to the alternate
route based on the CMS message. However, when a single sign displaying a credible
message was installed at the proper location, the diversion was the same as when several
advance signs were used. Therefore, this study concluded that repetition of messages is
not necessary if the messages are credible and located properly. Also, drivers are more
affected by messages which describe traffic conditions than with best-route messages.
Turner, Dudek, and Carvell (1978) conducted a study to measure the influence of
CMS during maintenance operations in Dallas, Texas. The traffic flow rates on freeway
exit ramps during incident conditions with and without messages displayed on a CMS
were analyzed on a case study basis. From the analysis results of this study, it was
concluded that more diversion was generated when a message is displayed than under

normal conditions (natural diversion). Table 2.5 shows the case study results in this study.

Table 2. 5 Case Study Results (Turner, Dudek and Carvell, 1978)

 

 

 

 

 

 

Case Change of 5-minute Flow Rates for Exiting Traffic (%)
No Message Information Sign Diversionary Sign
1 +190 +3247 +3438
2 - - -
3 +152.6 +176.3 +2273
4 +962 +125.9 +1473

 

 

 

 

 

 

30

The results are the 5-minute flow rates for exiting traffic. The exit traffic flow
increased significantly when a CMS message was provided compared with no messages.
Moreover, exit volumes increased more when the signs displayed messages
recommending diversion when compared to traffic information only messages. Therefore,
this would indicate that drivers prefer diversionary information to only incident
information.

Dudek, Stockton, and Hatcher (1982) evaluated the impact of CMS messages in
San Antonio, Texas. This study assessed the effectiveness of CMS in diverting traffic to
an alternate freeway route during incident conditions. The traffic volume on the freeway
and off-ramps were collected from field traffic data counters. Three different conditions;
(a) during normal conditions, (b) during an incident without the CMS messages, and (c)
during an incident with the CMS messages were compared using seven different incident
cases. Combining the results for all seven incident cases, the diversion volume during an
incident but with no CMS message was significantly higher (p<.05) than normal
conditions and the diversion volume during an incident with a CMS message was also
significantly higher (p<.05) than normal condition. However, there was no statistical
difference in the diversion volume during the incident conditions between with and
without CMS message cases.

The data were analyzed to determine whether diversion rates were affected by the
time of day when the incident occurred. During the peak-hours, the diversion rates both
with and without a CMS message, were significantly higher than normal condition.
However, there was no statistical difference in the diversion rate during the off-peak

period. According to the results from this study, the use of a CMS message during

31

incident conditions did not have a significant effect on drivers route diversion when
compared to natural diversion (without a CMS message). This study concluded that
drivers, generally, select their travel route based on the time of day, location of the
incident, and severity of congestion, and the rate is not affected much by the CMS
information.

Yim and anace (1996) investigated the effectiveness of the Systemed’
Information Routiere Intelligible aux Usagers System (SIRIUS). SIRIUS is the large
urban field traveler information and automated traffic management system in Paris,
France. It provides real-time traffic information via remotely controlled CMS operated
from regional Traffic Management Centers. Yim and anace assessed the performance
of SIRIUS with a link flow evaluation using loop detector data. This study evaluated one
link which is an access ramp connection between an arterial route (D45) and a freeway
route (A86). Freeway route A86 is the ring road that wraps around the suburbs of Paris,
and arterial route D45 is the frontage road serving the residential and industrial districts
along the freeway. Therefore, drivers can make a choice to either stay on the arterial
route or take the freeway route to avoid congestion based on the CMS message.

The traffic data such as volume, speed, and density at the access ramp connecting
D45 to A86 were collected using three loop detectors. A CMS was placed 300 m
upstream from the diversion point and displayed the length of queue on A86 at all times,
including free-flow conditions. Two different conditions were analyzed. The short-term
condition measured the traffic volumes during the 5 minute period before and after the
message changed. The long-term condition measured the traffic volumes during the 10

rrrinute period following the 5 minute period after the message changed.

32

When the message changes to indicate increased traffic congestion on A86, the
mean flow rate on the ramp to access A86 during the 5 minute period after the message
changed decreased by 3.68 percent. When the message changed to indicate a decreased
level of congestion on A86, no significant difference in the mean flow rate was observed.
Also, the data showed that the queue distance and diversion rate were positively related.
When the CMS displayed a queue length of l-km, 2-km, 3-km and 4-km, the traffic flow
on the access ramp reduced by 7, 10, 15 and 30 percent respectively. This reduction was
greater during the AM-peak hours than the PM-peak hours. This study indicated that the
CMS significantly impacts vehicle diversion and has the greatest effect when the
information is disseminated during periods of increasing congestion.

Krraan, Zijpp, Tutert, Vonk and Megen (1998) evaluated the performance of a
dynamic CMS system in Amsterdam, Netherlands. This study compared aggregate
performance measures such as severity of congestion, traffic performance, instantaneous
travel time delay and average travel speed before and after the CMS installation. The
measurements were conducted at seven CMS locations and only the queue length
information in each travel route was provided by the system. Congestion was measured
by the length and duration of queues. The Motorway Control and Signaling System
(MCSS) provided time and link based binary congestion information such as “1” if
congestion occurs and “0” otherwise.

The traffic performance was measured by vehicle-miles-traveled (VMT) which
was calculated based on traffic volume and the length of the links during each time
period. The instantaneous travel delay was measured by the difference between free flow

travel time and realized travel time, which was weighted by traffic volumes. In addition,

33

the standard deviation of Speed over the entire network was measured to analyze the
performance of the CMS system. This study assumed that improved network
performance leads to a decrease in the variation of travel speed and is an indicator of
more reliable travel times. From the analysis results, the severity congestion over the
entire network was slightly decreased and traffic performance improved after applying
the CMS systems. Even though the average speed over all the links did not increase, the
delay time decreased in both the morning and evening peak hours. Overall, the authors
concluded that the CMS system had a positive impact on network performance on the
Amsterdam freeway system.

Peeta and Gedela (2001) evaluated the performance of a proposed CMS system
with simulation experiments using DYNASMART, a mesoscopic traffic simulator. The
experiments used the Borman Expressway corridor network, which consists of 197 nodes
and 458 links and drivers’ CMS response attitudes collected by a stated preference (SP)
survey. The experimental simulation compared the network performance with and
without CMS information under different scenarios with variation in the number of
incidents, incident duration and congestion level. The simulation results showed that
CMS information improved System Optimal (SO) solutions ranging from 13 to 25
percent compared with the no-information case.

The second Simulation varied incident duration (5min, 10min and 20min). The
network performance gradually increased as incident duration increased. Therefore, it
could be interpreted that a CMS will be more effective with more severe incidents. The

last experiment compared the performance of CMS under different congestion levels.

34

With medium congestion, CMS provides the greatest improvement in the system
performance compared with the low and high congestion levels. At the extremes, the
opportunities to divert to better paths through CMS messages are reduced because of low
and high network congestion. Overall, this study indicated CMS control would provide

positive performance results in the real world.

2.6.2 User Impacts Based on Different Designs and Features of CMS

Many state Departments of Transportation (DOT) are currently operating two
different types of dynamic CMS messages, a static message and a flashing message for
attracting attention and emphasizing the importance of the message to drivers. Dudek
(2004) conducted a study comparing the effectiveness of using three different features of
the dynamic CMS messages; 1) Effect of flashing an entire one-phase message 2) Effect
of flashing one line of a one-phase message 3) Effect of alternating text on one line of a
three-line CMS while keeping the other two lines of text the same.

The results of flashing an entire one-phase message had no significant effect
upon driver comprehension compared with the static message. However, the flashing
feature required 1.5 seconds longer reading time to comprehend the message than the
static Sign. Flashing one line of a message reduced the ability of drivers to remember
parts of the message when compared with the static message. The average reading time
of the message increased 1.8 seconds compared with a line that was not flashed. The last
feature, which has three lines including redundant information by repeating the top two
lines on both phases of a two-phase message while changing the bottom line, was not

significantly different than with a message without redundancy. The average reading time

35

of the message that had redundant information was 2.8 seconds longer than the message
which did not include redundant information. From the test results, the flashing messages
requires a longer reading time, but only provides the same efficacy as the static message.

Hustad and Dudek (1999) conducted a study to evaluate and develop
abbreviations on Changeable Message Signs (CMS) using a human factors laboratory in
New Jersey. This study indicated that there were regional differences with respect to
driver understanding of some of the abbreviations. For example, Eighty-eight percent and
85 percent of the drivers tested in northern New Jersey understood the abbreviations EXP
CLSD and LOC LNS for “Express Closed” and “Local Lanes”, whereas less than 70
percent of the drivers studied in southern New Jersey understood these abbreviations.
Also, the abbreviations for some of the facilities/structures were generally understood by
a very high percentage of drivers who live near the facility/structure. For example, the
abbreviations studied for “Mount Tabor” and “Sandy Hook National Par ” which are
located in north New Jersey were understood by 88 percent of the drivers tested in that
part of the state. In contrast, only 58 and 65 percent of the drivers tested in south New
Jersey understood the abbreviations.

Brian G. Benson (1996) studied motorist attitudes toward the message content of
CMS, using as a case study the CMS system of Northern Virginia. The case study was
carried out using seven focus groups and an opinion survey. This study indicated that a
distinct negative correlation (-0.25) was found between motorists who had experienced
inaccuracies on CMS and those who are likely to use alternate routes recommended on a
CMS. It is twice as great as that between motorists who had experienced CMS

inaccuracies and those who are often influenced by CMS (- 0.12). This result reflects that

36

those who use a recommended route based on CMS will be more negatively affected by
having experienced CMS inaccuracies. From the survey responses to questions about
posting delays in travel time from heavy congestion, respondents were evenly divided
between two groups: one prefers information to be quantitative and the other prefers
information to be descriptive. Among the respondents who prefer quantitative
information, half want a range estimate (e. g., 10 — 20 minute delay) and the remaining

half want a point estimate (e. g., 15 minute delay).

2.7 Summary

From the literature, there is evidence that most drivers decide their driving route
before departure (pre-route choice) and have one regular route, especially for their
commute trips. The factors used to select a regular route are different based on the
characteristics of the trip. Generally, commuters place a higher priority on a faster and
shorter route while the non-commuter trips place priorities on convenience and
accessibility. The literature leads to the conclusion that the preference in route choice or
diversion is not only affected by the drivers’ characteristics, but also by factors such as
the trip purpose, geographic and network location, traffic conditions (risk of delay),
visual confirmation of traffic congestion and severe weather conditions.

The results concerning the effectiveness of CMS on the drivers’ route diversion
behavior are not consistent. Some of the research concluded that the CMS do not have a
significant affect on the drivers’ route diversion behavior, while other research indicated
that the CMS have a significant affect on route diversion behavior, especially during an

incident condition.

37

There are several possible explanations for these inconsistent results in the previous
research. First is the study technique. Some of the previous research evaluated the
effectiveness of CMS with Stated or Revealed Preference approaches conducted by
survey or simulators. These methods analyzed the respondent’s behavior based on the
participant response, and then assume the responses are the same as the actual choices
drivers would make under real conditions. However, respondents may vary their answers
at different times or locations and they often over-state their actual behavior. Therefore,
results based on survey data in a given situation may not be reliable.

Second, questionnaires are created under hypothetical scenarios constructed by the
researchers, and they may not include all valuable considerations. Therefore, the results
are always measured under the factors included by the researchers.

Third, the literature which analyzed the effectiveness of the CMS by observing
traffic diversion in the field, did not consider all the factors which potentially affect
drivers’ route diversion; Drivers’ route diversion attitudes are not consistent at all times
of day; drivers’ route diversion behavior is different for different trip purposes; and
drivers route diversion varies with familiarity with the location. The effect of a CMS
may also differ based on geographic locations. That is, a location with a freeway alternate
route and a location without a freeway alternate route may evoke different diversion
reactions from the same CMS information.

The literature indicated that drivers are very sensitive to anticipated versus
observed congestion. Therefore, an accident condition where drivers can observe the

congested queue and where they cannot observe the congestion queue might have a

38

different effect. However, this factor was not reported (or presumably considered) in
most of the study results.

Finally, the literature does not consider the effectiveness of CMS under expected
traffic conditions. When the demand exceeds the system capacity during the peak period,
congestion (called recurring congestion) is created. If drivers expect that recurring
congestion is likely to be present, based on their experience, they may react differently.

In this study, the effectiveness of CMS will be analyzed with field traffic data to
provide more reliable results and will consider those factors that potentially effect a

drivers’ route diversion behavior, but were not considered in the previous research.

39

CHAPTER 3

DATA COLLECTION

3.1 Introduction

For this research, the Michigan Intelligent Transportation System Center (MITSC)
in Detroit provided traffic data from Inductive Loop Detectors (ILD) and the CMS
message log from May - December 2001 and February - December 2002. This research

utilized the data to analyze the effectiveness of the CMS.

3.2 Description of Inductive Loop Detector Data

The ILD traffic data consists of volume, speed and occupancy and is summarized
and reported in one minute increments by lane 24-hours a day and 365 days a year. Each
report consists of five different fields, loop—ID, volume, occupancy, average speed, and
date/time. Each loop-ID consists of one or more lanes and each lane has an individual
detector to measure traffic. Information about each of the study sites, including the
number of lanes, type of loop, site name and corridor information was developed for each
loop-ID. The aggregation of the individual loop-IDs at each site, called the Rep_ID,

contains data on all lanes passing the location. The Rep_ID was depicted on the files

provided by MITSC. Table 3.1 shows the format of the ILD traffic database.

40

Table 3. 1 Format of Inductive Loop Detector Data

 

Field

Contents

 

Field 1

Loop-ID associated with Rep_ID (site-ID).
- Typical ID is a 6 or 7-digit integer.

 

Field 2

Volume

- An integer count of vehicles during the past minute

- Typical values are 0 to 70

- A blank means that no data was received from the Inductive Loop

 

Field 3

Occupancy

- Percent of time the loop was occupied by a vehicle during the past
minute

- Typical values are integer 0 to 100

- A blank means that no data was received from the Inductive Loop

 

Field 4

Speed

- The average vehicle speed in miles per hour

- Typical values are integer 0 to 100

- A blank means that no data was received from the Inductive Loop

 

Field 5

 

 

Date/Time stamp YYYY-MM-DD hhzmmzss denoting the end of the
minute

 

3.3 Description of CMS Message Log Data

The CMS message log database consists of date and time of operation, operator

name, sign hardware, sign number, activity type, and message information. The message

to be displayed when accidents or other events occur are selected by the MITSC traffic

engineer/operator based on information received from a responsible authority. Depending

on the direction, location, and nature of the incident, the message is selected from a

message library developed by MDOT, or message guideline, or it can be composed by

the operator. To avoid diminution of system credibility, overly precise descriptions were

not provided by the system. Depending on the impact of an incident or other events,

41

 

messages can be displayed at more than one CMS upstream location. The messages
included accidents, disabled vehicles, construction, maintenance activities (road work,
lane closure, debris pickup), severe weather (heavy rain, heavy snow, fog, thunderstorm),
amber alter, ozone-action days, and holiday traffic information. The displayed message
was discontinued when traffic conditions returned to normal. Table 3.2 shows the

database format of the CMS log message information.

Table 3. 2 Format of CMS Message Log Database

 

Field Contents
Field 1 Recording Date and Time

 

 

Field 2 User/Operator Name
Field 3 Sign Hardware Number
Field 4 CMS Number

Field 5 Abbreviation of Message

 

 

 

 

Field 6 Activity Types

 

Field 7 Detail Activity Information
Field 8 Line 1 Text (What)

Field 9 Line 2 Text (Where)

Field 10 Line 3 Text (Effect)

Field 11 Message II)

Field 12 Message Starting Date and Time

 

 

 

 

 

 

Field 13 Message Ending Date and Time

 

 

 

 

3.4 Study Site Selection

There are 59 CMS locations in the study area. From these 59 CMS locations, this

study selected five locations which were expected to incorporate the variables determined

42

to effect diversion in the literature review. The most important geographic feature was the
existence of an alternate freeway route. An alternate route was defined as “the same level
of state- or interstate-highway as the original route". Traffic conditions were classified
as “existing and non-existing recurring congestion Recurring congestion was defined
as congestion which occurs at a specific location and a specific time period causing the
average speed to fall below 35 mph (CALTRANS methodology) for at least 15—minutes.
Figure 3.1 shows the five different sites selected based on these criteria. Also, Table 3.3

contains brief information about each study site.

 

 

 

 

 

Figure 3. 1 Selected Study Sites

43

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

- - - oZ 8% 3A 02 8% we; 02 mm mEU m 8%
- - - oz 8% GTE 8% 8% mméz 02 m mEU v cam
oo 98ch
8% 8% an -6 oz 8% A - m: 02 8% co; 8% mm m—ZU m 8%
co H co H
8% 8% 3-2 - <2 2-2 oz 8% 9-2 8% N 920 N 2%
oz 8% pm; 8% 8% 0%.“ oz 8% co; 8% mm £20 ~ 2%
:oz8waoU San— :ocmoweoo 8am 858980 San—
. N 8.5m w ~ 88% m 830m 850m
mama—seem Din 52.58% Gd . Ebsoom at: .3385 a £20 25
8>uanz< anbchoQ E8583 8853?.

 

 

 

 

 

53555 ea :35 8.88 an 2.3.

 

3.5 Data Collection Method

3.5.1 ILD Traffic Data Collection

To measure the diversion, traffic data were collected at two or more different ILD
locations at each selected site: one upstream and the others downstream. ILDs which are

located slightly before or after the CMS message board were selected for the upstream

data collection. Depending on whether the site has an alternate route or no

one or two downstream locations were selected. Figure 3.2 shows the downstream ILD
locations based on these two conditions. In case (a) of Figure 3.2, route 1 and route 2 are
alternate routes. Therefore, the diversion rate will be different based on where the

accident occurred. In case (b) in Figure 3.2 no alternate freeway route exists but entrance

and exit ramps exist between the upstream ILD and downstream ILD.

alternate route,

 

 

 

 
 

Route 1

Down

Route 2

(a) A Site with a freeway alternate route

 
 
  
 
 
  

_ILD (1)

DownJLD (2)

 

 

DOWILILD

(b) A Site without a freeway alternate route

 

“a

V

 

Figure 3. 2 ILD Data Collect Locations

45

 

3.5.2 ILD Traffic Data Conversion
The Inductive Loop Detector (ILD) counted and stored traffic volume, speed, and
percent occupancy at each lane during each one-minute time period. This ILD traffic data

from each lane was aggregated into spot traffic data as follows.

Volume
Minute based lane volume was converted to a spot total lane volume by the

following arithmetic equation:

Q(t)= i410)
i=1

where,

t = time (minutes)

i = i"' lane

n = number of lanes

qi (I) = i'h lane volume at time t

Q(t) = total volume at time t

Speed

In a moving traffic stream, each vehicle travels at a different speed. Thus, the
traffic stream does not have a single characteristic speed so the mean speed is used to
characterize the traffic stream as a whole. There are two types of mean speeds: time mean
speed and space mean speed. Time mean speed is defined as the arithmetic mean of
individual spot speeds that are recorded for vehicles passing an observation point over a

selected time period. Space mean speed is defined as the harmonic mean of individual

46

speeds which are recorded for vehicles passing an observation point over a selected time
period. The harmonic mean is calculated by converting the individual spot speeds to an
individual travel time, then calculating the average travel time, and finally inverting the
average travel time rate to obtain an average speed. To measure the harmonic average

spot mean speed from the lane mean speed, this study used the following equation:

__1___
_1_ n 1
"i=1vz'(t)

ZSMS (I) =

 

where,
t = time( minutes)
i = i'h lane

n = number of lanes

F; (t) = i'h lane speed at time t (mph)

zSMS (t) =space mean speed at time t (mph)

Occupancy

Percent occupancy is defined as the percent of time a point or short section of
roadway is occupied. Occupancy can vary from 0 percent (the absence of vehicles
passing) to 100 percent (a vehicle completely stopped over a point). Each lane provided
the minute based percent occupancy. Lane density was estimated from the lane by lane

percent occupancy by the following equation:

47

" 52.8

2 =—————

i=1 LV + LD
n

X %0CCi(I)
K(t)=

 

where,
t = time (minutes)
i = i'" lane

= total lanes

K (t ) = density at time t

ZV =average vehicle length (feet)

LD =detection zone length (feet)

%0cc,- (t) =1“ lane percent occupancy at time t

3.5.3 Traffic Data Classification

This study determined the volume of diverted traffic resulting from the CMS
message by comparing the downstream through traffic volume based on the upstream
volume during times when the CMS was used with the same time period under normal
conditions. Therefore, data for two different conditions, the accident condition with CMS
message and the normal condition, were collected at each study site. The data
classification was based on the CMS message log information database from the MITSC.
The CMS message log information database provides the time of day, operating CMS ID,
message operator, displayed message text, accident or event location, and beginning and
ending time the CMS message was activated. The accident and normal conditions were

identified from this information.

48

3.5.3.1 Normal Condition Data Collection

Traffic flow on a section of a roadway varies from month to month and from day
to day. However, during a specific time period if traffic flow is collected over days or
months, the traffic pattern will be similar under normal conditions. Traffic patterns from
Monday through Thursday tend to be similar but Friday has more traffic than other
weekdays (Highway Capacity Manual 2000).

This study defined the normal condition as a weekday (Monday through Thursday)
without accidents, events, construction, maintenance, or severe weather. The days which
had no CMS message record on the CMS message log information database are
considered normal condition days. However, when there was major construction (lane
closed) or severe weather conditions (e. g., ice, fog, and thunder storm), these days were
eliminated. Also, national holidays’ such as New Year’s Day, Independence Day, Labor

Day, Thanksgiving and Christmas were not included as normal conditions.

3.5.3.2 Normal Condition Data Screening and Filtering

Even among the normal condition days as defined above, the database may
include unannounced/undetected accident conditions or abnormal traffic patterns from a
sports (or other) event. If these abnormal conditions are included in the normal condition
days, it could bias the results. Therefore, this study screened the normal condition days.
Each 10 minutes of data on the upstream and downstream volume and speed were plotted
and examined for outliers. The days, which have an abnormal pattern of volume or speed
either at the upstream or at the downstream detector stations were eliminated. Figure 3.3

(a) and (b) shows the volume and speed before filtering and figure 3.4 (a) and (b) shows

49

these parameters after filtering. Table 3.4 shows the total number of days processed, and

the filtered samples at each site.

Table 3. 4 Normal Condition Samples at the Study Site

 

 

 

 

 

 

 

 

 

 

Site Site 1 Site 2 Site 3 Site 4 Site 5
Total normal condition samples
based on CMS Message Log 44 50 40 55 40
Information Database
Filtered AM-peak sample 35 38 34 43 34
Filtered Non-peak sample 33 36 33 42 24

 

50

 

 

 

 

 

 

 

 

 

 

mm 2 85.
a4. 2 Q4?
8 2 - 8. 2
”Ham 0.. . O92.
are. r Grow
. 8. 2 . 85.
. 8a . 85
. «45 . e45
was . .
4 . one
. 3a a - as
f 25 m... . are
4 85 o . 8.9
. r .
f 3.» P . o3
mvfi C r wvum
. 85 m . on.
as m . no
. 25 o - 94..»
. . V are
. 8.0 m . .
8; n m . 8.»
. as» .u . 8K
8s ..m . 21.
8; m . 8k
. are C . was
. 88 m . 8x
. 9.5 o . ammo
- 85 N . 84a
as ) - 8mm
. . a mwb
are ( . .
. 85 2e
. 8e. 85
H $8 . an...
. omum . 3mm
«mum . mmum
. Se - 9.8
88 - 96
m m m m m o q d a d lfi - a. [1 4 “Com
1 m w m m w .0. m m m 0
22¢ oE:.o> 3&5 been»

 

 

 

Time

(b) Normal Condition Speed Profiles
5 1

Figure 3. 3 Normal Condition Profiles Before Filtering at Site 1

 

 

 

 

\ I... a.
a. H v
.1 .1
sawté
4m. av
. .f
c...
4..
3M...
9..., .
. a. x]
. 5r .g
. i‘n , ..__ .
7, ..

 

 

mmmmm

22c oE:_o>

 

Time

 

 

(a) Normal Condition Volume Profiles

 

. one
. a}:

89
8qu
25.
$8

. m3
8”

 

 

 

 

Time

 

 

(b) Normal Condition Speed Profiles

Figure 3. 4 Normal Condition Profiles After Filtering at Site 1
52

3.5.3.3 Accident Condition Data Collection

The literature review (Benson, 1996) reported that driver responses varied by
message type. Drivers are more likely to divert for an “Accident” message than for a
“Congestion ahead” message. It follows that the use of different types of accident related
messages (such as disabled vehicle ahead, congestion ahead, freeway or lane closed, and
ramp closed) might have differing affects on driver response. Therefore, this study
analyzed data only at times when the CMS displayed “Accident” in the first line of the
CMS text message. This study also used only accidents which occurred downstream of

the downstream ILD but before the next interchange on the link.

3.5.3.4 Accident Condition Data Screening and Filtering

The objective of this study is to determine the effectiveness of CMS in diverting
traffic by comparing the through traffic volume ratio between the upstream and
downstream 11st for accident conditions and normal conditions. If the CMS message
does not influence the drivers’ route, the through traffic volume ratio during the accident
conditions will be similar to the same time period under normal conditions. However, if
the CMS influences the drivers’ route change behavior, the through traffic volume ratio
will be reduced. In addition, if congestion from the accident extends upstream beyond the
diversion interchange, the drivers’ decision to divert to another route will be at least
partially based on encountering the congestion as well as seeing the CMS. Thus, the
database of accident days was filtered to remove all accidents where the congestion
queue extends upstream of the diversion point. Table 3.5 shows the number of accident

cases analyzed at each study site.

53

Table 3. 5 Accident Condition Samples at the Study Site

 

 

 

 

 

 

 

 

 

 

Site Site 1 Site 2 Site 3 Site 4 Site 5
Accident Sample 18 8 l3 8 13
3.6 Summary

The fundamental hypothesis for this research was that a CMS will not provide the
same benefit at all locations and at all times of the day. However, the evaluation results
from the literature were generally based on a particular location and time which may
explain the inconsistent results reported in the literature. To address this problem, this
research selected and analyzed the effectiveness at five CMS locations which have

different geographic and traffic conditions.

54

CHAPTER 4

MEASUREMENT METHOD

4.1 Introduction

Previous research on driver behavior indicated that drivers have a propensity to use
one regular route and are not likely to change their regular route under normal conditions.
These studies also indicated that even though drivers are hesitant to change from their
regular route in the absence of any information, some would use an alternate route if
presented with CMS information. These conclusions, however, are based on drivers
stated preference, and have not been fully evaluated with field data.

This study was conducted to identify and quantify the effectiveness of CMS using
field traffic data. The measure of effectiveness (MOE) used in this study is the percent of
traffic that diverts to an alternate route when presented with CMS information. The MOE
is measured by comparing the ratio of downstream traffic volume to upstream traffic
volume with and without CMS information. The variable used for this comparison is

called the “diversion ratio”.

The MOE was obtained by measuring the traffic volume at points upstream and
downstream from an interchange immediately downstream from a CMS location.
Volume data were collected and analyzed both when the CMS was actuated and at these

same stations under normal conditions (CMS was not activated).

55

4.2 ILD Traffic Data Collection Method

When an accident occurs, drivers who are informed of the accident upstream of an
interchange must decide to either stay on their current route or divert to an alternate route.
However, the true diversion ratio can not be obtained simply by comparing the upstream

and downstream traffic volumes during the time the CMS is activated for two reasons:

1) Vehicles still queued between the accident and the interchange when the
message is discontinued would not be recorded at the downstream location, and
these would be counted (incorrectly) as diverted vehicles; and

2) If the queue formed by the accident reaches the interchange, drivers may
choose to take the alternate route because they observe the congestion rather

than because they observed the CMS.

Therefore, it was necessary to develop a technique to identify all of the vehicles
(and only those vehicles) that responded to the CMS by choosing an alternate route.
Figure 4.1 shows a time-space vehicle trajectory diagram between an upstream and
downstream Inductive Loop Detector (H.D) location. When the CMS provides accident
information at the upstream location between T1 and T2, the first drivers who received the
message and remained on the current route arrive at the downstream location between
time T3 (if the accident induced congestion has not reached the downstream ILD location)
and T4 (if the accident induced congestion has reached the downstream ILD location).

The last drivers who observed the activated CMS message arrive at the downstream

56

location between time T5 (if all accident induced queued vehicles have been released) and
T6 (if not).

If the accident induced queue does not reach the downstream ILD location during
the CMS message time period, the upstream vehicles will arrive at the downstream
location with the same travel time as under normal conditions. However, if the accident
induced queue reaches the downstream location, the travel time to reach the downstream
ILD location is increased amount of time a or B. Therefore, an accurate count of traffic
that passes the downstream location must consider the increased travel time resulting

from the congestion.

 
 
   

 

 

 

 

 

   

 

g FrrstCllB LastCVS
5 irlomndvah‘des flanedvelides
.QA g
1:
Mean . .
l 5
time 5 E 5 E E E
EOVSoperaimE 5 : : :
: ti penod' : r a i l :

—||
“A

T2 T3 T4 T5 fine

Figure 4.1 Vehicle Trajectory Time-Space Diagram

57

The equation used to determine the traffic volume at the upstream and

downstream ILD are shown below,

Accident Condition

2
Qup__acc = unp_acc (Ti)

r=1
5+,6

Qdown_acc = Z qdown_acc (Ti)
i=3+a

Where,
i = time (minutes)

Qup _ ace = total upstream volume during CMS accident message (vehicles)
Qdown _ ace = total downstream volume during CMS accident message (vehicles)
qup a“ (Ti) = accident condition upstream volume at time T; (vehicles/minute)

qdown _ acc (T,) = accident condition downstream volume at Time T,- (vehicles/minute)

a, ,B = congestion delay time (minutes)

Normal Conditions

2
Qup_nor(j) = unp_nor(Ti)

i=1
5

Qdown_nor(j) = Z qdown_nor (Ti)
i=3

Where,
i = time (minutes)

j = a normal condition day j

58

Qup _ "0, (j) = total upstream volume during nomzal condition day j (vehicles)

Qdown _ no, (j) = total downstream volume during normal condition day j (vehicles)
qup_ n0r(7;') =normal condition upstream volume at time T,- (vehicles/minute)

qdown_,,0,(Ti) =normal condition downstream volume at time T,- (vehicles/minute)

Diversion Ratio Measurement

The diversion ratio was calculated as:

Qdown nor (j) X Qup acc
Exp — Qdown _ acc : - . -
Qup _ nor (1 )

 

Exp __ Qdown _ a“. — Re al _ Qdown _ acc = diverted trafiic volume

diverted trafiic volume
EXP — Qdown __ acc

 

% diversion ratio =

Where,

Exp _ Qdown _ ace = expected total downstream volume during accident condition with

CMS message (vehicles)

Re al _ Qdown _ ace = real total downstream volume during accident condition with CMS

message (vehicles)

Q“p _ acc = total upstream volume during CMS accident message (vehicles)

Q“p _ no, (j) = total upstream volume during normal condition day j (vehicles)

59

4.3 Travel Time Measurement Method

The time intervals T3 - T1 and T5 - T2 depend on the speed of traffic and the distance
between the upstream and downstream ILDs. Since continuous speed data over this
distance is not available, the travel time must be estimated from the speed data collected
at the two ILD locations. When traffic demand approaches or exceeds reduced system
capacity resulting from the accident, congestion is generated and travel time increases.

From the minute by minute speed data collected at the upstream and downstream
ILD locations, link travel time must be estimated. Depending on the downstream traffic
conditions and the ILD locations shown in Figure 4.2, travel time was estimated as

follows:

 

 

 

 

 

 

 

 

 

Figure 4.2 ILD Location Diagram

Case 1: No Accident Induced Con estion Exists at the Downstream Location

 

a) Link Travel Time 5

6O

#up_ A (ti) E :udown_ 8014a)? deSign speed

 

_ D
luup_A(ti)
Then
LTAB : __P__
taup_A(ti)

b) Link Travel Time A_C

:uup_A(ti) E :udown_C (ti+a') Z deSign speed

 

 

_ d1
.uup _ A (ti)
Then
LTAC 2 d1 + d2
#up_ A (ti) tudown_C (ti+k)
Where,

i = time (minute)

D = link distanceAE (mile)

d1 = link distance AA- (mile)

d2 = link distance ATE( mile)

”up-A (ti) = upstream ILD A vehicle speed at time i (mph)
[140me (1;) = downstream ILD B vehicle speed at time i (mph)
[1.10me (t,) = downstream ILD C vehicle speed at time i (mph)
LTAB = link travel time E (minute)

LTAC = link travel time A—C( minute )

61

Case 2: Accident Induced Congestion at Downstream Location
a) Link Travel Time XE
”up_ A (ti) tie ”down _ B (ti+a)

D
flup_A(ti)

 

k:

,uup _ A (ti) 2 design speed

”down _ B (tHa) = congestion delay speed

LTAB= D_QD + QD

#up_ A (ti) :udown __ B (ti+k)

 

a) Link Travel Time A?

#up_A(ti) ¢ tudown _C(t1+a)

_ d1 + d2
.uup_A(ti) ”down_C (ti)

 

 

,uup_ A (ti) 2 design speed

* a
”down _C (ti) : dang" speed
,udown _ C “H a) = congestion delay speed

Then
d1 + dz—QD + QD

LTAC= *
auup_A(ti) :udown_C(ti) fldown_C(ti+k)

 

 

62

Where,

i = time (minute)
D = link distance-A—B (mile)
d; = link distance A? (mile)

d2 = link distance A‘B (mile)
QD = congestion queue distance

pup_A (ti) = upstream ILD A vehicle speed at time i (mph)
[.410me (t,-) = downstream ILD B vehicle speed at time i (mph)
leown_C (ti) = downstream ILD C vehicle speed at time i (mph)

LTAB = link travel time E (minute)

LTAC = link travel time R (minute)

4.4 Queue Distance (QD) Estimation Method

When a sudden reduction in capacity occurs due to an accident, the flow on the
road suddenly changes from volume (q 1) to a lower value of volume (qz), with a
corresponding change in density from k, to a higher value k2. When congestion exists, the
density on the road is relatively large and the speed of the vehicles is relatively low. A
shockwave (u...) is generated as a result of this change in conditions. When the density
downstream is lower than upstream (u,,. is positive), we have a diffusion of flow similar to
that observed when a queue is discharging. When the downstream density is higher than
upstream (u... is negative), then shockwaves are generated and queues move in an
upstream direction.

The shockwave speed can be estimated using the volume and density upstream
(inflow) and downstream (outflow) by the following equation. The congestion distance at

any time (T,) can be obtained from the shockwave speed:

63

_ qdown _Qup

 

“w

kdown _ kup

Where,

u,,, = shockwave speed (mph)

q“p = upstream (inflow) volume

kup = upstream( inflow) density

qdown = downstream (outflow) volume

kdown = downstream (outflow) density

On an uninterrupted segment of roadway for which a flow-density relationship is
known, the congestion distance is equal to pw(t). However, in this study, the [IDS are not
located on an uninterrupted segment of roadway, since there is an interchanges and
entering/exit ramps between the two locations; and, for a site which has an alternate route,
the [IDS may not be located on the same roadway. Therefore, the upstream and
downstream ILD locations may have different flow-density relationships from each other.
The volume profiles for the upstream and downstream 11st at Site 1 under normal

condition are shown in Figure 4.3.

64

 

1600 -
1400 -
1200 ~

Volune
§

 

 

assaaaaéaaaaaéaaaaseaaaa8
mmmmmmmowmomuhthKmémmomd

939
9.49“
959 ‘
1009 ‘
1019 ‘
10.29 ‘
10.39 ‘
1049:
10.59

11:09 ‘
11:19 ‘
11:29 ‘
11:39 '
11:49 ‘
11:59]

9:19
929:

Time

 

[—0— Upstream —a—- Downstream]

 

 

 

 

Figure 4.3 Upstream and Downstream Volume Profile at Site 1

Therefore, it is not appropriate to use the upstream ILD traffic data as the inflow
traffic for measuring shockwave speed during the accident time period. Instead of using
the upstream flow-density data as inflow traffic at the downstream location during an
accident condition, this study used the flow-density that occurs under normal conditions
at the downstream ILD as the inflow traffic. Figure 4.4 shows volume and speed profiles
at the downstream location for an accident condition and the average normal condition.
As can be seen in the figure 4.4, the volume and speed are similar between the accident
and average normal condition except during the congestion time period resulting from the
accident. Therefore, it can be assumed that if the accident had not occurred, the inflow
traffic of volume and speed during the accident time period would be similar to the
average normal conditions. Using this assumption, the congestion distance at any time

following the accident occurrence can be calculated.

65

 

 

Congestion period
from accident

1600 ~
1400 1

2000 i
1800 1

1 mm”:
1 ad”:
. an":
r am”:
. mt:
mo”:
. mmuow
. manor
. manor
. 85.
- who?
. 86..
- mmum
. mama
. 8mm
- mNnm
0.0.0.0...ko mwum
. mono
. ammo
. meme
. 8mm
. mNHQ
I ate
oeeeeoeeeeoeeeeoowmoum
. mmK
. was
a 8K
. emu.
- mth
- 8K
fwmuo
- $30
- 8mm
. awe
, mic
i wonm
. mmnm
- 93m
. 8a
1 mNNm
x. mwnm
8mm
0

         

 

400‘
200*

600 .

 

[-0- Accident -¢— Ave. Normal I

 

 

 

(a) Volume Profile

 

 

Congestion period
from accident

     

i1-

 

 

90~
eo«
70<
eo~
504
404
so~
20~
10«

3

Accident —.— Ave. Normal ]

 

F.

 

 

 

(b) Density Profile

Figure 4.4 Volume and Density Profiles at Site 1 Downstream Location (I-696 EB)

66

As shown in Figure 4.5, the shockwave speed was calculated for three different
conditions: (a) a backward shockwave which occurs as a result of the normal flow-
density conditions encountering the flow-density conditions of the queue formed as a
result of the accident, (b) a backward shockwave resulting from the flow-density
conditions that exist with some traffic diverted in response to the CMS encountering the
flow-density condition of this queue, and (c) a forward shockwave during the queue

dissipation period.

67

Speed

 

 

 

 

 

 

 

 

 

1:: -. AA C.-°' '
0%. o... ..
....... e 3........e'
StatimaryLmerSJeedCbndfim
ta tb is td TIITB
Dom ta tn ta ta
ILD l 11 :
uw(A) "'3
"“00 uw<C>
Uw(B)
V

 

Figure 4.5 Shockwave Speed Diagram

68

The shockwave speed and congestion distance were estimated using the equations

 

 

 

 

below:
m m
_ 24mm) _ 2km (’1')
_ t _ = i=n k' t _ ___i_={t__
qm(n m) tm—tn "A" m) m_tn
m m
_ Z qout (t i ) _ 2 k0“, (ti )
qout(tn-m)= 1:" t k0ut(tn_m)= 1:"
m _ n tm —t"
_ 6O quut (ta- b)_ qin (til— I?)
uw(A) ‘—
number of lanes>< kou,(ta_ b)- kin (ta— b)
uw(B)= 6O xqout(tb- -C) qin(tb c)
number of lanes kout(tb_ c)- kin (11,— c)
t t
uW(C) : 60 xqout( C- -—d) qm( C" ’d)

number of lanes><

kout(tc— d)“ kin(tc— d)

QD(A) = uw(A)x1.47x60x(tb —ta)
QD(B) = uw(B)x1.47x60x(tc —z,,)
QD(C) =uw(C)x1.47x60x(td —tc)

Max_ QD = QD(A) + QD(B)

Where,

i = time (minute)
qin (ti )
kin (ti)

= inflow traflic volume at time t,- (veh/min)

= inflow traffic density at time t,- (veh/mile)

€101.10.) = outflow trafiic volume at time t,- (veh/min)

69

kout(ti) = outflow traflic density at time t, (veh/mile)

21in (tn—m) = average inflow traffic volume during time n - m (veh/min)
kin (tn—m) = average inflow traffic density during time n - m (veh/mile)
q out (I n—m) = average outflow traffic volume during time n - m (veh/min)
k0“; (tn—m) = average outflow traffic density during time n - m (veh/mile)
uw(A) = shockwave speed in group A trafi‘id mph )

uw( B ) = shockwave speed in group B traffic (mph)

uw( C) = shockwave speed in group C trafi‘ic (mph)

QD (A) = congestion queue distance during condition A (feet)

Q0 (8) = congestion queue distance during condition B (feet)

QD (C) = congestion queue distance during condition C (feet)

Max_QD = maximum congestion queue distance (feet)

4.5 Summary

Using data from the upstream and downstream ILD data under normal conditions,
the ratio of these two volumes can be used to estimate the number of vehicles that would
be expected to pass the downstream ILD under various conditions using upstream ILD
data from days when an accident occurred and the CMS was activated. The equations
developed in this section are used to estimate the number of vehicles that actually passed
the downstream ILD. The difference between the expected number of vehicles and the
actual number of vehicles that passed the downstream ILD is a measure of the traffic

diverted to an alternate route due to the CMS.

7O

CHAPTER 5

DATA ANALYSIS AND RESULTS

5.1 Introduction

These analyses were performed to accomplish three primary goals: first, to measure
the route diversion effectiveness of CMS; second, to determine the sensitivity of the
diversion factors which exert the greatest effect on a drivers’ route decision; and last, to
identify the characteristics of locations where the CMS will be most effective. To
accomplish this, the diversion ratio for each of the five sites was determined under
several conditions and the influence of factors which affect the diversion ratio were then

determined from these results.

5.2 Sensitivity Analysis Factors
5.2.1 Familiarity and Time Constraint Sensitivity

From the literature, it was determined that a drivers’ route diversion is influenced
by their familiarity with the street network and their time constraints. The drivers who are
familiar with an area are more likely to divert to an alternate route. Also, drivers who
have time pressure are more likely to switch their route.

In general, in the AM-peak period there is a larger percentage of home-to-work
drivers in the traffic stream than the same location at other time periods. This group of
drivers is more willing to divert from their usual route to an alternate route because of
two factors: they are likely to be familiar with alternative routes, and they are under time

pressure. In the non-peak period there are a larger proportion of shopping trips and non-

71

work trips, and past research has found that these trips are less influenced by these two
factors. The diversion behavior of drivers is hypothesized to be significantly different
between these two time periods when the same CMS message is displayed. To test this
hypothesis, this study compares the diversion ratio during the AM-peak period (06:00 —
09:00) with the Non-peak period (10:00 — 16:00) to determine the effect to these two

factors.

5.2.2 Visual Sensitivity

Previous surveys reported a relationship between drivers’ route diversion
behavior and traffic congestion. Most of the results indicated that drivers are more likely
to change routes when they observe traffic congestion than when they are simply
informed of the fact that it exists. To test this hypothesis, this study determined the
difference in the diversion ratio when drivers encounter the queue formed from the
incident and when they only observe the CMS. To do this, the models to estimate the
length of the congestion queue for each accident were developed, and the incidents were
classified into two groups: an accident condition in which the congestion moves upstream
beyond the diversion point and an accident condition where drivers could not observe the

congestion prior to the diversion point.

5.2.3 Traffic Condition Sensitivity

Previous surveys also reported that drivers expressed a higher willingness to

divert if an accident occurs at a time and location where recurring congestion exists. The

72

third sensitivity analysis compares the change in the route diversion ratio on routes with

and without recurring congestion.

5.2.4 Geographical Location Sensitivity

The final sensitivity analysis compared the diversion ratio base on the geographic
location. Some sites have an alternate freeway route, while other locations have only
arterial road alternate routes. Three sites (Site 1, Site 2 and Site 3) have freeway-based
alternate routes. The remaining two (Site 4 and Site 5) only have exits to arterial roads.
This study examined the diversion ratio between locations with and without freeway-

based alternate routes.

5.3 Analysis of the CMS Effect at Site 1
5.3.1 Description of Site 1

Site 1 is located at the junction of I-96(I-275)/I-696 on the Novi/Farmington Hills
boundary in southern Oakland County. Both I-96 and I-696 are major east/west freeways
in the Metropolitan Detroit area. CMS 23 provided information on downstream traffic
conditions to eastbound drivers in advance of an interchange which provides three
alternate routes: M-5 south-east bound, I-96(I-275) southbound and I-696'eastbound. I-
96(I-275) and I-696 have a 70mph speed limit and provide nearly equal travel distances
to the center of downtown Detroit. M-5 provides the shortest distance to the center of
downtown; however, it has a 65mph speed limit before the junction at Grand River

Avenue and a speed limit of 45mph from there to the center of downtown.

73

As shown in Figure 5.1, traffic data are collected at three ILD locations. The
upstream ILD (A) was located on I—96 west of Novi Road. The two downstream ILD
used at this site are located on I-696 east of Farmington Road (B), and on I-275 south of
Grand River Avenue (C). The CMS provides information on accidents that occur on I-
696, L96 (I—275) or M-5 downstream of the interchange. The downstream section of I—
696 has 20 - 30 minutes of recurring congestion during the AM-peak period, but the

downstream sections of I-96 (I-275) and M-5 do not experience recurring congestion.

 

 

Juniuwtfiu J 5' r“ = i

1' 4111 c! U] if : KK‘r’g, £0031”?

12m MRO ‘." ‘0‘
gamut”? :
a . c. FARMINGTQN
= K »..,

_ a: .2. ’2 “5. --~—-=—__*—::':2; ‘ff““n“"‘m
I

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Figure 5.1 Site 1 Location

Data on eighteen accidents were collected at site 1. Table 5.1 shows the date, time,
and location of each of these accidents. The congestion distance for each accident case

was calculated using the queue length measurement technique described in Chapter 4.

74

From the data collection procedure in chapter 3, filtered normal condition sample data for

each accident case was constructed.

There was a construction project on this freeway segment in July, August, and
September 2001. Among the collected accident cases, four accidents (cases 1, 2, 6, and
17) occurred during the construction period. The normal condition sample data used as a

comparison for these four accidents were collected during the construction time period.

75

 

mm

mm V 30132

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

god 9:2 w 3 mm SN-— EoEoo< emu: I cmnfl 8.3-2 2
9 n5 v DOI§2 98¢ 0:2 m 283 mm 2N-— Eo28< N: ”m: I N2 “2 5.3.8 5
mm n5 v QOJS2 30m 828—5 283 mm 0%-“ E0203. m _ x: I m 2 "E 86on 2
mm mm v Q0Ixa2 30¢ :3an2 283 mm 0%; EoEou< 3 no: I E “m _ 8.2-3 2
mm do v QOJS2 98% 8:3 99:80 3 mm 0%; Eo28< owns I on”: 8.2-8 2
mm n5 v Q0I§2 30m :5 scam—item 883 mm 0%; E0203. 9&2 I 9am— 5-2-: 2
mm mm v DOIxm2 god 0:2 o 3 mm mum; EoEou< wmuwo I wmfio 8-3-2 N:
mm mm v Q0Ixm2 28m 0:2 o 3 mm mm; EoEuo< Omnwo I Omfio No-om-wo 2
mm A5 A Q0I§2 98% 0:2 n 3 mm pm; 2ch3.. omnwo I omfio 8.3-5 2
mm mm A DOJS2 30m 0:2 o 3 mm mum; E0203. omuwo I 008 8.9.8 0
mm as v Q0I§2 30m e:2 m 3 mm mum; EoEoo< 3qu I vans No; Two w
mm mm A QOJS2 30m 0:2 o 3 mm RNA EoEoo< 3qu I emuco 5.5.”: n
9 n5 v Q0Ixa2 2.6% 0:2 m 3 mm mum; EoEou< omnwo I omfio 5-3-00 0
mm mm v Q0I§2 god 8:5 “:2qu H mm 0%-— EoEoo< m: ”we I M: “no No-8; 2 m
mm mm A DOJS2 28M £9523. 888 mm 0%.“ E0203. 2 Uwe I S “no 8-3-5 9
mm mm v DOJS2 25% 59336.: a mm 0%; EoEoo< 068 I omfio 3-2-: _ m
v ma v Q0132 god 8.3 3280 3 mm coo-“ EoEoo< mmnoo I Nmnwo 5-3-8 N
v A5 A Q0Ixm2 68m 2.30622 3 mm 0%; EoEoo< Sumo I Sumo 5-8-5 2

295m
.. .._......_......

 

 

 

 

 

 

 

c 85 a .9982 $6 5? 22:28 .828... 38:8 e 2.33.38: c .m 2...?

76

Previously, it was indicated that traffic flow at a section of a roadway varies from
day to day, however a certain time period of flow patterns and volumes are similar to
each other. Even though an accident occurred at a downstream location, traffic flow at an
upstream would be similar as the same time period under normal conditions. Figure 5.2
shows the upstream traffic volumes of the eighteen accident cases and the related normal

conditions at site 1.

 

Condmon
man-
. 0 Normal
wand -*Eknwm

Up_Vol

3,33% iii: lift;
40001 if ggfigfig

O

 

 

 

IIIIlllllIlTFlllll
123456789101112131415161718

Group

Figure 5. 2 Upstream Volume Distribution at Site 1

It is shown that the upstream volumes of all eighteen accident cases are similar to
the related normal condition. Therefore, it is expected that if the CMS has not effected

drivers’ route diversion behavior, the through traffic volume ratio should be similar to the

77

normal conditions. Figure 5.3 helps to visualize the distribution of the downstream to

upstream volume ratio between accident and the related normal conditions.

78

2823800 :53— oE=_o> E85335 Qm 9.53..—

 

 

 

 

 

 

 

52.280
“co—300.1 _uFtoZ
_ _
lode
E O l0.0N a
O .
lodn m
.8
lodv m.
0
I. lodm
969.0
53.280
«£03006. _octoZ
_ _
com
A... m
_
.. .. lo om oA
rah _m
m.
0
II. loos.

 

 

 

N “dado

COEVCOO
«co—£006. _choZ
_ _

 

 

0

 

 

lo.ON

lodn _

oma'IOA ma

l
q
o
V

 

lodm

 

O
.a. .
n 6ch
52980
E02001: _octoZ

 

 

.mT
a

l
.0.
0
an

I I I
O O O
O O O

to
0993 IOA’UMOG

0

 

l
O.
o
I\

 

 

79

32:5:ch 382385 8.5— 2.5.9» Egan—.Bcn Wm 259m

coEvCOO
«coEuox‘ _uctoZ
_ _

 

 

 

 

 

 

COED-.00
acoEoo< _uctoZ
_ _

 

 

.a. .m—
._._

 

 

 

Iod.
G
Iodm _w
m
lodn .8
ma
0
Iodv
lo.o
m
_
Io or m
.8
m.
0
lodm

 

 

m ”ammo

 

 

 

 

 

 

 

 

coEucoo
EoEuu< .8502
_ _
ImT 10.0 w
G
lodw _m
.m
.lo.on U
m.
0
rled?
h “ammo
coEUcoo
«£02094 _uctoZ
_ _
10.0—.

gal ,1

fi.

N

I I l
q o o
o o o
v n

0998 |°/\ W00

 

l
q
o
to

80

92.5258 macaw—595.0 on“: 2:23» 533235 rem 8:9,.—

 

 

 

 

 

 

 

 

cozficoo
acoEuod‘ .3502
_ _

[0.9
m

lodm
Wu _A
.9. I _w
[069 N.
m.
o

Iodv

Nr ”ammo
55.2.8
522091 .9502
_ _

m

10.3
_A
, I Fodw _W
I, I I, no
I_. led» o.

 

or Noam-"0

 

 

 

 

 

 

 

 

coaucoo
«co—2001‘ _choZ
_ _
T.|O.O —.
G
.53 _m
_ w m
., lo.on _H
H m.
1.0.0.?
w w ”ammo
52.280
«co—£004 _uctoZ
_ _
10.0
E O
..l0.0_.
. m
WE: IIO ON _p
E a
[0.00 m.
0
[0.0%
m ”omnO

 

 

 

 

81

Gas—5:09 38259.80 23% 2:23» Samba—.30: m.m 953,.—

 

 

 

 

 

 

 

 

COEVCOO
«co—3004‘ _uctoZ
_ _
Iodn
O
o m
193
_A
:9.- lodm _P
a
m.
10.8 0
OP “0m00
COEUCOO
«c00_00< .0532
_ _
193
.mT m
Iodm
_A
lode _ol
8
M.
[eds o
VF ”0000

c0£0c00

2228.1 .2502
_ _

 

 

O
0

A

 

 

m_. ”0000

£035.50

502004. .0532
_ _

 

 

 

¢ m.

 

 

 

 

m _. ”0000

82

80:52::an Satan—~80 0:3— 05:15 Sacha—Eon Qm 23E

53.800

.5203
_

_quoz

 

 

.lodv
[0.00
[0.00 _
lodh _
lodm

 

[0.0m

 

 

 

0A umoq

0998

52.280

52203
_

_uctoz
_

 

 

.m:

E

._I

 

 

N. w ”0000

I
q
o
to

l
q
0
V

 

l
0.
0
ID

0998 IOA’umoo

83

From Figure 5.3, it is shown that the accident conditions of downstream volume
ratio during the AM-peak period are outside of the lower boundary of the box (represents
the 25‘h percentile limits of the ratio) from the normal conditions while the accident

conditions in the non-peak period are inside the lower boundary of the box.

5.3.2 Diversion Ratio Analysis

As defined in Chapter 4, the difference between the expected and the actual
downstream volumes resulting from the CMS display were considered as diverted traffic.
Table 5.2 shows the CMS operation time period for each of the crashes along with the

message, the queuing conditions and the number of days in the normal operation sample.

Table 5.2, lists the upstream and downstream volume for each case along with the
diversion ratio. The diversion ratio ranges between -1.96% and 16.85%. Because there
are daily variations in the ratio of the downstream volume to the upstream volume, a
statistical test was conducted to determine the significance of the change in the
downstream traffic volume on the accident date. A t-Test was conducted using the
percent reduction from each normal condition day as an independent sample point. The

null hypothesis is that there is no reduction on these days;

H0:d=0
Ha:d¢00rd>0

Where d is the diversion ratio on the accident date

84

Table 5. 2 Accident and Related Normal Condition Traffic Volume at Site 1

Upstream Downstream Expected Volume %
Downstream

Volume Volume Reduction Reduction
Volume

Case Condition
Case 1 Normal 4551 7224

Case 2 Normal 3799 6155

Case 3 Normal 5789 7895

Case 4 Normal 5828 7882

Case 5 Normal 5814 8008

Case 6 Normal 5638 6050

Case 7 Normal 1 1344 13860

Case 8 Nomial 5735 7217

Case 9 Normal 5539 6881

Case 10 Normal 5407 6396

Case 1 1 Normal 5326 6708

Case 12 Normal 5232 6537

Case 13 Normal 3304 4952

Case 14 Normal 2751 4326

Case 15 Normal 3814 6026

Case 16 Normal 3515 5356

Case 17 Normal 3703 5058

Case 18 Normal 3905 6543

Normal 5055 6837

Average

 

85

The result of this test is shown in Table 5.3. The mean through traffic percent
reduction resulting from the CMS information on the 18 accident cases was 6.07% and
the standard deviation of the difference is 7.72%. The 95% confidence interval for the
average difference is 5.39% to 6.75%. Since the confidence interval does not include the
value of 0, therefore it rejects the null hypothesis that the downstream through traffic
percent reduction during the CMS display is 0. As it was expected, the significance level
(p=0.000) is smaller than .05, leading to a rejection of the null hypothesis. Therefore, the
results indicated that the downstream through traffic volume reduction during the CMS

operation time period was significantly different from 0.

Table 5. 3 One Sample t-Test (p < .05) for Downstream Volume Reduction

 

 

 

 

Test Value = 0
95% Confidence Interval
Std- of the Difference Sig.
N Mean Deviation ‘ (2-tailed)
Lower Bound Upper Bound
% . 496 6.07 7.72 5.39 6.75 17.51 0.000
reduction

 

5.3.3 Sensitivity Analysis at Site 1
5.3.3.1 Familiarity and Time Constraint Sensitivity Analysis

Twelve and six accidents occurred during the AM—peak and Non-peak period,
respectively. The percent reductions between these two time periods were compared to

determine the effect of familiarity and time constraints. As shown in Figure 5.4, the

86

average percent reduction in the AM-peak period is more than five times higher (8.45%)

than in the non-peak period (1.47%).

 

_A
O

 

%ReductIon
O inwammumo

 

AM-Peak Non-Peak
Time Period

 

 

Figure 5. 4 Average Through Traffic Percent Reduction at Site 1

To examine the significance of the mean difference between the two groups, a t-
test was conducted. The results are given in Table 5.4. There is a statistically significant
difference at the 95% confidence level. This finding is consistent with the literature

where drivers stated they would be more willing to divert in the AM-peak period.

87

 

Table 5. 4 Independent Sample t-Test for Familiarity and Time Constraint (p < .05)

 

 

. . Std. Std. Error Sig.
Condition N Mean Deviation Mean t (2-tailed)
AM-Peak 327 8.45 7.76 0.43

l 1.94 0.000
Non-Peak 169 1.47 5. 16 0.40

 

5.3.3.2 Visual Sensitivity Analysis

Depending on the congestion queue distance, the data was classified into two
groups: days when the queue reached the diversion point (QD 2 DP) and days when it did
not (QD < DP). There were no accident cases where the congestion queue reached to the
diversion point during the non-peak period. Therefore, only the AM-peak period data

were classified into these two groups and analyzed.

The results are shown in Table 5.5. The average through traffic reduction when
drivers observe the congestion before reaching the diversion point is almost three times
higher (13.26%) than where they can not see the congestion (4.55%). The mean
difference is statistically significant at the 95% confidence level. The results indicated
that when drivers can observe the queue, in addition to seeing the message on the CMS,

there is a change in drivers’ behavior.

88

Table 5. 5 Independent Sample t-Test for Visual Sensitivity (p < .05)

 

 

. . Std. Std. Error Sig.
C°“d"‘°" N Mean Deviation Mean ‘ (2-tailed)
QD < DP 183 4.66 7.71 0.57

11.90 0.000
QD 2 DP 144 13.26 4.47 0.37

 

QD = queue distance, DP = diversion point

5.3.3.3 Traffic Condition Sensitivity Analysis

As previously noted, downstream I-696 has a 20 - 30 minute period of recurring
congestion during the AM-peak period, but downstream I-275 does not experience
recurring congestion. To measure the sensitivity of the diversion ratio with and without
recurring congestion, the data were separated into the two groups depending on the
location of the accident. Since recurring congestion does not exist in the non-peak period

only the AM-peak period was compared.

The results are shown in Table 5.6. Almost twice as many drivers diverted to I-
275 when an accident occurred on I-696 (11.72%) than diverted to I-696 when an
accident occurred on I-275 SB (6.73%). This difference is statistically significant with a
95% confidence interval. Therefore, the results show that the existence of recurring

congestion affects the drivers’ route diversion behavior.

89

Table 5. 6 Independent Sample t-Test for Traffic Conditions (p < .05)

 

 

Std. Std. Error Sig.
Route N Mean Deviation Mean t (2-tailed)
L696 113 11.72 6.11 0.58
5.80 0.000
I-275 214 6.73 7.99 0.55

 

However, in the previous sensitivity analysis, it was determined that drivers are
more likely to divert when they can see the congestion queue. Therefore, the AM-peak
period accident data was further divided into four groups. Figure 5.5 shows the average
through traffic diversion ratio for each group. When the accident occurred on
downstream I-696, the diversion ratio was not much different between those who
observed the queue (QD 2 DP, 13.09%) and those who did not (QD < QD, 11.00%). The
difference was not statistically significant at a 95% confidence interval (p = 0.084, Table

5.7 (a)).

90

 

% reduction

 

 

 

l-696

13.09 13.33

0.36

 

OD < DP
l-275

 

Condition/Route

 

Figure 5.5 Average Percent of Through Traffic Reduction Based on Visual

Sensitivity at Site 1

However, when the accident occurred downstream on I-275 SB, the diversion

ratio was 13.33% when drivers observed the queue at or before the diversion point, and

0.36% when they did not. This difference is statistically significant with a 95%

confidence interval (Table 5.7 (b)).

Table 5. 7 Independent Sample t-Test for Traffic Condition Sensitivity (p < .05)

(a) Accident on I-696

 

 

Std. Std. Error Sig.
Rm“ N Mean Deviation Mean ‘ (2-tailed)
QD < DP 74 11.00 6.69 0.78
1.74 0.084
QD 2 DP 39 13.09 4.60 0.74

 

91

 

(b) Accident on I-275

 

 

Std. Std. Error Sig.
Route N Mean Deviation Mean t (2-tailed)
QD < DP 109 0.36 4.88 0.47
20.32 0.000
QD 2 DP 105 13.33 4.43 0.43

 

One additional analysis of the impact of the presence of downstream congestion
was conducted. Table 5.8 shows the test results. When drivers could not observe the
queue at the diversion point, the diversion ratio between I-696 and I—275 was statistically
different at a 95% confidence interval (p = 0.000) with a greater diversion when the
accident occurred on I-696. However, when drivers observed the delayed queue at the
diversion point, the diversion ratio was higher for both routes, and the difference between

the two routes was not statistically significant with a 95% confidence interval (p = 0.774).

Table 5. 8 Independent Sample t-Test for Route Traffic Condition Sensitivity (p
< .05)

 

 

(a) When QD < DP
Std. Std. Error Sig.
Rome N Mean Deviation Mean ‘ (2-tailed)
I-696 74 1 1.00 6.69 0.79
12.43 0.000
I-275 109 0.36 4.88 0.47

 

92

(b) When QD 2 DP

 

 

Std. Std. Error Sig.
Route N Mean Deviation Mean t (2-tailed)
L696 39 13.09 4.60 0.74
0.29 0.774
I-275 105 13.33 4.43 0.43

 

From these results, it was determined that at this site, drivers who anticipate
recurring congestion on their normal route are more likely to divert to an alternate route
when presented with CMS information than drivers who use a route that does not
experience recurring congestion. Moreover, drivers who use a route that does not
experience recurring congestion are less likely to respond to CMS information unless
they observe the congestion queue before reaching the diversion point. This is consistent
with the finding that neither route has a significant diversion in the non-peak period,
when drivers do not anticipate recurring congestion, nor the congestion queue never

reaches the diversion point.

5.4 Analysis of the CMS Effect at Site 2
5.4.1 Description of Site 2

Site 2 is located on M-lO on the border between Oakland County and Wayne
County. CMS 2 provides information to eastbound and southbound drivers prior to an
interchange which provides drivers with a choice to either stay on southeast bound M-10
or take M-39 southbound and then I-94 eastbound. M-lO provides a slightly shorter route
to downtown than does M-39, but the difference is small enough that M-39 is a

reasonable alternate route. Both routes have a 55 mph speed limit. The CMS is located on

93

SB M-lO at Mt. Vernon Road and the upstream ILD (A) is located on M-lO east of
Lahser Road. The downstream ILD is located on M-IO south of 8 Mile Road (B) and on
M-39 north of 7 Mile Road (C) as shown in Figure 5.6. There is recurring congestion on
M-39 during the AM-peak period. The congestion is severe (30 — 50 minutes) and the
delay reaches upstream to the M-39/M-10 interchange during a normal AM-peak period.

The downstream ILD on M-lO south of 8 Mile Road was not reporting data during this
study period.

      
  

 

hu.lll I?"

' l
Villa . E
~ 11 “.2

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

i
.5.
is— i- i
35 = [I
i3 4 '3. I
via 4' ‘a l
-. f ”4 a; /
-:.\i’9 ‘r 102
§ 4V8 :3 L -:
_,\4 1*
a , \J a
'22-. Q k D)"
. Loop Detectorm

 

 

 

 

 

 

Figure 5.6 Site 2 Location

Due to the malfunctioning of the downstream ILD, traffic data were not available

downstream on M— 10 EB. Therefore, only data for accidents that occurred downstream

94

on M-39 SB were collected and analyzed to determine the CMS effect at Site 2. Eight
accident cases were analyzed at this site. Table 5.9 provides information on each of the
accidents. The same analyses as those conducted at site 1 for comparing the upstream
volume and downstream volume ratio between accident and the related normal conditions

were conducted. The results are shown in the appendix.

95

 

 

 

 

 

 

 

 

 

 

mm A5 v motes: Box 6:2 w 3 mm 3-2 3528.... :9 I :o_ 32.2 w
on A5 v 00222 8:62 ooze 29.0 3 mm on: 86263. 8”: I See 52-: 6
mm mm v ooIeaE 2:62 ooze 6520 a mm on: 65262. owns I es: 538 o
mm mm A 00222 6202 63m oeeo 6% mm on: 35662 $8 I Sumo 338 m
mm A5 A 00222 6:52 66$. 235 3 mm 3.2 3528.34 omnwo I onus 32.3 4
mm mm A soda: 6262 ooze 280 a mm on: 3528.4 38 I memo 53.8 m
mm mm A ooIaE 88% .55 280 3 mm 32 26262 38 I ones 52-8 N
mm .5 A 0322 68m 6:2 e 3 mm on: .8262 38 I 88 58-8 _

:MHMWMMU 028:5 owammoz Eo?oo< mEU DEF 8mm— 080

13002 550950 . :28»qu mEU EoEoo<

 

 

 

 

 

 

 

N 35 a owes: £20 5? 22:25 EoEoe< 38:60 2.. 53.380 am 2%...

96

5.4.2 Diversion Ratio Analysis

Table 5.10 shows the upstream and downstream volume resulting from days with
the CMS display and on normal condition days. As can be seen in the table, the through
traffic volume reduction ranged from 1.47% to 9.23% when the CMS was displaying an
accident message. The average through traffic percent reduction due to the CMS
information was 5.93% with 4.55% of standard deviation. The 95% confidence interval
for the average difference is 5.41% to 6.75%. The confidence interval does not include
the value of 0 so it rejects the null hypothesis. The significance level is smaller than .05
so the reduction is statistically significantly different than “0” at a 95% level of

confidence. The statistical test result is shown in Table 5.11.

Table 5. 10 Accident and Related Normal Condition Traffic Volume at Site 2

Upstream Downstream %

Volume Volume Reduction Reduction
Volume

Case Condition

Case 1 Normal 4031 4427

Case 2 Normal 4909 5609

Case 3 Normal 3871

Case 4 Normal 4399

Case 5 Normal 4352

Case 6 Normal 2440

Case 7 Normal 2430

Case 8 Normal 3266

Normal 3712

Average

 

97

Table 5. 11 One Sample t-Test (p < .05) for Downstream Volume Reduction

 

 

 

 

Test Value = 0
95% Confidence Interval
Std- of the Difference Sig.
N Mean Deviation t (2-tailed)
Lower Bound Upper Bound
% . 295 5.93 4.55 5.41 6.45 22.37 0.000
reduction

 

5.4.3 Sensitivity Analysis at Site 2

Site 2 has recurring congestion on M-39 during the AM-peak period and the
congestion backed up through the M—39/M-10 interchange even on normal days. Also,
there was a malfunction on the M-lO downstream ILD during the research period so
traffic data were not collected downstream on M-lO. Therefore, the only sensitivity test
that could be conducted at this site was the AM-peak versus the non-peak period

difference.

5.4.3.1 Familiarity and Time Constraint Sensitivity Analysis

Five accidents occurred during the AM-peak period and three accidents occurred
during the non-peak period. Figure 5.7 shows the average through traffic percent
reduction at the downstream location. There is a higher percent reduction in the AM-peak

period (7.06%) than in the non-peak period (3.89%).

98

 

_L
O

7.06

  

 

%Reductlon
o—smmhmmumo

 

AM-Peak Non-Peak
Tirre Period

 

 

 

Figure 5.7 Average Through Traffic Percent Reduction at Site 2

From the result of independent sample t-test, the mean difference between these
two groups is statistically significant at a 95% confidence level (Table 5.12). This result

is consistent with the results from Site 1.

Table 5. 12 Independent Sample t-Test for Familiarity and Time Constraint (p
< .05)

 

 

. . Std. Std. Error Sig.
C°ndm°n N Mean Deviation Mean ‘ (2—tailed)
AM-Peak 190 7.06 4.25 0.31

6.06 0.000
Non—Peak 105 3.89 4.38 0.43

 

99

5.5 Analysis of the CMS Effect at Site 3
5.5.1 Description of Site 3

Site 3 is located on I-96 in Wayne County. 1-96 has a 6 mile long express/local
configuration in the western part of the Detroit Metropolitan area. A CMS located on EB
[-96 at Beech Daly Road provides information on downstream traffic conditions to
eastbound drivers prior to the junction where drivers select either the express or the local
lanes. Both highways have a 65 mph speed limit and are exactly the same length.
However, the drivers on the local lanes can access the surface streets via exit rampa, but
the express lanes cannot.

The upstream ILD (A) was located on I-96 approximately one quarter mile west
of Outer Drive. Two downstream ILDs located on I-96 east of Evergreen Road collected
traffic data at locations situated the same distance from the CMS, after the freeway
separates into the express lanes (B) and the local lanes (C). The CMS 27 located on the
EB 1-96 express lanes at Burt and CMS 28 located on the EB I-96 local lanes at
Evergreen provide the same downstream traffic information to drivers in advance of the
I-96/M-39 interchange. The location of the upstream ILD on I—96 covers six lanes and
both downstream ILDs cover three lanes each. Figure 5.8 shows the location of study Site
3. In a normal AM-peak period, the upstream and the downstream express lanes do not
experience recurring congestion but the downstream local lanes have severe recurring

congestion.

100

9i.
a:
w.
6

 

Loop Detector

Figure 5.8 Site 3 Location

Data on thirteen accidents were collected at Site 3. Table 5.13 shows the
information for each accident and Table 5.14 shows the upstream and downstream
volumes for both accident and normal conditions. There was construction during August
and September in 2001 and two of the thirteen accidents occurred during this
construction period. The related normal conditions for these two accident cases (case 4
and 10) were also collected during the same construction period. The upstream volume
and downstream volume ratio between accident and the related normal conditions are

also provided in the appendix.

101

 

mm

cosmowaoo oz

cosmowcoo oz

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Rom 2.55;» a _83 mm 82 2538.4 2 no: I 2 2 8-2-5 2
mm 8:658 oz cocooweoo oz Rom season a .83 mm 65.: 35284.. $2 I 32 52.8 2
mm 8:658 oz 856980 oz 852 one I 408 5-2-2 2 _
. 25225 5 mmoaxm mm ca; EoEoo<
oaom . I . - -
2 8:658 oz coon-3250 oz 5822 a soaxm mm 8-: 26262 $2 on: 5 2 8 2
5 38250 8:658 oz 2:62 65: 2.20 3 mm 82 828a. 2 ”5 I 2 5 5-8-5 a
4m ooeoweoo 8:82.60 oz Rom 36:85 2... mm 8.: Boone-a. 2 H5 I 2 U5 8-5-8 x
4m oaaomeoo 8:858 oz 35 2.20 3 mm 8.: 2628.4 35 I 35 5-2-: s
on ooeoweoo 8:858 oz 85% e838 3 mm 52 3528.4 2 5 I 2 H5 5-5-: o
.3 ooaomeoo 8:858 oz 262 2:8? a :83 mm 8.: 36262 2 “we I 2 5 Sow-8 m
: eoaoweoo oz 8:858 oz Rom mafia; a _83 mm on: 562833.. 55 I wwoo 5.8.5 o
Rom . I . - -
em ooaoweoo 8:658 oz 23:85 6% aoaxm mm 82 25262 2.5 2 .5 S 2 8 m
98...: . I . - -
em ooaomeoo 8:858 oz wees; a aoooxm mm 82 26262 2.5 o: .5 5 om 8 N
Rom . I . - -
em ooeomeoo 8:658 oz wees; a soaam mm 82 2628.... N08 N25 5 o: S 2
295m 5:680 5:580 DEF QED
2.03950 Etta:- oEm-F ownwmoz EoEoo< mEU :osmcomo mEU EoEoo< 030
35.52 8:3 .804 853 $8me . .

 

 

 

 

 

 

 

 

m 85 .a omen-.2 220 5.3 9.2.5.50 2.2.62 38:60 do 8:388: 22 2...?

102

Table 5. l4 Accident and Related Normal Condition Traffic Volume at Site 3

Upstream Downstream Expected Volume %
Downstream

Volume Volume Reduction Reduction
Volume

Case Condition
Case 1 Normal 7313 3844

Case 2 Normal 7887 4329

Case 3 Normal 10042 5837

Case 4 Normal 6387 2923

Case 5 Normal 9977 5809

Case 6 Normal 7543 4042

Case 7 Normal 7045 3619

Case 8 Normal 7735 4195

Case 9 Normal 10013 5831

Case 10 Normal 5498 2072

Case 11 Normal 4625 1883

Case 12 Normal 4850 2916

Case 1 3 Normal 6824 4125

Normal 7365 3956

Average

 

From the table, the through traffic volume reduction ranged from 4.58% to

7.67% during the time the CMS was displaying an accident message.

103

5.5.2 Diversion Ratio Analysis

From the data provided in Table 5.15, the mean through traffic percent reduction
due to the CMS information was 0.54% with 4.04% of standard deviation. The 95%
confidence interval for the average difference is 0.14% to 0.94%. The confidence interval
does not include the value of 0. The significance level (p=0.008) is smaller than .05
leading to the rejection of the null hypothesis. Therefore, the result indicated that the
downstream through traffic volume reduction was significantly different from 0 at a 95%
confidence level. However, this diversion was very small compared to the previous two

sites.

Table 5. 15 One Sample t-Test (p < .05) for Downstream Volume Reduction

 

 

 

 

Test Value = 0
95% Confidence Interval
Std- of the Difference Sig.
N Mean Deviation ‘ (2-tailed)
Lower Bound Upper Bound
%. 392 0.54 4.04 0.14 0.94 2.65 0.008
reduction

 

Three of the accidents studied occurred in the peak period on the express lanes,
and the CMS provided the accident information to the upstream drivers (cases 1, 2 and 3).
In these three cases, instead of drivers diverting to the local lanes as might have been
expected, more drivers used the express lanes than on normal days. The percent of
through traffic on the express lanes increased (case 1 = 4.58%, case 2 = 2.14%, and case
3 = 3.68%) compared with the same time period on normal days. Figure 5.9 provides the

downstream speed profiles for one of these accident days (case 3).

104

As can be seen in Figure 5.9, even though there is an accident on the express lanes,
there was no congestion or speed reduction on these lanes. However, even though the
accident occurred on the express lanes, congestion on the local lanes started earlier than
under normal conditions. The other two accidents cases where the volume increased on

the express lanes showed similar traffic speed characteristics.

 

CMS Activated

      

 

 

ao- :
.
704 —————————————
so:
so-
{40-
0’ 5 .
30‘ g g
C C
. .
204 : :
O .
C .
. O
104 : :
. .
C C
. C
0 j I I l I l I 1 rt 7 Tel T I fl l‘AI I l l 7 T 1 fr r r1 rt I l 7 F! T t l l 1
Gamma) 0,030,030) mma’a’mmmma’mm 0,030,030) 030103020) 0)
aeweswgtwoa98:99:99:weawngevmatwmvmavgggg
mmmmmtO‘DCOCDCDCDCDNN7‘NNNQQQQQNO’QQO’O’QOOOOOOPPPv-v-v-
v-v-v-v-v-Pv-v-v-v-v-v-
Tlme

 

 

 

[ —o— Nor_Local + Nor_Exp +031802_Loca| —o— 031802_Exp]

 

 

Figure 5. 9 Downstream Speed Profile between Accident and Normal Conditions
(Case 3: Accident on [-96 EB Express after Greenfield Road 07:10 — 08:10, 03-18-02)

Based on the results of these three accident cases, it is clear that even though the
CMS provides accident information on their preferred route, drivers tend to ignore the
CMS information when the alternate route is known to experience recurring congestion,

even an normal days.

105

The mean reduction for accidents occurring on the local lanes in the AM-peak
period or on the express lanes in the non-peak period is shown in Table 5.16. In both of
these conditions, the alternate path would not be experiencing recurring congestion. The
mean through traffic reduction was 1.95% for these 10 accidents. This reduction is

statistically significantly different than zero, as shown in Table 5.16.

Table 5. 16 One Sample t-Test (p< .05) for Downstream Volume Reduction without
Case 1, 2 and 3

 

 

 

 

Test Value = 0
95% Confidence Interval
Std- of the Difference Sig.
N Mean Deviation t (2-tailed)
Lower Bound Upper Bound
% . 290 1.95 3.40 1.56 2.34 9.78 0.000
reductron

 

5.5.3 Sensitivity Analysis at Site 3
5.5.3.1 Familiarity and Time Constraint Sensitivity Analysis

As was done at the previous sites, the AM-peak period and the non-peak period
diversion ratios were compared. Table 5.17 (a) shows that the mean reduction due to the
CMS information was - 0.11% for the AM-peak period and 2.23% for the non-peak

period. These differences are statistically significant at a 95% confidence level.

However, this result included the three accident cases which occurred on the
express lane in the AM peak period (case 1, 2 and 3). Table 5.17 (b) shows the
comparison of the diversion ratio without these three cases. The results show that there

was more diversion during the non-peak period (2.23%) than during the AM-peak period

106

(1.78%). However, this difference is not statistically significant at a 95% level of

confidence. The diversion ratio sensitivity based on familiarin and time at site 3 was not

consistent with the results of the previous two sites. However, this is at least partially

explained by the fact that drivers on the local lanes may plan to leave I-96 prior to the

point when the local and express lanes merge.

Table 5. 17 Independent Sample t-Test for Familiarity and Time Constraint (p

 

 

 

 

 

< .05)
(a) All Accident Cases
. Std. Std. Error Sig.
comm“ N Mean Deviation Mean ‘ (2-tailed)
AM-Peak 283 -0.1 l 4.04 0.24
‘ 5.61 0.000
Non-Peak 109 2.23 3.55 0.34
(b) All Accident Cases exclude case 1, 2, and 3
. . Std. Std. Error Sig.
Condrtron N Mean Deviation Mean t (2-tailed)
AM-Peak 181 1.78 3.30 0.25
1.07 0.284
Non-Peak 109 2.23 3.55 0.34

 

5.5.3.2 Visual Sensitivity Analysis

Visual sensitivity was not tested at Site 3 because of lack of comparable cases.

The local lanes were always congested in the peak period, and never congested in the

non-peak period.

107

5.5.3.3 Traffic Condition Sensitivity Analysis

As mentioned before, the downstream local lanes have recurring congestion
during the AM-peak period but the express lanes do not. As expected, the percent
reduction on the congested lanes and the non congestion lanes was significantly different _
at a 95% confidence level as shown in Table 5.18. This is consistent with the finding at

site 2.

Table 5. 18 Independent Sample t-Test for Traffic Condition (p < .05).

 

 

. . Std. Std. Error Sig.
comm“ N Mean Deviation Mean ‘ (2-tailed)
Express 102 -3.47 2.89 0.29
15.63 0.000
Local 45 4.54 2.78 0.42

 

5.6 Analysis of the CMS Effect at Site 4
5.6.1 Description of Site 4

Site 4 is located on M-39 in Wayne County. This is the closest site to the
downtown Detroit area. This site does not have an alternate freeway route, therefore
drivers can only divert to surface streets via an exit ramp in response to the CMS
information. The CMS is located on SB M-39 at Chicago Road and provides information
on downstream traffic conditions to southbound drivers. The upstream ILD (A) is located
on M-39 north of Wadsworth Road which is 0.3 miles before drivers receive information
from the CMS. The downstream ILD (B) is located on M-39 north of Cathedral Road

which is 1.2 miles south of the upstream ILD. The upstream ILD is located immediately

108

 

downstream from the I-96/M-39 interchange. The entering ramp traffic from eastbound I-
96 to southbound M-39 at the interchange is merging with the southbound M-39 traffic at
this location. This merge results in severe congestion during the AM-peak period, while
the downstream location has less congestion. Figure 5.10 shows the geographic location

of Site 4.

Loop Detector

 

Figure 5.10 Site 4 Location

Data from eight accident cases were collected and analyzed to determine the
effectiveness of the CMS at this site. Table 5.19 provides information on each accident.
Table 5.20 shows the upstream and downstream volume under accident conditions and

for the same time period under normal conditions. The upstream volume and downstream

109

 

volume ratio distribution between accident and the related normal conditions are shown

in the appendix.

 

 

110

 

 

 

 

 

 

 

 

 

 

NV 003030 82 “50% Bow 5 mm am-E E0203. mmumfi .- mm”: 8.8-2 m
NV 032030 82 03m 0003050 5 mm 3-2 E0200< 3”: 1 _NHS 5.8-: n
we 00:02.0 82 0::0>< :05? 5 mm 3.2 50203. oz: .- ciao 5.3-mo 0
me 003030 0==0>< c0253 2025 mm 3.2 E0203. v68 1 vmuoo SIB-mo m
mv 602030 50m Eon..— 5 mm 3-2 50200.4. emfio 1 ombc 5.5-ac v
mv 09.030 03:03.. :053 5 mm 3-2 E0200< 35o I Sumo SIQ-ao m
mv 003030 0==0>< 55? 2025 mm 3.2 E0200< mmnoo I mmnwo 5.3-S N
me 03.2030 0==0>< :25? 2025 mm 3-2 E0203. 35o 1 3qu S; We 2

295m

:ofiwcou >200 05802 50203. 920 08F 059 050
5:202 cosm0w=oo 2.25.0.5 mEU E0200<

 

 

 

 

 

 

.. 35 .o 8882 220 5? 83.580 0.5283. 808:8 2o counts-on 2.0 asap

1]]

 

Table 5. 20 Accident and Related Normal Condition Traffic Volume at Site 4

Expected
Downstream
Volume

Volume %
Reduction Reduction

- - U stream D w
Case Condrtron p 0 nstream
Volume Volume

Case 1 Normal 3784 5979

Case 2 Normal 3805 5943

Case 3 Normal 3233 5268

Case 4 Normal 4750 7371

4
Case 5 Normal 243 6947

 

Case 6 Normal 2700 4369 ;.

Case 7 Normal 4341

Case 8 Normal 4602

Normal 5603

Average

 

5.6.2 Diversion Ratio Analysis
When the CMS provided accident information, an average of 2.72% of the drivers
changed their route at Site 4. This diversion ratio is statistically different from “0” within

a 95% confidence interval. Table 5.21 shows the test results.

112

 

Table 5. 21 One Sample t-Test (p < .05) for Downstream Volume Reduction

 

Test Value = 0

 

 

 

95% Confidence Interval
Std- of the Difference Sig.
N Mea“ Deviation ‘ (2-tailed)
Lower Bound Upper Bound
%. 341 2.72 4.60 2.23 3.21 10.90 0.000
reductron

 

5.6.3 Sensitivity Analysis at Site 4

5.6.3.1 Familiarity and Time Constraint Sensitivity Analysis

 

As before, the data was classified into two groups to determine the familiarity and
time constraint sensitivity. Using the data in Table 5.22, the average through traffic
reduction in the AM-peak period is almost two times higher (3.31%) than in the non-peak
period (1.71%) at this site. The mean reduction is statistically significantly different with
a 95% confidence interval. This sensitivity result is consistent with the results at Site]

and Site 2.

Table 5. 22 Independent Sample t-Test for Familiarity and Time Constraint (p
< .05)

 

 

. . Std. Std. Error Sig.
comm“ N Mean Deviation Mean ‘ (2-tailed)
AM-Peak 215 3.31 4.73 0.32

3. 13 0.002
Non-Peak 126 1.71 4.21 0.37

 

 

113

The upstream location on this site has severe recurring congestion during the
normal AM-peak period. Therefore, the drivers can always observe the congested delay

at the upstream location.

5.7 Analysis of the CMS Effect at Site 5
5.7.1 Description of Site 5

Site 5 is located on 1-94 in Wayne County. This site also does not have a freeway
alternate route. The CMS is located on I-94 at 10 Mile Road and provides information on
the downstream traffic condition to the southwest bound drivers. The upstream ILD (A)
is located north of 10 Mile Road and the downstream ILD (B) is located south of 8 Mile
Road. The distance between the upstream and downstream ILD is approximately 2.1 mile.
There are surface street exits at 10 Mile, 9 Mile and 8 Mile Roads. Figure 5.11 shows the
location of Site 5.There is no recurring congestion at the upstream or downstream ILD

locations during the AM-peak period.

Highway construction was conducted during September, October, and November
2002. The construction was conducted with one driving lane closed at the downstream
location and this lane closure induced recurring congestion at the downstream location
during the AM-peak period, which was not present during the non-construction time

period. Therefore, traffic data were not collected and analyzed during this time period.

114

 

COUNTY Harper

 

Loop Detector

Figure 5.11 Site 5 Location

Thirteen accident conditions were collected and analyzed at site 5. Table 5.23
provides information on the traffic conditions for each accident, and Table 5.24 shows
the upstream and downstream volumes on accident days and normal condition days. The
upstream volume and downstream volume ratio distribution between accident and the

related normal conditions are shown in the appendix.

115

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5 8:88 oz 88% 8:80 0.. 55 5-2 2828». 52 I 52 5.2.5 2
5 ootoso .oz 2:52 2:0 2.2 a ma 5-2 2828... 55 I 52 35.5 2
5 833.8 .oz 88% M880 3 m3 5-2 286% 52 I 5.: 52-: :
5 8:38 2oz 88% 8220 3 m3 5-2 2828... 52 I 52 5-3: 2
5 8288 2oz 85:. 8580 3 55 5-2 .528... 52 I 5.2 5-2-2 5
5 8:88 “oz 2.82 8.8 oe> a ma 5-2 2528... 52 I 5;; 55.5 5
5 ootomoo “oz 882 22520 2... m3 5-2 3862. 5.2 I 52 5835 a
5 ootoaoo 88% 8:80 3 E» 55.2 E88... 55 I 555 $35 o
5 8:38 .oz 882 29520 3 $5 52 2828... 55 I 585 5.35 m.
5 88.8 .oz 2.82 .3280 a m? 5-2 .528... 58 I 5.5 35.5 a
5 88.8 oz oaom 88: so m3 5._ 0828... 82 I 8.5 32.5 m
5 8880 85>... 20520 3 m3 5-2 2828... 2.5 I 2.5 55.5 N
5 ootoso oz 882 02520 a 55 5-2 2883. 2.5 I 9.85 52.5 2

BQENW Dogm—Q DEF—r Quma
:MWWHND 2.2509.on 03502 50205. $20 2225208 £20 50203. 050

 

 

 

 

 

 

 

m 35 a 59.82 was 5...: 82.580 .523. 038:8 .o 52.585 mam 2%...

116

Table 5. 24 Accident and Related Normal Condition Traffic Volume at Site 5

Upstream Downstream Expected Volume %
Downstream

Volume Volume Reduction Reduction
Volume

Case Condition

Normal 543 1 4924
Case 1

Case 2 Normal 5223 4282

Case 3 Normal 3550 2882

Case 4 Normal 3742 3028

4
Case 5 Normal 5015 933

Case 6 Normal 5181 4954

Case 7 Normal 4638 3 180

2
Case 8 Normal 43 0 3254

Case 9 Normal 4642 3 190

Case 10 Normal 4641

Case 1 1 Normal 4540

Case 12 Normal 3859

Case 13 Normal 3891

Normal 4513

Average

 

5.7.2 Diversion Ratio Analysis

An average 0.47% of the through traffic was diverted when the accident message
was presented on the CMS (Table 5.25). On average only twenty five vehicles diverted to
exit ramps due to the message display. The mean difference is statistically significantly

different from 0 at the 95% confidence level.

117

Table 5. 25 One Sample t-Test (p < .05) for Downstream Volume Reduction

 

 

 

 

Test Value = 0
95% Confidence Interval
Std- of the Difference Sig.
N Mean Deviation ‘ (2-tai1ed)
Lower Bound Upper Bound
% . 372 0.47 3.35 0.13 0.81 2.70 0.007
reduction

 

5.7.3 Sensitivity Analysis at Site 5
5.7.3.1 Familiarity and Time Constraint Sensitivity Analysis

The data were classified into two groups and the familiarity and time constraint
sensitivity test was conducted. The AM-peak period (0.78%) had more diversion than the
non-peak period (0.09%) when the CMS message was displayed. However, the difference
in the mean diversion ratio is not statistically significantly different at a 95% confidence

level. The results of this test are given in Table 5.26.

Table 5. 26 Independent Sample t-Test for Familiarity and Time Constraint (p
< .05)

 

 

. . Std. Std. Error Sig.
C°“d‘”°“ N Mean Deviation Mean ‘ (2-tailed)
AM-Peak 204 0.78 3.77 0.26

1.97 0.05
Non-Peak 168 0.09 2.73 0.21

 

118

 

5.7.3.2 Visual Sensitivity Analysis

The sensitivity of diversion to the drivers’ ability to see the back of the congestion
queue was also tested at Site 5. For this sensitivity analysis, the data were classified into
two groups: congestion observed and congestion not observed. Among the thirteen
accident cases, two accidents had congestion where drivers encountered the traffic
congestion before passing the last downstream exit ramp before the downstream ILD.
Therefore, the downstream ratios for these two accident conditions were compared with
the others. These accidents occurred during the AM-peak period, therefore, only the AM-
peak period accidents when no congestion was observable were used as a comparison

group. The results are shown in Table 5.27.

The result is consistent with the previous results. A greater route diversion was
found when drivers observed traffic congestion and the difference is a statistically

significant at the 95% confidence level.

Table 5. 27 Independent Sample t-Test for Visual Sensitivity (p < .05)

 

 

. . Std. Std. Error Sig.
Condition N Mean Deviation Mean t (2-tailed)
Not
Observed 136 0.34 3.55 0.30
2. 37 0.019
Observed 68 l .65 4.05 0.49

 

119

 

5.8 Geographic Location Sensitivity Analysis

Figure 5.12 shows the through traffic percent reduction resulting from the CMS
display compared with same time period on normal days. As noted in the site descriptions,
Site 1 and Site 2 have freeway alternate routes, while Site 4 and Site 5 do not. When
drivers are notified of downstream congestion, the through traffic reduction ratio at Site 1
and Site 2 were 6.07% and 5.93% respectively compared to Site 4 and Site 5 reductions I
of 2.72% and 0.47% respectively. Site 3 was not included in this sensitivity analysis

based on the fact that the diversion was to a route where recurring congestion is a daily

 

 

 

 

occurrence.
10
9
8
7
C
5.3 6
§ 5
9 4
°\° 3
2
1 0.47
0 @
Site1 S1192 Sites

 

 

 

 

Figure 5.12 Average Through Traffic Percent Reduction Based on Geographic
Locations

To examine the significance of the mean reduction in the through traffic volume

between a group with an alternate route and a group without an alternate route, a t-Test

120

 

was conducted. The results are given in Table 5.28. The table shows that the mean
reduction due to CMS information was 6.02% for the drivers who have an alternate
freeway route and 1.54% for drivers with only surface street alternate routes. These

differences are statistically significant at a 95% confidence level.

Table 5. 28 Independent Sample t-Test for Geographic Location (p < .05)

 

 

 

. . Std. Std. Error Sig.
Condition N Mean Deviation Mean t (2-tailed)
With Alt 791 6.02 6.72 0.24
15.71 0.000
Without Alt 7 13 1.54 4. 15 0.16

 

121

 

CHAPTER 6

CONCLUSIONS AND RECOMMENDATIONS

6.1 Conclusions

This study was conducted to determine the effect of various parameters on the
diversion potential of CMS in southeast Michigan by analyzing traffic flow data obtained
when a message was displayed. This study used data from five CMS locations on the
Detroit freeway system which have different geographic and traffic conditions. The
measure of effectiveness (MOE) used to quantify the CMS effect is the percentage of
drivers that diverted to an alternate route when they observed a message that an accident
had occurred downstream on their current route.

The percentage of vehicles that passed a detector downstream from a diversion
opportunity after having encountered a message was compared to this same ratio on days
when no message was displayed. The difference between the number of vehicles passing
the downstream detector and the number that would have passed this detector under
normal conditions was deemed to be the effect of the CMS information. This difference
in the ratio of downstream volume to upstream volume was tested for statistical

significance.

From the statistical analysis of 60 different accident cases at five different locations,
this study determined the characteristics under which a CMS is effective in inducing
drivers to change their travel route to avoid a downstream accident. The average percent
diversion when a CMS displays a message was 3.43%. However, the diversion ratio is

not equal at all CMS sites, or at different times at any given CMS site. The average

122

 

diversion ratio ranged from a low 0.47% to a high of 6.07% across the five sites, and
from a low of -1.96% to a high of 16.44% at different times at a single site. To better
understand this variation in the diversion potential, four different factors were considered:
time of day; a driver’s ability to see the queue resulting from the accident; historical

traffic conditions on the alternate route; and geographic location.

For the sensitivity analysis based on the time of day, two different groups; the
AM-peak period (06:00 — 09:00) and a non-peak period (10:00 - 16:00) were compared.
The AM-peak period traffic has a high percentage of commuters, who are both familiar
with the alternate routes to their destination and under some pressure to reach their
destination on time. The non-peak period has a higher percentage of drivers that use this
route less frequently than commuters, and do not have the obligation to arrive at work at

a prescribed time.

From the analysis results, it was determined that drivers during the AM-peak
period are more responsive to CMS information than non-peak period drivers. The
average diversion ratio during the AM-peak period is as much as five times greater than
the Non-peak period (1.47% to 8.45%) at Site 1. Similar differences were also found at

the other sites.

To determine the sensitivity of the route diversion to conditions observed by the
drivers, a calculation of the congestion distance for each accident case was conducted,
and the crash events were categorized into two groups. The first was an accident
condition where drivers encountered the congestion upstream from the diversion point,
and the second was an accident condition where drivers could not observe the congestion

prior to the diversion point. From this analysis, it was shown that the visual observation

123

 

of congestion impacts the drivers’ route diversion behavior. When drivers observe the
congestion before reaching the diversion point, the diversion ratio is almost three times
higher at Site 1 (4.66% and 13.26%) and over four times higher at Site 5 (0.34% and
1.65%) than when the congestion can not be observed. The difference is statistically
significant at the 95% confidence level.

The willingness to divert to an alternate route is sensitive to the drivers’
perception of the likelihood of encountering recurring congestion on each route. At site 1,
where recurring congestion occurs on I-696 EB but not on I-275 SB, almost twice as
many drivers diverted when an accident occurred on I-696 EB (11.72%) than when the
accident occurred on I-275 SB (6.73%) during the AM-peak period. In fact, the
expectation of congestion has almost as much of an effect on diversion behavior as
encountering the congestion. When the accident occurred on the route with recurring
congestion (I-696) the diversion ratio is not much different between observing the queue
(13.09%) and not observing the queue (11.00%). However, when the accident occurred
on the route without recurring congestion (I-275), the diversion ratio was 13.33% when
the queue is observed prior to the diversion point, and only 0.36% when it is not observed.
This same phenomenon was observed at Sites 4 and 5. Site 4, which has recurring
congestion, had a greater diversion (3.31%) than Site 5 (0.78%), which has no recurring

congestion during the AM-peak period.

These results led to the conclusion that drivers who anticipate recurring
congestion on their normal route are more likely to divert to an alternate route when
presented with CMS information than drivers who use a route that does not experience

recurring congestion. Drivers who use a route that does not experience recurring

124

congestion are less likely to respond to CMS information unless they observe the
congestion queue. This factor may also partially explain the difference between the AM-
peak period and the non-peak period diversion ratio, because during the non-peak period

drivers would not anticipate recurring congestion on any of the alternate routes.

The final sensitivity analysis was based on the sites geographic location. A
comparison was made between sites with and without a freeway alternate route. The
average diversion ratio in Site 1 and Site 2, where there is an alternate freeway, were
6.07% and 5.93% respectively compared to Site 4 and Site 5 (where the alternate routes
are arterials) reductions of 2.72% and 0.47% respectively. The route diversion ratios due
to CMS information were almost four times higher at sites with a freeway alternate route
(6.02%) than a site with no alternate route (1.54%). The difference is statistically

significant at a 95% confidence level.

6.2 Recommendation

While the percentage of drivers diverting to an alternate route is relatively low, the
number of vehicles diverted can be fairly large: At Site 1, when an accident occurs on I-
696 in the AM-peak period, between 283 and 1382 vehicles diverted to I-275 as a result
of the CMS message. The result of this diversion reduces the network delay and improves
safety and the environmental impact of an accident in two ways. First, the diverted
vehicles avoid the congestions resulting from the accident, and second, the time in queue
for those vehicles that chose to remain on their primary route is reduced. These benefits

of a CMS system, however, are not generated equally at all locations on a freeway. The

125

 

 

 

diversion ratio is significantly different based on the existing traffic condition, the
geographic location and the drivers’ perception of the primary and alternate routes.

NCHRP Synthesis 237 published in 1997 includes a summary of State guidelines
for CMS. These guidelines include visibility and readability distances, lateral and vertical
placement, placement relative to the closest interchange and the message design. This
report concludes that research is necessary to gain a better understanding of CMS
potential includes:

“Additional field studies to evaluate message effectiveness in term of motorist
response would be useful. The number of documented studies that measured motorist
response to CMS messages in real -world operational settings is extremely small and most
were conducted -in the mid I970s. ”

The existing guidelines for the placement of CMS are based on visibility and the
time to understand the messages displayed. However, they do not consider the cost
effectiveness of CMS. The capital cost (without installation) for a full matrix, LED, 3-
1ine walk-in freeway CMS is $48,000 — $120,000, and the operating cost is $2,400 -
$6,000 per year (http://www.benefitcost.its.dot.gfiofl. Therefore, identifying the
parameters which influences diversion, and thus the fuel and travel time savings, is
important to establish the optimum deployment of a CMS system.

This analysis, based on field observation of drivers’ behavior, has contributed to
the identification of factors to be considered in determining the potential cost

effectiveness of prospective CMS locations.

126

 

 

APPENDIX

127

 

 

 

 

 

6000-
‘ o
5000—
‘ o
0
3
>
I 4000-
a.
D
3000+ g
2000-
r I I I I I r
2 3 4 5 6 7 8

Group

Figure A. 1 Upstream Volume Distribution at Site 2

128

Condition

0 Normal
* Accident

 

 

N 3% .0 8823800 0:5— 0:53» 50052525 «é. 0.5ME

coEucoo

«co—“good. _uctoZ
_ _

 

 

 

 

 

 

 

 

E II 0.0
IIo.o_.
lodN

III I 0.0m
v ”ammo
coEvcoO
«c00_uo< _chOZ
_ _
II 0.0
. Tod _.
_mm.
1.0.0N
110.00

 

 

 

 

N “OWNO

onaa'IOA ma

omu'IOA’umoo

COED—.00
«co—2004 _nctoZ
_ _

 

 

 

 

 

 

 

 

 

 

 

.Io.o
.EI . a
.8. 10 o. _m
_m
o Iodu W.
o
1000
m ”mono
coEucoo
acoEuud‘ _quoZ
_ _
E 10.0
.131. . a
.8”. IS: _m
m
o 10.00 .8
WI.
0
III Iodn
_. Homno

 

129

 

 

 

€255.88 N 35 «a m:8taQEcU 8.3— 0533» 88.52th ~.< 95w?—

cogucoo

acoEoo< _uEhoZ
_ _

 

 

0
0

.mT “En

 

 

lodn
Iodv G
0
Fodm _w
10.8 m
IQ? _U
lodm o
rno.8

 

 

m ”930

55.230

«tongue/x _ochZ
_ _

 

 

lodn
lodv O
lodm .m
Iodm m
[0.2 .8
Iodm o
Iodm

 

 

 

 

 

 

 

 

 

 

coEucoO
«cou_uu< _nctoZ
_ _
10.8
16.3 0
.EI lodm.
lodm m
. 10.2 .8
m.
lodm o
rno.8
N. “Omflo
cozficoo
EOEUU< _uctoZ
_ _
E lo.o
m
[0.3 _m
A
_ol
m lodu 8
m.
O
[0.8

 

 

 

 

m ”ammo

130

Up_Vo|

 

11000-

10000-

9000-

8000‘

7000 -

6000-1

5000‘

4000‘

3000-

 

 

 

Group

Figure A. 3 Upstream Volume Distribution at Site 3

131

Condition

 

 

 

m gm 3 macaw—2530 83: 2:33» E35325 vé. 0.53,.—

COEDCOO

acoEuoxx _lctoZ
_ _

 

 

.mT Q.

 

I
o.
o
V

I I
o. o.
O O
m
oneu'IOA ma

0

 

I
O.
O
[x

 

 

v “mono

COEUCOO
#5203». .2502
_ _

 

 

IE! :8"

I
o_
o
V

I I
O O
o' o'
to
0993 IOA'UMOG

0

 

l
q
0
[x

 

 

 

COED—.00

«:23qu .9502
_ _

 

 

 

 

I
O.
o
V

ID
0993 IOA'umoa

I I
O. 0.
O O
(D

 

Iodh

 

m “mono

coaucoo

«£02091 .9:qu
_ _

 

 

 

 

 

I
q
o
V

In
0993 IOA'UMOCI

I I
O. O.
O O
(D

 

I
q
0
[x

 

132

 

€025:ch n 03m 3 2.8.3an0 .55— 052.5 53.32.ch v.< 0.53%

55.800

acoEuud‘ _IEEZ
_ _

 

 

 

cog—5:00
“co—3001‘ _choZ
_ _

I
o_
o
V

0993 IOA‘uma

I I
O. O.
O O

U)

l
O.
O
h

 

 

 

m ”Gmflo

Iodv

In
0923 IOA-UMOG

I I
o, 0.
O O
(D

 

I
O.
O
l\

coEucoo

EoEuu< .2502
_ _

 

 

.mT B”

 

 

I
O.
0
V

ID

0993 IOA'UMOII

I I
O. 0.
O O
0

I
q
C
[x

 

 

N ”0300

COED—.00
acoEuu< _ugoZ
_ _

 

 

IodV

T I

o. o.

o o

co In
onvu‘IOA'umoa

 

I
o.
o
Is

133

 

Gas—550V m 3mm 3 magtaafioo :53— 25.?» 58.32.35 wé 953m

coEvcoo

acoEuox.‘ _quoZ
_ _

 

 

.9. mm?

 

NF ”owflo

COEVCOO
uc0200< _choZ
_ _

I 7
0. 0.
O O
V (‘0

I I
o, c.
O O
(D ID
09sz |°A “M00

I
9
O
[\

 

 

 

 

.g. a,

 

O F ”Omflo

 

COEUCOO

Eafluud‘ _nctoz
_ _

 

 

.mT WM.

0

 

I. no. «no

coEvcoO
50200.4 _oFcOZ
_ _

I I I I
0. O. o, O.
O O O O

ID V (0
0938 |O/\ UMOQ

(O

 

I
o.
o
[s

 

 

 

Iodn

I
q
o
to v
0998 |°/\ “M00

I I
0. o,
O O
(O

 

I
0.
o
h

134

Goa—5.88 m 35 an Eamtaafiau .53— 2:23, Sacha—Eon v.< 953m

5:650

“co—2004‘ _ufltoZ
_ _

 

.9. my

 

 

mwuolmdvaII

I I I I
0. 0. O. O.
O O O D

ID V (0
0998 |°A UMOIJ

(D

 

I
q
C
[s

135

 

 

 

 

O
5000-4 8
g o
3 4000- fi' S
>I *
3'
, 8 a
sooo- g
2000—
I I I I I I r
1 2 3 4 5 e 7 a
Group

Figure A. 5 Upstream Volume Distribution at Site 4

136

Condition

0 Normal
* Accident

 

 

v mam «a agar—«950 .55— oE=_c> 83.52.ch e.< 23E

55.280

Hem—200.1 _ucEOZ
_ _

 

 

.Qi -._.

 

 

 

I

. q
o

ID V

I I I
O O O
O O O

to
0993 IOA‘UMOCI

 

 

 

 

 

 

 

 

. n
Iodm
.30qu
32.280
acoEooxx _octoZ
_ _
Iodv
Iodm O
.a. W... _m
fin. Io om M
_ua
Iodn m.
o
Iodm
N ”omuO

coEuCOO
E02004. .uctoZ
_ _

 

 

¢ a
Q

m HomuO

 

coEvcoo

«coEuu/x _uctoZ
_ _

I I I I
0. O. 0. 0.
O O O O

(D ID V
0998 |°/\ IMO

I»

 

I
o.
o
co

 

 

 

 

 

IodV

I
q
0
ID

Iodw

I
o.
O
[s

 

I
0.
o
co

onea'IOA'UMOCI

137

€255.89 9 8mm 3 maomtanficu 9:5— 2:2?» 58.52an e.< 9.53m

COEDCOO

acoEou/x _urchoZ
_ _

 

 

 

w
m ”mono
coEvcoO
acoEooxx _nEboZ

_ _

 

 

 

 

 

 

fir mm.

 

 

 

 

coaucoo
acoEoo< _nctoZ
_ _
IO.Om IO.Om
lode Wu lodm
Iodh _ E Iodh
oA
Iodm .l [0.8
“a
10.8 m. Iodm
lode lode
hummuo
com—5:00
unoEou< _nctoZ
_ _
Iodm Iodv
lodx. _ AG! -
oA Iodm
lodm _I
M. 0.9.
10.8 m.
V0.03 Todm

 

 

 

 

 

 

 

m HomoO

onaa‘IOA won

0993 IOA'UMOG

138

3000 '-

 

 

 

 

-
—
—
—I
-
-
(04
_
q

Group

Figure A. 7 Upstream Volume Distribution at Site 5

139

Condition

0 Normal
* Accident

 

m 8mm 8 3823590 83% oE=_c> Samba—Ea: mi... 253%

5:650

«£02091 .5502
_ _

 

 

.9. a.

 

v “owuo

COEUCOO
50300.1 _quoZ
_ _

Iodb

Todm

I
o.
o
a)

I
o.
o
o
P

 

Io.o:.

 

 

.mT .8...

 

N Homoo

I I I
O. c. 0.
O O O
O) m l\

I
o.
o
o
P

 

Tod:

Onta'IOA'umoo

onvu‘IOA'umoo

con—5:00
#52091 .3502
_ _

 

 

 

 

 

 

 

 

10.?
..El a [0.8
Iodm
lode
1.0.0 F F
m HomuO
coancoo
“co—good. .5502
_ _ Iodn
lodm
umT fl lodm
[0.8 P
lodw F
_. ”omuo

OHBH'IM woo

oneu'IOA'umoa

140

Goa—550v m 35 3 3823—50 .53— 2:23» 58.52.ch m4 0.53%

 

 

 

 

 

 

 

COED—50
EOEUO< _uctoZ
_ _

Iodm
O
Feds _m
.... a. W.
. iodm U
m.
0

[0.8

m ownO
35.800
«502004 .2502
_ _
199.
G
lo.om

_m
IEI Io om M

“E“ _
. ua
lode m.
0

led:

 

 

COEUCOO
«£02004 .nFcoZ
_ F

 

 

 

I
O.
o
to

onva'IOA'umoo

I I
o, o,
O O
(D l‘

 

I
Q
o
m

 

a
h ”Odo
cofivcoo
502006. .9502

_ _

 

 

 

m:- ”QMNO

I
q
0
[x

I I
o o.
O O
O) m
oneu'IOA'umoo

I
0.
o
o
P

Io.o_.r

 

141

GEES—SUV m 8% 3 3823500 .53— 2:23» Egamcsoa w.< 95$..—

coaucoo

#52006. .9502
_ _

 

 

.m. E

 

NP umwflo

EOE—2.50
«coEuoxx _uctoZ
_ _

 

 

 

.a. a.

&

 

0.. “mono

 

lode
O
[0.3 _m
_m.
[0.8 8
M"
o
Todm
Iodm
_H
o.o\. M
r
.A
. ol
lo cm .8
m1.
0
Iodm

5.5800

«£02001». _oFtOZ
_ _

 

 

s. 5..

 

PF ”QWMO

confined

acoEou< _ufitoz
_ _

 

 

 

Iodm
O
[0.2. .m
m
Iodm .8
m"
o
10.8
Iodm
0
I92. _m
m
Iodm .8
W.
o
Iodm

 

 

 

142

€255.38 m 8% «a muomtanfiou 33y— oE=_o> 53583.5 wé. 95»...—

5.3.230

502004. .9502
_ _

 

IEI “fin

 

 

m _. HomuO

I I I
0. O. O.
O D O
m l\ (D

I
0.
o
m

 

onaa‘IOA'umoa

143

REFERENCES

144

.W

 

REFERENCES

l. Conrad L. Dudek and Carroll J. Messer, “Study of Design Considerations for
Real-time Freeway Information Systems”, Highway Research Record 363,
Highway Research Board, Washington, DC, 1971, pp. 1-17.

2. Kenneth W. Heathington, Richard D. Worrall, and Gerald C. Hoff, “Attitudes and
Behavior of Drivers Regarding Route Diversion”, Highway Research Record 363,
Highway Research Board, Washington, DC, 1971, pp. 18-23.

3. R. Dale Huchingson and Conrad L. Dudek, “Delay, Time Saved, and Travel Time
Information for Freeway Traffic management,” Transportation Research Record
722, Transportation Research Board, Washington, DC, 1979, pp. 36-39.

4. R. Dale Huchingson, John R. Whaley, and Nada D. Huddleston, “Delay Message
and Delay Tolerance at Houston Work Zones,” Transportation Research Record
957, Transportation Research Board, Washington, DC, 1984, pp. 19-21.

5. Hani S. Mahmassani, Christopher G. Caplice, and C. Michael Walton,
“Characteristics of Urban Commuter Behavior: Switching Propensity and Use of
Information”, Transportation Research Record 1285, Transportation Research
Board, Washington, DC, 1990, pp. 57-69.

6. Edward Daniels, Moshe Levin and Joseph M. McDermott, “Improving
Commercial Radio Traffic Reports in the Chicago Area,” Transportation Research
Record 600, Transportation Research Board, Washington DC, 1976, pp. 52-57.

7. R. Dale Huchingson, R.W. McNees, and Conrad L. Dudek, “Survey of Motorist
Route-Selection Criteri a,” Transportation Research Record 643, Transportation
Research Board, Washington, DC, 1977, pp. 45-48.

8. Stephanedes, Y., E. Kwon, and P. Michalopoulos, “Demand Diversion for
Vehicle Guidance, Simulation, and Control in Freeway Corridors,” Transportation
Research Record 1120, Transportation Research Board, Washington, DC, 1988,
pp. 12—20.

9. Mannering, F., “Poisson Analysis of Commuter Flexibility in Changing Routes
and Departure Times,” Transportation Research, Vol. 23 B, No. 1. 1989, pp. 53-
60.

10. Decek, C., C. Messer, and H. Jones., “Study of Design Considerations for Real-

Time Freeway Information Systems,” Highway Research Record 363, Highway
Record Board, Washington, DC, 1971.

145

ll.

12.

13.

14.

15.

16.

17.

18.

19.

20.

Elham Shirazi, Stuart Anderson, and John Stesney, “Commuter’s Attitudes
Toward Traffic Information Systems and Route Diversion,” Transportation
Research Record 1168, Transportation Research Board, Washington DC, 1988,
pp. 9-15.

Roper, D., R. Zimowski, and A. Iwamasa, “Diversion of Freeway Traffic in Los
An geles: It Worked,” Transportation Research Record 975, Transportation
Research Board, Washington, DC, 1984.

Asad, J. Khattak, Joseph L. Schofer, and Frank S. Koppelman, “Factors
Influencing Commuters’ En Route Diversion Behavior in Response to Delay”,
Transportation Research Record 1318, Transportation Research Board,
Washington, DC, 1991, pp. 125-136.

Amalia Polydoropoulou, Moshe Ben-Akiva, and Isam Kaysi, “Influence of
Traffic Information on Drivers’ Route Choice Behavior”, Transportation
Research Record 1453, Transportation Research Board, Washington, DC, 1994,
pp. 56-65.

Asad Khattak, Adib Kanafani, and Emmanuel le Collectter, “Stated and Reported
Route Diversion Behavior: Implications of Benefits of Advanced Traveler
Information System,” Transportation Research Record 1464, Transportation
Research Board, Washington, DC, 1994, pp. 28-35.

Samer M. Madanat, C. Y. David Yang, and Ying-Ming Yen, “Analysis of Stated
Route Diversion Intentions Under Advanced Traveler Information Systems Using
Latent Variable Modeling,” Transportation Research Record 1485, Transportation
Research Board, Washington, DC, 1995, pp. 148-154.

Mohamed A. Abdel-Aty, Ryuichi Kitamura, and Paul P. Jovanis, “Exploring
Route Choice Behavior using Geographic Information System-Based Alternative
Routes and Hypothetical Travel Time Information Input,” Transportation
Research Record 1493, Transportation Research Board, Washington, DC, 1995,
pp. 74-80.

Asad Khattak, Amalia Polydoropoulou, and Moshe Ben-Akiva, “Modeling
Revealed and Stated Pre-trip Travel Response to Advanced Traveler Information
Systems,” Transportation Research Record 1537, Transportation Research Board,
Washington, DC, 1996, pp. 46-54.

Brien G. Bension, “Motorist Attitudes about Content of Variable-Message Signs,”
Transportation Research Record 1550, Transportation Research Board,
Washington, DC, 1996, pp. 48-57.

Youngbin Kim and Jean-Luc anace, “Like-Flow Evaluation Using Loop
Detector Data: Traveler Response to Variable-Message Signs”, Transportation

146

21.

22.

23.

24.

25.

26.

27.

28.

29.

30.

Research Record 1550, Transportation Research Board, Washington, DC, 1996,
pp. 58-64.

Aemal J. Khattak and Asad J. Khattak, “Comparative Analysis of Spatial
Knowledge and En-Route Diversion Behavior in Chicago and San Francisco:
Implications for Advanced Traveler Information Systems”, Transportation
Research Record 1621, Transportation Research Board, Washington, DC, 1998,
pp. 27-35.

Ronald K00 and Youngbin Yim, “Commuter Response to Traffic Information on
an Incident,” Transportation Research Record 1621, Transportation Research
Board, Washington, DC, 1998, pp. 36-42.

Srinivas Peeta, Jorge L. Ramos, and Raghubhushan Pasupathy, “Content of
Variable Message Signs and On-Line Driver Behavior,” Transportation Research
Record 1725, Transportation Research Board, Washington, DC, 2000, pp. 102-
108.

Srinivas Peeta and Shyam Gedela, “Real-Time Variable Message Si gn-Based
Route Guidance Consistent with Driver Behavior”, Transportation Research
Record 1752, Transportation Research Board, Washington, DC, 2001, pp. 117-
125.

Fred Hall, Sarah Wakefield, and Ahmed Al-Kaisy, “Freeway Quality of Service;
What Really Matters to Drivers and Passengers?” Transportation Research
Record 1776, Transportation Research Board, Washington, DC, 2001, pp. 17-23.

Srinivas Peeta, Kamalakar Poonuru, and Kumares Sinha, “Evaluation of Mobility
Impacts of Advanced Information Systems,” Journal of Transportation
Engineering, June 2000.

Jeffrey L. Alder, Wilfred W. Recker and Michael G. McN ally, “Using Interactive
Simulation to Model Driver Behavior under ATIS,” University of California,
Irvine, 1992.

Conrad L. Dudek, R. Dale Huchingson, and R. Quinn Brackett, “Studies of
Highway Advisory Radio Messages for Route Diversion,” Transportation
Research Record 904, Transportation Research Board, Washington, DC, 1983,
pp. 4-9.

Conrad L. Dudek. “Human Factors Considerations for In-Vehicle Route
Guidance,” Transportation Research Record 737, Transportation Research Board,
Washington DC, 1979, pp. 104-107.

Conrad L. Dudek, William R. Stockton, and Donard R. Hatcher, “Real-Time
Freeway-to—Freeway Diversion: The San Antonio Experience,” Transportation

147

31.

32.

33.

34.

35.

36.

37.

38.

39.

40.

41.

42.

Research Record 841, Transportation Research Board, Washington, DC, 1982,
pp. 1-14.

D.H. Roper, R.F. Zimowski, and A.M. Iwamasa, “Diversion of Freeway Traffic in
Los An geles: It Worked,” Transportation Research Record 957, Transportation
Research Board, Washington, DC, 1984, pp. 1-4.

Conrad L. Dudek, Graeme D. Weaver, Donald R. Hatcher, and Stephen H.
Richards, “Field Evaluation of Message for Real-Time Diversion of Freeway
Traffic for Special Events,” Transportation Research Record 682, Transportation
Research Board, Washington, DC, 1978, pp. 3745.

Stephen H. Richards, William R. Stockton, and Conrad L. Dudek, “Analysis of
Driver Responses to Point Diversion for Special Events,” Transportation Research
Record 682, Transportation Research Board, Washington, DC, 1978, pp. 46-52.

J. Michael Turner, Conrad L. Dudek and James D. Carvell, “Real-Time
Diversion of Freeway Traffic During Maintenance Operations,” Transportation
Research Record 683, Transportation Research Board, Washington, DC, 1978,
pp. 8-10.

Development Traveler Information Systems Using the National ITS Architecture.
Report FHWA-JPO-98-O31. FHW A, US. Department of Transportation, 1998.

David Schrank, and Tim Romax, The 2003 Annual Urban Mobility Report, Texas
Transportation Institute, The Texas A&M University Systems, 2003

Raktim Pal, Investigation on Latent Factors Affecting Route Diversion Intentions,
Journal of Transportation Engineering, J uly/August 1998

Michael J. Wenger, Jan H. Spyridakis, Mark P. Haselkom, Woodrow Barfield,
and Loveday Conquest. Motorist Behavior and the Design of Motorist
Information Systems, Transportation Research Record 1281, Transportation
Research Board, Washington, DC, 1990, pp159-167. '

C.L Dudek. Guidelines on the Use of Changeable Message Signs. Report No.
FHWA-TS-90—043, May 1991.

CL Dudek. Changeable Message Signs. NCHRP Synthesis 237. National
Cooperative Highway Research Program, Transportation Research Board.

Washington DC. 1997.

Robert B. Schiesel and Michael J. Demetsky. Evaluation of Traveler Diversion
Due to En-Route Information. Report No. UVA/29472/CE00/103, May 2000.

http://www.benefitcost.its.dot. gov

148

IIIIIIIIIIIIIIIIIIII