MD-13-SP109B4C

Martin O’Malley, Governor Darrell B. Mobley, Acting Secretary

Anthony G. Brown, Lt. Governor Melinda B. Peters, Administrator

STATE HIGHWAY ADMINISTRATION

RESEARCH REPORT

EVALUATION OF DYNAMIC MESSAGE SIGNS AND

THEIR POTENTIAL IMPACT ON TRAFFIC FLOW

ALI HAGHANI
MASOUD HAMEDI
ROBIN L. FISH
AZADEH NOURUZI

UNIVERSITY OF MARYLAND
COLLEGE PARK

Project Number SP109B4C

FINAL REPORT

April 2013
The contents of this report reflect the views of the author who is responsible for the facts and the
accuracy of the data presented herein. The contents do not necessarily reflect the official views or
policies of the Maryland State Highway Administration. This report does not constitute a standard,
specification, or regulation.
 Technical Report Documentation Page
1. Report No. 2. Government Accession No. 3. Recipient's Catalog No.
MD-13-SP109B4C
4. Title and Subtitle 5. Report Date
Evaluation Of Dynamic Message Signs And Their Potential Impact On
Traffic Flow 6. Performing Organization Code

7. Author/s 8. Performing Organization Report No.

Ali Haghani, Masoud Hamedi, Robin Fish, Azadeh Nouruzi
9. Performing Organization Name and Address 10. Work Unit No. (TRAIS)
University of Maryland
1179 Glenn L. Martin Hall 11. Contract or Grant No.
College Park, MD 20742 SP109B4C
12. Sponsoring Organization Name and Address 13. Type of Report and Period Covered
Maryland State Highway Administration Final Report
Office of Policy & Research 14. Sponsoring Agency Code
707 North Calvert Street (7120) STMD - MDOT/SHA
Baltimore MD 21202

15. Supplementary Notes

16. Abstract

Maryland State Highway Administration (SHA) has a rich data archive of the messages posted to the
Dynamic Message Signs (DMS) and the time stamps when they were posted and taken down. The archive
also contains traffic information surrounding the DMS signs, such as traffic volumes and speeds from
various point or probe-based sensors. In this project, the research team used this data archive to study the
impact of DMS messages on traffic conditions. Multiple Bluetooth sensors were deployed on a major
travel corridor in the Baltimore Metropolitan Area to determine vehicle travel times and to monitor traffic
diversions.

17. Key Words 18. Distribution Statement: No restrictions

Dynamic Message Sign, Safety, This document is available from the Research Division upon
Accident request.
19. Security Classification (of this report) 20. Security Classification (of this page) 21. No. Of Pages 22. Price
None None

Form DOT F 1700.7 (8-72) Reproduction of form and completed page is authorized.
Executive Summary

The need to convey accurate travel information to motorists has become increasingly important
with the increase in traffic volume and the lack of additional roadway capacity. Knowledge of
rapidly changing traffic conditions gives drivers the option to modify their behavior in order to
avoid delays and dangerous situations. Highway Dynamic Message Signs (DMS) are often
referred to as the most visible form of ITS technology. Installed in conjunction with other
technologies of an Advanced Traveler Information System (ATIS), they enhance drivers’
knowledge of the highway network. In Maryland, there are more than 80 DMS installed on
major Interstates, highways and arterial roads.

The Maryland State Highway Administration’s Coordinated Highways Action Response Team
(CHART) routinely posts messages on both portable and fixed DMS. While most agree that
DMS are a valuable way to reach motorists and convey important information, there has long
been speculation that DMS messages may adversely affect traffic conditions. Recent publicity
surrounding the new travel time messages on DMS have rekindled this debate. The question
remains: Will a message posted onto a DMS adversely affect traffic? If so, do all types and
lengths of messages have this potential, or do only certain types and lengths of messages pose a
threat? This study attempted to answer this important question.

Another important measure of the value of a DMS message is its credibility. It is vital that
travelers believe messages displayed on a DMS are factual and accurately describe roadway
conditions. Without consistently valid information, road users will begin to ignore DMS
messages. In the case of travel-delay messages, phrases such as “Major Delays,” “Heavy
Delays,” and “Expect Congestion” have been used to describe traffic conditions. The most recent
trend has been to post messages containing travel-time estimates. In order to determine the
accuracy of such messages, this study examined the traffic conditions under which they were
displayed. Specifically, Bluetooth travel-time and route-diversion data were collected for the
analysis.

This project took advantage of a data analysis framework equipped with a database capable of
importing data from CHART, RITIS and Bluetooth detectors and employing data-mining
techniques to perform before-and-after analysis of traffic conditions with respect to a message
display. In addition to traffic pattern analysis, the system can visualize data from several sources
in accordance with message timelines.

This study also investigated claims that DMS messages have the potential to cause congestion
and safety risks. The Remote Traffic Microwave Sensor (RTMS) speed data and logs of DMS
messages were collected and analyzed. Messages were categorized into three types based on the
ideas proposed by Ridgeway (2003): Danger/Warning (Type 1), Informative/Common Road
Conditions (Type 2) and Regulatory/Non-Traffic-Related (Type 3). The primary analysis
consisted of examining the effect of message display (off-on), removal (on-off), and switching
i
(between any two types of messages) on traffic speeds over two consecutive five-minute periods.
In all three cases, the speeds in the first five minutes were compared to the speeds in the
following five minutes via paired t-tests. Overall, 2,268 cases were identified and examined. The
results showed that when messages were displayed (off-on), users slowed down most often in
response to Type 1 messages, followed by Type 2, and then Type 3. It can be speculated that the
higher incidence of slow-downs in response to Type 1 messages may be because of the low
frequency with which they were displayed or the conditions that caused the message to be
displayed. The average decrease in speed over all off-on cases was -3.13 mph; decreases
occurred in 17.1 % of cases. Speeds increased or were unaffected in 82.9% of cases. Also, DMS
displaying travel time messages did not show a higher propensity for slow-downs than DMS
displaying other types of messages. The on-off analysis indicated that average speeds increased
more often than they decreased in response to message removal. When broken down by message
type, no clear pattern was observed. Under message-switching condition, average traffic speeds
increased as often as they decreased. The overall findings from the before-after analysis indicate
traffic is unaffected by message appearance, removal or switching in the majority of cases, with
the remaining cases representing relatively small effects on traffic speeds.

The secondary analysis examined average speeds over 12 two-week periods to determine
aggregate effects of message display on traffic speeds. Type 1 messages have the largest effect
on average traffic speeds. Type 2 messages at most result in average speeds 4 mph below the
overall average during periods with no messages. In only three cases did this reduction represent
an average speed below the posted speed limit, and only one of these had a corresponding overall
average speed above the speed limit. Type 3 messages had the smallest negative impact on
average speeds. In most cases, average speeds were higher during these messages than during
times of no messages. The results of this study indicate that DMS message display is not likely to
cause congestion. It is important to note that speed changes observed in this study cannot be
wholly attributed to DMS because there are many other factors unaccounted for, such as weather.

The research team also evaluated localized safety impacts of highway DMS. The accident data
from 2007 to 2010 served as the baseline for analyses of traffic collisions in Maryland. The
accident data, DMS locations and Annual Average Daily Traffic (AADT) database are projected
onto Maryland roadway map in ArcGIS 10.1. In order to perform spot analysis to evaluate
whether DMS influenced drivers’ operational performance, an impact area of 900 feet was
defined for each DMS based on the maximum visibility distance for the average font size on
electronic signs. The accident database included 38,718 records, which were filtered, cleaned and
purged of data gaps and outliers. After data processing, the number of accidents considered
decreased to 23,842 for the four-year study period. The accident database consisted of accident
type (property damage, personal injury, and fatality), accident location and county, time and date
of the accident and coordinates of accident location. Due to confidentiality concerns, access to
police records and accident causes was not possible.

ii
As a part of this project, a case study was performed on Interstate 95 in Maryland. A sample of
70 road segments was chosen based on geometrical homogeneity. Regression analysis was
performed based on whether the segment was an impact area or not, if the segment included
interchanges or not and what the AADT of the segment was. An unbalanced two-way ANOVA
was used to compare mean accident rate in impact areas and other segments. To determine the
effects of DMS messages on the rate of accidents, accident rates in DMS impact areas and
adjacent segments were compared using paired t-tests. The difference in accident rates was tested
on two DMS operation statuses (when they displayed messages and when they were blank),
using a one-way ANOVA with a pairwise comparison test. Statistical analyses on DMS
characteristics, message types, weather conditions and accidents in impact area were performed.

The statistical analyses of accidents in conjunction with weather conditions showed that there
were only four accidents in all impact areas that occurred in rainy and snowy conditions. Thirty-
two out of 43 accidents were in wind gusts of 0-10 mph, nine were in wind gusts of 10-20 mph,
and two were in wind gusts of 20-30 mph.

Additionally, among 50 accidents in DMS impact areas and impact-adjacent areas, 35 collisions
were classified as property damage and 15 as personal injury. Analyses on displayed messages
showed 11 accidents occurred while Danger/Warning messages were displayed on DMS, 22
occurred during displays of Informative/Common Road Condition messages, and 11 during
displays of Regulatory/Non-Traffic-Related messages. Although there are some concerns that
Danger/Warning messages cause drivers to slow down, the least number of accidents in DMS
impact areas and impact-adjacent areas occurred during the period when Danger/Warning
messages were displayed. The findings from all evaluations converge to indicate DMS are a safe
tool for disseminating real-time travel information to motorists and do not have significant
adverse effects on driver’s operation and traffic safety.

In summary, the findings from these evaluations indicate DMS can be an accurate, effective, and
safe tool for disseminating real-time travel information to motorists. This research focused on
Maryland DMS, so the findings may not extend to DMS operations in other states. Nevertheless,
the methodology for evaluating data is applicable beyond Maryland’s borders.

This report has been prepared in two volumes. The first volume includes methodology and
results of an empirical analysis of the quality, effectiveness and localized impacts of DMS. The
second volume covers GIS data processing efforts for mapping accidents and weather data, and
the statistical analysis of DMS’ impact on accident occurrence.

iii
 STATE HIGHWAY ADMINISTRATION

RESEARCH REPORT

EVALUATION OF DYNAMIC MESSAGE SIGNS AND

THEIR POTENTIAL IMPACT ON TRAFFIC FLOW

Volume I: Empirical Analysis of the Quality, Effectiveness, and Localized

Impacts of Highway Dynamic Message Sign Messages

ALI HAGHANI
MASOUD HAMEDI
ROBIN L. FISH
AZADEH NOURUZI

UNIVERSITY OF MARYLAND, COLLEGE PARK

Project number SP109B4C

FINAL REPORT

February 2013
 Abstract

The need to convey accurate, real-time travel information to road users has long been

recognized by transportation engineers. One of the primary means to accomplish this

is the operation of highway Dynamic Message Signs (DMS), which have been in use

for more than 50 years. However, the quality of messages used, the extent of their

influence on motorists’ behavior and their localized impacts are not well documented.

This project introduced Bluetooth traffic detection sensors as a new tool to evaluate

the quality of DMS messages and their resulting influence on motorists’ route

choices. In addition, highway speed sensors were used to determine whether DMS

affected traffic speeds. Results indicate DMS messages are accurate in

communicating prevailing conditions and can influence drivers’ route choices. Speed

analyses indicated certain types of messages exert greater influence on traffic patterns

than others. Additionally, the majority of message types do not negatively affect

traffic speeds.

2
This document contains information that includes data owned by Traffic.com, Inc. or
is derived from data owned by Traffic.com (“Data”). The Data is proprietary,
confidential and protected by copyright (© Traffic.com), as well as other rights,
including but not limited to, trade secret rights. Traffic.com retains all right, title and
interest in the Data. The Data is used by the University of Maryland and the
Maryland Department of Transportation with the permission of Traffic.com under a
license with the State of Maryland dated July 13, 2005 and under an agreement with
the University of Maryland dated April 16, 2009. Use of this Data by any other
persons or entities requires the prior written consent of Traffic.com. Consent should
be requested from webmaster@traffic.com. The restrictions on the use and disclosure
of the Data shall survive the license and agreement.
3
 Table of Contents

List of Tables ................................................................................................................ 5

List of Figures ............................................................................................................... 6
Chapter 1: Introduction ................................................................................................. 7
1.1: Motivation and Background .............................................................................. 7
1.2: Literature Review .............................................................................................. 9
1.2.1: Message Quality ......................................................................................... 9
1.2.2: Driver Response and Diversion ................................................................ 11
1.2.3: Speed Impacts ........................................................................................... 13
1.2.4: Summary ................................................................................................... 14
1.3: Scope................................................................................................................ 15
1.4: Organization..................................................................................................... 15
Chapter 2: Detection Technology and Data ................................................................ 17
2.1: Bluetooth Technology...................................................................................... 17
2.2: Bluetooth Detectors and Data .......................................................................... 17
2.3: Dynamic Message Sign Data ........................................................................... 19
2.4: Traffic Speed Data ........................................................................................... 19
Chapter 3: Message Quality and Effectiveness .......................................................... 21
3.1: Deployments and Study Area .......................................................................... 21
3.1.1: Study Area ................................................................................................ 21
3.1.2: Sensor Deployment Considerations .......................................................... 22
3.1.3: Deployment Details .................................................................................. 23
3.2: Message Quality .............................................................................................. 27
3.2.1: Deployment 1 ............................................................................................ 28
3.2.2: Deployment 2 ............................................................................................ 35
3.2.3: Travel Time Messages .............................................................................. 49
3.3: Message Effectiveness ..................................................................................... 56
3.3.1: Sensors ...................................................................................................... 56
3.3.2: Diversion Analysis .................................................................................... 59
Chapter 4: Localized Impacts .................................................................................... 63
4.1: Motivation........................................................................................................ 63
4.2: Methodology .................................................................................................... 63
4.2.1: Data Sources and Preparation ................................................................... 63
4.2.2: Consecutive Five Minute Data Analysis .................................................. 66
4.2.3: Aggregate Two Week Speed Analysis ..................................................... 67
4.3: Findings ........................................................................................................... 67
4.3.1: Consecutive Five Minute Data Analysis .................................................. 67
4.3.2: Aggregate Two Week Speed Analysis ..................................................... 81
Chapter 5: Conclusions and Future Work .................................................................. 86
Bibliography ............................................................................................................... 90

4
List of Tables

Table 3.1. Selected Cases for Deployment 1 .............................................................. 28

Table 3.2. Deployment 2, Case I Messages ................................................................ 36
Table 3.3. Deployment 2, Case II Messages ............................................................... 40
Table 3.4. Deployment 2, Case III Messages ............................................................. 45
Table 3.5. Case I Travel Time Differences ................................................................. 53
Table 3.6. Case I Travel Time Difference (Outliers Removed) ................................. 53
Table 3.7. Case II Travel Time Differences ............................................................... 55
Table 3.8. Traffic Diversion Share Between I-95 and I-895 North Deployment 1 .... 60
Table 3.9. Traffic Diversion Share Between I-95 and I-895North Deployment 2 ..... 62
Table 3.10. Traffic Diversion Share Between I-95 and I-695East Deployment 2 ...... 62
Table 4.1. DMS Locations and Distance to RTMS .................................................... 64
Table 4.2. Message Categorization Summary and Examples ..................................... 65
Table 4.3. # Cases by DMS and Operational Condition ............................................. 68
Table 4.4. Off-On Summary by DMS ........................................................................ 70
Table 4.5. Off-On Summary by DMS and Message Type ......................................... 71
Table 4.6. On-Off Summary by DMS ........................................................................ 74
Table 4.7. On-Off Summary by DMS and Message Type ......................................... 76
Table 4.8. Switching Summary by DMS .................................................................... 78
Table 4.9. Switching Summary by DMS and Message Types ................................... 80

5
List of Figures

Figure 2.1. Bluetooth Detector Internals..................................................................... 18

Figure 2.2. Bluetooth Detection Concept of Operation .............................................. 19
Figure 2.3. DMS-RTMS Configuration ...................................................................... 20
Figure 3.1. Study Area: I-95 and I-895 ....................................................................... 21
Figure 3.2. June-July 2009 Deployment-Pickup Locations & Times ......................... 24
Figure 3.3. Sketch of Sensor Deployments labeled with TMC letter designations .... 26
Figure 3.4. March-April 2011 Deployment-Pickup Locations & Times .................... 27
Figure 3.5. Deployment 1, Case I Speed Data for Link AF........................................ 29
Figure 3.6. Deployment 1, Case I Speed data for Link FP ........................................ 30
Figure 3.7. Deployment 1, Case I Speed Data for Link OP........................................ 30
Figure 3.8. Deployment 1, Case II Speed Data for Link QR ...................................... 31
Figure 3.9. Deployment 1, Case II Speed Data for Link ST ....................................... 32
Figure 3.10. Deployment 1, Case III Speed Data for Link QR .................................. 33
Figure 3.11. Deployment 1, Case III Speed Data for Link ST ................................... 34
Figure 3.12. Deployment 2, Case I Speed Data for Link ST ...................................... 37
Figure 3.13. Deployment 2, Case I Speed Data for Link QR ..................................... 37
Figure 3.14. Deployment 2, Case I Speed Data for Link AF...................................... 38
Figure 3.15. Deployment 2, Case I Speed Data for Link FL ...................................... 38
Figure 3.16. Deployment 2, Case I Speed Data for Link LP ...................................... 38
Figure 3.17. Deployment 2, Case II Speed Data for Link ST ..................................... 41
Figure 3.18. Deployment 2, Case II Speed Data for Link OP .................................... 41
Figure 3.19. Deployment 2, Case II Speed Data for Link QR .................................... 42
Figure 3.20. Deployment 2, Case II Speed Data for Link FO .................................... 43
Figure 3.21. Deployment 2, Case II Speed Data for Link AF .................................... 43
Figure 3.22. Deployment 2, Case III Speed Data for Link ST ................................... 46
Figure 3.23. Deployment 2, Case III Speed Data for Link QR .................................. 46
Figure 3.24. Deployment 2, Case III Speed Data for Link AF ................................... 46
Figure 3.25. Deployment 2, Case III Speed Data for Link FO ................................... 47
Figure 3.26. Deployment 2, Case III Speed Data for Link OP ................................... 47
Figure 3.27. Sample Travel Time Message for DMS #7701 ...................................... 50
Figure 3.28. Case I, Displayed vs. Actual Travel Time.............................................. 51
Figure 3.29. Case I, Displayed vs. Rounded and Capped Travel Time ...................... 52
Figure 3.30. Case II, Displayed vs. Actual Travel Time ............................................ 54
Figure 3.31. Case II, Displayed vs. Rounded and Capped Travel Time..................... 55
Figure 3.32. I-95 and I-895 North Diversion Point .................................................... 58
Figure 3.33. I-95 and I-695 Diversion Point ............................................................... 58
Figure 3.34. Traffic Share During Message Cases Deployment 1 ............................. 59
Figure 3.35. Traffic Share During Message Case Deployment 2 ............................... 61
Figure 4.1. Sample DMS-RTMS Pair ......................................................................... 64
Figure 4.2. Graph of Off-On Summary by DMS ........................................................ 70
Figure 4.3. Graph of Off-On Summary by Message Type ......................................... 72
Figure 4.4. Graph of Off-On Summary by DMS ........................................................ 74
Figure 4.5. Graph of On-Off Summary by Message Type ......................................... 76
Figure 4.6. Graph of Switching Summary by DMS ................................................... 78
Figure 4.7. Aggregate Average Speeds for Two Week Analysis ............................... 83
Figure 4.8. Aggregate Speeds Normalized by Overall Average Speeds .................... 84
6
Chapter 1: Introduction
1.1: Motivation and Background

The need to convey accurate travel information to motorists has become

increasingly important with the increase in traffic volume and the lack of additional

roadway capacity. Knowledge of rapidly changing traffic conditions gives motorists

the option to modify their behavior in order to avoid delays and dangerous situations.

Many states, as part of an Advanced Traveler Information System (ATIS), have

installed Dynamic Message Signs (DMS) to help provide motorists this information.

Also known as Variable Message Signs (VMS) and Changeable Message Signs

(CMS), these electronic signs can display various messages that can be specified by a

remote operator or updated automatically. Among others benefits, this capability

allows roadway administrators to communicate with motorists about accidents,

delays, and in some jurisdictions, travel time.

An important measure of the value of a DMS message is its credibility. It is

vital that travelers believe messages displayed on a DMS are factual and accurately

describe roadway conditions. Without consistently valid information, road users will

begin to ignore DMS messages.

The Maryland State Highway Administration’s (SHA) Coordinated Highways

Action Response Team (CHART) operates nearly 80 DMS. The signs are located on

major highways and their arterial roadways. The DMS often inform motorists of

delays, incidents, road closings, and recently, real-time travel times. In the case of

travel-delay messages, phrases such as “Major Delays,” “Heavy Delays” and “Expect

Congestion” describe traffic conditions. Ambiguous phrases such as these do little to

inspire confidence in the DMS system unless their meanings are easily understood,

consistent and appropriate for the given road conditions.

To determine the meaning and accuracy of such messages, the road conditions

during which they are displayed are examined. Specifically, Bluetooth travel-time

data was collected and analyzed during the display of various DMS on Maryland’s I-

95 and I-895 corridors. This project presents the first attempts to use Bluetooth

ground truth data to determine the timeliness and accuracy of the DMS messages.

Another important aspect of DMS is their effectiveness. DMS messages

should accurately inform motorists of road conditions and if necessary, induce

changes in motorists’ behavior. A good measure of whether or not a message yields

such a change is if users divert or change routes during a period in which a message

suggests the same. The unique identification and re-identification capability of

Bluetooth sensors allows for an estimate of these diversion rates. We compared

detection rates between the current and suggested routes during the period of study to

determine the effectiveness of DMS messages in influencing motorist behavior.

Although the quality and effectiveness of messages are important for DMS

systems, some are concerned that displaying messages causes localized speed

reductions and congestion, increasing danger to motorists. To investigate this

concern, DMS systems in close proximity to Remote Traffic Monitoring Sensors

(RTMS) were identified. The speed data from these detectors is used to analyze any

impacts the display of messages had on the traffic streams during their display.

The findings from these analyses should give comprehensive insight into the

performance, quality, effectiveness and effects of DMS in Maryland. State officials

8
will be able to apply these findings and methods to analyze and improve their DMS

operations.

1.2: Literature Review

The following sections present a summary of literature relevant to the study of DMS.

Study of existing publications will give insight into the previous methods, relevant

findings and any benefits or shortcomings others encountered.

1.2.1: Message Quality

DMS are a relatively new but frequently changing technology. As such, a unified

standard for displaying messages has not yet been developed. The Manual on

Uniform Traffic Control Devices (MUTCD) suggests some formatting requirements

such as text size and message length, but it does little to address what warrants the

display of certain messages.

In order to be effective, a displayed message must contain a combination of

the following elements: problem, location, effect, attention and action (1). These

components must be combined in a way that conveys enough information to be useful

to motorists while fitting within the limited space on a DMS. The MUTCD specifies

that a message be readable at least twice while traveling at the posted speed limit (2).

This guideline means motorists have approximately 8 seconds at most to read

messages on a DMS in normal weather and roadway conditions (3). These restrictions

can be complicated by the occurrence of multiple incidents or less-than-optimal

weather or roadway conditions.

9
 Several states developed message hierarchies that rank the relative importance

of various message categories should a conflict in message choice arise. In general,

messages requiring a change in motorists’ behavior (e.g., emergencies, incidents and

roadway closures) are near the top of such hierarchies (1, 4, 5). Messages of moderate

importance in the rankings tend to be related to congestion, travel time or weather

conditions. If none of the previous conditions occur, some jurisdictions display public

service or safety messages, whereas others display nothing. The three levels of the

hierarchy are termed Danger & Warning Messages, Informative Messages and

Regulatory Messages (6).

In jurisdictions where quantitative travel time information is not available,

terms such as “Heavy Delay” and “Major Delay” are often used. There is little

explanation about how to define or use these terms. However, Dynamic Message

Sign Message Design and Display Manual reports the average motorist in Texas

interprets “Heavy Delay” as being between 25 and 45 minutes, whereas a “Major

Delay” is interpreted as a delay greater than 45 minutes (7). Similarly, a study in

England to determine driver response to DMS messages found motorists interpreted

“Long Delays” as being between 35 and 47 minutes, whereas they perceived “Delays

Likely” as indicating a 10 to 31 minute delay (8). The Minnesota Department of

Transportation’s Guidelines for Changeable Message Sign (CMS) Use specifies that

“Major Delay” indicates an incident causing more than two miles of traffic backup

and not a length of time (4). These conflicting definitions alone demonstrate the need

to evaluate how well DMS messages match the conditions during which they are

displayed.

10
 Since state transportation agencies introduced travel time messages on DMS,

there have been attempts to validate the accuracy of these messages. In Oregon, travel

time messages were derived from loop detector data. To validate the displayed travel

times, researchers utilized 87 probe vehicles outfitted with GPS devices. Using paired

t-tests, the researchers compared what they called the “ground truth” data to the

displayed travel times. Using this method, they determined the travel times were

accurate in many cases but suffered from deficiencies during incidents or when

detectors were placed poorly (9). Researchers in California used probe vehicles to

validate the travel times after designing a model using loop detector data to predict

and automatically display travel times on DMS. Eighty-eight probe vehicle runs were

made on two different roads. The authors found good agreement between travel times

and probe data when sufficient data existed. As a result, the authors concluded it is

necessary to validate travel times using probe data prior to deploying DMS travel-

time messages (10). These studies demonstrate previous attempts to validate DMS

travel-time message using data from probe vehicles. Although the collected data were

of high quality, neither investigation produced more than 100 data points. This project

utilizes Bluetooth travel time collection for the validation of DMS travel times. As a

result we were able to collect a large data set, which yields a higher quality analysis.

1.2.2: Driver Response and Diversion

Revealed preference (RP) and stated preference (SP) surveys of drivers have been

used in numerous studies to determine the influence DMS have on drivers. A RP

survey combined with an ordered logit model suggested that the propensity of drivers

to divert due to a DMS message was correlated to how often drivers encountered a

11
DMS and whether or not they believed DMS contain useful and trustworthy

information (11). In Beijing, a SP survey found that diversion increased as the speed

of traffic decreased. Specifically, at speeds under 20 km/h (indicated as serious

congestion on VMS) 21.45 percent of drivers say they will divert, whereas when

traffic is moving between 20-35 km/h (common congestion) a mere 7.02 percent of

drivers expect that they would divert (12). Canadian and British drivers were

compared in a SP survey to determine perceived effectiveness of DMS information.

The survey revealed evidence to suggest more exposure to DMS leads to an increase

in appreciation of the information displayed (13). A combined SP and RP survey

performed by researchers at the University of California, Berkeley found that en route

travelers were not inclined to divert in response to an Advanced Traveler Information

System (ATIS) device unless the device specifically recommended such action or

provided specific information about delay time on the preferred route (14). Similarly,

a SP survey of Borman Expressway drivers in Indiana revealed a strong correlation

relating the type of message displayed to the driver response. It was concluded that

message content is an “important control variable for improving system performance”

(15). As expected, the importance of trust and specific information weigh heavily on

the effectiveness of DMS.

Another method of determining effectiveness of DMS is the examination of

loop detector data. A study of DMS effects on traffic was performed in the Hampton

Roads area of Virginia. In order to assess these effects, loop detector data from two

alternative routes were collected and analyzed along with DMS messages displayed

regarding travel delays on the routes. The diversion rates found were very low, which

12
the researchers believed were caused by weak messages, unwillingness to divert and

distance from the secondary route. A secondary analysis under a new message system

found higher diversion rates; however, there was not enough data to make any

conclusions (16). In Ontario, Canada, three years of loop detector data were collected

along with DMS messages on the highway 401 express collector. The study was

interested in finding the response of traffic to a change in DMS message. The study

found the initial diversion reaction to a change in DMS message is significant and the

occurrence of a message change plays a vital role in influencing downstream

diversion (17). Using loop detector and message characteristic data as inputs,

researchers in Minnesota estimated a probit model to estimate diversion as a function

of message content. Through this method it was determined VMS messages can

significantly influence route diversion. Specifically, when warned by a message,

users are more likely to divert than if confronted with congestion (18).

Loop detector data analyses have shown DMS can potentially impact

diversion. One caveat to these findings is that loop detector data are unable to identify

individual vehicles, so their individual paths cannot be determined with certainty.

1.2.3: Speed Impacts

Several researchers investigated the effects of DMS messages on traffic speed using

various methods. At the University of Iowa, researchers used a full-size traffic

simulator to investigate the dynamics of travelers’ speed in response to DMS and

other in-vehicle information systems. They found those who saw DMS messages

slowed down in the areas the messages correspond to, but once out of range of the

message tended to compensate by increasing their speeds (19). A simulation study by

13
researchers in Sweden found all participants reduced their speeds in response to

Incident Warning Systems in the simulation (20). Researchers in Finland found

drivers reduced speeds 1-2 km/h in response to a DMS warning of slippery conditions

(21). A field study of two DMS by researchers in Norway found vehicles showed

“large speed reductions.” Through video recording, researchers observed “large

proportions” of the traffic stream-braking in advance of the DMS (22). To determine

the effects of DMS on traffic slowdowns, researchers at the University of Rhode

Island used five-minute interval speed data during the nearest periods when messages

switched from off to on and from on to off. They found slowdowns occurred in more

than half of the cases examined and particularly during cases of danger messages,

although not all instances were statistically significant (23).

These findings seem to indicate that DMS may cause localized speed

reductions, but examination of more cases and higher quality data would be useful to

further understand these patterns.

1.2.4: Summary

The need for DMS to present accurate, timely and useful messages has been

recognized since their inception. Many methods have been used to determine whether

these needs are being met. Surveys, simulators and loop detector data have been the

most common of these methods in the past and have shown promising results. This

project presents Bluetooth detection as an emerging method for evaluating DMS. The

ability to anonymously identify and re-identify individual vehicles and users to track

travel time and diversion was previously unavailable or prohibitively expensive. The

method used in this study should provide a higher quality analysis method than

14
previously available. In addition, the use of high quality one-minute interval speed

data for analysis of localized effects will provide finer results than previous attempts.

1.3: Scope

This project covers the Bluetooth analysis of two separate DMS case studies on the

same segments and an examination of speed data in proximity to six DMS. The

Bluetooth case studies consist of data collected in June-July 2009 and March-April

2011. Both deployments were completed on the same segments of I-95 & I-895 for

the examination of DMS # 7701 & #7702. In the first deployment, 20 Bluetooth

sensors were used; due to technical difficulties, 19 were used for the second. For both

deployments, specific message types were selected and analyzed for timeliness and

accuracy. For cases that suggested diversions, analysis of diversion rates as

represented by Bluetooth detection sampling were performed. Finally, the localized

effects of the DMS were studied through analysis of highway speed data. Two

specific analyses were undertaken: the first investigates the effects of message display

on speed in two consecutive five-minute periods; the second investigates the speeds

over several two-week periods.

1.4: Organization

The organization of the report is as follows: Chapter 2 provides a brief review

of Bluetooth technology, the specific sensors used in this study and the data that were

analyzed. Chapter 3 presents the efforts and results of the quality and effectiveness

analyses based on the Bluetooth analysis. Chapter 4 examines the localized speed

15
effects of DMS message display. Finally, Chapter 5 provides conclusions and

recommendations for future work.

16
Chapter 2: Detection Technology and Data

2.1: Bluetooth Technology

The primary data for this study came from Bluetooth device detection. Bluetooth is a

short-distance wireless networking protocol found in many modern electronic devices

including vehicles, cell phones, laptops and earpieces. Depending on the power rating

of the device, transmission distances range from one up to 100 meters. Consumer

devices most commonly use class 2 radios, which have a range of approximately 10

meters.

Each Bluetooth device is assigned a unique identifier known as a Machine

Access Control (MAC) address. These MAC addresses allow for the management and

proper handling of data. When operating, Bluetooth devices continuously transmit

their MAC addresses to locate other devices with which to pair and transmit data.

This transmission forms the basis for Bluetooth traffic detection technology, as it

allows identification and re-identification of individual devices without depriving

motorists of their anonymity. The Bluetooth Special Interest Group provides more

detailed information about Bluetooth technology.

2.2: Bluetooth Detectors and Data

In order to take advantage of the traffic information that can be obtained using

Bluetooth devices, a specialized detector is required. For this study, detectors

developed by the University of Maryland were used for data collection. The detectors

are considered off-line because they do not transmit collected data in real time. The

main components of the detectors include a large battery, antenna, computer board,

17
GPS unit, and a memory card slot (Figure 2.1). The antenna detects Bluetooth MAC

addresses from up to 100 meters and stores them and detection times on a memory

card. When the sensors were retrieved, the memory cards were removed and the data

was downloaded.

Figure 2.1. Bluetooth Detector Internals

The main processing effort consisted of matching MAC addresses from

detector to detector and calculating the elapsed time (Figure 2.2). Because the

locations of the detectors are known, the distance between them can be calculated.

These data were then used to calculate travel times and space mean speeds. A more

detailed explanation of Bluetooth travel time detection for freeway segments can be

found in (26). Additionally, the specific processing efforts for this project are

discussed in Chapter 3.

18
 Figure 2.2. Bluetooth Detection Concept of Operation

2.3: Dynamic Message Sign Data

The DMS data used in this study were provided by the Maryland SHA and

retrieved through the University of Maryland Center for Advanced Transportation

Technology (CATT). Messages are provided in the Markup Language for

Transportation Information (MULTI), along with indication of beacon status and

timestamps for start and end times. MULTI tags allow users to determine the

formatting and the number of lines and panes of messages as they were originally

displayed. Using this information, relevant messages could be selected for evaluation

based on content and display time. The same message logs were manipulated as

described in Chapter 4 in order to assess the impacts of messages on traffic speeds.

2.4: Traffic Speed Data

In order to analyze the localized effects of message display on traffic speeds, high

quality speed data were required. The data used to complete this analysis were

collected from the Center for Advanced Transportation Technology (CATT) lab and

consisted of one-minute interval speed data provided by pole-mounted, side-fired

19
Remote Traffic Monitoring Sensors (RTMS). In each case, DMS were selected such

that the corresponding RTMS was within forward sight distance of the DMS (Figure

2.3).

Figure 2.3. DMS-RTMS Configuration

20
Chapter 3: Message Quality and Effectiveness

3.1: Deployments and Study Area

The following sections describe the study area, sensor deployment considerations and

descriptions of the deployments used.

3.1.1: Study Area

Before deployment of sensors, identification of appropriate locations is required. To

maximize the available information from the data, the study area should contain at

least one frequently used Dynamic Message Sign. In addition, the roadway should

have reasonably high traffic volumes and have available alternate routes and major

junctions. For the deployments in this study, sections of Interstate 95 Northbound and

its parallel route Interstate 895 were selected (Figure 3.1).

Figure 3.1. Study Area: I-95 and I-895

21
 In Figure 3.1, yellow pins represent Bluetooth sensors deployed for travel

time detection, red pins represent Bluetooth sensors deployed for diversion tracking

and blue pins represent Dynamic Message Sign locations.

This study area represents a major commuting corridor with three major

parallel routes through and around Baltimore (I-95 N, I-895 N, and I-695 E). The

DMS selected for evaluation in this area were #7701 and #7702. In the initial

deployment, the signs most commonly referenced delays on either I-95 or I-895 and

in some cases suggested alternative routes. In the second deployment, the signs

adopted real-time travel time information as their primary messages, while displaying

delay and other messages as necessary.

3.1.2: Sensor Deployment Considerations

When selecting Bluetooth sensor locations, several factors have to be

considered. Primarily, the locations must be in a safe, accessible and secure location.

Because the sensors are deployed manually, there must be a shoulder where a vehicle

can stop so the operator can safely activate sensors and lock them to a permanent

object. The next consideration is the distance between sensors. Due to the 300-foot

sensing buffer, an error of up to 600 feet may be encountered. In order to reduce

overall errors in travel time and space mean speed, it is desirable to place travel-time

detection sensors at least one mile apart. (More information on this error can be found

in (26).) Sensors must also be placed on the major diversion routes to detect any

vehicles that exit the main road. Therefore, sensors must be placed on diversion

routes so they are as close to the main road as possible without being close enough to

detect the vehicles on the main road.

22
3.1.3: Deployment Details

On the morning of June 29, 2009, the deployment team drove to the Maryland

Welcome Center rest stop on I-95 N just before the interstate’s junction with

Maryland Route 32. There, they gave the sensors to an SHA employee and briefed the

employee on the deployment plan. The deployment team gave the driver sufficient

warning prior to each deployment site in order to allow for safe exiting from the main

travel lanes. The Bluetooth sensors were turned on once at their site of deployment.

The team waited for the sensor to acquire a GPS signal and then tethered and locked

the sensors in position. To supplement the internal GPS, a handheld unit was used to

collect latitude and longitude coordinates of the sensor deployments (Figure 3.2).

The UMD team again collaborated with the SHA to retrieve the sensors. On

July 7, 2009 around 9 a.m., the deployment crew met with an SHA employee at the

same Maryland Welcome Center rest stop. Upon arrival at the sensor locations, the

deployment team unlocked the sensors and powered them down, noting any unusual

operating conditions (e.g. GPS no longer locked on, powered off prematurely). The

Micro SD memory cards were then removed from the sensors and carefully sorted

into corresponding cases.

In total, 20 sensors were deployed, each with a corresponding letter from A-T,

resulting in 65 links that were designated as virtual Traffic Message Channels

(TMCs). For example, I95+XXXAF would represent the link between sensor A and

sensor F. These virtual TMCs were later used to match and analyze travel time

23
 Bluetooth Deployment Proposed Actual Deployment (6/29/09) Pick Up (7/7/09)
Number Location Latitude Longitude Sensor ID Actual Latitude Actual Longitude Time (AM) Status Time (AM)
1 Rest stop prior to DMS #7701 on I-95 N 39.14188433 -76.84554586 AJ 39°08.570' N 76°50.960' W 9:36 ON 9:55
2 Prior to exit 38B for MD Route 32 39.15477028 -76.82876430 T 39°09.357' N 76°49.890' W 9:43 ON 9:59
3 MD 32 East off of I-95 39.15046400 -76.82431700 2 39°08.972' N 76°49.423' W 9:49 ON 10:01
4 MD 32 West off of I-95 39.16308300 -76.83394400 AD 39°09.753' N 76°50.024' W 9:56 ON 10:07
5 Prior to exit 41B for MD Route 175 39.17542200 -76.79310800 AH 39°10.341' N 76°47.924' W 10:05 ON 10:15
6 Prior to exit 43A for MD Route 100 39.19247490 -76.77071154 4 39°11.529' N 76°46.291' W 10:11 ON 10:19
7 MD 100 East off of I-95 39.19192500 -76.76087800 S 39°11.516' N 76°45.653' W 10:15 ON 10:21
8 MD 100 West off of I-95 39.20278900 -76.77265300 AE 39°12.134' N 76°46.335' W 10:21 ON 10:27
9 I-95 .5 miles prior to Montgomery Rd 39.20734643 -76.74653437 AG 39°12.441' N 76°44.792' W 10:28 ON 10:33

24
10 Prior to exit 46 I-95/I-895 Interchange 39.22034200 -76.72338600 AF 39°13.220' N 76°43.403' W 10:34 ON 10:36
11 Prior to exits for MD Route 166/I-195 39.23410882 -76.70973274 Q 39°13.785' N 76°42.842' W 10:39 OFF 10:38
12 Prior to exits for I-695 39.24679896 -76.68531180 1 39°14.828' N 76°41.064' W 10:44 ON 10:43
13 I-695 East off of I-95 39.24643600 -76.67552200 3 39°14.675' N 76°40.516' W 10:48 ON 11:02
14 I-695 West off of I-95 39.23822800 -76.68652200 O 39°15.394' N 76°41.568' W 10:54 ON 10:54
15 Prior to I-95/I-295 Interchange 39.27061100 -76.64474700 AB 39°16.223' N 76°38.578' W 11:05 ON 11:13
16 Prior to Ft. McHenry Tunnel 39.26612500 -76.59782800 AC 39°15.953' N 76°35.771' W 11:12 ON 11:18
17 I-895 N .25 miles prior to Washington Blvd 39.21853900 -76.71059400 AM 39°13.207' N 76°42.891' W 11:54 ON 11:47
18 I-895 N .25 miles prior to Toll 39.24070300 -76.58812200 AL 39°14.455' N 76°35.369' W 11:30 ON 11:33
19 I-895 after toll, before Tunnel 39.24257800 -76.57896900 AA 39°14.555' N 76°34.378' W 11:33 ON 11:36

Figure 3.2. June-July 2009 Deployment-Pickup Locations & Times

20 I-895 after Harbor Tunnel 39.26500956 -76.56088968 AK 39°15.921' N 76°33.701' W 11:39 ON 11:26
data. A total of 893,582 travel time samples were collected. After processing and

aggregation by two-minute intervals, 362,901 data points were available for analysis.

In March 2011, more sensors were deployed in a method identical to that used

in the 2009 deployment. As in the first deployment, the team met with an SHA

employee to transfer and place sensors. Sensors were placed as closely as possible to

the locations used in the 2009 deployment in order to make valid comparisons and to

simplify post processing (Figure 3.4). Unfortunately, one of the sensors

malfunctioned prior to deployment and was not used in the 2011 deployments as a

result. Therefore, only 19 sensors were deployed in the 2011 data collection. The

sensors were retrieved on April 12, 2011, 14 days after their March 29 deployment.

The omitted sensor corresponded to only one missing virtual TMC. This allowed for a

total of 64 virtual TMCs. A sketch of sensor placements for both deployments was

produced for conceptualization purposes (Figure 3.3). For the second deployment,

sensor N was not used.

25
Figure 3.3. Sketch of Sensor Deployments labeled with TMC letter designations

26
 Bluetooth Deployment Proposed Actual Deployment (3/29/11) Pick Up (4/12/11)
Number Location Latitude Longitude Sensor ID Actual Latitude Actual Longitude Time (AM) Status Time (AM)
1 Rest stop prior to DMS #7701 on I-95 N 39.14188433 -76.84554586 S 39.14216667 -76.84536667 10:05 OFF 10:06
2 Prior to exit 38B for MD Route 32 39.15477028 -76.82876430 F 39.15578333 -76.83215 10:12 OFF 10:08
3 MD 32 East off of I-95 39.15046400 -76.82431700 P 39.15061667 -76.82433333 10:22 OFF 10:10
4 MD 32 West off of I-95 39.16308300 -76.83394400 A 39.16303333 -76.83391667 1:09 OFF 12:08
5 Prior to exit 41B for MD Route 175 39.17542200 -76.79310800 D 39.17221667 -76.79861667 10:29 OFF 10:17
6 Prior to exit 43A for MD Route 100 39.19247490 -76.77071154 I 39.19255 -76.7711 10:35 OFF 10:21
7 MD 100 East off of I-95 39.19192500 -76.76087800 G 39.19193333 -76.76105 10:39 OFF 10:23
8 MD 100 West off of I-95 39.20278900 -76.77265300 O 39.20273333 -76.77278333 12:57 OFF 12:01
9 I-95 .5 miles prior to Montgomery Rd 39.20734643 -76.74653437 H 39.2076 -76.74655 10:48 OFF 10:36

27
10 Prior to exit 46 I-95/I-895 Interchange 39.22034200 -76.72338600 E 39.22023333 -76.72375 10:56 OFF 10:30
11 Prior to exits for MD Route 166/I-195 39.23410882 -76.70973274 C 39.2298 -76.7141 11:57 OFF 11:25
12 Prior to exits for I-695 39.24679896 -76.68531180 T 39.2464 -76.68575 12:04 OFF 11:28
13 I-695 East off of I-95 39.24643600 -76.67552200 L 39.24635 -76.67558333 12:09 OFF 11:30
14 I-695 West off of I-95 39.23822800 -76.68652200 - - - - OFF -
15 Prior to I-95/I-295 Interchange 39.27061100 -76.64474700 M 39.27061667 -76.64478333 12:20 OFF 11:38
16 Prior to Ft. McHenry Tunnel 39.26612500 -76.59782800 Q 39.26585 -76.59623333 12:29 OFF 11:41
17 I-895 N .25 miles prior to Washington Blvd 39.21853900 -76.71059400 K 39.21853333 -76.71068333 11:07 OFF 10:43
18 I-895 N .25 miles prior to Toll 39.24070300 -76.58812200 B 39.24073333 -76.5883 11:15 OFF 10:52
19 I-895 after toll, before Tunnel 39.24257800 -76.57896900 R 39.2425 -76.57906667 11:21 OFF 10:55

Figure 3.4. March-April 2011 Deployment-Pickup Locations & Times

20 I-895 after Harbor Tunnel 39.26500956 -76.56088968 N 39.26526667 -76.56178333 11:28 OFF 10:58
3.2: Message Quality

Analyses of message quality and timeliness of selected cases for both deployments

are presented in the following sections.

3.2.1: Deployment 1

After collecting and examining message logs, three interesting cases from the

first deployment were selected for evaluation (Table 3.1). In all cases, the messages

were the same on the two DMS and were shown to drivers for a relatively long period

of time. For the evaluation, the Bluetooth travel times were converted to space mean

speed and graphs were produced for observation purposes.

Table 3.1. Selected Cases for Deployment 1

Case # Time Period Duration Message Displayed DMS #
I 7/2/2009 58 I-95 MAJOR DELAYS 7701
15:4516:44 minutes ALT I-895 NORTH 7702
(PM) OR I-695 EAST
II 7/2/09 2 hours I-895 MAJOR DELAYS 7701
16:5919:18 18 ALT I-95 NORTH 7702
(PM) minutes OR I-695 EAST
III 7/1/09 42 I-895 NORTH 7701
10:2011:02 minutes EXPECT CONGESTION 7702
(AM) AND DELAYS

In the first case, the DMS displayed “Major Delays” on I-95 for

approximately one hour during the afternoon peak traffic period on July 2, 2009. The

traffic conditions recorded by the Bluetooth sensors on I-95 were examined for a time

period slightly before and after this message was displayed to determine the

conditions that gave rise to the choice to display and, later, to remove the message.

To determine the validity of the message, links between the first DMS and the

Harbor tunnel were examined. Graphs of space mean speed for virtual TMC links AF,

28
AB, BE, EF, FP, FI, IJ, JK, KL, LO and OP, over a time period of 15 minutes before

and after the message, were inspected for disturbances. Traffic speed on the link AF,

from just before DMS #7701 until just past DMS #7702 was below 35 mph before,

during and after the display of the message (Figure 3.55). The links between sensors

A and F (AB, BE, EF) displayed similar reductions in traffic speeds throughout the

duration of the message, with link AB being the least affected.

Figure 3.5. Deployment 1, Case I Speed Data for Link AF

Link FP, which covers the overall path from DMS #7702 to the harbor tunnel,

shows no major disturbances in space mean speed during the display period (Figure

3.66). Similarly, links FI, IJ, JK, KL, and LO remain relatively stable and maintain

speeds above 55 mph for the duration of the message. Link OP, the link closest to the

tunnel, shows a slight disturbance from 15:50 to 16:00 in which the speed drops to

around 45 mph (Figure 3.7).

29
 Figure 3.6. Deployment 1, Case I Speed data for Link FP

Figure 3.7. Deployment 1, Case I Speed Data for Link OP

In case I, the message appears to be misleading. The speed on link AF was

below 35 mph, which could indicate a major delay, however, on the links beyond

sensor F the speed of traffic is stable and relatively high. The first DMS (#7701)

accurately portrays travel conditions between itself and the next DMS (#7702), but

travelers who see the message on the second DMS would not have experienced any

30
congestion between that point and the beginning of the Harbor tunnel, a distance of

about 11 miles. Furthermore, the suggestion on the first DMS to use I-895 or I-695E

as possible alternatives is not helpful because neither of those choices become

available until beyond the second DMS, where the congestion had cleared. The same

suggestion on the second DMS is not only inaccurate but may have led to degradation

of trust in the DMS system because users continuing on I-95N in spite of the DMS

warning would have experienced no reason to divert.

The second case alerts drivers of “Major Delays” on I-895 for two hours and

18 minutes during the afternoon peak period on July 2, 2009. The DMS message is

displayed at 16:59 and turned off at 19:18.

In this case, both virtual TMC links on I-895 reveal major disturbances in

speed data. Link QR undergoes a speed reduction to below 35 mph from 16:20 to

17:45 (Figure 3.8). Similarly the speed on link ST remains below 25 mph between

16:20 and 18:20 (Figure 3.9). The speeds on both links appear to have returned to

relatively normal levels and stabilized by 18:30.

Figure 3.8. Deployment 1, Case II Speed Data for Link QR

31
 Figure 3.9. Deployment 1, Case II Speed Data for Link ST

For case II, the message appears to have been appropriate given the prevailing

traffic conditions. The “Major Delay” message seems to have been prompted by the

severity and duration of the drops in traffic speed. However, data reveal the

deployment time of the message was at least 25 minutes after the traffic conditions

began to deteriorate. In addition, up to 15 minutes prior to deployment of this

message the signs were warning of “Major Delays” on I-95 and displayed I-895 as a

suggested alternative (Case I). Drivers complying with this suggestion would have

found themselves in congestion on I-895 and would likely be displeased with the

DMS system. Although the message was displayed appropriately for over an hour, the

message was left on for nearly 45 minutes after the link speeds had rebounded to 45

mph or higher. Though drivers seeing the “Major Delays” message may have been

happy to find no congestion during these 45 minutes, the credibility of the DMS

system would improve if the message was removed in a timely manner.

32
 In the third case, the message states that the road users should “Expect

Congestion and Delays” on I-895 North. This message is displayed for 42 minutes

approximately one hour after the end of the morning peak period on July 1, 2009.

The Bluetooth derived space mean speed data were examined for links QR

and ST on I-895. The data for link QR appears stable and above 55 mph for the time

period from 10 minutes before the message until 10 minutes after the message,

although the number of data points is limited (Figure 3.10). On link ST, a speed drop

occurred 20 minutes prior to the message display (Figure 3.11). Speeds went from

above 50 mph to below 25 mph and remained below 25 mph for 10 minutes. When

the message came on at 10:20, the speed began to return to normal and stabilized

between 45 and 55 mph by 10:40.

Figure 3.10. Deployment 1, Case III Speed Data for Link QR

33
 Figure 3.11. Deployment 1, Case III Speed Data for Link ST

The message displayed under these conditions appears to be in reaction to the

slowdown in speed in the Harbor Tunnel (link ST) and possibly further north. The

message accurately alerts motorists of congestion and delays occurring on I-895;

however, it appears to have been posted just as the congestion was beginning to clear.

The delay in display of the message may have resulted in some drivers experiencing

little or no congestion after having seen the message, once again resulting in

devaluation of the DMS system.

In general, the findings from the 2009 deployment reveal that DMS operations

could benefit from some adjustments. Although all the messages were warranted by

the prevailing traffic conditions and would provide some benefit to drivers, their

benefits were diminished by non-timely display and removal. For the DMS system to

maintain its credibility, the road conditions experienced by users should match the

descriptions on the signs as closely as possible. When messages do not appear in a

timely manner, users may experience congestion without warning. Conversely, a

34
message removed late will result in users experiencing no delays when a message

warns that a delay exists. Another consideration is the specificity of the messages. In

these cases, all of the messages warned of delays on I-95 or I-895 “North”. This

description is very vague and could potentially refer to immediate delays or delays

that are miles away. More useful messages should contain clear indication of the

affected roadway and a specific description of the location of delays.

3.2.2: Deployment 2

In the second deployment, DMS messages often operated independently of

each other and displayed travel-time messages by default. During disruptive traffic

events, however, the signs tended to act in unison as in the previous deployment.

Where differences in content during message display existed, the cases are split by

DMS identification number. Several of these cases are analyzed using the techniques

in the previous deployment. In addition, some periods in which travel-time messages

were displayed were analyzed to assess the accuracy of these messages.

In the first case, the sequence of messages begins at 16:33 and ends at 18:45

on March 31, 2011 (Table 3.2). The first message appears on DMS # 7702 and refers

to “Major Delays” on I-895 North of the tunnel. This message persists for 16 minutes

until a second pane is added that mentions “Major Delays” prior to the tunnel on I-95

North. At the same time, this single-pane message is displayed on DMS # 7701. At

approximately 17:16, both signs begin displaying a message warning of “Major

Delays” on both I-95 and I-895 North and recommends I-695 East as an alternate

route. In order to analyze these messages, links AF, FL, LP, QR and ST were

examined.

35
Table 3.2. Deployment 2, Case I Messages
CASE I - DMS # Time Period Duration Messages
3/31/2011 MAJOR DELAYS
16:5017:16 26 minutes I-95
(PM) PRIOR TO TUNNEL
7701
3/31/2011 1 hour 29 MAJOR DELAYS
17:1618:45 minutes I-95 AND I-895 NORTH
(PM) ALT. ROUTE I-695 E.
3/31/2011 MAJOR DELAYS
16:3316:49 16 minutes I-895 N
(PM) NORTH OF TUNNEL

MAJOR DELAYS
3/31/2011 I-895 N
16:4917:16 27 minutes NORTH OF TUNNEL
(PM) MAJOR DELAYS
I-95 N
7702 PRIOR TO TUNNEL

MAJOR DELAYS
3/31/2011 I-895 N
17:1617:17 1 minute NORTH OF TUNNEL
(PM) MAJOR DELAYS
I-95 AND I-895 NORTH
ALT. ROUTE I-695 E.

3/31/2011 MAJOR DELAYS

17:1718:45 1 hour 28 I-95 AND I-895 NORTH
(PM) minutes ALT. ROUTE I-695 E.

The initial message displayed on DMS #7702 appears to be appropriate as the

speeds on link ST (Figure 3.12) are 30 mph below free flow at the message onset

(solid green line). This indicates that the delays north of the tunnel on I-895 are

spilling back and causing delays in the tunnel itself. Speeds on link QR during the

same time show that the delays do not extend below the tunnel (Figure 3.13).

36
 Figure 3.12. Deployment 2, Case I Speed Data for Link ST

Figure 3.13. Deployment 2, Case I Speed Data for Link QR

At 16:50, both signs began warning of delays on I-95 prior to the tunnel. On

link AF, speeds decreased as the message was deployed (Figure 3.14), although

speeds are steady on links FL and LP (Figure 3.15, Figure 3.16). The message

displayed on DMS #7701 is accurate and appears fairly soon after conditions begin to

deteriorate. On DMS #7702, however, there is no indication that the message is yet

necessary as the links after it remain unaffected.

37
Figure 3.14. Deployment 2, Case I Speed Data for Link AF

Figure 3.15. Deployment 2, Case I Speed Data for Link FL

Figure 3.16. Deployment 2, Case I Speed Data for Link LP

38
 For approximately 90 minutes (beginning at 17:16), both signs displayed a

message about the delays on I-95 and I-895 North. In addition, they suggested that I-

695 East be used as an alternate route. In all figures, this activation is represented by

the dashed green line. As observed in Figure 3.14 and Figure 3.15, the negative speed

trends on links AF and FL warrant the warning of delays on I-95 North. The speed

trends on links QR and ST as seen in Figure 3.13 and Figure 3.12 are seen to decrease

or are already low at the onset of the message, supporting the delay warning for I-

895. Though the delay warnings are warranted for both roads, the diversion message

may be inappropriate. Examination of link FP on I-95 North beyond I-695 reveals no

apparent delay. This indicates that continuing on I-95 rather than diverting onto I-695

may be preferred, depending on the condition of I-695.

Messages about a delay on I-95 North were activated just as delays were

beginning and were removed as conditions were recovering, except in the case of link

LP where no delays were observed during the period. The first message displayed

about I-895 North appeared after speeds were already low on link ST. At the time of

removal, speeds on I-895 were at normal levels for 20 minutes on link ST and for 30

minutes on link QR. It appears that the message continued until conditions had

recovered on both roads, rather than changing the message to refer to only the

persisting delays on I-95 North.

The messages in this first case of the second deployment attempted to inform

motorists of delays on I-95 and I-895 North. All messages displayed were at least

partially warranted and were, for the most part, displayed and removed in a timely

fashion. The speed data suggests that diversion onto I-695 East was unnecessary;

39
although avoiding I-895 by continuing on I-95 North toward I-695 would have been

preferred.

The second case occurs during the afternoon peak period on April 1, 2011

(Table 3.3). At 16:27 a message was posted on DMS #7701 and #7702, alerting

motorists of Major Delays on I-95 and I-895, north of their respective tunnels. On

DMS #7701 this message persisted until 19:14. The message on DMS #7702 was

updated at 16:57 with a second pane that noted Major Delays prior to the I-895 tunnel

in addition to the delays north of the tunnel. This two-pane message continued until

19:13, when the message reverted to the original one-pane message for one minute.

At 19:14, both signs began displaying their default travel time messages.

Table 3.3. Deployment 2, Case II Messages

CASE II - DMS # Time Period Duration Messages

4/1/2011 MAJOR DELAYS

7701 16:2719:14 2 hours 46 I-95 AND I-895 N
(PM) minutes NORTH OF TUNNEL

4/1/2011 MAJOR DELAYS

16:2716:57 29 minutes I-95 AND I-895 N
7702 (PM) NORTH OF TUNNEL

MAJOR DELAYS
4/1/2011 I-95 AND I-895 N
16:5719:13 2 hours 16 NORTH OF TUNNEL
(PM) minutes MAJOR DELAYS
I-895 N
PRIOR TO TUNNEL
1 minute
4/1/2011 MAJOR DELAYS
19:1319:14 I-95 AND I-895 N
(PM) NORTH OF TUNNEL

40
 Links ST and OP, the northernmost links on I-895 and I-95 respectively, show

space mean speeds at or below 25 mph at the time of message activation (solid green

lines), indicating spillbacks from the posted delays north of the tunnels (Figure 3.17,

Figure 3.18). It is also evident that these spillbacks persisted for at least 25 minutes

on each of these links prior to the message activation. These delays, though, were

accounted for by the previously posted travel time messages that indicated higher

than normal travel times.

Figure 3.17. Deployment 2, Case II Speed Data for Link ST

Figure 3.18. Deployment 2, Case II Speed Data for Link OP

41
 When DMS #7702 began warning of delays on I-895 prior to the tunnel

(dashed green lines), it was observed that speeds on link QR (Figure 3.19) fell

approximately 10 mph since the posting of the original message. The message

appeared to be in reaction to this increased congestion.

Figure 3.19. Deployment 2, Case II Speed Data for Link QR

Examining link FO (Figure 3.20) reveals no apparent delays on I-95 between

DMS #7702 and the interchanges just prior to the Ft. McHenry tunnel. The delays

north of the tunnel on I-95 did not spill back as they had on I-895. Therefore, the

messages displayed on DMS #7702 were accurate and useful, as there were no

unaccounted-for delays occurring on I-95 prior to the tunnel. If motorists chose to

avoid I-895 by remaining on I-95 as a result of the DMS message, they would not

experience unexpected delays.

42
 Figure 3.20. Deployment 2, Case II Speed Data for Link FO

On DMS #7701, the message warned only of the delays north of the tunnels

for the entire period. Although this message accurately described those conditions,

the message failed to warn motorists of the delays on link AF (Figure 3.21) from

DMS #7701 to DMS #7702. In this case, users may have benefited from continued

display of the default travel time message on DMS #7701, which would have taken

into account these delays. Since the information displayed on the sign would not have

been useful until after DMS #7702, where it was repeated, users may have found the

information inadequate given the prevailing conditions.

Figure 3.21. Deployment 2, Case II Speed Data for Link AF

43
 At the time of message removal at 19:14, all of the examined links had

returned to near free-flow speeds. Although many of the links had been stable for at

least 30 minutes, link QR remained unstable until 10 minutes before message

removal, meaning the message was maintained until all links had stabilized, as

observed in the previous case. Both signs resumed displaying travel-time messages at

the end of the period and both indicated free flow conditions.

This second case shows that the DMS communicated accurate and timely

information to motorists. The conditions posted were apparent in the data and would

have been useful to motorists, although the first DMS could have been used to inform

users of the delays prior to the second DMS as well.

The third case was an all-day event resulting from a closure of the Harbor

Tunnel on I-895 during the morning peak hour (Table 3.4). At 7:31. both DMS begin

alerting drivers of the tunnel closure on I-895 and recommend I-95 North or I-695

East as alternate routes. After 15 minutes, the message was removed and both signs

displayed their respective travel time messages until 9:32. At this time, both signs

displayed a message that informed users to expect congestion and delays on I-895

North. After approximately three hours, this message was removed from both signs.

DMS #7701 resumed displaying travel time messages. Meanwhile, DMS #7702

warned motorists of Major Delays on I-895 and suggested the same alternate routes

as in the morning message. The “Major Delay” message persisted on DMS #7702 for

approximately six hours and was removed at 18:23.

44
Table 3.4. Deployment 2, Case III Messages
CASE III - DMS # Time Period Duration Messages

4/2/2011 I-895 TUNNEL

7:317:46 15 minutes CLOSED
(AM) ALERNATE ROUTES
7701/7702 I-95 N. OR I-695 E.
4/2/2011
9:3212:28 2 hours 56 I-895 NORTH
(AM/PM) minutes EXPECT CONGESTION
AND DELAYS

7702 4/2/2011 I-895 MAJOR DELAYS

12:3218:23 5 hours 51 ALT I-95 NORTH
(PM) minutes OR I-695 EAST

When the initial message was displayed, link ST experienced near free flow

conditions (Figure 3.22). For the next 15 minutes there was no Bluetooth data

available, which indicated that no traffic passed through the tunnel. This finding

corresponds with the message advising of the tunnel closure. During the following 15

minutes, traffic speeds rapidly dropped, stabilizing around 20 mph. At the same time,

link QR appeared to be unaffected by the tunnel closure (Figure 3.23). In addition, the

recommendation to use I-95 North as an alternate route appeared to be sound, as there

were no apparent delays on links AF, FO, or OP (Figure 3.24, Figure 3.25, Figure

3.26).

45
Figure 3.22. Deployment 2, Case III Speed Data for Link ST

Figure 3.23. Deployment 2, Case III Speed Data for Link QR

Figure 3.24. Deployment 2, Case III Speed Data for Link AF

46
 Figure 3.25. Deployment 2, Case III Speed Data for Link FO

Figure 3.26. Deployment 2, Case III Speed Data for Link OP

Between 7:46 and 9:32, both signs resumed display of travel time messages.

All links on I-95 North and I-895 North prior to the tunnel were unaffected by the

tunnel delays, so the signs displayed free flow speed-limited travel times.

Unfortunately, no warning was given during this time of the delays occurring in the

Harbor Tunnel. At 9:32, the congestion in the Harbor Tunnel appeared to have

backed up onto link QR, which resulted in speeds dropping to 40 mph. A message

warning of “congestion and delays” appeared on both DMS for the next three hours.

47
During those three hours, speeds appeared to steadily drop on link QR, eventually

falling to approximately 25 mph.

At 12:28, the message was removed from both signs and replaced with travel

time messages. DMS #7701 continued displaying travel time messages for the

remainder of the case period. At 12:32, DMS #7702 replaced its travel time message

with a “Major Delay” message relating to I-895 with I-95 North and I-695 East as

alternate routes (dashed lines). The message continued for approximately six hours

until it was removed at 18:23. The delays cleared on both links QR and ST about 15

minutes before the removal of the message, indicating a timely reaction to traffic

conditions. During the same period, the traffic conditions were near free flow and

were steady on all links on I-95 North, making it a viable alternate route. In addition,

the choice to display travel time on DMS #7701 during this period gave users more

information to make their decision on whether to continue on I-95 North.

In this third case in the 2011 deployment, the DMS communicated changing

conditions through the day. At each change of message, the conditions observed

through the data matched the descriptions displayed. The messages were also updated

and removed rapidly with the conditions to which they corresponded. One shortfall

during this third case was the morning period, during which both signs reverted to

travel-time messages, ignoring delays in the Harbor Tunnel. Because DMS #7702

displayed equal travel times to the Harbor Tunnel and the Fort McHenry Tunnel

during this time, motorists may have taken I-895 North only to find heavy delays in

the tunnel. Although travel-time information displayed on the DMS was accurate,

motorists could have benefitted by being warned of the delays in the tunnel. Overall,

48
this case demonstrated a sound operation of the DMS system through the timely

display of high quality messages with useful information.

The cases from 2011 deployment indicate an improvement in the quality and

timeliness of DMS messages over the 2009 deployment. In these cases, the messages

specifically indicated certain sections (e.g. before or after the tunnel) when necessary,

and the Bluetooth-observed conditions supported the accuracy of the messages. In

some instances, messages were left on longer than necessary, which meant users

experienced no delays even though they were warned of them. On the other side,

travel-time messages displayed by default alleviated some problems arising from

delay messages being displayed long after conditions had deteriorated. Users would

be at least somewhat aware that conditions were worsening as the displayed travel

time would be above normal. Again, Bluetooth detection has been demonstrated as a

viable tool for analysis of Dynamic Message Signs.

3.2.3: Travel Time Messages

During the 2011 deployment, the DMS were used by default to display real-time

travel time information to various destinations. Using the Bluetooth-derived ground

truth travel times, the travel times displayed on the DMS can be analyzed for

accuracy and timeliness. DMS #7701 displayed travel time from itself to I-695, a

stated distance of 11 miles (Figure 3.27). To analyze this segment, the ground truth

travel time on virtual-TMC segment AL, from DMS #7701 to the first I-695 Exit

ramp, was used.

49
 Figure 3.27. Sample Travel Time Message for DMS #7701

The timestamps from the Bluetooth data were matched to timestamps from the

DMS message log; the displayed travel time was then extracted from each message.

The Bluetooth travel times were matched in raw format to the displayed travel times

and converted to minutes. The average difference between these two data sets and the

standard deviations was then determined.

In addition, a comparison to the rounded and capped Bluetooth travel times

was made. The rounded and capped data is differentiated from the raw data by taking

into account the speed limit and the integer restriction on the signs. On a segment 11

miles long with a speed limit of 65 mph, the minimum legal travel time is

approximately 10.15 minutes. Since the signs only display integer values, the

minimum travel time displayed is 11 minutes because display of a travel time of 10

minutes or lower implies traffic speeds above the posted speed limit. For this reason,

any travel times below 11 minutes are rounded up to 11 minutes. All other travel

times are rounded to the nearest integer value. In order to demonstrate the analysis

ability of the Bluetooth data, two travel time message cases were selected for

evaluation.

50
 Displayed vs Actual Travel Time 3/30/2011

Travel Time (Minutes) 28

23

18

13

8
7:12 8:24 9:36 10:48 12:00 13:12 14:24 15:36 16:48 18:00 19:12 20:24 21:36
Time Displayed Travel Time
Actual Travel Time

Figure 3.28. Case I, Displayed vs. Actual Travel Time

In both cases, the messages began display in the morning and indicated free-

flow travel times. They changed to the display travel times as conditions began to

deteriorate. The first case begins at 8:32 March 30, 2011, and ends at 19:52 the same

day. Travel times remained at or above free-flow levels until approximately 16:22, at

which time travel time on the segment began to increase. The first sign update

occurred at 16:30. Travel time continues to increase until approximately 18:30, at

which point it leveled off and returned to free-flow conditions by 19:08. (Figure

3.28).

51
 The graph indicates the displayed travel time very closely matched the

ground truth travel time with some lag during the period of travel time increase. This

lag may be attributed to data acquisition and processing time prior to display on the

DMS. It is also notable that the Bluetooth data displays several very high travel times

during the free flow period. We speculate the sensors detect vehicles that make stops

or diversions between the matched detectors, causing these outliers. As previously

mentioned, much of the data during free-flow conditions is below the displayed travel

time because of traffic exceeding the speed limit. To account for this, a similar graph

where the actual travel time is converted to rounded and capped travel time is

produced (Figure 3.29).

Displayed vs Rounded and Capped Travel Time

3/30/2011
28
Travel Time (Minutes)

23

18

13

8
7:12 8:24 9:36 10:48 12:00 13:12 14:24 15:36 16:48 18:00 19:12 20:24 21:36
Time
Displayed Travel Time
Rounded & Capped Travel Time
Figure 3.29. Case I, Displayed vs. Rounded and Capped Travel Time

With this manipulation, it is clear travel times displayed during free-flow

conditions are accurate. During the congested period, the same lag between actual and

displayed travel times is observed. To determine the numerical discrepancies between

52
the displayed and ground truth travel times, the difference between them at each time

was calculated as follows:

𝐷𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑐𝑒 = 𝑇𝑇𝑎𝑐𝑡𝑢𝑎𝑙,𝑡 − 𝑇𝑇𝑑𝑖𝑠𝑝𝑙𝑎𝑦𝑒𝑑,𝑡

The average and standard deviation of this difference was calculated for both

the actual and capped travel times (Table 3.5).

Table 3.5. Case I Travel Time Differences

Actual Capped
Average Difference 0.2616441 0.72865854
Standard Deviation 2.4018081 2.22362256

The average difference in both cases indicates actual travel times are slightly

higher than the displayed travel times. In the capped case, the higher average value

probably results from the free flow times being rounded up. The standard deviations

are somewhat high, although certainly a result of the outliers during the free flow

period. With the outliers removed, the results change (Table 3.6).

Table 3.6. Case I Travel Time Difference (Outliers Removed)

Actual Capped
Average Difference -0.01027 0.464174
Standard Deviation 1.424859 1.180357

53
 The second case occurs the following day, March 31, 2011, between 5:00 and

16:48. The period primarily consists of free-flow conditions, with increases in travel

time beginning at 15:12. (Figure 3.30). At the end of this period, the DMS message

was changed to a non-travel time message.

Displayed vs Actual Travel Time 3/31/2011

24
22
Travel Time (Minutes)

20
18
16
14
12
10
8
3:36 4:48 6:00 7:12 8:24 9:36 10:48 12:00 13:12 14:24 15:36 16:48 18:00
Time Displayed Travel Time
Actual Travel Time
Figure 3.30. Case II, Displayed vs. Actual Travel Time

During free-flow conditions, the actual travel times are very close to the

displayed travel times with only a few outliers. When travel time began to increase,

the gaps observed are small. Overall, the messages appear to accurately represent the

true travel times. The rounded and capped travel time shows similar trends (Figure

3.31).

54
 Displayed vs Rounded and Capped Travel Time
24
3/31/2011
22
Travel Time (Minutes)

20
18
16
14
12
10
8
3:36 4:48 6:00 7:12 8:24 9:36 10:48 12:00 13:12 14:24 15:36 16:48 18:00
Time Displayed Travel Time
Rounded & Capped Travel Time
Figure 3.31. Case II, Displayed vs. Rounded and Capped Travel Time

Again it is clear that the display of 11-minute travel time for the majority of

the period was justified. There are several instances where travel times go above 11

minutes during this period, but none persist long enough to influence the messages.

There is more visible lag between the ground truth and displayed times when

rounded, but none appear to be unreasonable. The difference between the actual and

displayed travel times was calculated as previously described (Table 3.7).

Table 3.7. Case II Travel Time Differences

Actual Capped
Average Difference -0.0998325 0.3193548
Standard Deviation 1.38101214 1.2005714

These results show the influence of speed limit-capped travel time. When

compared with the actual travel time, the displayed times are slightly higher, as

expected. By removing the influence of speed limits, the average difference shows the

55
displayed travel times during congested periods were lower than the actual travel

times. The standard deviations in both cases are small, indicating a tight spread in

travel time differences.

Overall, these two cases show the data and updating system used for DMS

travel-time messages provided accurate and mostly timely information to motorists.

On average, the difference between the actual travel time and the displayed travel

time was less than one minute, with standard deviations, outliers removed, of less

than two minutes. These cases also demonstrate that Bluetooth sensors are capable of

providing high-quality data of DMS travel times. The methods used are repeatable

and applicable to systems in other jurisdictions regardless of their data sources and

updating systems.

3.3: Message Effectiveness

The following sections describe the methodology and findings from the using

of Bluetooth sensors to evaluate traffic diversion resulting from Dynamic Message

Sign messages.

3.3.1: Sensors

Previous attempts to empirically analyze traffic diversion in response to DMS

messages have used loop detector data. This type of data gives only traffic counts, so

it is impossible to determine the specific path of a given vehicle. On the other hand,

Bluetooth detectors are capable of providing a sample of origin-destination data

through identification and re-identification of the individual vehicles at consecutive

sensors. The drivers’ response to the messages displayed on DMS can be studied by

56
analyzing Bluetooth origin destination data (as a proxy for the actual origin

destination data) before and after the DMS message display The downside of this

approach is that Bluetooth is only a sampling technology with an average 3.5 percent

penetration rate (26).

Several sensors were deployed to track vehicle diversion. In order to do this,

the sensors were placed such that they would detect vehicles shortly after major

diversion or exit points. The primary diversion from I-95 recommended by the DMS

was I-895 North. To determine the share of traffic on these alternative routes,

detections were matched between sensor J and sensors K and Q (Figure 3.32). In

addition, the messages often recommended I-695 East as an alternate route. For these

cases, the detections between sensor L and sensors M and O were compared (Figure

3.31).

57
Figure 3.32. I-95 and I-895 North Diversion Point

Figure 3.33. I-95 and I-695 Diversion Point

58
3.3.2: Diversion Analysis

In the first deployment, Cases I and II both displayed messages recommending

diversion. Case I recommended using I-895 as an alternate route for 58 minutes. After

being blank for 15 minutes, Case II recommended using I-95 for the next 2 hours.

The share of traffic diverting on each link is analyzed during the times of the day in

which the signs were blank, during the message cases, and during the time between

the two messages (Figure 3.34).

Figure 3.34. Traffic Share During Message Cases Deployment 1

During periods in which the signs were blank, approximately 80% of vehicles

continued on I-95 while the other 20% used I-895. When the message in Case I

recommended diversion onto I-895, it was observed that the share of traffic

continuing on I-95 North dropped by 5 percent. Similarly, when Case II suggested

59
use of I-95 instead of I-895, there was a 7 percent increase in utilization of I-95

(Table 3.8).

Table 3.8. Traffic Diversion Share Between I-95 and I-895 North Deployment 1
Average Average
I-95 I-895 Standard
Time interval Share Share Deviation
(%) (%)
All times with no message on display 80.4 19.6 10.2
Case I: divert to I895 North or I695 East 75.5 24.5 10.4
Time between removal of message in case I and display 80.3 19.7 7
of message in case II

Case II: divert to I95 North or I695 East 87.4 12.6 8.4

In the second deployment, similar diversion messages were posted. On

April 6, 2011, DMS #7702 posted three messages recommending drivers to divert

away from I-95 through the use of I-895 or I-695. In this deployment, the default

posted messages displayed travel time. To determine the baseline diversion shares,

times when the DMS displayed free flow travel times were used. Traffic shares were

calculated during periods of diversion message display (Figure 3.35).

60
 Figure 3.35. Traffic Share During Message Case Deployment 2

During the times the signs indicated free flow travel times, the share of drivers

using I-95 instead of I-895 was approximately 89 percent. The proportion of drivers

choosing I-95 instead of I-695 East during the same periods was approximately 80

percent. When diversion messages were posted, the average share of drivers using I-

95 instead of I-895 dropped approximately 10 percent (Table 3.9). Similarly, those

choosing I-95 instead of I-695 East dropped nearly 18 percent (Table 3.10).

61
 Table 3.9. Traffic Diversion Share Between I-95 and I-895North Deployment 2
Average Average
Standard
Time Interval I-95 I-895
Deviation
Share (%) Share (%)
Free Flow Travel Time 88.7 11.3 6.04
Divert to I-895 or I-695 78.5 21.5 12.03

Table 3.10. Traffic Diversion Share Between I-95 and I-695East Deployment 2
Average Average
Standard
Time Interval I-95 I-695
Deviation
Share (%) Share (%)
Free Flow Travel Time 80.1 19.9 10.51
Divert to I-895 or I-695 62.3 37.7 20.90

These findings provide preliminary support for the contention that DMS have

modest effect on drivers’ route choices. When the messages suggested specific

diversions, the Bluetooth detection data showed corresponding shifts in diversion

patterns. It must be noted that these numbers serve only as a proxy to the drivers’

response because only a fraction of the traffic can be detected using Bluetooth

sensors. Although one cannot conclude with certainty that drivers changed their

original route as a result of DMS recommendations, the change in the traffic pattern

at the time of message display is noticeable and can be interpreted as the effectiveness

of the dynamic message signs. These results confirm parts of the findings in (17).

62
Chapter 4: Localized Impacts

4.1: Motivation

The State of Maryland began providing motorists with nearly real-time travel

time information using DMS in January 2010. Although much of the public response

to these messages was positive, some motorists and media outlets renewed complaints

that DMS messages caused vehicles to slow down, which resulted in congestion and

caused safety issues. To investigate these claims, the effects of several highway DMS

were evaluated by their proximity to one-minute interval RTMS speed sensors. In

total, six DMS-RTMS pairs were selected for evaluation. In all of the cases, the

RTMS were installed prior to and within sight distance of the DMS.

The evaluation process consisted of two analyses. The first analysis compared

average traffic speeds of vehicles in consecutive five-minute periods during which the

DMS operational condition changed. In the second analysis, traffic stream speeds

were averaged in two-week periods to determine the impact on traffic under different

DMS operational scenarios. The purpose of this study is to determine whether using

DMS on Maryland highways presents significant localized safety hazards or

congestion problems. The data used and methods are described in detail below.

4.2: Methodology

4.2.1: Data Sources and Preparation

The data used in this analysis were collected from the University of Maryland

Center for Advanced Transportation Technology (CATT). The data were DMS

message logs for each DMS and one-minute interval speed data provided by pole-

63
mounted, side-fired Remote Traffic Monitoring Sensors (RTMS). In each case, DMS

were selected so the corresponding RTMS was within forward sight distance of the

DMS (Figure 4.1). Six cases are included in this study (See Table 4.1). For each

DMS-RTMS pair, data was retrieved for a period starting January 1, 2010, and ending

February 28, 2011. In some cases, data gaps existed such that all months in the range

were not available for analysis.

Table 4.1. DMS Locations and Distance to RTMS

DMS # Distance from RTMS Location
839 150 ft I-95 SB @ Exit 55
3316 1800 ft I-95/495 NB Outer Loop North of MD 202
3317 1900 ft I-95/495 SB Inner Loop @ Good Luck Road
4401 785 ft I-695 SB Outer Loop @ Exit 12B
4403 50 ft I-695 SB Outer Loop @ Exit 10
8557 50 ft I-895 NB past Ritchie Spur

Figure 4.1. Sample DMS-RTMS Pair

64
 In order to analyze these data, the DMS and RTMS data needed to be in the

same units. The raw DMS data were in inconsistent time intervals because the time

interval was controlled by when messages were initiated, changed or removed. In

order to match the DMS data to the one-minute interval RTMS speed data, an Excel

Macro code was written to increment the DMS data in one-minute intervals. The

resulting minute-by-minute DMS logs were then matched by their timestamps to the

RTMS speed data and the corresponding quality scores. Speed data receiving quality

scores other than zero (zero indicates a valid score) were discarded. Due to observed

inconsistencies in the data and low traffic volumes, the data were filtered to remove

observations between the hours of 19:00 and 6:00 the next day. For the first analysis,

data from weekends were also excluded.

When necessary, as described in the following sections, messages were

categorized into three types based on the ideas proposed by Ridgeway (6). The types

are as follows: Danger/Warning Messages, Informative/Common Road Conditions

and Regulatory/Non-Traffic Related. Some common messages falling into each

category can be seen in Table 4.2.

Table 4.2. Message Categorization Summary and Examples

Message Category Common Examples
Type 1 Accidents, Disabled Vehicles, Non-recurring Slow-
Danger/Warning Downs, Roadway Debris, Unplanned Lane/Tunnel/
Bridge Closures
Type 2 Roadwork Closures, Major & Minor Delays,
Informative/Common Congestion, Travel Time, Other travel related
Road Condition messages (Fog, Ice, Snow Plowing, Major Events)
Type 3 Work Zone Speeds, Seatbelt Use, Cell Phone
Regulatory/Non-Traffic Regulations, Motorcycle Awareness, Amber & Silver
Related Alerts, Homeland Security Messages

65
4.2.2: Consecutive Five Minute Data Analysis

This study examined speed changes in consecutive five-minute periods in

which the DMS operational condition changed. The types of operational conditions

considered were off-on, on-off and switching. In the off-on condition, the DMS is off

in the first five minutes and on with a displayed message in the following five

minutes. The on-off condition is the exact opposite (i.e., on and displaying a message

for the first five minutes, off for the following five minutes). The final condition,

switching, is a situation in which the DMS is on for the entire ten minute

investigation period. The two five-minute periods are differentiated by a significant

change in the message content.

Cases were selected manually by combing through the minute-by-minute

DMS-speed datasets and isolating those instances when the DMS operational

condition changed. Each case was then sorted and stored into one of the three

operational conditions (off-on, on-off or switching). Cases with congestion (indicated

by low traffic speeds) were not included for analysis.

To determine the effects of the changing DMS operational condition on traffic

speeds, the one-minute interval speeds in each consecutive five-minute period were

compared using paired t-tests at 95 percent confidence level. The null hypothesis is

that there is no difference in mean speeds between consecutive periods. On the other

hand, the alternative hypothesis is that the difference between the means is some

value not equal to zero. They are written as follows:

𝐻0 : 𝜇2 − 𝜇1 = 0

𝐻1 : 𝜇2 − 𝜇1 ≠ 0

66
 The total number of significant speed changes was tabulated. For each sample

case, the difference in average speed between the first five minutes and the following

five minutes was calculated. For significant cases, the overall average speed change

was calculated. Each case was then assigned a category per Ridgeway’s previously

described scheme in order to examine differences in effects over message types.

4.2.3: Aggregate Two Week Speed Analysis

To assess the effects of DMS messages on absolute travel speed over longer

periods, two 14-day periods were selected for analysis for each DMS. Each message

was assigned into one of the three categories. The messages were then run through

the minute-by-minute incrementing macro, and then matched by their timestamps to

the one-minute RTMS speed data. As in the previous analysis, speed data and the

corresponding messages with a quality score other than zero were discarded.

Using the categorization, average speeds over the two-week periods for each

message type were determined. The five averages taken for each two-week period

were the overall speed, speed during all messages, speed during no messages, and

speeds during type 1, 2, and 3 messages. In addition, the fraction of the observations

that fell into each message type was recorded. Using this information, any trends that

exist over message types could be identified.

4.3: Findings

4.3.1: Consecutive Five Minute Data Analysis

67
 In total, 2,268 cases of consecutive five-minute DMS operational condition

change were analyzed. This total was broken down over the three condition types:

off-on, on-off and switching. 842, 701, and 725 cases were available, respectively.

Table 4.3 shows the complete breakdown by DMS # and operational condition. As

discussed in (21), in the off-on condition we expect speeds will tend to decrease due

to the added task of message comprehension. Conversely, we would anticipate speeds

increasing in the on-off condition since the traffic in the second five-minute period

would no longer be influenced by the message. The switching condition presents a

situation in which the expected effects are dependent on the messages in the

consecutive periods. For instance, it would be expected that a change from a message

related to seatbelt use to a message informing drivers of a nearby road closure would

result in a speed reduction.

Table 4.3. # Cases by DMS and Operational Condition

DMS # Off-On On-Off Switching Total
839 96 83 76 255
3316 74 65 146 285
3317 151 76 93 320
4401 215 163 259 637
4403 101 88 68 257
8557 205 226 83 514
842 701 725 2268

To test these hypotheses, the number of statistically significant cases of speed

increases and decreases were tabulated for each DMS and operational condition.

These numbers were used to find the percent of cases in which statistically significant

increases or decreases were observed. The average speeds over the significant cases

were also calculated to determine the extent of the impact.

68
Off-On

There were significant speed decreases in 144 cases and significant speed

increases in 101 cases in which the DMS condition changed from off to on. These

numbers represent 17.1 percent and 12.0 percent of the 842 total cases, respectively.

In terms of speed, the average decrease over significant cases was -3.12 mph, while

the average increase was 2.34 mph. The breakdown over DMS is shown in Table 4.4

and Figure 4.2.

We can infer from these results that in the case of the Off-On condition,

drivers tended to slow down more often than they speed up, confirming the general

hypothesis. It is observed that the lowest ratios of significant changes in speed occur

for the DMS-RTMS pairs that are the furthest apart (i.e. 3316 & 3317). Interestingly,

there also appears to be a tendency of those DMS with higher incidence of significant

decreases to have a higher incidence of significant increases. This may indicate that

the cause of the increases or decreases is not a function of the DMS, but rather the

general heterogeneity of the traffic stream.

While the percentage of significant speed changes may suggest a problem

exists with respect to message display, the average changes in speed appear to

mitigate this concern. Overall, the average speed change for significant decreases is -

3.13 mph. Over a ten-minute period, this change in speed is unlikely to cause the

congestion reported by some users. Another important consideration is that in 82.9

percent of all cases, there is either no significant change in traffic speeds or there is a

significant increase in traffic speed.

69
Table 4.4. Off-On Summary by DMS
DMS # 839 3316 3317 4401 4403 8557 Total
Total Cases 96 74 151 215 101 205 842
# of Significant
Decreases 15 9 15 41 13 51 144
15.63 12.16 9.93 19.07 12.87 24.88 17.10
% Cases Significant % % % % % % %
Weighted Average
Decrease -1.80 -2.28 -5.90 -3.15 -3.30 -2.79 -3.13
# of Significant
Increases 19 6 13 24 11 28 101
19.79 8.11 8.61 11.16 10.89 13.66 12.00
% Cases Significant % % % % % % %
Weighted Average
Increase 1.89 2.92 2.85 2.51 3.50 1.69 2.34

Significant Speed Changes for Message

Activation
30.00%
25.00%
% Cases Significant

20.00%
15.00%
Significant Speed Decrease
10.00%
Significant Speed Increase
5.00%
0.00%
839 3316 3317 4401 4403 8557 Overall
DMS #

Figure 4.2. Graph of Off-On Summary by DMS

Because many of the concerns about the DMS messages stemmed from a

particular message type, namely travel time messages, the overall cases must be

broken down into more specific categories. Table 4.5 and Figure 4.3 show the off-on

cases disaggregated into message types.

70
Table 4.5. Off-On Summary by DMS and Message Type
DMS # 839 3316 3317 4401 4403 8557 Total
# Type 1 Cases 11 20 49 33 4 38 155
# of Significant
Decreases 2 5 6 12 0 9 34
18.18 25.00 12.24 36.36 0.00 23.68 21.94
% Significant % % % % % % %
# of Significant
Increases 1 2 1 2 0 2 8
9.09 10.00 2.04 6.06 0.00 5.26 5.16
% Significant % % % % % % %
# Type 2 Cases 84 35 45 127 68 167 526
# of Significant
Decreases 12 1 7 25 9 29 83
14.29 2.86 15.56 19.69 13.24 17.37 15.78
% Significant % % % % % % %
# of Significant
Increases 18 1 5 15 5 20 64
21.43 2.86 11.11 11.81 7.35 11.98 12.17
% Significant % % % % % % %
# Type 3 Cases 1 19 57 55 29 56 217
# of Significant
Decreases 1 3 2 4 4 13 27
15.79 13.79 23.21 12.44
% Significant 100.00% % 3.51% 7.27% % % %
# of Significant
Increases 0 3 7 7 6 6 29
15.79 12.28 12.73 20.69 10.71 13.36
% Significant 0.00% % % % % % %

71
 Percent of Significant Speed Changes by
Message Type for Message Activation
25.00%
% Significant Cases
20.00%
15.00%
10.00% Significant Decrease
5.00% Significant Increase
0.00%
1 2 3
Message Type

Figure 4.3. Graph of Off-On Summary by Message Type

When grouped this way, the data show the message type that causes

significant decreases in speed most often are Danger/Warning messages, followed by

Informative/Common Road Condition messages, and finally, Regulatory/Non-

Traffic-Related messages. This hierarchy is as expected because Danger/Warning

messages are commonly urgent and safety-related and should therefore draw the most

attention. Additionally, Danger/Warning messages usually indicate an incident that

would create congestion downstream, such as road closures or accidents.

Informative/Common Road Condition messages, which include travel-time messages,

are usually less urgent and caused a lower proportion of disruptions than

Danger/Warning messages, as expected. Interestingly, the two DMS in the study that

display travel-time messages, 839 and 3317, did not show an increase in the

percentage of significant cases of speed decrease. In fact, such data is lower than the

average, and DMS 839 is the only DMS that shows a higher percentage of significant

increases than significant decreases for Informative/Common Road Condition

messages. The data for Regulatory/Non-Traffic-Related messages indicate these

72
messages either go unnoticed, or users interpret them to mean there are no disruptions

ahead, both of which result in speed increases. This finding may be because

Regulatory/Non-Traffic-Related messages include information that most drivers

already know (e.g., seatbelt and cell phone laws).

The findings from the off-on analysis indicate that in the majority of cases,

traffic speeds are either unaffected or increase when messages appear on a DMS.

When traffic does respond negatively to the messages, the average decrease in speed

is just over three miles per hour. When broken down by message type, the data

showed that DMS that include travel-time messages do not produce higher fractions

of significant speed decreases than their counterparts.

On-Off

Similar to the off-on analysis, cases were examined for situations in which the DMS

switched from on to off. From this analysis, we find that traffic speed decreases

significantly in 11.98 percent of cases and increases significantly in 19.69 percent of

cases. This finding supports the general hypothesis that drivers will increase speeds as

a result of the removal of a message. Looking closer at the data (Table 4.6, Figure

4.4), we find that in four of the six DMS, the discrepancy between significant

increases and decreases is much smaller. In fact, in the two DMS where the difference

is quite large (i.e. 4401 & 8557), the differences in the off-on analysis were also

relatively large compared to the other four. One interpretation from this finding is that

those two locations have traffic streams that are much more sensitive to

environmental changes than other locations.

73
Table 4.6. On-Off Summary by DMS
DMS # 839 3316 3317 4401 4403 8557 Total
Total Cases 83 65 76 163 88 226 701
# of Significant
Decreases 9 10 14 10 11 30 84
10.84 15.38 18.42 6.13 12.50 13.27 11.98
% Cases Significant % % % % % % %
Weighted Average
Decrease -2.01 -2.35 -5.26 -2.70 -3.05 -1.63 -2.68
# of Significant
Increases 7 9 13 28 13 68 138
8.43 13.85 17.11 17.18 14.77 30.09 19.69
% Cases Significant % % % % % % %
Weighted Average
Increase 2.23 2.91 4.77 2.75 3.36 2.18 2.70

Significant Speed Changes for Message Removal

35.00%
30.00%
% Significant Cases

25.00%
20.00%
15.00% Significant Speed Decrease

10.00% Significant Speed Increase

5.00%
0.00%
839 3316 3317 4401 4403 8557 Overall
DMS #

Figure 4.4. Graph of Off-On Summary by DMS

74
 To see if these effects were a function of message type, a similar breakdown

was performed as in the off-on analysis. Table 4.7 and Figure 4.5 show the case

summary broken down by message categorization.

75
Table 4.7. On-Off Summary by DMS and Message Type
DMS # 839 3316 3317 4401 4403 8557 Total
# Type 1 Cases 9 18 30 20 3 12 92
# of Significant
Decreases 0 4 4 1 0 1 10
0.00 22.22 13.33 5.00 0.00 8.33 10.87
% Significant % % % % % % %
# of Significant
Increases 1 3 1 6 3 2 16
11.11 16.67 3.33 30.00 100.00 16.67 17.39
% Significant % % % % % % %
# Type 2 Cases 73 37 46 103 62 159 480
# of Significant
Decreases 9 6 6 6 8 23 58
12.33 16.22 13.04 5.83 12.90 14.47 12.08
% Significant % % % % % % %
# of Significant
Increases 6 4 6 17 6 57 96
8.22 10.81 13.04 16.50 9.68 35.85 20.00
% Significant % % % % % % %
# Type 3 Cases 1 10 40 40 23 55 169
# of Significant
Decreases 0 0 4 3 3 6 16
0.00 0.00 10.00 7.50 13.04 10.91 9.47
% Significant % % % % % % %
# of Significant
Increases 0 2 6 5 4 9 26
0.00 20.00 15.00 12.50 17.39 16.36 15.38
% Significant % % % % % % %

Percent of Significant Speed Changes by

Message Type for Message Removal
25.00%
% Significant Cases

20.00%
15.00%
10.00% Significant Decreases
5.00% Signficant Increases
0.00%
1 2 3
Message Type

Figure 4.5. Graph of On-Off Summary by Message Type

76
 The findings in this case are less clear. Although all three message types show

a higher percentage of cases where drivers significantly increased their speeds than

when drivers significantly decreased their speeds, differences among message types

are unclear. Informative/Common Road Condition messages show the highest rate of

significant speed increases, followed by Danger/Warning messages and finally

Regulatory/Non-Traffic-Related. This finding may indicate removing

Informative/Common Road Condition messages reduces drivers’ cognitive load.

Conversely, it could be interpreted that these messages relate to less severe influences

on the traffic stream, and thus, speeds are expected to recover more quickly when the

messages are no longer applicable. Similar arguments could be made for the other

two messages types. In the case of Regulatory/Non-Traffic-Related messages, the

relatively low percentage of significant increases is expected due to the low

percentage of significant decreases found in the previous analysis.

The findings from the on-off analysis indicate removing a message tended to

increase traffic speeds in approximately 20 percent of cases. The average speed

increase across these cases was 2.7 mph. Again, these results indicate in most cases,

traffic is unaffected by the messages displayed on DMS, and any influence on overall

traffic speeds is relatively small.

Switching

In many cases, especially on signs displaying travel-time messages, a DMS

message may be supplanted by a more important message, or reverted from an

applicable message to a default message. Analysis of these cases revealed drivers

77
significantly sped up in 13.52 percent of cases but significantly slowed down in 11.72

percent. The corresponding changes in speed were -3.14 and 2.44 mph. The similar

rates indicate the switching condition does not tend to influence traffic conditions one

way more than it does the other.

Table 4.8 and Figure 4.6 show that in four of the six cases, the rates of

significant increase and decrease are either identical or nearly so. The other two cases

show traffic tends to decrease speed in response to a change in message.

Table 4.8. Switching Summary by DMS

DMS # 839 3316 3317 4401 4403 8557 Total
Total Cases 76 146 93 259 68 83 725
# of Significant Decreases 16 18 8 30 10 16 98
21.05 12.33 8.60 11.58 14.71 19.28 13.52
% Cases Significant % % % % % % %
Weighted Average
Decrease -2.21 -2.56 -3.38 -2.60 -3.47 -5.41 -3.14
# of Significant Increases 9 17 8 30 4 17 85
11.84 11.64 8.60 11.58 5.88 20.48 11.72
% Cases Significant % % % % % % %
Weighted Average
Increase 2.11 1.97 2.97 2.28 4.50 2.62 2.44

Significant Speed Changes for Message Switching

25.00%

20.00%
% Cases Significant

15.00%

10.00% Significant Speed Decrease

Significant Speed Increase
5.00%

0.00%
839 3316 3317 4401 4403 8557 Overall
DMS #

Figure 4.6. Graph of Switching Summary by DMS

78
 In the message-switching condition, there are nine sub-conditions that can

occur: 1-1, 1-2, 1-3, 2-1, 2-2, 2-3, 3-1, 3-2, 3-3, where the first number is the starting

message type and the second number is the ending message type. (For example, a 2-1

condition could be a switch from a travel time message to an accident message. A

breakdown by these sub-conditions is made in Table 4.9. Overall, the percentage of

cases of significant increase and decrease are nearly the same. When examined over

each DMS, there do not appear to be any appreciable patterns within the data. This is

likely due to the low number of cases for each message-switching condition.

After examining 2,268 cases, the data indicate the majority of traffic streams

are unaffected by the display, removal or change of a DMS message. In the cases

where message initiation influenced a significant decrease in speed, drivers were most

sensitive to Danger/Warning messages, followed by Informative/Common Road

Condition messages and Regulatory/Non-Traffic-Related messages, respectively.

DMS locations that display travel-time messages were not found to be more sensitive

to message appearance than those that do not. In the on-off analysis, traffic was found

to speed up more often than it slowed down. It is not clear whether this was a result of

message removal or of dissipation of the conditions the message described. The

switching analysis indicated more evenly split results, indicating no appreciable bias

in speed impact during a change from one message to another.

79
Table 4.9. Switching Summary by DMS and Message Types
DMS # 839 3316 3317 4401 4403 8557 Total
# Type 11 Cases 1 3 7 1 1 0 13
# of Significant Decreases 0 1 0 0 0 0 1
% Cases Significant 0.00% 33.33% 0.00% 0.00% 0.00% - 7.69%
# of Significant Increases 0 0 2 0 0 0 2
% Cases Significant 0.00% 0.00% 28.57% 0.00% 0.00% - 15.38%
# Type 12 Cases 14 24 8 41 3 12 102
# of Significant Decreases 4 4 0 7 0 5 20
% Cases Significant 28.57% 16.67% 0.00% 17.07% 0.00% 41.67% 19.61%
# of Significant Increases 2 2 0 8 0 4 16
% Cases Significant 14.29% 8.33% 0.00% 19.51% 0.00% 33.33% 15.69%
# Type 13 Cases 0 5 11 11 3 6 36
# of Significant Decreases 0 0 1 2 0 0 3
% Cases Significant - 0.00% 9.09% 18.18% 0.00% 0.00% 8.33%
# of Significant Increases 0 2 1 0 0 0 3
% Cases Significant - 40.00% 9.09% 0.00% 0.00% 0.00% 8.33%
# Type 21 Cases 15 32 10 42 3 11 113
# of Significant Decreases 2 2 4 6 0 2 16
% Cases Significant 13.33% 6.25% 40.00% 14.29% 0.00% 18.18% 14.16%
# of Significant Increases 1 7 1 5 0 2 16
% Cases Significant 6.67% 21.88% 10.00% 11.90% 0.00% 18.18% 14.16%
# Type 22 Cases 35 28 16 92 21 35 227
# of Significant Decreases 8 5 1 6 4 5 29
% Cases Significant 22.86% 17.86% 6.25% 6.52% 19.05% 14.29% 12.78%
# of Significant Increases 3 2 1 10 3 9 28
% Cases Significant 8.57% 7.14% 6.25% 10.87% 14.29% 25.71% 12.33%
# Type 23 Cases 6 24 15 26 13 3 87
# of Significant Decreases 1 3 1 4 2 1 12
% Cases Significant 16.67% 12.50% 6.67% 15.38% 15.38% 33.33% 13.79%
# of Significant Increases 2 1 0 2 0 1 6
% Cases Significant 33.33% 4.17% 0.00% 7.69% 0.00% 33.33% 6.90%
# Type 31 Cases 0 6 17 11 4 5 43
# of Significant Decreases 0 1 1 2 1 1 6
% Cases Significant - 16.67% 5.88% 18.18% 25.00% 20.00% 13.95%
# of Significant Increases 0 0 1 1 0 0 2
% Cases Significant - 0.00% 5.88% 9.09% 0.00% 0.00% 4.65%
# Type 32 Cases 5 23 9 33 15 10 95
# of Significant Decreases 1 2 0 3 3 2 11
% Cases Significant 20.00% 8.70% 0.00% 9.09% 20.00% 20.00% 11.58%
# of Significant Increases 1 2 2 4 1 1 11
% Cases Significant 20.00% 8.70% 22.22% 12.12% 6.67% 10.00% 11.58%
# Type 33 Cases 0 1 0 2 5 1 9
# of Significant Decreases 0 0 0 0 1 0 1
% Cases Significant - 0.00% - 0.00% 20.00% 0.00% 11.11%
# of Significant Increases 0 1 0 0 0 0 1
% Cases Significant - 100.00% - 0.00% 0.00% 0.00% 11.11%

80
4.3.2: Aggregate Two Week Speed Analysis

For the same DMS used in the five-minute analysis, two two-week periods were

selected for aggregate analysis. These findings should indicate whether the display of

certain types of messages result in congestion. Figure 4.7 shows the 12 two-week

periods along with their average speeds under different message conditions. Figure

4.8 shows these values normalized over their corresponding overall average speeds.

For the most part, the trends show that traffic is most influenced by Type 1 messages.

Overall, Danger/Warning messages accounted for 1.5 percent of the total

study times, Informative/Common Road Condition messages for 34 percent and

Regulatory/Non-Traffic-Related messages for 12.5 percent. In the remaining time, the

signs were blank. The low proportion of time Danger/Warning messages were

displayed indicates these messages are unlikely to appear on a daily basis. Therefore,

drivers would not be used to the messages and may reduce speeds more to attend to

them. In two cases (DMS 3316 and DMS 3317 in January 2011), Regulatory/Non-

Traffic-Related messages seem to have had a significant impact on traffic speeds.

However, in both cases these messages were displayed for less than one percent of

the overall two-week period (approximately 1.5 hours each). This time-length

information indicates the messages in these cases could not have caused recurring

congestion.

Informative/Common Road Condition messages have the greatest potential to

create congestion because of the large proportion of time they are displayed. The

findings show speeds during these types of messages ranged from four mph below to

one mph above the speeds found when no messages were displayed. These speeds

81
indicate either light or no congestion during message displays. Also, only three of 12

cases yielded speeds more than one mph below the posted speed limit during the

times these types of messages were displayed. In two of these three cases, the overall

average speeds were already below the speed limit.

82
 Aggregate Average Speeds During 2 Week Periods for Selected DMS
70

60

50

40

30

83
Average Speed (mph)
20

10
839 Jan '11 839 July '10 3316 Jan '11 3316 May '10 3317 Jan '11 3317 May '10 4401 Jan '11 4401 May '10 4403 Dec '10 4403 Mar '10 8557 Jan '11 8557 May '10
Overall 58.8 56.06 53.43 48.93 55.64 52.21 56.43 56.81 57.24 60.19 60.16 61.22
No Message 59 55.99 53.3 49.87 56.61 51.67 57.67 56.52 56.02 60.53 61.59 61.79
All Message 58.67 56.25 53.66 48.87 52.66 52.79 54.06 56.98 57.8 60.09 56.15 57.95
Type 1 59.32 54.35 49.93 42.03 54.55 49.14 39.6 50.42 54.7 54.25 58.81 60.49
Type 2 58.66 56.29 54.54 47.47 53.15 49.22 54.37 56.77 54.6 60.13 56.02 57.21
Type 3 0 0 18 53.86 24.94 54.12 55.5 57.48 58.35 0 59.57 60.84

Figure 4.7. Aggregate Average Speeds for Two Week Analysis

 Aggregate Speeds Relative to Overall Average During 2 Week Period for Selected DMS
1.2

0.8
No Message
All Message
0.6

84
Type 1
Type 2
0.4 Type 3

Fraction of Overall Average Speed

0.2

0
839 Jan '11 839 July '10 3316 Jan '11 3316 May '10 3317 Jan '11 3317 May '10 4401 Jan '11 4401 May '10 4403 Dec '10 4403 Mar '10 8557 Jan '11 8557 May '10

Figure 4.8. Aggregate Speeds Normalized by Overall Average Speeds

 In summary, the aggregate analysis shows average speeds during

Danger/Warning message display are generally lower than speeds during blank sign

conditions. However, the occurrence of these messages is rare and therefore could not

be the cause of recurring congestion. Informative/Common Road Condition messages

are displayed more often and in some cases resulted in decreased traffic speeds. In the

majority of the cases, however, the speeds were not below the posted speed limit.

Speeds during Regulatory/Non-Traffic-Related messages are usually higher than the

other message types. In the cases where they were much lower, they accounted for

less than one percent of the overall study time.

85
Chapter 5: Conclusions and Future Work

This project presented empirical evaluations of the quality, effectiveness and

localized effects of highway Dynamic Message Signs (DMS). This project used

Bluetooth sensor technology as a new method for evaluating DMS messages’

accuracy and influence on motorist behavior. Two sensor deployments were

undertaken for this purpose and several message cases were selected from each for

evaluation. To determine whether DMS messages cause localized effects (i.e., drivers

change speed), 2,268 cases of message activation, removal and switching were

analyzed using RTMS speed data. In addition, the cases were sorted into categories to

determine if any trends exist with respect to message types.

The first deployment revealed that the Bluetooth data was an effective tool for

evaluating DMS messages. It was determined the messages being displayed

accurately described many of the prevailing conditions, although they suffered from

late display and removal of messages. In addition, the messages used vague location

descriptors, giving drivers no indication of where traffic disruptions were occurring.

The second deployment confirmed the effectiveness and repeatability of

Bluetooth traffic detection as a tool to evaluate DMS messages. The second

deployment found the DMS system had improved. Messages used more specific

terms to describe congestion locations. The messages also provided travel-time

messages. The travel-time messages alleviated some of the concerns with late

message display because motorists could observe travel times increasing in response

to congestion. In some cases, messages were left on longer than necessary, although

there were some mitigating circumstances.

86
 In addition to evaluating congestion and delay messages, Bluetooth travel

times were used to validate the travel times displayed on the DMS. Analyses of

nearly 24 hours of data from two travel time cases revealed the average difference

between the displayed and true travel times was less than one mile per hour. This

confirms the source data and updating system used on DMS in Maryland is high

quality. Nevertheless, the Bluetooth travel time evaluation is applicable to any DMS

travel time system independent of the source of the DMS travel-time data.

To determine the effectiveness of DMS messages, counts of Bluetooth

detections on alternate routes suggested by the messages were compared. Analyses of

three cases showed a 5-20 percent increase of traffic diversion rates on alternate

routes during periods when DMS messages recommended drivers take those alternate

routes. It can be inferred from this finding that DMS messages influence drivers’

route choices. However, because of the sampling rate of Bluetooth sensors, this

conclusion must be tentative. Even with this caveat, Bluetooth detection can be used

as a powerful tool for evaluation of traffic diversion.

Evaluation of RTMS speed detector data in proximity to DMS revealed that in

some cases, traffic streams do decrease speed in response to message activation.

Danger/Warning message displays were associated with the highest percentage of

speed decreases. An aggregate analysis showed that overall traffic speeds were slower

during Danger/Warning messages, although these messages appeared infrequently.

Similar analyses on message removal and switching were performed. Overall, the

majority of traffic streams either increased in speed or did not change speed in

response to DMS messages. The most frequently appearing message was

87
Informative/Common Road Condition messages. These messages tended to

correspond with lower traffic speeds than periods with blank signs. It is unclear

whether these reduced traffic speeds are a result of the message display or of the

conditions (e.g. accident ahead) to which the messages correspond.

In summary, the findings from these evaluations indicate DMS can be an

accurate, effective, and safe tool for disseminating real-time travel information to

motorists. This project focused on Maryland DMS, so the findings may not extend to

DMS operations in other states. Nevertheless, the methods employed for evaluation

are extendable without regard for the DMS location.

In the future, Bluetooth sensors can be used to evaluate DMS in other

locations throughout Maryland and other states. More deployments will strengthen

the reputation of Bluetooth as a DMS evaluation tool and will build broader

knowledge about DMS operations. In order to validate diversion patterns observed

through Bluetooth detection, a deployment could be undertaken in parallel with

license plate-reading technology. If the findings from such a study showed a strong

correlation between Bluetooth and license plate detections, the use of Bluetooth for

tracking origin destination and diversion would be strengthened.

To further investigate the localized effects of DMS on traffic streams, the use

of small, portable speed sensors could be employed. By spacing such sensors at fixed

intervals in proximity to DMS, speed profiles could be built as vehicles approach the

signs. In addition, accident data could be analyzed to determine if there are any

indications that the existence of a DMS results in higher accident rates.

88
 The methods employed in this project can be applied to any DMS operation

and could be used to evaluate locations before and after installation of DMS. It would

be informative to learn what effect the installation of a DMS has on a traffic stream in

terms of travel time and diversion patterns.

Ultimately, findings from future studies can be used to calibrate traffic

simulations and build automated message display and incident detection systems.

These technologies would help transportation engineers and planners improve DMS

operations and in turn overall network conditions. The broad range of future study

will provide challenges and opportunities for many researchers in the coming years.

89
Bibliography
1. Agah, M. (2002). Guidelines on the Use of Permanent Variable Message
Signs. Arizona Department of Transportation, Transportation Technology
Group. Retrieved June 22, 2009, from TMC Pooled Fund Study Web Page:
http://tmcpfs.ops.fhwa.dot.gov/cfprojects/new_detail.cfm?id=25&new=2

2. Federal Highway Administration. (2003). Manual on Uniform Traffic Control

Devices (MUTCD). Sec. 6F.55

3. Dudek, C. L. (2004a). Changeable Message Sign Operation and Messaging

Handbook. Report No. FHWA-OP-03-070, FHWA, United States Department
of Transportation, pp. 2-3.

4. Minnesota Department of Transportation. (2000). Guidelines for Changeable

Message Sign (CMS) Use. Retrieved June 9, 2009, from MnDOT publications
web site: http://www.dot.state.mn.us/trafficeng/otepubl/index.html

5. Miller, J., Smith, B., Newman, B, Demetsky, M. (1995). Development of

Manuals for the Effective Use of Variable Message Signs. Virginia
Transportation Research Council Report # VTRC 95-R15, Charlottesville,
Virginia.

6. Ridgeway, R. Framework for harmonized Implementation of Variable

MessageSigns in Europe. Rijkswaterstaat, The Netherlands, 2003.

7. Dudek, C. L. (2004b). Dynamic Message Sign Design and Display Manual.

Report No. FHWA/TX-04/0-4023-P1, Texas Department of Transportation,
pp. 4-11.

8. Wardman, M., Bonsall, P. W., Shires, J. D. (1997). Driver Response to

Variable Message Signs: A Stated Preference Investigation. Transportation
Research-C, Vol. 5, No. 6, pp. 389-405.

9. Monsere, C., Breakstone, A., Bertini, R., Deeter, D., McGill, G. (2007).
Validating Dynamic Message Sign Freeway Travel Times with Ground Truth
Geospatial Data. Transportation Research Record 1959, TRB, National
Research Council, Washington, D.C., pp. 19-27.

90
10. Chen, C., Skabardonis, A., Varaiya, P. (2003). A System for Displaying Travel
Times on Changeable Message Signs. 83rd Annual Meeting of the
Transportation Research Board, Washington, D.C.

11. Peng, Z. R., Guequierre, N, Blakeman, J. C. (2004) Motorist Response to

Arterial Variable Message Signs. Transportation Research Record 1899, TRB,
National Research Council, Washington, D.C., pp. 55-63.

12. Chen, S., Liu, M, Gao, L, Meng, C., Li, W., Zheng, J. (2008) Effects of
Variable Message Signs (VMS) for Improving Congestions. IEEE
International Workshop on Modeling, Simulation and Optimization. Paper #
978-0-7695-3484-8/08, pp. 416-419

13. Cheng, J. J., Firmin, P. E. (2004) Perception of VMS Effectiveness: A British

and Canadian Perspective. 12th IEE International Conference on Road
Transport Information and Control 20th-22nd April 2004, The IEE, Savoy
Place, London.

14. Polydoropoulou, A., Ben-Akiva, M., Khattak, A., Lauprete, G. (1996)

Commuters’ Normal and Shift Decisions in Unexpected Congestion: En Route
Responses to Advanced Traveler Information Systems Volume 2. California
PATH Research Report UCB-ITS-PRR-96-21, Institute of Transportation
Studies, University of California, Berkeley.

15. Peeta, S., Ramos, J. L., Pasupathy, R. (2000) Content of Variable Message
Signs and On-Line Driver Behavior. Transportation Research Record 1725,
TRB, National Research Council, Washington, D.C., paper no. 00-0970

16. Schiesel, R., Demetsky, R. J. (2000) Evaluation of Traveler Diversion Due to

En-Route Information. Center for Transportation Studies Report
UVA/29472/CE00/103, University of Virginia, Charlottesville, Virginia

17. Foo, S., Abdulhai, B., Hall, F. L. (2008) Impact on Traffic Diversion Rates of
Changed Message on Changeable Message Sign. Transportation Research
Record: Journal of the Transportation Research Board, No. 2047,
Transportation Research Board of the National Academies, Washington, DC

91
18. Levinson, D and Huo, Hong (2003), Effectiveness of VMS Using Empirical
Loop Detector Data presented at Transportation Research Board Conference,
January 12 – 16 2003 Washington DC (Session 565).

19. Boyle, L.N., and F. Mannering (2003). Impact of Traveler Advisory Systems
on DrivingSpeed: Some New Evidence, Transportation Research – C, Vol.12,
pp. 57-72.

20. Alm, H. and L. Nilsson. Incident Warning systems and traffic safety: a
comparison between the PORTICO and MELYSSA test site systems.
Transportation Human Factors, Vol.2, No.1, 2000, pp. 77-93.

21. Luoma, J., Rama, P., Penttinen, M., Anttila, V. (2000), Effects of variable
message signs for slippery road conditions on reported driver behavior.
Transportation Research – F, Vol. 3, pp. 75-84.

22. Erke, A., Sagberg, F., and R. Hagman (2007), Effects of route guidance
variable message signs (VMS) on driver behavior. Transportation Research –
F, Vol. 10, pp. 447-457.

23. Wang, J.H., Keceli, M., and V. Maier-Speredelozzi (2009), Effect of Dynamic
Message Sign Messages on Traffic Slowdowns. Transportation Research
Board Annual Meeting 2009, Paper #09-1964.

24. Traffic Technology Today, Traffax deploys its BluFax real-time traffic
monitoring systems in Arizona and Maryland (2011),
http://cms.ukintpress.com/UserFiles/Image/TTT%20images/BluFax-unit.jpg

25. University of Maryland Center for Advanced Transportation Technology,

Bluetooth Traffic Monitoring Technology – Concept of Operation and
Deployment Guidelines (2008), http://www.catt.umd.edu/documents/UMD-
BT-Brochure_REV3.pdf

26. Haghani, A., M. Hamedi, K.F. Sadabadi, S. Young, and P. Tarnoff (2010)
Freeway Travel Time Ground Truth Data Collection Using Bluetooth Sensors.
Transportation Research Board of the National Academies, Annual Meeting,
Washington, D.C.,

92
 STATE HIGHWAY ADMINISTRATION

RESEARCH REPORT

EVALUATION OF DYNAMIC MESSAGE SIGNS AND

THEIR POTENTIAL IMPACT ON TRAFFIC FLOW

Volume II: Impact of Dynamic Message Signs on Occurrence of Road Accidents

ALI HAGHANI
MASOUD HAMEDI
ROBIN L. FISH
AZADEH NOURUZI

UNIVERSITY OF MARYLAND, COLLEGE PARK

Project number SP109B4C

FINAL REPORT

February 2012
 Abstract

Dynamic Message Signs (DMS) are key components of Advanced Traveler Information

Systems (ATIS). DMS is used to manage transportation networks, reduce congestion and

improve safety by providing motorists with real-time information regarding downstream

traffic conditions. Although DMS are intended to improve efficiency and safety of road

networks, there have been few studies on the effect of the signs on driver safety or the

signs’ localized safety effects. This project employs ground truth data as the basis to

investigate these matters in Maryland over a four-year period (2007-2010). The results

show no significant difference between the accident pattern in the proximity of DMS and

onward adjacent segments. An on-and-off study was also conducted on DMS operation

status (on/off). These results converge with the previous analysis to suggest there is no

meaningful relationship between occurrence of accidents and presence of DMS.

Statistical analysis on DMS characteristics and accidents in impact areas were performed.

2
 Table of Contents

Chapter 1: Introduction ....................................................................................................... 7

1.1. Research Motivation and Objectives ................................................................ 7
1.2. Organization of the report ........................................................................................ 9
Chapter 2: Background and Literature Review ................................................................ 11
2.1. Dynamic Messages Signs ............................................................................... 11
2.2. DMS Process and Operations ......................................................................... 11
2.3. DMS Types ..................................................................................................... 12
2.3.1. Portable vs. Permanent Signs .................................................................... 12
2.3.2. Dynamic Features ..................................................................................... 13
2.4. Message Types ................................................................................................ 13
2.5. Danger/Warning Messages ............................................................................. 15
2.5.1. Incident Messages ..................................................................................... 15
2.5.2. Road and Vehicle Unpredicted Condition Warning Messages................. 15
2.6. Informative/Common Road Condition Messages ........................................... 15
2.6.1. Travel Time Messages .............................................................................. 15
2.6.2. Congestion Messages ................................................................................ 16
2.6.3. Queue Warning Messages......................................................................... 16
2.6.4. Weather-Related Messages ....................................................................... 17
2.6.5. Railroad Crossing Messages ..................................................................... 17
2.7. Regulatory/Non-Traffic Related Messages..................................................... 18
2.7.1. Public Service Announcement Messages ................................................. 18
2.7.2. AMBER Alerts.......................................................................................... 18
2.8. Inappropriate uses of DMS ............................................................................. 19
2.8.1. Traffic-Related Messages ......................................................................... 20
2.8.2. Non Traffic Related Messages .................................................................. 21
2.8.3. Reasons Motorists Disregard the DMS..................................................... 21
2.9. Location of Dynamic Message Signs ............................................................. 22
2.10. DMS Performance Metrics ............................................................................. 24
2.11. Studies Related to Designs of DMS................................................................ 24
Chapter 3: Driver Response Behavior to Messages and Localized Impact of DMS ........ 27
3.1. Drivers’ Response to Displayed Messages ..................................................... 27
3.1.1. Route Diversion in Response to Messages ............................................... 29
3.1.2. Speed Reduction in Response to Messages .............................................. 32
3.2. Effect of DMS Design on Driver Response .................................................... 32
3.2.1. Text-based and Graphic-aided Messages.................................................. 32
3.2.2. Flashing and Static Messages ................................................................... 34
3.3. Localized Impact of DMS ............................................................................... 34
3.3.1. Traffic Speed Slow Down for Perception of Messages ............................ 34
3.3.2. Driver Distraction and Collision Occurrence ........................................... 37
3.4. Summary ......................................................................................................... 38

3
Chapter 4: Investigation on Possible Relationship between DMS and Occurrence of Road
Accidents......................................................................................................... 43
4.1. Problem Statement and Motivation for Research ........................................... 43
4.2. Methodology ................................................................................................... 44
4.2.1. Data Sources and Preparation ................................................................... 44
4.2.2. Accident Database .................................................................................... 45
4.2.3. DMS Database .......................................................................................... 47
4.2.4. AADT Database ........................................................................................ 49
4.3. Data Processing and Preparation Challenges .................................................. 50
4.4. Defining the Impact Area of DMS.................................................................. 51
4.5. Case Study on I-95 .......................................................................................... 51
4.5.1. Analysis of Case Study and Preliminary Results ...................................... 53
4.6. Weather Conditions Database ......................................................................... 64
4.7. Log of Messages Database.............................................................................. 67
4.8. Analysis and Results ....................................................................................... 69
4.9. Analysis on Impact Areas and Following Segment ........................................ 70
4.9.1. Findings..................................................................................................... 77
4.10. On-and-off Analysis........................................................................................ 80
4.10.1. Findings..................................................................................................... 83
4.11. Accidents in DMS Impact Areas and Weather Conditions ............................ 85
4.12. Accidents in DMS Impact Areas and DMS Characteristics ........................... 88
Chapter 5: Conclusions and Directions for Further Research .......................................... 91
5.1. Summary and Conclusions ............................................................................. 91
5.2. Future Research .............................................................................................. 95
References ......................................................................................................................... 96

4
List of Tables

Table 2.1. Message Categorization ................................................................................... 14

Table 2.2. Example Performance Indicators for Dynamic Message Signs....................... 25
Table 3.1. Results from driver surveys (Wendelboe, 2008) ............................................. 28
Table 3.2. Literature Summary on Driver Response to Diversion Messages ................... 39
Table 3.3. Literature Summary on Driver Response to Speed Reduction Messages ...... 41
Table 3.4. Driver Distraction and Speed Slow Down for Perception of Messages .......... 42
Table 4.1. Variables used in case study ............................................................................ 55
Table 4.2. I-95 Case Study Samples ................................................................................. 56
Table 4.3. Tower stations assigned to each weather region .............................................. 64
Table 4.4. Tabulated facts of impact areas and forwarding segments .............................. 75
Table 4.5. Tabulated facts of on and off study ................................................................. 81
Table 4.6. Accidents in DMS areas and precipitation ...................................................... 85
Table 4.7. DMS accidents and wind gust ......................................................................... 86

5
List of Figures

Figure 2.1. Permanent vs. portable DMS .......................................................................... 13

Figure 2.2. A Queue Warning Message ............................................................................ 17
Figure 4.1. The databases and sources of data used in the research ................................. 44
Figure 4.2. Study Area ...................................................................................................... 45
Figure 4.3. First shape of accident data and pointing location of accidents on road
network map...................................................................................................................... 46
Figure 4.4. First shape of DMS database and projection to road map .............................. 48
Figure 4.5. Map of accidents and DMS locations ............................................................. 49
Figure 4.6. An example of the volume map AADT (SHA 2011) ..................................... 50
Figure 4.7. Impact Area .................................................................................................... 52
Figure 4.8. I-95 along with the DMSs along this highway ............................................... 52
Figure 4.9. Accidents in I-95 ............................................................................................ 53
Figure 4.10 . Projection of AADTs to road map............................................................... 54
Figure 4.11. Multiple Buffers along I-95 .......................................................................... 55
Figure 4.12. SAS outcomes of unbalanced two-way ANOVA for case study in I-95 ..... 58
Figure 4.13. SAS outcomes of Poisson regression for case study in I-95 ........................ 60
Figure 4.14. SAS outcomes of Negative Binomial regression for case study in I-95 ...... 62
Figure 4.15. Weather Database Format ............................................................................ 66
Figure 4.16. Log of Messages Database ........................................................................... 68
Figure 4.17. Projection of Integrated Database ................................................................ 71
Figure 4.18. Close up shot of projected map .................................................................... 72
Figure 4.19. Accident rate for impact area of 900 feet compared to their subsequent900
feet segment ...................................................................................................................... 74
Figure 4.20. Difference of the accidents rates between the impact area and its subsequent
segment ............................................................................................................................. 76
Figure 4.21. SAS outcomes for comparison of impact areas and following section ........ 78
Figure 4.22. Comparison of accident rates while DMS are on and while blank .............. 82
Figure 4.23. Difference of the accidents rates in on and off study ................................... 82
Figure 4.24. SAS outcomes for on and off study.............................................................. 83
Figure 4.25. Frequency of accidents in different precipitation conditions.................... 86
Figure 4.26. DMS accidents and wind gust ...................................................................... 87
Figure 4.27. Type of accidents in DMS area # ................................................................. 88
Figure 4.28. Number of accidents versus Beacon status .................................................. 89
Figure 4.29. Number of accidents for DMS message types ............................................. 90

6
1. Chapter 1: Introduction

1.1. Research Motivation and Objectives

Increasing traffic volumes over recent decades is the compelling motivation to manage

transportation networks, increase capacity, enhance communication capabilities of

transportation systems, improve safety and reduce congestion. Physically increasing the

capacity of roadways and arterials by adding lanes is usually not economically and

environmentally justified, and it is an ineffective long-term solution. One of the most

popular alternate strategies is to provide travelers with real-time information about

downstream traffic conditions using Advanced Traveler Information Systems (ATIS).

Two of the main technologies employed in the ATIS effort are Highway Advisory Radio

(HAR) and Dynamic Message Signs (DMS). DMS are often regarded as the most visible

form of ATIS because they are available equally to all motorists. Maryland State

Highway Administration’s (SHA) Coordinated Highways Action Response Team

(CHART) operates 184 DMS. The signs, located on major highways and arterials, are

often used to inform motorists of delays, incidents, road closings and real-time travel

estimates. The most popular types of messages displayed on DMS are weather

conditions, travel time, construction information, speed limits, incident locations and

various other public service announcements, including AMBER alerts. Although DMS

are intended to improve the efficiency and safety of road networks, little has been done to

study whether the signs affect driver safety. The purpose of this study is to determine if

drivers exposed to DMS are distracted by what the signs display and, if so, whether that

distraction leads to their involvement in traffic accidents.

7
Accident data and log of messages data in the study period were acquired from the Center

for Advanced Transportation Technology (CATT) Laboratory at the University of

Maryland, College Park and from Coordinated Highway Action Response Team

(CHART) reports. Quality control and consistency check were conducted on the

database. The DMS inventory was obtained from the CATT Laboratory. The DMS types

in this investigation include permanently-mounted overhead signs, roadside models and

portable signs operated by CHART or Maryland Transportation Authority (MdTA). The

roadway network map and AADT of roadway segments were obtained from Maryland

Department of Transportation and the SHA, and weather conditions databases were acquired

from DOT archival data.

The locations of accidents and DMS and the AADT data were projected onto a Maryland

roadway map in ArcGIS 10.1. An impact area was defined to perform spot analysis to

evaluate whether DMS influences drivers’ operational performance.

A case study was performed on a portion of Interstate 95 in Maryland, a roadway

regarded as a major highway. A sample of 70 road segments was chosen based on

homogeneity of their geometry. Regression analysis was performed; relevant predictor

variables included the segment’s status as an impact area (yes/no), whether the segment

connects with interchanges (yes/no) and the segment’s AADT. Additionally, an

unbalanced two-way ANOVA was used to compare mean accident rates in impact areas

and other segments.

The study area was divided into five regions, and the nearest central weather

station in each region represented the weather condition in each region. The weather

database was aggregated for the four-year study period and was joined to the main

8
database based on proximity of weather station to the time and location of the accident.

The matching process was performed using SQL queries coded in C++.

The message log database was imported in the SQL server and the main database.

If an accident was in an impact area, the assigned DMS was matched with the message

displayed at the time of the accident. This matching process was conducted using SQL

queries coded in C++.

The integrated database was analyzed in several respects. Accident rates in DMS

impact areas and adjacent segments were compared using paired t-tests in order to

determine the effects of DMS on accidents occurrence.

An on-and-off study compared the results from the previous study. The difference

in accident rates was tested on two DMS operation statuses (message display/blank)

using a one-way ANOVA with pairwise comparison tests. Statistical analyses on DMS

characteristics, message types, weather conditions and accidents in the impact areas were

performed.

The findings of this project and the methods used to obtain them are widely applicable so

state officials and transportation and ITS agencies can analyze, evaluate and ultimately

improve their DMS operations. Although this project focused on DMS operations in

Maryland, the methods employed for evaluation can be applied to other locations.

1.2. Organization of the report

This report has five chapters. Chapter 2 reviews the literature on DMS operations, design

and type of messages displayed on DMS. Chapter 3 discusses driver behavior, drivers’

response to messages and localized safety effects of these signs. It also provides a

comprehensive review on study methods and research to evaluate effectiveness and

9
safety effects of DMS. Chapter 4 investigates the relationship between DMS and

occurrence of road accidents. It also describes the motivation, methodology, analyses and

results of this project. Conclusions and suggestions for future research are found in

Chapter 5.

10
2. Chapter 2: Background and Literature Review

2.1. Dynamic Messages Signs

The Maryland Manual on Uniform Traffic Control Devices defines a Dynamic

Message Sign as “a sign that is capable of displaying more than one message, changeable

manually, by remote control or by automatic control.” These signs are called Dynamic

Message Signs in the National Intelligent Transportation Systems (ITS) Architecture.

DMS are also known as Variable Message Signs (VMS) or Changeable Message Signs

(CMS) and can be used by transportation authorities and operating agencies to

disseminate travel information on a near real-time basis.

DMS are valuable instruments. The Deployment Tracking Database of Federal

Highway Administration has data indicating that more than $330 million has been spent

on deploying DMS in the United States (Dudek, 2008). The main goal of DMS is to

enhance motorist safety and provide real-time traffic information to motorists, thus

allowing them to make intelligent travel decisions.

2.2. DMS Process and Operations

The information displayed on DMS is gathered from a variety of traffic monitoring and

surveillance systems, including video detection systems, loop detectors, automatic

vehicle identification transponders and toll tags. All data are reported to Traffic

Management Centers (TMC). Travel-time messages are derived from applying an

algorithm that calculates distance covered in order to determine the estimated travel times

from a DMS to a specific destination. The destination is usually a major intersection or

interchange. In most jurisdictions, travel-time information is posted during morning and

evening peak travel times; the system is usually timed to begin and end at a certain time

11
of day. The TMC operator is responsible for monitoring and interpreting messages as

well as making decisions about posting the messages.

2.3. DMS Types

Dynamic Message Signs can be divided into two types regarding method of

installation: permanent and portable. DMS may also be equipped with beacons and/or

flashing messages.

2.3.1. Portable vs. Permanent Signs

DMS can be permanent (overhead or roadside) or portable. Both permanent and portable

DMS are used to manage incidents and inform motorists. Permanent DMS can be

installed above arterials, highways, bridges, tunnels and toll plazas. Portable truck-or-

trailer-mounted DMS can be dispatched by highway agencies to warn drivers of incidents

(e.g., accidents or work zones) in areas where permanent DMS are not practical or

available. Trailer-mounted DMS are used to alter traffic patterns near work zones and to

manage traffic during special occasions (e.g., sporting events and natural disasters) that

necessitate temporary changes in normal traffic patterns. Most manufacturers produce

trailers that comply with the National Transportation Communications for Intelligent

Transportation System Protocol (NTCIP), which allow the portable trailer to be

integrated with intelligent transportation systems. Trailer-mounted DMS can be equipped

with radar, cameras, and other sensing devices as part of a smart work zone deployment.

Figure 2.1 shows permanent and portable DMS signs.

12
 Figure 2.1. Permanent vs. portable DMS

2.3.2. Dynamic Features

DMS can be equipped with flashing beacons, which are typically installed on top

of the message panel. Beacons are usually yellow in color and meet NTCIP requirements

for size and shape. Messages displayed on DMS can also flash or blink and are used in

areas such as school zones. However, flashing line messages may have an adverse effect

on message comprehension of messages (Dudek, 2005), so these types of messages are

not very common.

2.4. Message Types

DMS warn motorists about different situations and provide real-time information

about traffic, roadway and environmental conditions, location and expected duration of

incident-related delays, alternate routes for a roadway closure, redirected routes for

diverted drivers and traversable shoulders in the event of a major incident to restore the

traffic flow safely (Farradyne, 2000). However, DMS are primarily used to display the

following messages (Dudek, 2008):

13
 - Random and unpredictable situations such as crashes, stalled vehicles and spilled

loads

- Temporary and preplanned activities such as construction, maintenance or utility

operations

- Adverse environmental situations such as fog, floods, ice and snow

- Special events such as road closures for sport games and parades

- Traffic flow operational initiatives such as high occupancy, reversible, exclusive or

contraflow lanes

- Certain design features such as drawbridges, tunnels and ferry services

- Travel-time information

- AMBER (America’s Missing: Broadcast Emergency Response) alerts to help locate

missing people

Ridgeway categorizes messages into three types: Danger/ Warning Messages,

Informative/Common Road Conditions and Regulatory/Non-Traffic Related messages.

Table 2.1 shows the type and example of messages in this classification.

Table 2.1. Message Categorization

Message Category Examples of Displayed Messages

Type 1: Danger/Warning Incidents, Disabled Vehicles, Non-recurring Slow-
Downs, Roadway Debris, Unplanned Lane/Tunnel/
Bridge Closures
Type 2: Informative/Common Roadwork Closures, Major & Minor Delays, Congestion,
Road Condition Travel Time, Other travel-related messages (Fog, Ice,
Snow Plowing, Major Events)
Type 3: Regulatory/Non-Traffic Work Zone Speeds, Seatbelt Use, Cell Phone Regulations,
Related Motorcycle Awareness, Amber Alerts, Homeland
Security Messages

14
2.5. Danger/Warning Messages

2.5.1. Incident Messages

One of the main functions of DMS is to alert motorists of lane closures that are

the result of unexpected situations, such as accidents or other incidents that reduce

roadway capacity. Messages can be displayed to warn of a traffic incident. However, no

message should be displayed if the sign is so far from the affected area that full capacity

will be restored before motorists who read the sign would be affected by the disruption

the sign warned them of. Conversely, if the incident is confined to an adjoining route

such that motorists in that route would be affected, a message should be displayed.

Depending on the location, severity and duration of the incident, messages may be

displayed up to several hundred miles in advance of the incident. If a situation arises

where multiple incidents have occurred downstream from a sign, DMS should alert

motorists to the closest incident unless conditions warrant otherwise (NJDOT, 2008).

2.5.2. Road and Vehicle Unpredicted Condition Warning Messages

These messages may inform drivers of special issues with respect to road and

vehicle conditions, including changes in roadway alignment or surface conditions,

disabled vehicles, vehicle restrictions and advance notice of new traffic-control device

installation (Walton et al., 2001).

2.6. Informative/Common Road Condition Messages

2.6.1. Travel Time Messages

Travel-time messages inform drivers any of the five following categories:

1. Travel time on freeways displays the number of minutes required to go from

one specified location to another

15
 2. Comparative travel times on the freeway and an alternate route

3. Time saved by taking an alternate route

4. Delays on the freeway

5. Delays avoided by taking the alternate route

2.6.2. Congestion Messages

DMS may be used to display information on traffic conditions when freeways

become congested. However, a problem arises because a multitude of situations exist

that may be difficult to describe on DMS. In jurisdictions where quantitative travel-time

information is not available, terms such as “Heavy Delay” and “Major Delay” are often

used. However, little information or guidance exists about how these terms should be

defined. The Dynamic Message Sign Message Design and Display Manual indicates the

average motorist in Texas interprets “Heavy Delay” as being between 25 and 45 minutes,

whereas “Major Delay” is interpreted as a delay greater than 45 minutes. A similar study in

England attempted to determine driver response to DMS; English drivers reported they

interpreted “Long Delays” as being between 35 and 47 minutes, whereas “Delays Likely”

indicated a 10 to 31 minute delay. The Minnesota Department of Transportation’s

Guidelines for Changeable Message Sign Use specifies a “Major Delay” is not indicative of

an amount of time but rather an incident causing more than 2 miles of traffic backup. These

conflicting definitions demonstrate the need for an evaluation of DMS messages and the

conditions to which they correspond (Fish et al., 2012).

2.6.3. Queue Warning Messages

Queue warning messages have been employed on several German motorways. Queue

warning messages vary in appearance, scope and complexity. A queue warning system

uses a small roadside DMS with flashers to indicate the length and location of the queue.

16
Germany Transportation Policy strongly emphasizes the comprehensive communication

of the queue warning on the message signs by using minimal wording and simple

imagery. The German queue warning system yielded benefits including fewer incidents,

reduced incident severity, closer headways, greater uniformity on all driver speeds and a

slight increase in capacity (Bolte, 2006). Figure 2.2 depicts a dynamic queue warning

message sign.

Figure 2.2. A Queue Warning Message

2.6.4. Weather-Related Messages

One of most common uses of DMS is to display weather information that affects

traffic. DMS advise motorists of severe weather or environmental conditions in the area,

especially situations that require a change in their driving behavior (NCDOT, 1996;

ORDOT, 2000).

2.6.5. Railroad Crossing Messages

DMS may also be applied where roadways and railroads meet. Finely et al. (2001)

argued that because traffic conditions can also be affected by rail systems, railroad grade

crossing information should be available via DMS. An example of application of DMS in

17
railroad crossing area is in San Antonio, where displaying real-time information on these

messages allows drivers to alter their routes to avoid a lengthy wait for a crossing train.

2.7. Regulatory/Non-Traffic Related Messages

2.7.1. Public Service Announcement Messages

The use of DMS for Public Service Announcements (PSA) is allowed by some

agencies. The types of PSAs permitted depend on the jurisdiction. PSAs may include

brief messages that do not require an immediate response but encourage drivers to alter

future driving behavior. Because PSAs do not provide drivers with real-time safety or

travel efficiency information and are not usually associated with any urgent response,

these messages are generally given low priority. PSAs provide motorists with information

that can be given more effectively through other methods such as media campaigns or

pamphlets (NCHRP, 2008). Another argument against displaying PSA messages focuses

on the concern that motorists who continually travel a specific route will become

accustomed to messages and will begin to ignore all DMS. For example, in State of

Oregon Department of Transportation the very lowest priority is given to PSAs. Oregon’s

DOT allows DMS to display PSAs only in off-peak periods for a maximum of five hours

a day and five days a month. In addition, these messages are generally restricted to

permanent DMS and not permitted on portable DMS (ORDOT, 2000).

2.7.2. AMBER Alerts

AMBER alerts are notification programs to help locate missing children who

authorities believe have been abducted. The Emergency Alert System (formerly known as

the Emergency Broadcast System) alerts the public about these children’s abductions by

means of television and radio (NCHRP, 2008). America’s AMBER Plan Program,

18
through which emergency alerts are issued to notify the public, is voluntary. The Federal

Highway Administration notes DMS are not always the safest or most effective methods

of disseminating information about child abductions because only a limited amount of

information can be conveyed on DMS. When there is a need to provide extensive

information to motorists, the Federal Highway Administration specifies that other types

of traveler information media (e.g., 511, HAR, informative Web sites or commercial

radio) should be used and that DMS should only supplement those media.

2.8. Inappropriate uses of DMS

A national policy specifying DMS use and message design does not exist.

Transportation authorities are responsible for creating and implementing their own

guidelines on the use, location, operation and evaluation of DMS in their jurisdictions.

Mounce et al. (2007) assessed current DMS applications and practices based on a

national literature reviews and agency surveys. They found the majority of respondents in

the survey believed one major benefit of DMS was to provide timely and important

information about travel routes. The survey revealed that although most DMS

applications are considered effective, respondents indicated some concerns including

information overload, adverse traffic effects and lost motorist confidence. The results of

the survey also indicated that although DMS evaluations are generally conducted in

conjunction with an entire ITS evaluation, very little has been done to evaluate DMS.

Additionally, special considerations should be given to DMS’ unique capabilities as well

as the message content, location and evaluation of DMS to aid in creating successful

DMS systems.

19
 According to Mounce et al. (2007) DMS messages should be prioritized in the

following order:

1. Safety-related: messages that are directly related to safety are given first

priority for display. Examples of this type of messages include winter traction

device requirements, mountain pass information and flammable restrictions.

2. Roadway closures: DMS are used to display road or ramp closures, regardless

of the reason for the closures (accident, construction, weather, etc.).

3. Minor traffic effects: DMS are used to display information about minor traffic

effects such as construction lane closures, blocking incidents and delay

information.

4. Public text messages: As mentioned in the previous section, the lowest priority

messages are transportation-related PSAs. These messages do not directly affect

motorists and therefore are not critical to the safe and efficient operation of the

transportation system. Examples of these messages are Click It or Ticket,

Rideshare information and announcements about traveler information phone

numbers like 511.

5. Test messages: These types of messages are used to perform sign operation or

maintenance checks and to ensure proper operation of new DMS.

2.8.1. Traffic-Related Messages

The Kentucky Transportation Center notes several inappropriate ways to use

DMS (Walton et al., 2001). A notable inappropriate application is using DMS to restate

or replace required permanent signage. This could result in serious problems of

20
information overload and driver inattention to DMS. Specifically, DMS messages should

not replace static signs, regulatory signs, pavement markings, standard traffic control

devices, conventional warnings or guide signs.

2.8.2. Non Traffic Related Messages

Policies governing the display of non-traffic-related messages on DMS are not

consistent. The Manual on Uniform Traffic Control Devices indicates DMS should not be

used to display information other than regulatory, warning and guidance information

related to traffic control. Some policies state messages displayed on DMS must require

motorists to take an action or alter their driving behavior (Johnson, 2001; NCDOT,

1996). There is a consensus that DMS should not be used to advertise commercial events

or entities. Additionally, DMS should not display tourist information (Jones et al., 2003;

NCDOT, 1996; ORDOT, 2000; Walton, 2001).

2.8.3. Reasons Motorists Disregard the DMS

Dudek (2008) identified problematic uses of DMS that erode motorists’

confidence:

- Displaying inaccurate or unreliable information

- Displaying information too late for drivers to take an appropriate response

- Displaying messages drivers do not understand

- Displaying messages too long for drivers to read

- Not informing drivers of major incidents

- Informing drivers of something they already know

- Displaying information not related to environmental, roadway or traffic conditions

or routing

21
 - Displaying garbled messages

If DMS operators commit these errors, motorists are likely to disregard DMS. Influencing

the decisions of motorists is necessary for DMS to be effective.

2.9. Location of Dynamic Message Signs

DMS locations are generally established through prior experience with local

traffic problems. Recently, however, researchers have experimented with computer

programs that can place signs more precisely. These methods have not yet been

implemented by any local traffic management agency responding to the survey. The

locations of DMS are often determined through unwritten historical practices and general

policies. Agencies seldom implement methods to ensure specific DMS locations are

optimal. Two methods used to optimize DMS locations include genetic algorithms and

integer programming. Abbas and McCoy (1999) researched using genetic algorithms.

They indicated several factors influenced their decision to implement genetic algorithms,

one of which being that genetic algorithms give several solutions instead of one “best”

solution. Additionally, the constraints required in genetic algorithms are less than those

necessary to find an integer programming solution (Abbas et al., 1999).

Chiu et al. (2001) researched the use of integer programming to optimize DMS

locations. With a specified number of DMS, possible locations were determined and

analyzed. Optimal locations were chosen so that the long-run expectation of benefits was

satisfied under stochastically-occurring incident scenarios. Chiu et al. found the main

benefit of correctly locating DMS was reducing total user travel time. Implementing the

programming required numerous inputs describing geometry and traffic patterns of the

highway network. The problem was simulated using a dynamic traffic assignment

22
algorithm, which aided in determining the effectiveness of DMS locations. Each location

needed to have a high probability of capturing the randomly-occurring incidents and then

demonstrate it could effectively divert traffic. The final solution generated by the integer-

programming model determined the optimal location for all incident scenarios on the

system, although given solutions might not be optimal for an individual incident (Chiu et

al., 2001).

Chiu and Huynh (2007) combined a mesoscopic dynamic traffic assignment

simulation with a tabu search heuristic to optimally locate DMS. Incidents were

randomly generated using a Monte Carlo scheme that specified some drivers would

switch routes if their path encountered an incident and a DMS sign. Based on the

resulting flow patterns, a set of DMS locations was determined to optimize some measure

of effectiveness (Chiu et al., 2007).

Huynh et al. (2003) used a similar analysis framework to find the optimal

locations of portable DMS in a real-time framework using the G-D heuristic. Although

the simulation approach allowed a rich set of traffic and behavioral effects to be modeled,

the computational burden associated with many simulation runs on a large network could

be troublesome. This limitation was realized by Henderson (2004), who paired a static

equilibrium framework for DMS location with a discrete choice model in order to

determine the proportion of drivers who were likely to switch routes in response to

learning about an incident. Henderson (2004) developed and compared several heuristic

techniques, including a genetic algorithm and a greedy approach based on sequential

location. While computationally faster, the approach implicitly assumed drivers did not

anticipate receiving information. This assumption means drivers’ initial route choice was

23
not affected by the DMS locations, so links with a DMS did not "attract" drivers who

anticipated benefitting from that information (Hendeson, 2004). Although this distinction

may seem subtle, this anticipation effect could lead to radically different route choices for

rational drivers, even from the origin (Boyles, 2006).

2.10. DMS Performance Metrics

Tarry (1996) defined performance indicators expressly to evaluate DMS. Table

2.2 presents examples. To produce appropriate driver response, the messages displayed

on DMS should be meaningful, accurate, timely and useful. Operators lose credibility if

the messages displayed on DMS do not adhere to the guidelines of Dynamic Message

Sign Message Design and Display Manual (Dudek, 2006).

2.11. Studies Related to Designs of DMS

Extensive human factors and traffic operations research have been studied to

develop fundamental principles and guidelines for DMS message design, including

alphanumeric messages, graphics and symbols. Using these fundamental principles,

guidelines for effective message design and display for TxDOT were published in Report

0-4023-P3 Dynamic Message Sign Message Design and Display Manual (Dudek, 2006).

European countries such as Germany and Spain have used graphics or symbols on DMS,

but this practice has not yet gained popularity in the United States.

Nygårdhs (2011) reviewed DMS literature 2006-2009. The literature review

reached the following findings about design of DMS:

24
 Table 2.2. Example Performance Indicators for Dynamic Message Signs

Evaluation Category Indicators

Technical Analysis • Reliability and correctness of information displayed

• Appropriateness of plans

• Operator interface usability

• Sensitivity to errors in inputs

• Level of operator intervention needed

Impact Analysis • Degree of diversion at nodes

• Reduction in delays and extent of queuing

• Change in travel time on individual routes

• Change in total travel times and journey distances in the network

• Reduction in the duration of congestion

• Reduction in emissions

• Driver response to: range of information types, travel cost differences on

alternative routes and driver familiarity with the network

• Reduction in traffic diversion through urban areas or on the undesirable

routes

• Number of accidents

Socioeconomic Analysis • User cost-benefit analysis of performance network

• Impact on non-road users

Legal/Institutional Analysis • Legal/institutional conflicts

Public Acceptance Analysis • User attitudes to DMSs

• Non-user attitudes to DMSs

1. Graphic-aided messages are significantly better than text-only messages (because

of motorists’ preference, response time and accuracy) and should be used as much

as possible.

25
 2. Red is not recommended for DMS messages.

3. Older drivers’ performance was significantly improved by graphic-aided messages.

4. Graphic-aided DMS messages enhanced message comprehension time for non-

native English speakers.

5. More research is required to determine the proper specifications and design

guidelines of graphical images accompanying DMS messages.

6. The number of lines on DMS should be kept to a minimum.

7. Bilingual signs should only be used when absolutely necessary.

8. If bilingual signs are used, different colors or type fonts should separate the

languages.

9. The number of information units may be better correlated to DMS reading time

than the number of lines displayed.

10. A blank “off-screen” of a short duration may enhance motorists’ information

processing when successive DMS frames are used.

11. Right-justified text on DMS should be avoided.

12. Abbreviations could decrease understanding of DMS if they are not commonly

known.

13. Luminance class L3 is preferable for symbols on DMS.

14. A three-diode symbol thickness leads to better legibility than a symbol of one- or

two-diodes’ thickness.

26
3. Chapter 3: Driver Response Behavior to Messages and Localized Impact

of DMS

3.1. Drivers’ Response to Displayed Messages

Extant literature evaluating drivers’ responses to DMS focus mainly on route-

choice guidance, improving road network performance and speed slowdown. It is evident

motorists’ acceptance of DMS is associated with their perceptions and subjective attitudes

about information and how it is presented. Most of the studies found demographic and socio-

economic characteristics are important factors in controlling travelers’ satisfaction with

DMS. Travelers have specific preferences about the format and contents of messages and

information posted on the DMS. While most studies show travelers adopt DMS for their

traveler information needs, DMS do not necessarily change their travel behavior. Network

familiarity, proactive information and advisory information have been found to have different

effects at different locations studied (Rogers, 2005). Multinomial and binomial logit models

have been predominantly used to model driver diversion behavior under traveler information

scenarios with DMS. The effect of DMS has been found to vary at different study sites.

Wendelboe (2008) studied driver response to DMS in 2008 by surveying drivers. Table

3.1 shows the results and conclusions of the surveys.

27
 Table 3.1. Results from driver surveys (Wendelboe, 2008)

Respondents who:
Percent

Understood variable speed limits (VSL) correctly 82%

Perceived queue information correctly 88%

Perceived queue information correctly when information about 61%

distance to the rear end of the queue was added
Had a generally positive attitude to VSL 84%

Thought VSL had a positive effect on traffic flow 58%

Thought VSL had a negative effect on traffic flow 12%

Thought VSL had a positive effect on traffic safety 33%

Thought VSL had a negative effect on traffic safety 3%

Had a generally positive attitude to queue information 86%

Had a generally negative attitude to queue information 5%

The literature review conducted by Nygårdhs (2011) concluded the following about DMS

and drivers’ reaction to messages:

1. DMS effectively reroute traffic.

2. Supplementary DMS information may not increase drivers’ compliance with the

messages.

3. Reading and processing DMS messages lead to speed reductions.

4. Displayed delay times on DMS correlate to diversion patterns.

5. Factors correlating to drivers’ unwillingness to divert from the freeway include:

driving employer-provided cars, frequency of driving on the freeway and being

middle-aged.

28
There is some concern that more frequent use of non-incident and non-roadwork

transportation-related messages can compromise DMS credibility. If DMS distract

drivers from more critical tasks while traveling at prevailing speeds or if the messages are

erroneous or outdated, driver compliance can be compromised. In addition, if the

messages are too long, complex, and/or confusing to read and comprehend, drivers may

reduce speed to read the messages, which could result in a safety problem (Dudek, 2008).

3.1.1. Route Diversion in Response to Messages

Many researchers have studied drivers' attention and response to DMS. To evaluate

the effectiveness of DMS for route choice guidance, some researchers have tried to

estimate a route choice model that predicts how drivers respond to DMS-provided

information and whether the drivers will divert to avoid an incident or congestion on

road. Many researchers used surveys or simulations to gather the data about motorists’

behavior in response to DMS messages. The questionnaires asked motorists to state their

preference for actual or hypothetical situations (Abdel-Aty, 2000; Hao et al, 1999;

Khattak et al., 1993; Wardman et al., 1998;). Fish (2012) presented an empirical

evaluation of the quality and effectiveness of highway DMS. An additional innovation

Fish presented was to use Bluetooth sensor technology as a new method for evaluating

the accuracy and influence on drivers’ travel behavior of DMS messages. The results

showed diversion messages are effective in motorists’ route choice decisions.

Studies of incident effects on driver behavior focused on changes at the strategic

behavior level, particularly changes in their route choice behavior. Incident messages

include information about accidents, lane closures and traffic merges. Several researchers

have used the stated-preference approach in an attempt to determine the percentage of

29
travelers changing trip decisions in response to information disseminated by ATIS

devices such as DMS. The studies concluded the disseminated information could result in

up to 60-70 percent of motorists exiting the freeway ahead of a bottleneck, which yielded

a 30-40 percent reduction in congestion (Barfield et al., 1989; Benson, 1996; Chatterjee

et al., 2002; Madanat et al., 1995). However, limited information is available about actual

diversion behavior because traveler information was reflected by revealed preference and

not field measurements. A European field study found that DMS compliance rates range

between 27 and 44 percent (Tarry et al., 1995). Knopp et al. (2009) found that up to 50

percent of travelers take another route for major incidents. Schroeder et al. (2010)

investigated the effects of existing message strategies to determine which messages

maximize diversion for specific circumstances and to develop new messages for future

deployment.

Ullman et al. (2005) evaluated DMS messages to determine which message drivers

found most effective in emergencies. The authors concluded that during emergencies, DMS

messages should provide meaningful and straightforward messages that can be read and

responded to quickly because their impact on drivers can be great.

In a questionnaire survey, Benson (1996) investigated whether drivers noticed and

responded to DMS. Benson found that about 20 percent of respondents ignored active DMS

while driving. Interviews conducted by Bonsall (1993) in Paris revealed 97 percent of drivers

knew that DMS existed, 84 percent identified DMS as providing very useful information, but

only 46 percent had at least once detoured accordingly. Peng et al. (2004) conducted a similar

study in Wisconsin with results indicating that 62 percent of drivers responded to DMS

messages more than once per week and 66 percent changed their route at least once per

month as a result of the posted message. Khattak (1993) suggested diversion behavior was

30
influenced by the accuracy and detail of information including travel times, alternate

choices, knowledge of nature of the event and how to avoid incidents.

Roshandeh and Puan (2009) used archival traffic data from a freeway area in

Kuala Lumpur to assess the accuracy with which DMS displayed travel-time estimates

and driver response to messages of varying lengths and formatting. Results showed DMS

reduced the average travel times during the duration of the incident until the clearing of

the resulting congestion by a significant amount.

Levinson and Huo (2003) conducted an on-and-off study using data from

inductive loop detectors placed on different networks located in Minneapolis and St.

Paul, Minn. The purpose was to measure the effectiveness of DMS. Using the traffic flow

and occupancy data, a discrete choice model was developed to forecast the percentage of

vehicles diverting to an alternate route as a result of the message displayed. Results

showed drivers’ diversion increased when a warning message about the traffic conditions

was displayed, reducing total delay.

Peeta (1991) found the location of an incident and its duration also affected route

choice. Virginia drivers’ characteristics such as age, education, income and sex had no

significant influence on their attitudes about DMS messages (United States Department

of Transportation, 2002). In Dallas, 71-85 percent of surveyed drivers used the

recommended route. The factors influencing diversion included traffic conditions on the

alternate routes, familiarity with the alternate route and confidence in the information

(United States Department of Transportation, 2002).

Yang (1993) also found route choice behavior was affected by the characteristics

of the alternate routes. The results of this study, based on loop detector data, indicated

31
DMS could significantly affect drivers’ diversion, especially during congested times.

DMS had more influence on drivers during morning peak hours than during evening peak

hours. According to a survey conducted by Huo and Levinson (2002), drivers are more

willing to divert if there were fewer traffic stops on the alternate routes and if they were

familiar with the alternate routes. Their study also showed young, male and unmarried

drivers were more likely to divert than other drivers.

3.1.2. Speed Reduction in Response to Messages

Benekohal and Shu (1992) evaluated drivers’ behavioral responses to speed

reduction messages in construction work zone areas. They compared treatment and

control conditions when DMS displayed messages and when they were blank. They

found displaying the speed limits on DMS was an effective way to reduce average speed.

Their study showed displaying messages reduced drivers’ speed immediately after

passing the sign, but not at a point far from DMS. Cars and trucks reduced their speed by

as much as 5 and 4 mph, respectively, near the DMS.

3.2. Effect of DMS Design on Driver Response

Studies show DMS with different formats and designs could have different effects

on drivers’ behaviors. This section reviews the research comparing drivers’ responses to

text-based and graphic-aided messages as well as flashing and static messages.

3.2.1. Text-based and Graphic-aided Messages

Wang et al. (2007) studied the use of graphics on DMS. They found most drivers

preferred graphics to text and responded faster to graphic-aided messages than to text-

only messages. We suggest using graphics in some advisory signs to help enhance

32
drivers’ understanding and responses to messages and improve the effectiveness of these

signs.

In similar research, Bai et al. (2011) suggested traditional text-based messages

have several limitations, such as confusing drivers (which delays their responses), being

difficult to read for older drivers and non-English-speaking drivers, and having a short

range of legibility. Bai et al. (2011) said using graphic-aided and graphic messages on

portable DMS have many advantages over text-based DMS based on a number of

previous laboratory simulation experiments. They used field experiments and driver

surveys to determine the effectiveness of a graphic-aided and graphic portable DMS on

reducing vehicle speed in the upstream of a one-lane, two-way rural highway work zone.

They also compared the effectiveness of text, graphic-aided and graphic-portable DMS

on reducing vehicle speed in a highway work zone in Kansas using regression models of

the relationship between mean vehicle speed and distance under the three conditions. The

findings showed that:

1. Text, graphic-aided and graphic-portable DMS resulted in a mean vehicle

speed reduction of 13 percent, 10 percent and 17 percent, respectively.

2. Graphic-aided portable DMS reduced mean vehicle speed more effectively than

the text one from 1,475 feet to 1,000 feet in the upstream of a work zone.

3. The majority of drivers understood the work zone and flagger graphics and

believed the graphics drew their attention more to the work zone traffic

conditions.

4. Most drivers preferred the information to be presented in the graphic-aided format.

33
 3.2.2. Flashing and Static Messages

Average reading times for flashing messages were not higher than for static

messages (Dudek, 2005). However, results indicate flashing messages may have an

adverse effect on message comprehension for drivers unfamiliar with flashing DMS.

Average reading times for flashing line messages and two-phase messages with

alternating lines were significantly longer than the alternative messages. In addition,

message comprehension was negatively affected by flashing line messages.

3.3. Localized Impact of DMS

3.3.1. Traffic Speed Slow Down for Perception of Messages

Oh, Hong, and Park (2009) investigated 20-30-year-old drivers’ behavioral

responses to DMS when reading and processing the messages in a DMS influence zone.

Individual vehicle trajectories were studied via differential global positioning system

(DGPS); speed and acceleration rates were used as surrogate measurements to represent

driver behavior. DMS influence zones was divided into five sections of 100 meters long.

ANOVA results showed drivers’ average speed and acceleration were statistically

different in each section. Drivers tended to reduce their travel speed while reading and

processing DMS messages, and they tended to increase speeds again after they finished

reading the messages.

Rama and Kulmala (2000) investigated the effects of two DMS on drivers’ car-

following behavior. Results showed a sign for slippery road conditions reduced the mean

speed by 1-2 km/hour after controlling for the decrease caused by the adverse road

conditions.

34
 Wang et al. (2007) studied the effects of DMS messages on traffic approaching

and passing the signs. Traffic data gathered by several mobility technology units (MTUs)

near DMS along I-95 in Rhode Island were analyzed with the goal of understanding the

effects of various DMS messages on the speed variations on traffic approaching and

passing the signs. The researchers found a positive correlation between certain posted

DMS messages and traffic slow-downs, so the study next explored ways to improve

messages’ design and display on DMS. Results from a questionnaire indicated DMS

were among the top few that caused drivers to slow down, while Danger/Warning

messages attracted the most attention from drivers. Results also showed the majority of

drivers reduced their speeds when approaching active DMS; lengthy, complex or

abbreviated messages caused further slowdowns. Their study also employed a computer-

based questionnaire survey and a driving simulation experiment to measure drivers’

preferences and responses to various DMS displays and formats. The results showed

older drivers exhibit a higher tendency to slow down.

Fish et al. (2012) investigated 2,268 cases of message activation, removal and

switching on DMS using RTMS speed data to determine whether DMS messages cause

speed slowdown. The study confirmed that in some cases traffic streams decrease speed

in response to message activation.

Harder et al. (2003) used a computer-based driving simulation to test various message

types to see whether they could detect a slow-down. Results showed 21.7 percent of

participants slowed their speed by 13.9 mph as they approached AMBER alert DMS

messages. Alternatively, when a Crash alert DMS message was displayed, 13.3 percent of

participants slowed their speed by 12.7 mph.

35
 Boyle and Mannering (2004) used a driving simulation to determine the impact of

DMS on drivers’ speed. Although they found drivers slowed down when approaching active

DMS, they also found drivers sped up to compensate for their initial speed reduction.

Furthermore, the study demonstrated drivers were more likely to have a larger deviation in

speed when encountering a new DMS message. A possible explanation may be that drivers

notice a new message and as a result, take more time to process the information when a new

DMS message is presented. Moreover, when a DMS displayed the same message for a long

period of time, drivers became familiar with the information and thus required less time to

read it.

Several studies discussed in section 3.2.1 showed using graphics to convey meaning

on roadway signs provided many advantages over text-only messages. Graphic-aided

messages could be more easily and quickly identified compared to text-only messages at a

greater distance.

DMS including graphic information allowed faster responses than text-only

information (e.g., Bruce et al., 2000; Hanowski and Kantowitz, 1997; Staplin et al., 1990).

Wang et al. (2007) studied the use of graphics on DMS and found drivers tended to respond

faster to graphic-aided messages than text-only messages. All of these studies and practices

indicated adding graphics might help enhance drivers’ understanding of and responses to

DMS and ease slowdowns. Adding graphics to DMS messages could help enhance

drivers’ understanding of and responses to those messages and reduce their speed

variations that occur as a result of reading DMS; it might eventually help ease the slow-

downs.

36
 3.3.2. Driver Distraction and Collision Occurrence

Driver distraction plays a significant role in traffic safety. Driver distraction is a

factor in one in four car crashes, and of those crashes involving driver distraction, one in

four involves distractions outside the vehicle (NHTSA, 2009). Few studies have been

conducted on accident rates due to distractions associated with DMS. However, a definite

accident rate would be hard to determine. According to the Kiewit Center for

Infrastructure and Transportation (2003), accident rates for a section of road can be

determined by a ratio of accidents per million vehicle miles of travel. The normalized

formula allows for a comparison of various accidents to rates of other stretches of roads

that may not be the same length.

Many studies focus on the effects of DMS on driver behavior and the potential

benefit of using DMS to reduce downstream accidents. Chamberlain (1995) demonstrated

that the use of DMS coupled with a queue detecting system could reduce accidents for

upstream drivers who otherwise would be unprepared for queues downstream. According

to NHTSA’s Distraction initiative, 20 percent of all accidents are related to some kind of

distraction (2010). Many studies indicated that DMS take drivers' attention away from

their driving (Wang et al., 2007). Because drivers expect useful information from active

DMS, they slow to gain extra time to read and comprehend the messages. To compensate

for their speed reduction, drivers speed up after passing DMS. Crashes correlate to

driving speed, so this speed variation could pose a threat to other vehicles in the traffic

and lead to crashes.

Erke et al. (2007) conducted a field test and video observation study. Their research

messages were set to on-and-off in order to compare driver behavior such as route choice,

37
speed and braking behavior when they approached DMS displaying messages and when

DMS were left blank. Two DMS were used in this study and displayed road closure and

recommendations for alternative routes. Speed measurements of 3,342 vehicles showed

large speed reductions; video observations showed large proportions of vehicles braking

while approaching the DMS. This research finds speed reductions and braking maneuvers

can partly be attributed to attention overload or distraction due to the information on the

DMS. In addition, a proportion of the speed reductions was due to chain reactions where

one vehicle braked and forced the following vehicles to brake or change lanes in order to

avoid collisions. Safety problems may result from distraction or from the reactions of the

drivers to the distraction.

3.4. Summary

Many methods have been used to determine drivers’ responses when approaching

DMS. Data from surveys, simulators, video observations and loop detectors are the most

common methods to study these behaviors, and these methods have shown some

promising results. Table 3.2 and Table 3.3 present summaries of previous studies of

drivers’ responses to diversion and speed reduction messages, whereas Table 3.4

summarizes the reviewed literature on the localized effects of the signs. This project uses

ground truth data integrated database to evaluate the impact of the signs on occurrence of

road accidents.

38
 Table 3.2. Literature Summary on Driver Response to Diversion Messages

Author Source Year Country Study Results

Approach

Fish TRB 2012 US, Field Test/ • diversion messages are

et al. Maryland Bluetooth effective in route choice

sensors decisions.

Chen et al. IWMSO 2008 China, SP survey • diversion increases as the

2008 Beijing, traffic speed decreases. (<20

km/h).

• 21.45% of drivers divert

Foo & TRB 2008 Canada, Field • occurrence of a message

Abdullahi Ontario Test/Loop change plays a vital role in

detector influencing downstream

diversion

Cheng & 12th IEE 2004 U.K., SP survey • more exposure to DMS

Firmin Int. Conf. London increases appreciation of the

information displayed.

Peng Trans. 2004 US, RP survey • 75% are positive with

et al. Res. Wisconsin combined usefulness of VMS.

Rec. with • 16% don’t trust VMS

information and don’t

logit model
change their route.

39
Author Source Year Country Study Results

Approach

Levinson & TRB 2003 US, Field Test/ • a probit model to estimate

Huo Minnesota Loop diversion as a function of

Detector message content.

• ahead warning is effective

for diversion.

Chatterjee et Trans. 2000 U.K., Survey, • location of incident and

al. Res. Leeds Logistic message content influence

Part C Regression the probability of diversion.

40
 Table 3.3. Literature Summary on Driver Response to Speed Reduction Messages

Author Source Year Country Study Results

Approach

Alm & Trans. 2000 Sweden Simulation • all participants reduced their

Nilsson Human speed in response to incident

Factors warning messages

Luoma Trans Res. 2000 Finland Simulation • drivers reduced speed 1-2

et al. – Part F km/h in response to a DMS

warning of slippery condition

Benekohal& Civil Eng. 1992 US, Treatment • displaying the speed limits is

Shu Studies Illinoise control effective in reducing the

(DMS on speed.

and off)/ • speed of cars reduces

statistical immediately after passing the

analysis DMS, but not at a point far

from DMS.

• cars and trucks reduced their

speed by as much as 5 and 4

mph respectively near the

DMS.

41
 Table 3.4. Driver Distraction and Speed Slow Down for Perception of Messages

Author Source Year Country Study Results

Approach

Wang TRB 2009 US, Survey • DMS cause slowdown (specially

et al. Rhode danger warning messages).

Island • lengthy, complex or abbreviated

messages caused further slowdowns.

• elder drivers exhibit a higher

tendency to slow down

Erke Trans. 2007 Norway, Field • most of vehicles braked approaching

et al. Res. Oslo Test/video the DMS.

Part F observation • messages causes distraction and leads

to speed reduction and chain
(messages
collisions and safety problem.
on/off)

42
4. Chapter 4: Investigation on Possible Relationship between DMS and

Occurrence of Road Accidents

4.1. Problem Statement and Motivation for Research

Although DMS are intended to improve the efficiency and safety of road

networks, little has been done to study the effect of these devices on driver safety. In spite

of all advantages of DMS, some issues regarding the disadvantages of real-time travel

signs have emerged. Reports from WTOP and NBC are examples of the opposing side.

These reports claim these devices are expensive adversely affect drivers’ concentration,

and cause speed slowdown, which may lead to an increase in road crashes (HSM, 2010).

The purpose of this research is to investigate the problem and determine if there is any

meaningful relationship between occurrence of accidents and proximity of DMS in

regards to these accidents.

For this study, accident data and DMS locations in Maryland for 2007 to 2010

were mapped in ArcGIS to determine accident patterns on the state highway network. In

order to investigate the above-listed claims, all 184 highway DMS in Maryland were

evaluated to their proximity to nearby accident patterns. The purpose of this study is to

determine whether DMS on Maryland highways produce significant localized safety

issues. The data used and methods of research are described in detail in the following

sections.

43
4.2. Methodology

4.2.1. Data Sources and Preparation

The data used in this research were collected from three major sources: the Center

for Advanced Transportation Technology (CATT) Laboratory in the Department of Civil

and Environmental Engineering at the University of Maryland, Coordinated Highway

Action Response Team (CHART) reports for regions within the District of Columbia in

Maryland, and the Maryland Department of Transportation, State Highway

Administration (SHA) and DOT archival data. Figure 4.1 shows the databases and

sources that are used in the research.

Figure 4.1. The databases and sources of data used in the research

The study area was the roadway network in Maryland. Figure 4.2 depicts the

study area in the research.

44
 Figure 4.2. Study Area

4.2.2. Accident Database

The accident database included 38,718 records. A data cleansing process was conducted

to remove data gaps and outliers, which yielded a data set of 23,842 usable accident

records for the period of 2007 to 2010. The data set included accident type (property

damage, personal injury and fatality), geographic location, jurisdiction, accident time and

other related information. Because of confidentiality concerns, access to police records

and accident causes was not possible. Locations of accidents are pinpointed on road

network map for further analysis. Figure 4.3 shows the first shape of accident database

and the locations of accidents projected on the road map.

45
Figure 4.3. First shape of accident data and pointing location of accidents on road network map

46
 4.2.3. DMS Database

The DMS inventory was acquired from CATT Laboratory. The DMS inventory

includes all types of signs, including those permanently mounted overhead, roadside

models and portable signs operated by CHART or Maryland Transportation Authority

(MTA). The database listing Maryland’s 184 DMS includes the signs’ identification

number, longitude and latitude, address and type. Figure 4.4 shows the first shape of

DMS data and its projection onto the road network map. An accident’s longitude and

latitude were used to join the accident and DMS databases. Figure 4.5 shows a network

system is created with the three overlaid layers.

An impact area of 900 feet was defined for each DMS. Accidents that occurred

within the 900-foot impact zone were assigned to the relevant sign. The details of how

impact areas were defined are presented in next section. Accidents in the 900-foot

proximity to DMS were categorized as occurring in an impact area based on location

field, visual judgment and direction of DMS.

47
Figure 4.4. First shape of DMS database and projection to road map

48
 Figure 4.5. Map of accidents and DMS locations

4.2.4. AADT Database

The Highway Safety Manual (2010) specifies traffic flow is one of the most

important factors contributing to crashes. This research uses the average annual daily

traffic (AADT) of the road segments as an index for traffic flow. The AADT data were

retrieved from Maryland’s SHA volume maps for the four-year period of study. The

AADTs were collected from more than 3,000 program count stations and 79 automatic

traffic recorders located throughout Maryland. The shape file of AADT layer is projected

onto the road map and the accidents and DMS. An example of the map is shown in

Figure 4.6.

49
 Figure 4.6. An example of the volume map AADT (SHA 2011)

4.3. Data Processing and Preparation Challenges

This study uses a new approach to the problem dealing with several large

databases, each with a different data structure and coordination systems. Acquiring data

from different sources was another challenge for this project. In addition, some parts of

police accident reports, such as causes of the accidents, were not accessible because of

confidentiality concerns. In addition to the difficulties in obtaining the necessary data,

another issue encountered involved processing data sets that each had more than 10,000

records. That problem was resolved by using a data cleaning process that filtered and

removed outliers. Joining the databases by time and location was a challenge that was

50
resolved by pinpointing the locations through GIS tools and matching the time of events

through coding in a SQL environment.

4.4. Defining the Impact Area of DMS

The goal of projecting the DMS and accident locations on ArcGIS was to

determine the distance within which DMS might affect the occurrence of accidents. The

size of characters on electronic signs is the most important factor determining the

maximum viewing distance. In order to define the distance within which DMS may affect

occurrence of accidents, the visibility distance from DMS needed to be determined.

According to Maryland Manual on Uniform Traffic Control Devices (MUTCD), the

minimum character size of DMS fonts on major roads (55 mph speed limit) is 18 inches.

Based on the information provided by International Sign Association, the maximum

viewing distance for 18 inches character size sign is 900 feet. Figure 4.7 illustrates the

impact area for research.

4.5. Case Study on I-95

Interstate 95 in Maryland is a major highway that runs diagonally from northeast

to southwest, from Maryland's border with Delaware to the Woodrow Wilson Bridge. It

briefly enters the District of Columbia before continuing into Virginia. We chose this

freeway because the route is one of the most heavily traveled interstate highways in

Maryland, especially between Baltimore and Washington, D.C. Figure 4.8 shows I-95

and the DMS located on this highway. The light blue pushpins are DMS on northbound

and the dark one are the signs located on southbound.

51
 Figure 4.7. Impact Area

Figure 4.8. I-95 along with the DMSs along this highway

52
 The accidents along I-95 were projected onto the map. Figure 4.9 gives a

perspective of the accidents on I-95 and northbound and southbound DMS.

Figure 4.9. Accidents in I-95

Figure 4.10 shows the projected AADTs to road map. Because the impact area of DMS

was determined to be 900 feet, multiple ring buffer zones with radius of 900 multiplier-

feet (900, 1800, 2700, etc.) radius were performed for each DMS sign along I-95. This is

shown in Figure 4.11.

4.5.1. Analysis of Case Study and Preliminary Results

In this step, 70 geometrically homogenous segments with a 900-foot impact zone

were selected from Interstate 95. For each segment, accidents were counted and the total

number of crashes in each segment was counted for use in regression analysis

considering that the segment is an impact area or not as well as the existence of

53
interchange and AADT in the segment. Table 4.1 shows the variable used for the case

study.

Table 4.2 shows the 70 segments with their accumulated number of crashes; it also shows

the existence of DMS and interchanges in the impact areas. To analyze the data, an

unbalanced two-way ANOVA was performed using SAS software. The results show P-

value strongly rejects the hypothesis that Interchanges have no impact on the occurrence

of accidents. The significance level for the impact of DMS is not high and shows DMS

do not significantly contribute to occurrence of accidents. The results are shown in Figure

4.12.

Figure 4.10 . Projection of AADTs to road map

54
 Figure 4.11. Multiple Buffers along I-95

Table 4.1. Variables used in case study

Poisson regression analysis was conducted to predict the number of crashes

within 900 feet segments given DMS, interchanges and AADT data about the route. The

test strongly rejects the hypothesis that interchanges and AADT do not have significant

impact on the occurrence of accidents. Regression analysis also shows DMS are not

significant contributors in crash occurrence. The results of both methods converge point

out that while interchanges and AADT are important factors to accidents, there is no

relationship between presence of DMS and occurrence of accidents.

55
 Table 4.2. I-95 Case Study Samples

BufferID NumberCrash ImpactArea Interchange AADTVMT SouthORNorthBound

10 1 1 0 147581 S
20 1 1 0 147130 N
30 7 1 0 177981 S
40 1 1 0 206880 N
50 0 1 0 213841 N
60 0 1 0 213841 S
70 1 1 0 205142 N
80 28 1 0 205142 S
90 0 1 0 212261 N
100 0 1 0 188601 S
110 0 1 0 183961 S
120 3 1 0 188671 N
130 0 1 0 194069 N
140 0 1 0 192871 S
150 0 1 0 182473 N
160 40 1 0 182478 S
170 4 1 0 123232 S
180 0 1 0 129021 S
190 3 1 0 119161 N
200 2 1 0 165104 S
210 0 1 0 147341 N
220 1 1 0 147341 S
230 1 1 0 121581 N
240 0 1 0 121581 S
250 0 1 0 96951 N
260 0 1 0 96951 S
270 1 1 0 98941 N
280 2 1 0 98941 S
290 0 1 0 84721 N
300 0 1 0 91711 S
310 0 1 0 91711 N
320 43 0 1 191981 N
330 57 0 1 147581 S
340 2 0 1 147581 N
350 19 0 1 147130 N
360 10 0 1 213841 N
370 81 0 1 213841 S
380 8 0 1 231801 S
390 5 0 1 205142 N
400 15 0 1 221521 N

56
 410 4 0 1 188671 N
420 15 0 1 182473 N
430 3 0 1 123232 N
440 16 0 1 147341 N
450 9 0 1 98941 N
460 5 0 1 80571 N
640 3 1 1 187501 N
650 38 1 1 174051 S
960 18 0 0 147130 S
970 1 0 0 177981 N
980 9 0 0 177981 S
990 1 0 0 177981 N
1000 1 0 0 213841 N
1010 1 0 0 213841 S
1020 1 0 0 205142 S
1030 2 0 0 205142 S
1040 2 0 0 212261 N
1050 2 0 0 183961 N
1060 0 0 0 194069 S
1070 11 0 0 192871 N
1080 1 0 0 192871 N
1090 4 0 0 175027 N
1100 9 0 0 129021 N
1110 17 0 0 129021 S
1120 5 0 0 119151 N
1130 0 0 0 165104 S
1140 1 0 0 161521 S
1150 1 0 0 96951 S
1160 0 0 0 9651 N
1170 0 0 0 98841 N

57
Figure 4.12. SAS outcomes of unbalanced two-way ANOVA for case study in I-95

58
Recent studies have raised concerns about using a Poisson distribution for accident

frequency regression models. They state one characteristic of crash-frequency data is the

probability that the variance exceeds the mean of the crash counts (Dominique et al,

2010). A property of the Poisson distribution is the mean and variance are equal; because

of this, factor, characteristics of crash-frequency data could be problematic. To validate

results, a negative binomial regression was also performed. The results of negative

binomial regression also converges Poisson regression. The second analysis strongly

rejects the hypothesis that interchanges are not significant contributors, but suggests

DMS are not contributing factors to occurrence of accidents. The result for negative

binomial regression analysis agrees with the Poisson regression that DMS do not cause

accidents. The coding and outcomes are presented in Figure 4.13 and Figure 4.14.

59
Figure 4.13. SAS outcomes of Poisson regression for case study in I-95

60
Figure 4.13 (Continued). SAS outcomes of Poisson regression for case study in I-95

61
Figure 4.14. SAS outcomes of Negative Binomial regression for case study in I-95

62
Figure 4.14 (Continued). SAS outcomes of Negative Binomial regression for case study in I-95

63
4.6. Weather Conditions Database

An important factor that contributes to driver distraction is visibility. Because

precipitation, wind gusts and severe weather may have adverse effects on message

visibility, another factor that should be considered as contributing to accidents is climate

status. The factors analyzed for weather conditions include precipitation, wind gusts and

visibility factors. The weather data for this research were retrieved from DOT archival

databases. The data initially were formatted as month-to-month archival data from 49

weather tower stations. The initial data files contained the following fields: Date, time, air

temperature, humidity, average wind speed, wind gust, wind direction, precipitation type,

precipitation intensity (light, medium, heavy), precipitation accumulation, rate (rate per

hour in inches), visibility (miles) and surface temperature.

For simplicity, the area of research was divided into north, south, west, east and

Washington, D.C. regions. The nearest central weather tower in each region was

designated to represent weather conditions in that region. Table 4.3 shows these regions.

The data set covered the four-year period from 2007 to 2010. Figure 4.15 shows the

format of weather database.

Table 4.3. Tower stations assigned to each weather region

Weather Station Region Latitude Longitude

I-68 @ Cumberland West 39.70302 -78.63177
US 50 Kent Narrow Bridge East 38.97203
-76.25391
I-895 @ Levering Ave North 39.21854 -76.71071
US-301 at Potomac River South 38.36366 -76.983
I-270 @ I-370 Washington, DC 39.11946 -77.19593

If an accident in the database occurred within 900 feet of approach a DMS, the

accident is joined with the DMS and AADT associated with that accident. As a result, the

64
main database was integrated with the weather stations’ data by proximity to the closest

weather tower station and occurrence time of accident.

In order to integrate the weather database and the main database, the weather

database was imported into a SQL server and each accident was matched with the closest

weather tower and weather condition at the time of accident. The matching process is

performed using SQL queries coded in C++.

65
Figure 4.15. Weather Database Format

66
4.7. Log of Messages Database

The database containing the message logs was acquired from the CATT

Laboratory in the Department of Civil and Environmental Engineering at the University

of Maryland. This database contains entire messages displayed on all DMS in Maryland

during the time period from 2007 to 2010. This database includes 1,047,586 records of

messages, their DMS identification number, time messages were displayed and beacon

data fields. The beacon data field shows whether the beacon was on or off. Figure 4.16

shows an example of the messages database.

The syntax for message data field is based on the definitions in National Transportation

Communications for ITS Protocol, Object Definitions for Dynamic Message Signs

Version 02 (2007). The number of panes can be determined by interpreting the system of

coding that comes with each message. The main codes of messages are:

- [PT##O#]: This code is interpreted as Panel Time, ## in tenths of seconds on, # in

tenths of seconds off (normally this # is 0, otherwise the panel would be flashing)

- [JL#]: This code is for text justification. The number corresponds to various

justifications (i.e. 2 left, 3 center, 4 right)

- [NL] - New Line

- [NP] - New Pane

67
Figure 4.16. Log of Messages Database

68
The following example illustrates the message syntax:

[PT25O0][JL3]ACCIDENT AHEAD[NL][JL3][NL][JL3]PAST EXIT 51[NP][PT25O0][JL3] 2

LEFT LANES BLOCKED[NL][JL3][NL][JL3] EXPECT DELAYS

This message has 2 panes, alternating appearances for 2.5 seconds, with all lines center-

justified. It would appear as:

PANE 1:
ACCIDENT AHEAD

PAST EXIT 51

PANE 2:
2 LEFT LANES BLOCKED

EXPECT DELAYS

The message log database was imported into SQL server and the main database. If

an accident was in impact area, the assigned DMS was matched with the message

displayed at the occurrence time of accident. The same process matched weather data sets

to messages using SQL queries coded in C++.

4.8. Analysis and Results

The integrated database consisted of the integrated data for each accident. Every record

of an accident contains was associated with the following information: time and date of

accident, location (longitude and latitude) of accident, type of accident, AADT of the

roadway and weather condition at the time of accident (including air temperature,

humidity, average wind speed, wind gust, wind direction, precipitation type and rate, and

69
visibility). If the accident occurred in a DMS impact area, the following information is

also present: the assigned DMS and the message DMS displayed at the time of accident..

Figure 4.17 depicts the projection of integrated database for the entire study area in

ArcGIS and Figure 4.18 illustrate the close-up shot of the projected map.

The integrated database consisted of 23,842 records for accident during the period

from 2007 to 2010. There were 298 accidents within 900 feet of a DMS. Drivers of 50

accidents were exposed to active DMS that displayed messages. For other 248 accidents,

the DMS were blank. As the following sections present, multiple approaches were

employed to analyze different aspects of the data. A paired t-test analysis at a significant

level of 95 percent compared accident rate in impact areas with their onward 900-foot

segment. In addition, an on-and-off study was conducted to compare accident rates of 15

DMS with on-and-off display messages. Statistical analyses investigated the effects of

weather conditions, visibility and type of messages on accidents in impact areas.

4.9. Analysis on Impact Areas and Following Segment

To investigate the effects of DMSs on occurrence of road accidents, a paired t-tests

statistical analysis at 95 percent confidence level was used to compare accident rates for

the 50 accidents in impact areas of active DMSs with their subsequent 900 feet segment.

The null hypothesis states the difference in mean accident rate between two consecutive

900 feet segments is equal to zero. On the other hand, the alternative hypothesis suggests

difference between the means is not equal to zero:

H 0 : µ2 − µ1 =
0
H1 : µ2 − µ1 ≠ 0

70
Figure 4.17. Projection of Integrated Database

71
Figure 4.18. Close up shot of projected map

72
 The accident rates for the segments were calculated using the spot accident rate

formula recommended by both the FHWA Safety Program guidance and the Kiewit

Center at Oregon State University (2003). According to the formula, the accident rate for

a spot of a road is calculated by a ratio of accidents per million vehicles. A spot location

is generally defined as a location 0.3 miles or less in length. Because the segments

compared in this study were 900 feet length (0.17 miles), this formula was appropriate for

calculating the accident rate (Kiewit, 2003). The formula allows comparison of various

accidents rates. The equation for computing accident rate for a spot location is as follows:

Rsp = A/Exposure [million entering vehicles] (Equation 1)

or

Rsp = (C) (1,000,000)/AADT (365)(N)

Where:

Rsp = Accident rate at a spot in accidents per million vehicles,

C = Number of crashes for the study period,

N = Period of study (years or fraction of years),

AADT = Average Annual Daily Traffic (AADT) during the study period.

Figure 4.19 shows the accident rates for impact areas compared to their subsequent 900

feet segment.

73
 Figure 4.19. Accident rate for impact area of 900 feet compared to their subsequent900
feet segment

Table 4.4 shows the tabulated facts for the accident rates in both segments

compiled in a table including DMS identification number, AADT of segment, number of

accidents in segment and accident rates in segments.

74
 Table 4.4. Tabulated facts of impact areas and forwarding segments
#
# accidents Forward
accidents 900ft in 900ft
Impact in 900 Accident Forward Accident
Area DMS_id AADT feet Rate 900 feet Rate DMS
2 CHART_01010528004f00820047f02c76235daa 13974 7 0.343102945 5 0.245073532 0.098029
12 CHART_0c011090002d0067003f062c3d235daa 2364 1 0.28973414 2 0.57946828 -0.28973
29 CHART_1901170900050002003d242c3b235daa 61273 2 0.022356715 0 0 0.022357
32 CHART_1b010c38005200820047f02c76235daa 65821 2 0.020811945 1 0.010405972 0.010406
33 CHART_1b01212600da0008003d242c3b235daa 187920 1 0.003644804 27 0.098409699 -0.09476
34 CHART_1c000b26004c00820047e22c9e235daa 145780 1 0.004698391 2 0.009396783 -0.0047
41 CHART_1e01133800d90008003d242c3b235daa 57512 4 0.047637467 2 0.023818734 0.023819
46 CHART_2c00083a004b00820047e22c9e235daa 444336 5 0.00770736 7 0.010790304 -0.00308
55 CHART_39010a59005100820047f02c76235daa 88882 1 0.007706077 0 0 0.007706
64 CHART_40ff12d400c200820047e32c96235daa 8282 1 0.08270122 3 0.248103661 -0.1654
68 CHART_46010ade0036005a0039fc442f1f5daa 190391 1 0.003597499 3 0.010792498 -0.00719
70 CHART_4701165e00d90008003d242c3b235daa 74887 2 0.018292401 25 0.228655009 -0.21036
88 CHART_5f00077a004600820047e32c96235daa 245421 1 0.002790843 5 0.013954216 -0.01116
92 CHART_62000ff900a300e0003e062c3d235daa 121581 1 0.005633541 1 0.005633541 0
95 CHART_650113d6003d0067003f062c3d235daa 65214 2 0.021005659 2 0.021005659 0
104 CHART_6dff058b004500820047e32c96235daa 255882 1 0.002676748 1 0.002676748 0
105 CHART_6e00069600af0054003afc442f1f5daa 23726 8 0.230947149 4 0.115473574 0.115474
113 CHART_74000733009000d3003e062c3d235daa 98941 1 0.006922626 0 0 0.006923
124 CHART_89000cab00d80008003d242c3b235daa 147130 3 0.013965843 0 0 0.013966
137 CHART_aa01033e000c00630045152cea235d0a 23741 1 0.028850154 0 0 0.02885
139 CHART_ac0064d1002f00ae003ac7442f1f5daa 66761 1 0.010259455 1 0.010259455 0
162 CHART_d8ff030400b800c60047832c33235daa 153481 1 0.004462647 2 0.008925294 -0.00446
182 CHART_fdff03d9008000c80040062c3d235daa 8600 2 0.159286397 0 0 0.159286
50

75
 The graph shows for the majority of impact areas, rate of accidents is lower than

their onward adjacent segment. Figure 4.20 shows the difference of the accidents rates for

the two segments.

Figure 4.20. Difference of the accidents rates between the impact area and its subsequent
segment

The analysis of difference between the accident rates show 70 percent of the

impact areas have lower or equal accident rates compared to the 900 feet segments that

follow them. This finding indicates DMS do not have significant effects on increasing the

accident rate. The remaining 30 percent, or 7 impact areas, show a positive difference

between the accident rates. The case study of Interstate 95 supported the fact that

interchanges are contributing factor to accidents. Therefore, a simple qualitative analysis

of the locations of the DMS with the highest accidents rates showed they tended to occur

within short distances of interchanges. Additionally, those with lower rates tended to

76
occur further away from interchanges. The reason for positive accident rates could be

attributed to external factors such as existence of interchanges in DMS buffer zones and

roadway geometry that increase accident rates in these segments.

4.9.1. Findings

A paired t-test on the accident rates was performed to compare accident rates in the two

segments. Results suggest DMS do not increase accident occurrence. The mean

difference of the two accident rates is -0.013. The coding in SAS software and the results

are presented in Figure 4.21.

77
Figure 4.21. SAS outcomes for comparison of impact areas and following section

78
Figure 4.21 (Continued). SAS outcomes for comparison of impact areas and following
section

79
 4.10. On-and-off Analysis

An on-and-off study compared results obtained from the previous section. The

data were inputted into a table. Total numbers of accidents for 15 signs were counted for

periods when DMS displayed messages and when they were blank. The accident rates for

both situations were calculated using the formula articulated in the previous section. A

one-way ANOVA with pairwise comparisons assessed accident rates in impact area when

DMS were on and when they were off. Table 4.5 shows data used in the on-and-off

analysis, including DMS identification number, number of accidents in impact areas, and

AADT of segment and accident rates in segments.

Figure 4.22 depicts the comparison of accident rates when messages are

displaying on DMS and when these signs are blank.

To better determine how different the accidents rates are for on and off DMS, the

graph of the difference between the rates of the two conditions is shown in Figure 4.23.

As this graph shows, in all DMS impact areas the accident rate is lower when the sign

shows a message.

80
 Table 4.5. Tabulated facts of on and off study

# accidents With DMS

Impact in impact Message Accident DMS blank DMS Blank
Area DMS_id area Rate AADT before Accident Rate DMS Effect
2 CHART_01010528004f00820047f02c76235daa 7 0.343102945 13974 11 0.53916177 -0.196059
29 CHART_1901170900050002003d242c3b235daa 2 0.022356715 61273 7 0.078248503 -0.055892
32 CHART_1b010c38005200820047f02c76235daa 2 0.020811945 65821 8 0.083247779 -0.062436
33 CHART_1b01212600da0008003d242c3b235daa 1 0.003644804 187920 4 0.014579215 -0.010934
34 CHART_1c000b26004c00820047e22c9e235daa 1 0.004698391 145780 1 0.004698391 0
41 CHART_1e01133800d90008003d242c3b235daa 4 0.047637467 57512 19 0.22627797 -0.178641
46 CHART_2c00083a004b00820047e22c9e235daa 5 0.00770736 444336 16 0.024663552 -0.016956
55 CHART_39010a59005100820047f02c76235daa 1 0.007706077 88882 1 0.007706077 0
64 CHART_40ff12d400c200820047e32c96235daa 1 0.08270122 8282 6 0.496207322 -0.413506
68 CHART_46010ade0036005a0039fc442f1f5daa 1 0.003597499 190391 2 0.007194999 -0.003597
70 CHART_4701165e00d90008003d242c3b235daa 2 0.018292401 74887 7 0.064023403 -0.045731
105 CHART_6e00069600af0054003afc442f1f5daa 8 0.230947149 23726 21 0.606236266 -0.375289
113 CHART_74000733009000d3003e062c3d235daa 1 0.006922626 98941 1 0.006922626 0
124 CHART_89000cab00d80008003d242c3b235daa 3 0.013965843 147130 8 0.037242249 -0.023276
162 CHART_d8ff030400b800c60047832c33235daa 1 0.004462647 153481 1 0.004462647 0

81
 Figure 4.22. Comparison of accident rates while DMS are on and while blank

Figure 4.23. Difference of the accidents rates in on and off study

The results show accident rates for DMS displaying messages were less than or

equal to blank DMS for all cases analyzed. The results of this on-and-off study support

the outcomes of the previous sections – DMS are not contributing factors to accidents.

82
 4.10.1. Findings

A Repeated Measures Analysis of Variance was conducted to compare the mean

accident rates in two conditions. The F-value of 6.73 and P(F < 6.73) of 0.0212 for the

one-way ANOVA with paired comparison suggests null hypothesis is rejected with 98

percent level of confidence in favor of supporting the fact that the mean accident rate for

active DMSs was lower than the rate of accidents for inactive DMSs. The SAS coding

and the outcomes are presented in Figure 4.24.

Figure 4.24. SAS outcomes for on and off study

83
Figure 4.24 (Continued). SAS outcomes for on and off study

84
4.11. Accidents in DMS Impact Areas and Weather Conditions

This section summarizes and categorizes accident characteristics in DMS areas. As

mentioned before, weather conditions can contribute to accidents by reducing drivers’

visibility. According to FHWA Road Weather Management Program, visibility

impairments, precipitation, high winds and temperature extremes affect driver

capabilities and operational decisions, traffic flow, and crash risk. This research project

concerns driver response to DMS messages, which is known to have environmental

factors, so it was necessary to investigate the accident in conjunction with weather

conditions at the time of accident for active DMS. As Table 4.6 and

Figure 4.25 show, there are only four accidents in the entire set of accidents within the

impact area that happened in rainy and snowy conditions.

Table 4.6. Accidents in DMS areas and precipitation

Precipitation Accidents in Impact Area #

Rain 2
Snow 2
None 45
other 1
Total 50

85
 Figure 4.25. Frequency of accidents in different precipitation conditions

Despite the concerns about lack of visibility of messages during wind gust

condition, as shown in Table 4.7 and Figure 4.26, the statistical analysis regarding 43

accidents in impact area indicates there was not a significant number of accidents in these

adverse conditions.

Table 4.7. DMS accidents and wind gust

Wind Gust
Accidents in Impact Area #
(mph)
0-10 32
10-20 9
20-30 2
Total 43

86
Figure 4.26. DMS accidents and wind gust

87
4.12. Accidents in DMS Impact Areas and DMS Characteristics

This section details statistical analysis of accident types in DMS impact area, the

type of messages and beacon operational status (on/ off) of DMS. Figure 4.27 shows of

the 50 accidents in DMS impact areas, 35 collisions were property damage and 15 were

personal injury.

Researchers and laypersons have expressed concerns that flashing beacons could

distract drivers and negatively affect driving performance. As Figure 4.28 shows, 10

accidents (20 percent) occurred when beacons were on.

Figure 4.27. Type of accidents in DMS area #

88
 Figure 4.28. Number of accidents versus Beacon status

Analysis of displayed messages showed 11 accidents occurred when

Danger/Warning messages were displayed: 22 during Informative/Common Road

Condition messages and 17 during Regulatory/Non-Traffic-Related Messages. Although

concerns that accident-warning messages attract more attention from drivers than the

other types (Wang et al, 2007) and are thus more dangerous, the fewest number of

accidents happened during displays of Danger/Warning messages (see Figure 4.29).

89
Figure 4.29. Number of accidents for DMS message types

90
5. Chapter 5: Conclusions and Directions for Further Research

5.1. Summary and Conclusions

This project evaluated localized safety effects of highway Dynamic Message

Signs (DMS). Accident data from 2007 to 2010 was the basis for the analysis of road

collisions in Maryland. Accidents and message data in the study period were collected

from the Center for Advanced Transportation Technology (CATT) Laboratory in the

Department of Civil and Environmental Engineering at the University of Maryland and

Coordinated Highway Action Response Team (CHART) reports for regions within the

District of Columbia and Maryland. The roadway network map and AADT of roadway

segments were obtained from Maryland Department of Transportation State Highway

Administration (SHA), and weather conditions databases were gathered from DOT

archival data. This research project faced numerous challenges, including managing and

joining large databases with different data structures based on only time and location,

coordinating systems and working within confidentiality required of police accident

reports. Each was successfully overcome.

The accident database included 38,718 records, which were filtered and cleaned

and from which data gaps and outliers were removed. After cleaning, the number of

accidents decreased to 23,842 records for the four-year study period. The accident

database consisted of accident type (property damage, personal injury and fatality),

address location and county, time and date of accident and coordinates of accident

location. Due to confidentiality concerns, access to police records and accident causes

was not possible.

91
 The DMS types, obtained from CATT Laboratory, included permanently mounted

overhead, roadside models and portable signs operated by CHART or Maryland

Transportation Authority (MTA). The DMS database has 184 signs and the following

information associated with each: DMS ID, longitude and latitude, address location and DMS

type fields.

Traffic flow is another important contributing factor to crashes, so the AADT of

road segments was another factor this analysis accounted for. The AADT data was

obtained from Maryland SHA volume maps of the state of Maryland for study period.

The accidents, DMS locations and AADT database were projected onto a

Maryland roadway map to perform spot analysis in order to evaluate the influence of

DMS on drivers’ performance. An impact area of 900 feet was defined for each DMS

based on the average size of electronic signs character and maximum visibility distance

for the signs. A DMS was assigned to accidents within 900 feet of each DMS based on

location and direction of DMS.

A case study was performed on Interstate 95 in Maryland, a major highway. 70

samples of 900 feet segments along I-95 highway were chosen based on the homogeneity

of their geometry. The number of accidents were counted for each segment and

aggregated for use in regression analysis. Independent variables included whether the

segment was in an impact area or not, the existence of interchange in the segment and the

AADT of the segment. The results of unbalanced two-way ANOVA showed that

interchanges affect occurrence of accidents, whereas DMS do not. Results from Poisson

regressions supported this conclusion as well. The results for both methods converged on

the idea that interchanges and AADT are important factors on accidents, whereas DMS

are not.

92
 Another main factor that contributes to accidents is lack of visibility due to

adverse weather conditions. This project sought to determine whether adverse weather

conditions such as precipitation, wind gusts and severe weather negatively affected driver

performance by impairing visibility. For simplicity, the area of research was divided into

five regions: north, south, west, east, and Washington, DC. The nearest central weather

tower station in each region was assigned to represent the weather condition in each

region. The database was accumulated for the study period from 2007 to 2010. Each

accident was joined with its associated weather station. This weather database was joined to

the main database by the proximity of the closest weather tower station and the time and

location of each accident. The matching process was performed using SQL queries coded

in C++.

The database of messages was acquired from the CATT laboratory. This database

contained all the messages displayed on DMS in Maryland during the period study. The

database contained 1,047,586 records of messages, including their DMS ID, time of

displaying the message, the message text and beacon data fields.

The message log database was imported in SQL server and the main database. If a

record was located in impact area, the assigned DMS was matched with the message

displayed at the time of occurrence of accident. The same was done to match records with

weather data. The matching process was conducted using SQL queries coded in C++. The

integrated database consisted of 23,842 accident records during the study period. There

were 298 accidents within 900 feet of a DMS. 50 accidents occurred during times when

the DMS displayed messages. For the remaining accidents, the DMS were blank. The

data were analyzed in several aspects.

93
 The paired t-test analysis showed DMS do not increase the likelihood of accidents

occurring. The mean of the difference of the two accident rates was -0.013.

A one-way ANOVA using pairwise comparisons was used in an on-and-off analysis for

15 DMS. The results of this analysis showed the mean accident rate associated with

active DMS is lower than the inactive DMS.

The statistical analysis of accidents in conjunctions with weather conditions

showed that there are only four that occurred in rainy and snowy conditions. Thirty-twoof

43 accidents were in wind gusts of 0-10 mph condition, nine were in gusts of 10-20 mph,

and two were in wind gusts of 20-30 mph.

A statistical analysis of accidents revealed 35 of the 50 total collisions resulted in

property damage, and 15 in personal injury. There were no fatalities.

Ten accidents (20 percent) occurred while beacons were on. Analysis on

displayed messages showed 11 accidents occurred while Danger/Warning messages were

displayed, 22 occurred during Informative/Common Road Condition messages and 11

occurred during Regulatory/Non-Traffic-Related messages. Although some concerns

exist that accident-warning messages attract more attention from drivers, the fewest

accidents in DMS areas occurred when DMS displayed such messages.

In summary, the findings from all evaluations converge to indicate DMS are a safe tool

for disseminating real-time travel information to motorists because these signs largely do

not cause accidents by diverting drivers’ attention. This project focused on DMS

operations in Maryland, although the methods employed for evaluation can be extended

to other locations if suitable data are available.

94
5.2. Future Research

The broad range of subjects for future study provides opportunities and challenges for

researchers. The research could be further extended if future study areas include several

states. Future research in this area may be improved through investigating the issue

through simulation and site-human factor analysis. Also, it would be of interest to improve

DMS design (e.g., message design, size, color, length and number of panes and speed of

switching between messages) to enable drivers (especially older and bilingual drivers) to

more easily understand the content of DMS messages. Topics for future research include

investigations about the effects of displaying messages on newly installed DMS as well as

DMS on road curves. It would also be of interest to investigate differences in daytime and

nighttime situations. Another direction for future research concerns the extension of this

project to investigate the effect of incident messages and to provide motorists with

information about tailgating and secondary accidents close to the incident location.

Moreover, the integrated database could be used to investigate the impact of weather

conditions on occurrence of road accidents.

Finally, optimizing displayed messages and DMS location while accounting for

traffic flow, roadway geometry, and proximity to interchanges; and how to reduce

drivers’ mental processing time to perceive environmental factors and speed up drivers’

response could be other topics for future study. A cost-benefit analysis of installing DMS

could clarify concerns about expenses and values associated with the signs. These

directions for future studies would help transportation engineers and planners improve

DMS operations and eventually improve transportation network management and yield

smoother traffic flow.

95
6. References

1. Abbas, M.M. and McCoy, P.T. “Optimizing Variable Message Sign Locations on
Freeways Using Genetic Algorithms.” Presented at the 78th Annual Meeting of the
Transportation Research Board, Washington, D.C., 1999.

2. Abdel-Aty Mohamed A., Using ordered probit modeling to study the effect of ATIS
on transit ridership. Transpn. Res.-C 9, pp 265-277, 2000.

3. Adler J., Recker W. and McNally M. G., A conflict model and Interactive Simulator
(FASTCARS) for predicting enroute driver behavior in response to real-time traffic
condition information, Transportation 20, pp 83-106, 1993.

4. Bai, Y., Huang, Y., Schrock, S.D. and Li, Y., Determining the Effectiveness of
Graphic-aided Dynamic Message Signs in Work Zone, Smart Work Zone
Deployment Initiative/IDOT/FHWA, 2011.

5. Barfield, W., Conquest, L., Spyridakis, J. and Haselkorn, M. “Information

Requirements for Real-Time Motorist Information Systems.” Proceedings of the
Vehicle Navigation and Information Systems Conference (VNIS), New York,
IEEE, 1989, pp. 101-112.

6. Benekohal, R.F., Shu, J., Speed Reduction Effects of Changeable Message Signs in
a Construction Zone, Civil Engineering Studies, Transportation Engineeing, No. 72,
Illinois Cooperative Highway Research Program, Series No. 239, 1992.

7. Bonsell P. W., MERRALL A.C., Analyzing and Modeling the Influence of

th
Roadside Variable Message Displays on Drivers' Route Choice, 7 WCTR, Sydney,
1995.

8. Benson, B.G. Motorist Attitudes about Content of Variable-Message Signs.

Transportation Research Record, No. 1550, pp. 48-57, 1996.

9. Bolte, F. 2006. “Transport Policy Objectives: Traffc Management as Suitable

Tool,” Federal Highway Research Institute (BASt), Bergisch-Gladbach, Germany,
Presentation to Planning for Congestion Management Scan Team.
10. Borello, C. and Ornitz, S. E., Dynamic Message Sign Influence on Driver
Performance, ITE 2010 Annual Meeting and Exhibit, 2010.

96
11. Boyles, S. D. (2006). Reliable routing with recourse in stochastic, time-dependent
transportation networks. Master's thesis, The University of Texas at Austin.

12. Changeable Message Sign (CMS) Usage Procedure, New Jersey Department of
Transportation, 2008.

13. Chatterjee, K., Hounsell, N.B., Firmin, P. E. and Bonsall, P. W. Driver Response
to Variable Message Sign Information in London. Transportation Research Part C,
Vol. 10, No. 2, 2002, pp. 149-169.

14. Chien-Jung Lai, Effects of color scheme and message lines of variable message
signs on driver performance, Accident Analysis and Prevention 42 (2010) 1003–
1008, 2010.

15. Chiu, Y.-C. and N. Huynh, Location configuration design for dynamic message
signs under stochastic incident scenarios. Transportation Research Part C 15 (1),
33-50, 2007.

16. Chiu, Y.C., Huynh, N. and Mahmassani, H.S. “Determining Optimal Locations for
Variable Message Signs under Stochastic Incident Scenarios.” Presented at the 80th
Annual Meeting of the Transportation Research Board, Washington, D.C., 2001.

17. Dominique L. and Mannering, F., “The statistical analysis of crash-frequency data:
A review and assessment of methodological alternatives”, Transportation Research
Part A 44, 291–305, 2010.

18. Dos Santos, C., “Assessment of the Safety Benefits of VMS and VSL using the
UCF Driving Simulator”. Masters Thesis, University of Central Florida orlando,
2006.

19. Dudek CL, Changeable message sign displays during non-incident, non-
roadwork periods. NCHRP Synthesis 383. Transportation Research Board,
Washington, DC, 2008.

20. Dudek, C.L. Dynamic Message Sign Message Design and Display Manual. Report
No. FHWA/TX-04/0-4023-P3. Texas Transportation Institute, The Texas A&M
University System, College Station, TX, May 2006.

97
21. Dudek, C.L., S.D. Schrock and G.L. Ullman. Impacts of Using Dynamic Features
to Display Messages on Changeable Message Signs. Report No. FHWA-HOP-05-
069. FHWA, U.S. Department of Transportation, Washington, D.C., August 2005.

22. Dudek, C.L. Changeable Message Signs—Synthesis of Highway Practice 61.

National Cooperative Highway Research Program, 1979.

23. Erke, A., Sagberg, F., Hagman, R., Effects of route guidance variable message
signs (VMS) on driver behavior, Institute of Transport Economics, Gaustadalleen
21, NO-0349 Oslo, Norway, 2007.

24. Farradyne, P.B., Traffic Incident Management Handbook, Prepared for Federal
Highway Administration, Office of Travel Management, 2000.

25. Finley, M.D., Durkop, B.R., Wiles, P.B., Carvell, J.D. and Ullman, G.L. Practices,
Technologies and Usage of Incident Management and Traveler Information
Exchange and Sharing in Texas. Report 4951-1, Texas Transportation Institute,
December 2001.

26. Fish, R. L., Haghani, A. Hamedi, M., Analysis of Traffic Speed Response to
Display of Dynamic Message Sign Messages , Empirical Analysis Of The Quality,
Effectiveness and Localized Impacts Of Highway Dynamic Message Sign, TRB
Annual Meeting, 2012.

27. Guidelines for the Operation of Variable Message Signs on State Highways oregon
Department of Transportation, 2000.

28. Hao E., Taniguchi M., Sugie Y., Kuwahara M. and Morita H. (1999),
Incorporating an information acquisition process into a route choice model with
multiple information sources, Transport. Res.-C 7, pp 109-129, 1999.

29. Henderson, J. M. A planning model for optimizing locations of changeable

message signs. Master's thesis, University of Waterloo, 2004.

30. Highway Safety Manual (HSM), 1st edition, ASHTO, 2010.

31. HUO, H., Evaluation of Variable Message Signs Using Empirical Loop Detector
Data, Thesis UNIVERSITY OF MINNESOTA, 2002.

98
32. Huynh, N., Y.-C. Chiu and H. S. Mahmassani, Finding near-optimal locations for
variable message signs for real-time network traffic management. Transportation
Research Record 1856, 34-53, 2003.

33. Johnson, C.M. Use of Changeable Message Signs (CMS). FHWA Policy
Memorandums—Manual on Uniform Traffic Control Devices, January 19, 2001.

34. Jones, S.L., Jr. and Thompson, M.W. State of the Practice for Displaying Non-
traffic Related Messages on Dynamic Message Signs. Report No. 03413,
University Transportation Center for Alabama, August 2003.

35. Kiewit Center for Infrastructure and Transportation, Oregon State University
(2003), Analysis of Accident Statistics.

36. Khattak A.J., Schofer J. L. and Koppelman F. (1993), Commuters' Enroute

Diversion and Return Decisions: Analysis and Implications for Advanced Traveler
Information Systems, Transpn. Res.-A, Vol. 27A, No.2, pp.101-111.

37. Knoop, V.L., Hoogendoorn, S.P. and Van Zuylen, H.J. Route Choice Under
Exceptional Traffic Conditions. International Conference on Evacuation
Management, 23-25 September 2009, The Hague, the Netherlands.

38. Kraan M., Zijpp B., Tutert B., Vonk T. and Megen D. V., Evaluating Networkwide
Effects of Variable Message Signs in the Netherlands, Transportation Research
Record 1689, Paper No. 99-1126, 1999.

39. Levison, D. and Huo, H. (2003) “Effectiveness of Variable Message Signs.”

Presented at the 82nd Annual Meeting of the Transportation Research Board,
Washington, D.C.

40. Madanat, S., Yang, C.Y. and Yen, Y.M. Analysis of Stated Route Diversion
Intentions under Advanced Traveler Information Systems Using Latent Variable
Modeling. Transportation Research Record, No. 1485, 1995, pp. 10-17.

41. Manual on Uniform Traffic Control Devices (MUTCD). Federal Highway

Administration, U.S. Department of Transportation, Washington, D.C., 2000.

99
42. Molino, J., Wachtel, J., Farbry, J., Hermosillo, M., & Granda, T. The Effects of
Commercial Electronic Variable Message Signs (CEVMS) on Driver Attention
and Distraction: An Update (FHWA-HRT-09-018). Washington, DC: Federal
Highway Administration, 2009.
43. Mounce, J.M., Ullman, G., Pesti, G. and Pezoldt, V. Guidelines for the Evaluation
of Dynamic Message Sign Performance. Texas Transportation Institute, College
Station, 2007.

44. Operational Guidelines for the Use of Changeable Message Signs. North Carolina
Department of Transportation, 1996.

45. Peng, Z., Guequierre, N., Blakeman, J.C., Motorist Response to Arterial Variable
Message Signs Transportation Research Record: Journal of the Transportation
Research Board, No. 1899, TRB, National Research Council, Washington, D.C., ,
pp. 55–63, 2004.

46. Rama, P. & Kulmala, R.,Effects of variable message signs for slippery road
conditions on driving speed and headways. Transportation Research F, 3(2), 85-
94, 2000.

47. Ridgeway, R. Framework for harmonized Implementation of Variable Message

Signs in Europe, Rijkswaterstaat, Netherlands, 2003.

48. Rogers, John, “Evaluating the Impact of OOCEA’s Dynamic Message Signs
(DMS) on Travelers Experience Using the Pre-Deployment Survey.” University of
Central Florida, 2007.

49. Roshandeh, A.M. and Puan, A. C., "Assessment of impact of variable message
signs on traffic surveillance in Kuala Lumpur," in EEE International Conference
on Intelligence and Security Informatics, ISI 2009, p. 223-225.

50. Sara Nygårdhs, Literature review on variable message signs (VMS) 2006–2009,
VTI notat 15A-2011.

51. Schroeder, J. L. and Demetsky, M. J., Evaluation of Driver Reactions for Effective
Use of Dynamic Message Signs in Richmond, Virginia, Virginia Transportation
Research Council, 2010.

100
52. Sperry, R., McDonald, T., Nambisan, S. and Pettit, R., Effectiveness of Dynamic
Messaging on Driver Behavior for Late Merge Lane Road Closures, The Iowa
Department of Transportation, 2009.
53. Song, M., Maier-Speredelozzi, V., Wang, J-H, Cheung, S., Akdemir, M. K.,
Assessing The Slow-Down Effects Caused By Active Dynamic Message Signs
And Exploring Means To Ease The Slow-Down, available at:
www.onlinepubs.trb.org/onlinepubs/conferences/2011/RSS/2/Song,M.pdf.

54. SRF Consulting Group Inc. "Dynamic" signage: Research related to driver
distraction and ordinance recommendations (Report prepared for the City of
Minnetonka). Minnetonka, M, 2007.

55. Tarry, S. “A Framework for Assessing the Benefits of ITS.” Traffic Technology
International, Aug./Sept. 1996, pp. 25-30.

56. Tarry, S. and Graham, A.. The Role of Evaluation in ATT Development. Traffic
Engineering and Control. Vol. 36, No. 12, London, England, 1995, pp. 688- 693.

57. United States Department of Transportation, ITS Benefits and Unit Costs
Database, Access time: March, 2002:
http://www.benefitcost.its.dot.gov/its/benecost.nsf.

58. Walton, J.R., Barrett, M.L. and Crabtree, J.D. Management and Effective Use of
Changeable Message Signs. Report KTC-01-14/SPR233-00-1F, Kentucky
Transportation Center, 2001.

59. Wang, J.H., Keceli, M. and Maier-Speredelozzi, V., Effect of Dynamic Message
Sign Messages on Traffic Slow Downs, TRB 2009 Annual Meeting.

60. Wang, J-H,Collyer, C. E. and Clark A., Assisting Elder Drivers’ Comprehension
of Dynamic Message Signs, University of Rhode Island Transportation Center,
2007.

61. Wang, J-H, Hesar, S. and Collyer, C. E.. “Adding Graphics to Dynamic Message
Sign Messages.” Transportation Research Record 2018, (2007): 63-71.

62. Wardman M., Bonsall P. and Shires J., Driver Response to Variable Message
Signs: A Stated Preference Investigation, Transpn. Res.-C, Vol.5, No.6, 1998,
pp389-405, 1998.

101
63. Wendelboe, J., Toft., When Under Construction: Copenhagen's Motorway
Experience. Traffic Technology International, 2008, p. 74–76, 2008.

64. WTOP, Real-time travel signs become real pain for drivers (2010) , retrieved from:

http://www.wtop.com/?nid=&sid=1913047

102