ISSN 1055- 1425
May 2010
This work was performed as part of the California PATH Program of the
University of California, in cooperation with the State of California Business,
Transportation, and Housing Agency, Department of Transportation, and the
United States Department of Transportation, Federal Highway Administration.
The contents of this report reflect the views of the authors who are 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 State of California. This
report does not constitute a standard, specification, or regulation.
Final Report for Task Orders 5211/ 6211
CALIFORNIA PATH PROGRAM
INSTITUTE OF TRANSPORTATION STUDIES
UNIVERSITY OF CALIFORNIA, BERKELEY
Estimating Pedestrian Accident Exposure
UCB- ITS- PRR- 2010- 32
California PATH Research Report
UC Berkeley Traffic Safety Center
CALIFORNIA PARTNERS FOR ADVANCED TRANSIT AND HIGHWAYS
ESTIMATING PEDESTRIAN ACCIDENT EXPOSURE
Final Report
TO 5211/ 6211
for
California State of California Department of Transportation ( Caltrans)
Division of Research & Innovation
Estimating Pedestrian Accident
Exposure
TO’s 5211 & 6211
Final Report
UC Berkeley Traffic Safety Center
California Partners for Advanced Transit and Highways
for the
California Department of Transportation
The contents of this report reflect the views of the authors who are 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 State of California or the
Federal Highway Administration. This report does not constitute a standard,
specification or regulation.
This research was funded by the California Department of Transportation.
Executive Summary
We are pleased to present the final report of Caltrans Task Orders 5211 and 6211,
“ Estimating Pedestrian Accident Exposure.” The project focused on defining
pedestrian exposure and evaluating methods for measuring it within the State of
California. The project was funded by the California Department of Transportation
as part of the California Partners for Advanced Transit and Highways ( PATH)
Program of the University of California.
Deliverables associated with the project include ( I) a protocol report on assessing
pedestrian exposure, which is accompanied by a training curriculum and an
evaluation of manual pedestrian counting methods; ( II) an evaluation and test of
automated pedestrian counting methods; and ( III) a report on strategies to create a
statewide pedestrian exposure database and ( IV) a protocol for Pedestrian Exposure
Study in Alameda County. The deliverables are discussed in more detail below.
( I) Protocol report, training curriculum, and test of manual counting methods
The protocol report aims to assist transportation engineers and planners with the
task of measuring pedestrian exposure for a variety of purposes and contexts.
Purposes may include comparisons of the safety effects of pedestrian infrastructure;
comparisons of pedestrian risk among different population groups; or comparisons of
risk by mode of travel ( e. g. walking versus bicycling). The geographic contexts may
range from the entire state of California to a specific pedestrian crossing. Because
each possible purpose and context will have a unique set of considerations and
constraints, the protocol focuses on matching data collection methods with different
study needs.
The protocol report guides the user through the tasks of determining an appropriate
definition for pedestrian exposure; choosing the method of measurement that best
suits the data collection purpose; devising a sampling strategy; and estimating
annual pedestrian exposure from short samples of pedestrian volume. To
accompany the report, we created a six- module training curriculum in powerpoint
format. The course could be administered by Caltrans staff or local officials to
educate engineers and planners about the task of measuring pedestrian exposure.
We also conducted two supporting research efforts to support development of the
protocol. The first was a review of state- of- the- art pedestrian volume modeling
methods used to estimate pedestrian exposure, including sketch plan, network
analysis, and microsimulation models. The review was published in the 2006
Transportation Research Board Meeting CD- Rom as “ Pedestrian Volume Modeling
for Traffic Safety and Exposure Analysis: The Case of Boston, Massachusetts” and
is attached to this report.
The second supporting research effort we conducted was a detailed field test of
manual pedestrian counting methods. We compared the accuracy and
effectiveness of counts obtained from field observers and from manual review of
video recordings. The results of the test are attached as an appendix to the protocol
report, and was also published as “ Pedestrian Counting Methods at Intersections: a
Comparative Study” ( Section 1: Appendix B) in the 2007 edition of the
Transportation Research Record ( Vol. 2002).
( II) Evaluation and test of automated pedestrian counting methods
Several automated pedestrian detection technologies have emerged in recent years,
some of which can also be adapted for the purpose of pedestrian counting. These
devices have the potential to allow pedestrian data collection over extended periods,
and to reduce the labor costs associated with data collection.
We reviewed existing technologies using information from the literature, and
identified five technologies that could be adapted for the purpose of counting
pedestrians. We described each of these in our report on automated pedestrian
counting methods ( II). Based on the results of the review, we selected the passive
infrared sensing technology as the most promising candidate for further study,
because it is commercially available, not sensitive to lighting conditions, easy to
install, and has been used successfully in outdoor environments in the United
States. We conducted a test of this technology and included the results as an
appendix to the report on automated pedestrian counting methods.
( III) Pedestrian exposure database: Approaches to a Statewide Pedestrian Exposure
Database
Volume data is routinely collected for motorized modes but is not for non- motorized
modes. Such data is essential for tracking pedestrian exposure and for
infrastructure planning purposes. In this deliverable, we explore the possibility of
creating a formalized, institutionalized mechanism for pedestrian data collection
through a statewide pedestrian volume database. This database would meet a
variety of data needs for different stakeholder groups. One of its principal purposes
would be to allow safety professionals at the state and local levels to estimate
pedestrian exposure to risk at specific sites.
In the report, we discuss the technical and institutional challenges inherent in
creation of a pedestrian exposure database; possible sources of a pedestrian
network inventory; and possible approaches to data collection. In addition, we
recommend further steps for pursuing database development.
( IV) Alameda County Pedestrian and Bicycle Counting Protocol
This document describes the methods that will be used to collect pedestrian and
bicycle counts at a sample of roadway intersections in Alameda County. There are
two immediate purposes of this counting effort: a) obtain a sample of counts that can
be used as a basis for predicting the number of pedestrians and bicyclists at all 531
intersections of Caltrans roadways in the county, and b) demonstrate that the data
collection and modeling methods used in this pilot study have the potential to be
applied to Caltrans roadways statewide. Ultimately, the predicted pedestrian and
bicycle volumes can be used to represent exposure in a crash risk analysis. This will
allow Caltrans and Alameda County to evaluate and prioritize pedestrian and bicycle
safety needs more accurately at each intersection. The methods used in this effort
can be repeated by the County at regular intervals to track changes in pedestrian
and bicycle activity over time.
During the research process, we identified several areas for further research. Two
of these in particular stand out. First, we determined that the goal of a statewide
pedestrian database could be furthered through research into a pilot database. This
could be achieved either by collecting a sample of pedestrian volumes at locations in
the state highway network, which could then be entered into the TASAS database,
or by developing and sampling a GIS- based inventory of the pedestrian network in
one of the Caltrans districts ( e. g. District four). Second, we determined that the
phenomenon of pedestrian “ safety in numbers” has very important implications for
the measurement of pedestrian risk and deserves immediate study. This
phenomenon potentially undermines the usefulness of pedestrian collision rates as a
proxy for pedestrian risk. Further research is needed to determine whether the safety
in numbers phenomenon is a result of pedestrian or driver behavior; built
environment factors; or other sources.
The “ Alameda County Pedestrian and Bicycle Counting Project Summary” is a 5-
page document outlining the final effort of this task order. It contains the section
“ Extrapolating Weekly Pedestrian Intersection Crossing Volumes from 2- Hour
Manual Counts” and “ A Pilot Model for Estimating Pedestrian Intersection Crossing
Volumes,” highlighting the research design, findings, and considerations.
Keywords: Pedestrians, Exposure, Intersections, Pedestrian Counts, Pedestrian
Traffic, Pedestrian Accidents, Risk Analysis, Pedestrian Volume, Pedestrian
Movement
FINAL REPORT TABLE OF CONTENTS
I. Protocol Report
Appendix A: Example of a Tally Sheet Used to Count Pedestrians
Appendix B: Supporting Paper: Pedestrian Counting Methods at
Intersections: A Comparative Study
II. Automated Pedestrian Counting Devices Report
III. Pedestrian Exposure Database Report: Approaches to a Statewide
Pedestrian Exposure Database
Supporting Paper: Pedestrian Volume Modeling for Traffic Safety and
Exposure Analysis: The Case of Boston, Massachusetts
IV. Alameda County Pedestrian and Bicycle Counting Protocol
Alameda County Pedestrian and Bicycle Counting Project
Summary
ESTIMATING PEDESTRIAN ACCIDENT EXPOSURE
Protocol Report
MARCH 2007
Dan Burden
1
The mission of the UC Berkeley Traffic Safety Center is to reduce traffic fatalities and injuries through
multi- disciplinary collaboration in education, research, and outreach. Our aim is to strengthen the
capability of state, county, and local governments, academic institutions, and local community
organizations to enhance traffic safety through research, curriculum and material development,
outreach, and training for professionals and students.
ESTIMATING PEDESTRIAN ACCIDENT EXPOSURE
Protocol Report
Prepared for Caltrans under
Task Order 6211
Prepared by
RYAN GREENE- ROESEL
MARA CHAGAS DIOGENES
DAVID R. RAGLAND
University of California Traffic Safety Center
University of California
Berkeley, CA 94720
Tel: 510/ 642- 0655
Fax: 510/ 643- 9922
2
ACKNOWLEDGEMENTS
The University of California Traffic Safety Center ( TSC) appreciates and
acknowledges the contributions of the following participants.
TSC staff:
Noah Radford
Marla Orenstein
Judy Geyer
Kara MacLeod
Tammy Wilder
PATH staff:
Ashkan Sharafsaleh, Research Engineer
Steve Shladover, Research Engineer
Fanping Bu, Research Engineer
Specialized consultants:
Charles Zegeer
Funding for this project was provided by Caltrans.
LIST OF CONTENTS
LIST OF TABLES AND FIGURES ___________________________________________________ 4
1. PREFACE__________________________________________________________________ 6
2. PEDESTRIAN EXPOSURE ___________________________________________________ 10
3. AREA- WIDE METHODS _____________________________________________________ 30
4. SITE SPECIFIC METHODS ___________________________________________________ 41
5. DATA COLLECTION PLANNING AT INTERSECTIONS ____________________________ 53
6. ESTIMATING ANNUAL PEDESTRIAN VOLUMES_________________________________ 65
REFERENCES _________________________________________________________________ 82
APPENDIX A: Example of a Tally Sheet Used to Count Pedestrian ______________________ 92
APPENDIX B: Comparative Study between Manual Count Methods _____________________ 93
4
LIST OF TABLES AND FIGURES
Table 2.1: Exposure versus Risk____________________________________________________ 10
Table 2.2: Fatality Risks over Distance and Time for Travel Modes in the EU _________________ 15
Table 2.3: Common Metrics Used to Describe Pedestrian Exposure ________________________ 16
Table 2.4: Exposure Based on Population Data ________________________________________ 17
Table 2.5: Exposure Based on Pedestrian Volume______________________________________ 18
Table 2.6: Exposure Based on Trips_________________________________________________ 20
Table 2.7: Exposure Based on Distance______________________________________________ 22
Table 2.8: Exposure Based on Time_________________________________________________ 23
Table 3.1: Block Group Level A Summary of ACS Data Availability _________________________ 36
Table 3.2: Characteristics of Existing Pedestrian Related Surveys__________________________ 36
Table 3.3: Comparison of Modeling Methods __________________________________________ 39
Table 3.4: Comparison of Approaches to Pedestrian Volume Estimation_____________________ 40
Table 4.1: Comparison of Methods to Count Pedestrian at Crossings _______________________ 52
Table 5.1: Non- Probabilistic Sampling Techniques______________________________________ 56
Table 5.2: Probabilistic Sampling Techniques _________________________________________ 57
Table 5.3: Stratification Variables ___________________________________________________ 64
Table 6.1: Characteristics of Strata__________________________________________________ 69
Table 6.2: Categories of Area Type _________________________________________________ 71
Figure 2.1: Number of pedestrians injured or killed in New Zealand, 1988- 91 _________________ 13
Figure 2.2: Number of pedestrian casualties per million hours walked in New Zealand 1988- 91 __ 13
Figure 2.3: Assumed relationship between exposure and number of collisions ________________ 26
Figure 2.4 Non- Linear Relationship Between Exposure and Accidents ______________________ 27
Figure 4.1: Field Observation using clickers___________________________________________ 43
Figure 4.2: Video- image camera angle and resolution ___________________________________ 43
Figure 4.3: Path QuickTime Playback Tool____________________________________________ 44
Figure 4.4: Video Imaging for Counting Pedestrian _____________________________________ 51
Figure 4.5: Irysis Infrared Pedestrian Counting Device __________________________________ 51
Figure 5.1: Generalized Model of Sampling ___________________________________________ 53
Figure 5.2: Classification of Sampling Techniques ______________________________________ 55
Figure 5.3: Methodology for Planning Pedestrian Exposure Data Collection at Intersections _____ 60
Figure 5.4: Sampling Design Steps for Pedestrian Exposure Data Collection at Intersections ____ 61
Figure 6.1: 12- hour Pedestrian Volume Distribution Patterns at Sites in Washington, D. C._______ 70
Figure 6.2: Relationship between maximum expected sampling error and sampling time for various
levels of pedestrian activity ________________________________________________________ 75
Figure 6.3: Daily volume adjustment factors developed for CBD, Fringe, and Residential Sites ___ 79
5
Figure 6.4: Comparison of daily pedestrian crossing volume distributions in Israel, Germany, and
Australia_______________________________________________________________________ 80
1. PREFACE
1.1. Purpose of the Protocol
Walking is a healthful, environmentally benign form of travel, and is the most basic
form of human mobility. Walking trips account for more than 8 percent of all trips
taken in California, making walking the second most commonly used mode of travel
after the personal automobile ( Caltrans, 2002). In addition, many trips made by
vehicle or public transit begin and end with walking.
In spite of the importance and benefits of walking, pedestrians suffer a
disproportionate share of the harm of traffic incidents in California. As noted above,
walking trips make up just 8 percent of all trips in the state, but 17 percent of all
traffic fatalities are suffered by pedestrians. In 2004, 694 pedestrians were killed in
the state of California and 13,892 were injured ( California Highway Patrol, 2004).
To address this problem, significant resources are focused on countermeasures that
aim to reduce the risk of pedestrian injury. Because resources are limited, risk
analysis is necessary to develop cost- effective countermeasures ( Høj and Kröger,
2002).
In the field of pedestrian safety, risk analysis involves assessing factors that
contribute to the danger that a pedestrian is struck by a vehicle. These factors may
include physical characteristics of the street, such as lack of sidewalks; behavioral
issues, such as pedestrian or driver alcohol use; as well as other environmental
variables. In order to fully understand how these factors contribute to risk, it is
necessary to collect information on pedestrian exposure. Collection of pedestrian
exposure information is an essential component of risk analysis.
Pedestrian exposure is a concept that refers to the amount that people are exposed
to the risk of being involved in a traffic collision. In principle, pedestrians are exposed
to this risk whenever they are walking in the vicinity of automobiles. There are many
metrics that can be used to measure pedestrian exposure, but pedestrian volumes
are the most frequently used.
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 7
Although many state, regional, and local agencies have developed methodologies to
collect pedestrian volume data, there is no consensus on which method is best
( Schneider et al., 2005; Schweizer, 2005). This is because there is no “ one size fits
all” method of counting pedestrians. Rather, the choice of strategy depends on a
complex range of factors, including the characteristics of the area being studied; the
resources available for data collection; and the specific purpose of data collection.
This protocol aims to improve pedestrian data collection in the state of California by
providing information and guidance for each decision point in the data collection
process. Each chapter represents one of these decision points, and each will guide
the user through important considerations relevant to the data collection stage. In
addition, each chapter provides a combination of real- world and hypothetical
example scenarios to illustrate the issues discussed in the text.
The first chapter, “ Pedestrian Exposure,” discusses the issue of how to select a
definition of pedestrian exposure that is appropriate to the study purposes,
resources, and chosen counting method. It also discusses the meaning of pedestrian
exposure and its importance in pedestrian risk analysis.
The second chapter, “ Area- Wide Methods,” describes three general approaches to
measuring pedestrian exposure for defined geographic areas, such as cities or
counties. This chapter assists users in understanding the strengths and weakness of
different methods of measuring pedestrian exposure over wide areas, and introduces
users to existing sources of data on pedestrian activity.
The third chapter, “ Site- Specific Methods,” focuses on commonly used methods for
counting pedestrian activity directly at specific sites, such as intersections or
crossings. The performance of these methods is evaluated in terms of their relative
cost, convenience, accuracy, and ability to collect a range of data points.
The fourth chapter, “ Data Collection Planning at Intersections,” assists users with the
task of planning data collection at specific sites. It describes the statistical issues that
must be addressed when designing a pedestrian data collection strategy, such as
how to choose which sites to study and how to determine the number of sites to be
studied.
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 8
The fifth chapter, “ Estimating Annual Pedestrian Volumes,” describes a method for
converting short pedestrian counts into an annual measure of pedestrian volume
using statistical analysis of pedestrian flow patterns. This method can be used to
reduce the time and cost associated with developing an annual measure of
pedestrian exposure, which is necessary to determine the annual pedestrian risk at a
site.
Taken together, these chapters will assist the user in measuring pedestrian exposure
for a variety of purposes and contexts. The purposes may include comparisons of
the safety effects of pedestrian infrastructure; comparisons of pedestrian risk among
different population groups; or comparisons of risk by mode of travel ( e. g. walking
versus bicycling). The geographic contexts may range from the entire state of
California to a specific pedestrian crossing. Because each possible purpose and
context will have a unique set of considerations and constraints, this protocol
focuses on matching data collection methods with different study needs.
1.2. Who Should Use this Protocol
This protocol is intended to be used by traffic engineers and planners, consultants,
and researchers interested in measuring pedestrian exposure. Although unaffiliated
users will benefit from reading the protocol, it is most appropriate for those who are
associated with an institution that has the resources necessary to mount a data
collection program.
1.3. How to Use this Protocol
As discussed above, each chapter is aimed at a particular aspect of the data
collection process. Some users may wish to read only the section that is most
relevant to their needs. However, because the issues in the chapters are closely
inter- related, many users will benefit from reading the entire document.
Users should understand that this protocol is not a “ how- to” guide for measuring
pedestrian exposure. Although many specific methods and equations are provided,
the intention is to educate the user about the data collection process rather than to
provide a set of instructions. This is because, as mentioned above, measuring
pedestrian exposure is a complex task that is constrained by the study resources,
purposes, and context. This protocol aims to inform the user about the data
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 9
collection strategies available to them, and to assist them in choosing which one best
meets their needs.
2. PEDESTRIAN EXPOSURE
Before seeking to measure pedestrian exposure, it is important to have a clear
understanding of the concept and its relationship to pedestrian risk. This chapter
discusses the meaning of exposure in the context of risk analysis for pedestrian
safety, and presents several common measures of pedestrian exposure used in the
transportation safety field.
As this guide will demonstrate, there is no single best measure of pedestrian
exposure, but some measures are better adapted to specific needs and purposes,
such as comparing infrastructure; comparing risk among populations; or evaluating
the change in pedestrian risk over time. This chapter will assist users in selecting an
appropriate measure of exposure to match their needs.
2.1. Understanding Exposure and Risk
In epidemiology, exposure refers to a person’s contact with a potentially hazardous
situation or substance. For example, each time you fly in an airplane, you are
exposed to ionizing radiation. Each time you cross a street, you are exposed to the
possibility of being injured by a vehicle. Exposure can also be understood as a “ trial
event” in during which a harmful outcome might occur.
Risk is an abstract concept that refers to the probability a harmful event will occur
given a certain number of trials. In pedestrian safety, each “ trial” is a unit of exposure
such as a minute spent walking or a road crossing Table 2.1 describes the
relationship between exposure and risk.
Table 2.1: Exposure versus Risk
Exposure Contact or amount of contact with potentially
harmful situation ( x) ( x)
Risk Probability of collision/ injury/ fatality ( c) per unit of
exposure. P( c⎪ x)
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 11
The likelihood that any given trial event will result in a particular outcome is a
function of the “ chance set up”. In transport safety, the “ chance set up” is the
transportation system itself, including its physical characteristics, users, and
environment. Any one of these characteristics might influence the likelihood that a
given trial event – such as a pedestrian crossing – will result in a collision ( Hauer,
1982).
Risk and exposure are theoretical concepts that can only be indirectly estimated
through the use of proxy measures. In the field of traffic safety, risk is typically
represented by a simple ratio between collisions, injuries or fatalities, and exposure
for a specific geography and time period ( Chu, 2004). This ratio is referred to as the
“ collision rate” or the “ accident rate”. See Section 2.6 for a discussion of the
limitations of collision rates as a proxy for risk.
Collision rate = Number of collisions in a specified time and place ( 1)
Amount of exposure in a specified time and place
If one finds that risk is higher at one intersection than another, it suggests that
something in the “ chance set up” ( e. g. higher traffic speeds at one intersection)
explains the difference. In this way, risk analysis is used to identify dangerous
aspects of the transportation environment.
A short list of some of the factors thought to be associated with pedestrian risk
include:
􀀹 Pedestrian characteristics including age and gender ( Evans, 1991; Keall, 1995),
and socioeconomic status and ethnicity ( Ogden, 1997; Kraus et al., 1996). These
characteristics may be related to distance and time traveled; pedestrian behavior;
and awareness of the road environment.
􀀹 Pedestrian behavioral characteristics, such as risk- taking behavior, propensity to
jaywalk, etc ( Campbell et al., 2004).
􀀹 Trip characteristics: time of day/ year, purpose, time elapsed between drinking
alcohol and commencement of trip ( Keall, 1995).
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 12
􀀹 Area characteristics related to transportation service and land use ( Herms, 1970;
Ossenbruggen, 1999).
􀀹 Roadway features such crosswalks and alternative crossing treatments,
signalization, signing, pedestrian refuge islands, provisions for pedestrians with
disabilities, bus stop location, and school crossing measures ( Campbell et al.,
2004).
2.2. Incorporating Exposure into Risk Measurement
Exposure is a crucial component of risk measurement. If the absolute number of
injuries or fatalities is presented without controlling for exposure, it is easy to come to
erroneous conclusions about risk.
The following graphs are provided to illustrate the importance of incorporating
pedestrian exposure into measurement of risk. Figure 2.1 shows the number of
pedestrians killed in New Zealand between 1988- 1991, ordered by age and gender.
These “ raw” counts make it seem that children under twenty are most in danger of
being killed.
However, when the raw counts are presented as a function of exposure, measured
as the hours spent walking, a very different picture emerges ( Figure 2.2). The age
categories with the highest risk are those aged 80 and above and those ten and
younger. Adolescents aged 15- 20 do not have elevated risk levels; rather, the high
numbers of fatalities in this category are due the fact that adolescents spend more
time walking than other age groups.
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 13
Figure 2.1: Number of pedestrians injured or killed in New Zealand, 1988- 91 ( Keall, 1995)
Figure 2.2: Number of pedestrian casualties per million hours walked in New Zealand 1988- 91
( Keall, 1995)
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 14
When constructing a pedestrian safety risk measure, it is important to keep the
following points in mind:
􀀹 The numerator and denominator in a risk measure must be consistent ( Hauer,
2001); if exposure is in person- hours of pedestrian travel then the event in the
numerator should be the number of pedestrians that experienced a collision or
injury.
􀀹 The risk measure should reflect the type of risk being studied ( Hakkert and
Braimaister, 2002), such as whether the risk being studied is for an individual, or
for a defined social group ( Jorgensen, 1996).
􀀹 The denominator of the risk measure ( pedestrian exposure) must reflect the
intended purpose of the risk measure ( Hakamies- Blomqvist, 1998). For example,
a risk measure used to compare risk between different modes of travel should
have a denominator ( exposure measure) that is comparable across all modes.
􀀹 The denominator of the risk measure should reflect the target population being
studied.
2.3. Defining Pedestrian Exposure
Pedestrian exposure is an abstract concept that reflects the opportunity for a
potentially harmful pedestrian- vehicle interaction to occur; in other words, it is the
number of trial events that could result in an injury or collision. It is very difficult to
measure directly, since this would involve tracking the movements of all people at all
times.
Instead, pedestrian exposure must be approximated using an appropriate proxy
measure. Examples of measures used to represent pedestrian exposure at the micro
level include pedestrian volume ( Davis et al., 1988); the product of pedestrian and
vehicle volumes at an intersection ( Cameron, 1982) or roadway segment ( Knoblauch
et al., 1984); and the square root of that product ( TRL, 2001). Measures used to
represent exposure at the macro level in the U. S. include pedestrian distance
traveled and pedestrian trips made ( Pucher and Dijkstra, 2000, 2003); and the
number of streets crossed ( Roberts et al., 1996). In Europe, the most common
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 15
measures include the number of pedestrian trips made; time spent walking; and
distance walked ( ETSC, 1999).
In situations where travel- based measures of exposure are unavailable, population-based
measures are sometimes used to approximate exposure ( NHTSA, 2004).
These may include population density ( Qin and Ivan, 2001), and population divided
by the percent of workers who reported that they usually walked to work in the last
week ( STPP, 2002, 2004).
The choice of exposure measure strongly impacts the resulting calculation of risk.
For example, researchers at the Surface Transportation Policy Project used “ miles
traveled” as the denominator in estimating risk to pedestrians across the nation in
the 2004 Mean Streets report. They concluded that walking is about twenty times
more dangerous than riding in passenger cars, trucks, or on public transit ( STPP,
2002, 2004). This conclusion can be distorted by the fact that walking is much slower
per mile than other forms of transportation. If the researchers had used as the
measure of exposure the amount of time spent traveling, rather than miles traveled,
they may have reached different conclusions.
To illustrate further, Table 2.2 presents pedestrian collision rates in the European
Union calculated using two different exposure measures: person- kilometers traveled
and person- hours of travel. When person- kilometers walked is the measure of
exposure, pedestrian travel appears to be many times riskier than travel by car.
When person- hours spent walking is the exposure measure, then pedestrian travel
appears to have the same risk as vehicle travel.
Table 2.2: Fatality Risks over Distance and Time for Travel Modes in the EU
Travel mode 108 person km 108 person hours
Total 1.1 33
Bus/ Coach 0.08 2
Car 0.8 30
Foot 7.5 30
Cycle 6.3 90
Road
M/ C, MOPED 16.0 500
Trains 0.04 2
Ferries 0.33 10.5
Planes 0.08 36.5
Source: ETSC, 1999
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 16
2.4. Measures of Pedestrian Exposure
Presented in Table 2.3 is an exploration of some of the common ways that
pedestrian exposure is measured. For each of these exposure measures, an
explanation and examples are provided; common and appropriate uses are
discussed; and benefits and limitations are explored. Not all possible ways of
estimating pedestrian exposure are described.
Table 2.3: Common Metrics Used to Describe Pedestrian Exposure
Explanation
Population Number of residents of a given area, or number of people in a demographic
group.
Number of
pedestrians
Number of pedestrians observed in a given area during a fixed interval.
Trips Number of distinct trips taken by an individual pedestrian.
Distance traveled Total distance traveled by an individual pedestrian or aggregate distance
traveled by all pedestrians in a fixed area.
Time spent
traveling
Total time traveled by an individual pedestrian or aggregate time traveled by all
pedestrians in a fixed area.
These examples will illustrate that there is no single best definition of pedestrian
exposure. However, it is important to choose the definition of exposure that best
matches the needs and purposes of the study. The chosen exposure measure
should be compatible with the measurement devices being used and the target
population being studied within a geographic area. The choice of exposure measure
will also be determined in part by the amount of available resources, as some
measures of exposure are more costly to collect than others.
2.4.1. Exposure based on population data
Population refers to the number of people who live in a given area, or the number of
people who make up a particular demographic group. Because it is relatively easy
and cheap to estimate, population data is often used as a simple proxy for
pedestrian exposure.
There are a large number of issues that make the use of population highly unreliable
as an exposure estimate. First of all, actual physical exposure to traffic is unlikely to
be evenly distributed throughout the population. Second, time spent as pedestrians,
or distance traveled, are not represented or accounted for in any way. Third,
population does not necessarily relate directly to the actual number of people
walking on the streets.
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 17
For example, some tourist sites attract a large number of people who are not
accounted for by residential or employment population density, but who may still be
involved in traffic collisions ( Ivan et al., 2000). Models of pedestrian risk based on
population provide only the roughest approximation, and are probably unreliable.
Table 2.4 summarizes the issues related to exposure measures based on
population.
Table 2.4: Exposure Based on Population Data
APPROPRIATE
USES
􀀹 Used as an alternative to exposure data when cost constraints make
collecting exposure data impractical
􀀹 Used to compare jurisdictions over time because population data is
available for many geographies and time periods
HOW DATA IS
GATHERED
􀀹 Population data for most cities is available on an annual basis through the
American Community Survey ( ACS). The ACS is administered by the U. S.
Bureau of the Census and is accessible online ( U. S. Census Bureau,
2006)
PROS 􀀹 Easy and low- cost to obtain; available for most geographies and time
periods
􀀹 Adjusts for differences in the underlying resident population of an area – for
example, sparsely populated suburbs versus densely populated inner- city
areas
􀀹 Provides a crude adjustment for amount of vehicle traffic on the streets,
since areas where more people live also tend to be areas where more
people drive
􀀹 May be the only way to represent exposure if direct measurements cannot
be taken
CONS
􀀹 Does not accurately represent pedestrian exposure
􀀹 Does not account for the number of people who travel as pedestrians in the
area
􀀹 Does not provide information about amount of time or distance that
members of the population were exposed to traffic
COMMON
MEASURES
􀀹 Number of people in a given area: neighborhood, city, county, state or
country
􀀹 Number of people in a particular demographic group: by age, sex, race,
immigrant status or socioeconomic status
EXAMPLES
􀀹 In 2001, pedestrian collisions killed 20 people per million in California, but
only 7 people per million in Nebraska. ( FARS and U. S. Census data from
2001).
􀀹 In 2004, the male pedestrian fatality rate per 100,000 population in United
States was 2.22, while the female pedestrian fatality rate was 0.95 per
100,000 population ( NHTSA, 2004).
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 18
2.4.2. Exposure based on pedestrian volumes
Pedestrian exposure can be measured by the number of pedestrians that pass
through a fixed point during a specified time interval. This is a common exposure
metric, as it is relatively simple to assess through established manual and automated
counting methods. This exposure measure is explained in more detail on Table 2.5.
Table 2.5: Exposure Based on Pedestrian Volume
APPROPRIATE
USES
􀀹 Estimating pedestrian volume and risk in a specific location.
􀀹 Assessing changes in pedestrian volume or characteristics due to
countermeasure implementation at that site.
HOW DATA IS
GATHERED
􀀹 Manual or automated counts of pedestrians.
PROS 􀀹 Counts are simpler to collect than other measures such as time or
distance walked.
􀀹 Automated methods for counting number of pedestrians are improving.
CONS 􀀹 Does not differentiate pedestrians by walking speed, age, or other factors
that may influence individual risk.
􀀹 Does not account for the amount of time spent walking or the distance
walked
􀀹 Not easily adapted to assess exposure over wide areas ( for example, a
city).
COMMON
MEASURES
􀀹 Average number of pedestrians per day, sometimes called Average
Annual Number of Pedestrians ( Zeeger et al., 2005; Cameron , 1976,
Hocherman et al., 1988)
􀀹 Number of pedestrians per time period, e. g., hour ( Davis et al., 1988;
Cove and Clark, 1993)
EXAMPLES 􀀹 The average daily pedestrian traffic at marked crossings was 312
pedestrians per site ( Zeeger et al., 2005).
􀀹 Between 7: 00 am and 10: 00 am, 203 pedestrians crossed Rose Street at
the intersection of Shattuck Avenue.
While the “ number of pedestrians” is the term most frequently used to refer to this
exposure variable, that terminology is not, strictly speaking, accurate. A more precise
term is ‘ number of pedestrian crossings’, since a single pedestrian can contribute to
the count more than once if that person passes through the measurement point more
than one time during the observation period ( such as during an outbound journey,
and then again on the return). In addition, it is important to distinguish whether the
crossing is over a roadway or over an arbitrary line on a sidewalk. Statistics suggest
that crossing the street might be more dangerous than walking along the road, so
that crossing exposure should be distinguished from roadside or sidewalk exposure
( Evans, 1991; Ossenbruggen, 1999).
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 19
Key to the accurate measurement of the number of pedestrians is a good operative
definition of what constitutes an entry into the area, and what constitutes a
pedestrian. For example, should a mother pushing an infant in a stroller be counted
as one pedestrian, or two?
Any fixed point can be used. However, in practice, intersection crossings are often
used as the fixed point. The reason for this is that crossing the street is an activity
with a relatively high risk. In a study of pedestrian crash types across several states,
Hunter et al. ( 1996) found that about a third of crashes involving a pedestrian occur
at intersections, whereas only about 8 percent of all crashes occurred while the
pedestrian was walking along the roadway.
A major assumption made in using an intersection as a fixed point is that each
crossing represents a fixed unit of risk, independent of crossing distance or location
within the crossing.
2.4.3. Exposure based on trips
Exposure based on number of trips estimates the number of walking trips taken by
an individual, regardless of the distance or time the journey takes. Trips may be
taken for the purpose of commuting to work or school, for social visiting, for utilitarian
purposes such as shopping, for walking a dog, or walking purely for recreation. This
information is generally gathered by surveying a representative subset of a
population. Because other survey questions are usually asked at the same time,
each trip can be linked to information regarding trip purpose, time of day, etc.
Number of trips as assessed by survey is usually difficult to relate to pedestrian
collision data on a small- area scale. However, the data is useful to assess exposure
over wide areas, especially when combined with other datasets, such as U. S.
Census information or land use data, enabling additional analyses of factors
affecting walking patterns.
Number of trips may not be the most useful metric for risk analysis purposes, but it is
commonly used for assessing pedestrian behavior and activity, for making
comparisons between large jurisdictions, and for examining changes over time
( Table 2.6).
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 20
Table 2.6: Exposure Based on Trips
APPROPRIATE
USES
􀀹 Assessing pedestrian behavior in large areas, such as cities, states, or
countries.
􀀹 Examining changes in pedestrian behavior over time.
􀀹 Making comparisons between jurisdictions.
􀀹 Assessing common characteristics of walking trips, such as purpose,
route, etc.
HOW DATA IS
GATHERED
􀀹 Data is gathered through use of surveys, such as the National Household
Travel Survey ( 2001)
PROS 􀀹 Appropriate for use in large areas.
􀀹 Best metric to assess relationship of walking with trip purpose
􀀹 Trips can be assessed as a function of person, household and location
attributes.
CONS 􀀹 As with most surveys, a large number of respondents are needed to
adequately represent the underlying population.
􀀹 Unlikely to provide information at the level of detail needed to assess risk
at specific locations
􀀹 Pedestrian trips are often underreported in surveys ( Schwartz and Porter,
2000)
COMMON
MEASURES
􀀹 Average number of walking trips made by members of a population per
day, week or year.
􀀹 Proportion of walking trips taken for particular purposes, such as
commuting or shopping.
EXAMPLES 􀀹 In US, the percentage of all work trips made by walking fell from 10.3% in
1960 to only 2.9% in 2000 ( Pucher and Dijkstra, 2003).
􀀹 While in the Mid- Atlantic States 15.8% of all trips are made by the walking
mode, in the East South Central and West South Central states this
percentage is around 6% ( Pucher and Renne, 2003).
􀀹 In US, 38% of all pedestrian trips are made for social and recreational
purposes and 32% for going to school and church, while 10% represent
work trips ( Pucher and Renne, 2003).
2.4.4. Exposure based on distance
Exposure based on distance, or distance traveled, represents the distance that
pedestrians walk while exposed to vehicular traffic. This exposure measurement can
be assessed on the level of the individual or on the level of the geographic area. On
the individual level, exposure based on distance is expressed as the total or average
distance that an individual pedestrian travels in a fixed time period, such as a day,
week, or year. Typically the risk is stated in terms of the number of deaths per 100
million person miles traveled ( Chu, 2003). As with the measurement of number of
trips, assessment of this exposure measure is carried out through surveys of a
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 21
representative sample of the population. It is also possible to attach walking
measurement devices, such as pedometers, to a sample of pedestrians.
On the geographic level, distance traveled is measured directly by aggregating the
pedestrian distance traveled within a defined area during a fixed time period. This
version of distance traveled is defined as the number of pedestrians counted,
multiplied by the distance across the intersection. In this instance, the focus is on the
total pedestrian- miles traveled, not the number of unique individuals traveling, and
each individual may contribute distance more than once, if they pass through the
observation area more than one time.
Using exposure based on distance to estimate risk, through either of the methods
presented above, relies on the assumption that risk is a function of distance traveled.
That means that other things being equal, crossing a roadway with four lanes carries
twice the risk of crossing a roadway with two lanes.
The metric does not differentiate in terms of walking speed or other factors that could
moderate the risk associated with distance. This potentially distorts the risk
associated with walking when compared to other modes. One person- mile of walking
represents far more exposure to vehicle traffic than one person- mile of riding in a
passenger vehicle because of the differences in travel speeds between the modes
( Chu, 2003). Thus, using a distance- based measure of exposure when comparing
risk between modes may distort the results of the comparison. Table 2.7 presents
more details about exposure measure based on distance.
2.4.5. Exposure based on time
Time exposure data has long been used for measuring risk ( Jonah and Engel 1983;
Anderson et al., 1989; ETSC, 1999). It has also been used to compare risk in
different social groups or between travel modes. Keall ( 1995) estimated the risks of
traffic collision for different sex and age groups by combining road collision data with
survey data using the exposure measures “ time spent walking” and “ number of roads
crossed”. Chu ( 2003) proposed a time- based comparative approach to examining
the fatality risk of walking and vehicle travel because time- based measures take into
account the speed differences between walking and riding in a passenger vehicle.
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 22
Exposure based on time incorporates not only the distance traveled, but also adjusts
for walking speed. Like distance traveled, time traveled can be measured on the
individual level through surveys or through direct measurement at specific locations.
Time spent walking at a crossing, for example, might be measured by multiplying the
number of pedestrians by the average crossing time. It can also be measured by
adding the crossing times of each individual. In comparing two individuals, all other
characteristics being equal, the measure will account for different walking speeds.
To better characterize the exposure measure based on time, Table 2.8 presents its
appropriate uses and examples.
Table 2.7: Exposure Based on Distance
APPROPRIATE
USES
􀀹 Estimating exposure at the micro or macro level.
􀀹 Estimating whether risk increases in a linear manner with distance
traveled.
􀀹 Assessing how crossing distance affects risk
HOW DATA IS
GATHERED
􀀹 For individual level exposure, through surveys such as the National
Household Travel Survey ( 2001)
􀀹 For aggregate level exposure, measurement of the length of the area of
interest, combined with a manual or automatic count of the number of
pedestrians.
PROS 􀀹 Can be used to measure exposure at the micro and macro levels
􀀹 More detailed than pedestrian volumes or population data
􀀹 Can be used to compare risk between different travel modes
􀀹 Common measure of vehicle exposure
CONS 􀀹 Does not take into account the speed of travel and thus cannot be reliably
used to compare risk between different modes ( e. g. walking and driving)
􀀹 Assumes risk is equal over the distance walked
􀀹 Must typically assume that each pedestrian walks the same distance in a
crossing or along a sidewalk
COMMON
MEASURES
􀀹 Average miles walked, per person, per day.
􀀹 Total aggregate distance of pedestrian travel across an intersection.
EXAMPLES 􀀹 The 2001 fatality rate per 100 million miles traveled in the U. S. was 1.3 for
drivers and their passengers and 20.1 for pedestrians ( STPP, 2004).
􀀹 Between 1990 and 2000, the share of Americans walking to work fell
from 3.9% to 2.9% ( U. S. Census 2000 Summary File 3, Census 1990
Summary Tape File 3.)
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 23
Table 2.8: Exposure Based on Time
APPROPRIATE
USES
􀀹 Estimating total pedestrian time exposure for specific locations.
􀀹 Comparing risks between different modes of travel ( e. g. walking vs. riding in
a car).
􀀹 Estimating whether risk increases in a linear manner with walking time.
􀀹 Comparing risk between intersections with different crossing distances and
between individuals with different walking speeds.
HOW DATA IS
GATHERED
􀀹 The number of persons passing through an area multiplied by the time
traveled.
􀀹 Time spent on walking activities reported on surveys.
PROS 􀀹 Accounts for different walking speeds
􀀹 Allows for accurate comparison between different modes of travel.
􀀹 Can be used to measure exposure at the micro and macro levels
􀀹 More detailed than pedestrian volumes or population data
CONS 􀀹 Time based measures assume risk is equal over the entire distance of a
crossing. Only a small portion of time spent walking on roadways
represents real exposure to vehicle traffic. This portion would include time
spent crossing roads, walking on the road surface, or possibly walking
along the roadside where there are no curved sidewalks ( Chu, 2003).
􀀹 Time spent on walking can be over estimated in surveys, because people
perceive that they spend more time walking than they actually do ( Chu,
2003).
􀀹 Walking may also be under- reported in surveys, because people may forget
walk trips or may purposely choosing not to report. Both of these reasons
are related to the fact that walking trips are relatively short. These very short
trips may not register in the memory of respondents or the respondents may
think that these short trips are unimportant ( Chu, 2003)
COMMON
MEASURES
􀀹 Average time walked, per person, per day or year.
􀀹 Total aggregate travel time of pedestrian travel across an intersection.
EXAMPLES 􀀹 In 2001, the U. S. annual per capita minutes traveled was 2,139 minutes
( Chu, 2003).
2.5. Choosing an Appropriate Exposure Measure
Exposure can be estimated in a number of different ways for almost any situation, as
summarized in Table 2.3. These different ways of assessing exposure lead to
different risk estimates, each of which may be correct but each may convey a
different meaning. When determining the best exposure measure for a given
purpose, key considerations include:
􀀹 What is the chosen method of measuring exposure? Does it match the
study purpose? Surveys will yield individual- level measures of exposure such
as person- trips or person- distance walked, while direct observation will yield
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 24
geographic- level measures of exposure such as number of crossings or distance
walked within a defined area.
􀀹 Where is the exposure to be measured? If exposure is measured at a facility
such as a pedestrian crossing or along a sidewalk, then the exposure measure
should be a micro- level measure, such as number of crossings.
􀀹 What are the study resources? Some exposure measures, such as time and
distance, more accurately portray pedestrian risk than pedestrian counts alone.
However, time or distance spent as a pedestrian will likely be more costly to
collect than simpler measures of exposure.
The following section lists examples of study purposes and provides guidance on the
choice of exposure measure for each.
2.5.1. Comparing safety infrastructure and countermeasures
When comparing the effects of infrastructure and/ or countermeasure on pedestrian
risk, the ideal measure of exposure will be collected directly in the area where the
infrastructure and/ or countermeasure are in place. This will allow an objective
connection to be established between the site and pedestrian risk, and will allow a
consistent numerator and denominator in the pedestrian risk measure. That is, the
numerator will reflect the number of pedestrian- vehicle incidents occurring at the
specific site and the denominator will reflect the number of “ trials” occurring in the
vicinity of the countermeasure. It should be noted however that surveys can in
theory be used to track pedestrian use of infrastructure, although they are not well-adapted
for this purpose. For example, the New Zealand Travel Survey of 1988- 89
asked respondents to keep a diary recording the number of crossings made at
‘ zebra- style’ pedestrian crossings ( Keall, 1995).
The exposure measure should also be appropriate to the type of infrastructure being
studied. If the effect of enhanced crossing devices is being studied, than the
pedestrian crossing is an appropriate measure of exposure. Zeeger et al. ( 2005), for
example, used the number of pedestrian crossings as the unit of exposure in a study
comparing risk at marked and unmarked crossings. If the effect of new sidewalks
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 25
along the length of a block are being studied, then pedestrian distance walked along
the block would be a better measure of exposure.
2.5.2. Compare risk between groups of pedestrians
If the purpose of the study is to compare risk among different groups of pedestrians,
the measure of exposure should be linked to individual- level attributes such as age;
racial or ethnic group; income category; and so on. For example, Keall ( 1995)
estimated the risks of collision for different sex and age groups by combining road
collision data with survey data using the exposure measures “ time spent walking”
and “ number of roads crossed”. These attributes are most easily collected through
surveys, although it is possible to estimate certain pedestrian characteristics such as
age and gender through direct observation.
2.5.3. Compare risk among different modes of travel
When comparing risk among different modes of travel, the best exposure measure
reflects the different travel speeds of the modes being compared. For that reason, it
is best to use time spent traveling to compare risk among different travel modes.
Because different modes use different infrastructure, it may be difficult to record and
compare geographic- level measures of time spent traveling by various modes such
as automobiles, airplanes, bicycles, and pedestrians. Recording the individual- level
use of these modes by survey is more commonly used to compare risk.
2.6. Collision Rates as a Proxy for Risk
Although an in- depth discussion of risk measurement is outside the scope of this
paper, it is important to be aware of possible pitfalls associated with using exposure
data in simplistic risk analysis.
As noted above, exposure data is commonly used to calculate collision rates, namely
the number of collisions in a given time and place divided by an exposure measure.
The calculation of collision rates rests on the assumption that the number of
collisions is proportional to exposure. In other words, it assumes that, all other things
being equal, a place with more pedestrians should have more pedestrian- vehicle
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 26
collisions, and that the number of collisions should increase at a constant rate as the
number of pedestrians increases. Figure 2.3 illustrates this assumption.
Figure 2.3: Assumed relationship between exposure and number of collisions
Although the assumption that collisions increase as a linear function of exposure is
commonly made, there is substantial evidence to suggest that it is erroneous.
Jacobsen ( 2003) has shown that pedestrian- vehicle collisions vary non- linearly with
the number of pedestrians. In other words, risk appears to drop off when more
pedestrians are present. Similarly, Lee and Abdel- Aty ( 2005) showed that
pedestrian- vehicle collisions vary non- linearly with vehicle volumes. Collisions
increase when more vehicles are present, but the rate of increase declines at high
traffic volumes. The non- linear relationship may be due to more cautious driver
behavior or reduced speed when many road users are present.
The calculation of collision rates without taking into account the non- linear
relationship between exposure and collisions can lead to spurious conclusions in
safety studies.
Hauer ( 1995) illustrated the pitfalls of collision rates using the following diagram
( Figure 2.4). Accidents increase with exposure, but the rate of increase is not
constant. The resulting curve is referred to as the “ Safety Performance Function” of
Collision rate= c / x
Quantity of exposure in a given time ( x)
Number of collisions in a given time
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 27
the roadway. It may be empirically measured over time with the collection of accident
data in periods of differing exposure.
Hauer ( 1995) shows how the collision rate ( the slope of the curve) at point “ B” in the
diagram is lower than that at point “ A” simply by virtue of the fact that the exposure
has risen from 3,000 to 4,000 vehicles. If this fact is not taken into account, one
could incorrectly conclude that a safety countermeasure was the cause of the
decline in accident rates, when a change in exposure was alone responsible.
Figure 2.4 Non- Linear Relationship Between Exposure and Accidents ( Hauer, 1995)
The best method of coping with the problems of accident rates is to discard them in
favor of more complex models of risk. However, since risk modeling is often too
costly for practical applications, accident rates are likely to remain common currency.
Given that fact, it is sufficient to be aware that the usefulness of accident rates in
measuring risk may be undermined in situations where exposure has changed
substantially. Future studies of the relationship between pedestrian volumes and
collisions are needed to define typical safety performance functions for pedestrian
collisions. This will help identify the level of pedestrian exposure associated with a
decline in collision rates.
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 28
2.7. Converting Between Exposure Measures at Pedestrian Crossings
As noted above, study resources may constrain the choice of exposure measure.
For example, in areas with large numbers of pedestrians, recording the actual time
each pedestrian spends at a crossing will require multiple observers, whereas
recording the pedestrian volume will require fewer observers. In many cases,
however, the estimated time a pedestrian spends crossing a street will provide a
better indication of exposure than will a simple volume measurement.
In these cases, it is possible to convert the pedestrian crossing volume into an
estimate of the aggregate distance crossed or time spent crossing. This can be
achieved through the following equations ( 1) and ( 2).
Ped distance traveled ( feet) = no. of crossings * distance crossed ( ft) ( 2)
Ped time walked ( seconds) = Ped distance traveled ( ft) / 4 ( ft/ s) 1 ( 3)
Transforming pedestrian volume into time spent traveling or distance traveled at a
crossing should be conducted for estimation purposes only. It should not be
considered the “ true” time spent traveling for the following reasons.
􀀹 Pedestrian crossing speed is not static but varies by pedestrian age; gender;
pedestrian compliance with intersection controls; weather conditions; and signal
cycle length ( Knoblauch et al., 1996). One study noted that as many as 19
percent of pedestrians actually run across the intersection ( Fitzpatrick et al.,
2006).
􀀹 Pedestrians crossing distance is not static because some pedestrians may cross
at an angle or walk outside the painted crossing.
􀀹 Pedestrian crossing speed alone does not fully account for crossing time
because pedestrians who wait for signals to change require a “ startup” time of
approximately 3 seconds to begin walking ( Knoblauch et al., 1996).
It should also be noted that this conversion should only be attempted for constrained
areas where pedestrian distance walked can be estimated with reasonable accuracy.
1 Pedestrian speed as indicated in the Federal Highway Administration 2003 Manual on Uniform
Traffic Control Devices with Revision 1 Incorporated, published 2004
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 29
Observing pedestrian distance walked along a roadway, for example, is prone to
error because individual pedestrians can stop, change directions, or enter and exit
buildings, thus changing their distance traveled.
3. AREA- WIDE METHODS
The previous chapter illustrated the fact that there are several possible definitions of
pedestrian exposure, and that the definition used in any given study is, to some
extent, a function of the measurement instrument and the geographic context. This
report identifies two main geographic contexts where measurement of pedestrian
exposure takes place: wide areas, such as neighborhoods, cities, or the state, and
specific sites, such as intersections or pedestrian crossings. These contexts can
overlap when pedestrian exposure at specific sites is sampled in order to estimate
exposure over a wide area.
This chapter discusses three general approaches to estimating area- wide pedestrian
volumes. The first strategy involves directly sampling pedestrian activity at a
representative set of sites throughout an area. The second strategy involves using
surveys to gauge how much individuals report having walked in a given area.
Surveys of this kind have already been implemented in some metropolitan areas and
on the state level in California. The third strategy involves using modeling techniques
to estimate pedestrian volumes from a combination of direct counts, surveys, and
secondary data. The strengths and weaknesses of each of the methods listed above
are discussed, and examples of each are provided.
3.1. Direct Sampling
Direct samples of pedestrian volume can be used to estimate pedestrian activity over
a wide area. To achieve this, it is necessary to develop a strategy to sample volumes
systematically through time and space. A systematic sampling design could be used
to develop an estimate of the average volume at intersections in an area, for
example. An in- depth discussion of representative sampling methods may be found
in chapter 5, “ Data Collection Planning at Intersections.”
The direct sampling approach to measuring area- wide pedestrian volumes has some
distinct advantages. Direct measurements of pedestrian activity are based on real
observations, rather than reported behaviors, so they avoid the problem of under-reporting
of short pedestrian trips common to surveys ( Schwartz and Porter, 2000).
Direct measurements capture the activity of all pedestrians at the sampled site,
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 31
regardless of age or economic status, although they do not capture the rich
demographic information typically included in surveys. Direct measurements allow
the linkage of pedestrian activity to site- specific factors such as intersection design.
Despite these advantages, there are very few examples of direct measurement
approaches. This may be because of the lack of good inventories of the pedestrian
network, which are necessary to devise a sampling scheme. The Institute of
Transportation Engineers Pedestrian and Bicycle Council, with the assistance of Alta
Planning and Design, have attempted to implement a program of pedestrian volume
sampling over wide areas. This effort, known as the National Pedestrian and Bicycle
Documentation Project, aims to establish a nationally consistent methodology for
performing pedestrian and bicycle counts; to promote the performance of counts on
official counting days during the second week of September; and to input counts into
a national database ( Alta Planning and Design, 2006). The project has resulted in
collection of pedestrian volumes in a few cities throughout the nation. However,
since there is no spatial sampling scheme associated with the project, the resulting
volumes cannot be used to estimate pedestrian volumes over wide areas. The
likelihood that the project will generate systematic, routinely collected pedestrian
counts is small given its voluntary nature.
The best example of direct volume sampling comes from outside the pedestrian
realm. The Federal Highway Administration has developed a Traffic Monitoring
Guide to aid states in the systematic sampling of vehicle volumes. The guide
describes a method for sampling every roadway section at least once within a six-year
period, and for converting a point- measure of volume ( Average Daily Traffic)
into a distance- based measure ( Vehicle Miles Traveled) based on the length of the
roadway segment ( FHWA, 2001). Although many states use the methods in the
Traffic Monitoring Guide, some states, such as California, use a combination of
direct counts and modeling to estimate vehicle volumes ( Caltrans, 2005).
3.2. Surveys
Unlike direct sampling methods, surveys conducted at the local, state, and national
level are commonly used to quantify pedestrian activity over wide geographic area.
Because surveys are able to capture detailed pedestrian characteristics and
preferences, they are very useful for studying the pedestrian behavior of specific
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 32
groups. Surveys are also able to capture detailed trip characteristics such as the
number and length of walking trips made by an individual.
In direct sampling, by contrast, it is very difficult to determine the origin and
destination of each pedestrian trip, or to determine detailed pedestrian
characteristics. However, surveys have certain weaknesses. Surveys do not
generally link pedestrian activity to specific infrastructure, such as roadway or
sidewalk width, so it is difficult to determine the relationship between infrastructure
and pedestrian activity from surveys alone. It is also difficult to determine whether
the walking trips reported in surveys were made in areas where the pedestrian was
exposed to traffic. Lastly, walking trips are commonly underreported in surveys,
because individuals do not always remember short walking trips ( Schwartz and
Porter, 2000). For example, individuals may not report walking to access transit as a
separate trip.
Survey data is available for many different types of geographies and time periods.
When seeking information about pedestrian exposure over a wide area, it is
important to know whether relevant survey data has already been collected. For that
reason, this section focuses on describing existing pedestrian- related surveys and
the type of information available from each. Three types of existing surveys are
identified and evaluated: ( i) health- related surveys; ( ii) travel surveys; and ( iii) the
Journey- to- Work portion of the U. S. Census. These characteristics are also
summarized in Table 3.2.
There will be cases where existing surveys will not always meet the data needs of
the user. For example, there is no existing data source that provides an estimate of
pedestrian exposure for the state of California as a whole on a frequent basis. In
these cases, institutional support and resources are needed to implement more
frequent or new data collection efforts.
3.2.1. Health- Related Surveys
Health surveys aim to track health conditions and risky behaviors. Since walking is a
form of physical activity, some of these surveys include walking- related questions,
which tend to be focused on whether the respondent obtained a healthy amount of
physical activity. Therefore, these types of surveys may not contain information on
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 33
they exact amount of walking or whether walking took place in areas where
pedestrians were exposed to traffic.
For example, the California Department of Health Services and the California
Department of Transportation sponsored the Pedestrian Characteristics in California
Survey in 2003 in order to track health trends. The survey included a question on the
amount of time spent walking in a typical week ( Schneider et al., 2005). Because the
survey is not conducted on a regular basis, it is limited in its ability to track
pedestrian volume trends over time, and it does not provide information about the
total amount of exposure to traffic.
The Behavioral Risk Factor Surveillance System ( BRFSS), an annual telephone
survey administered by the Centers for Disease Control, is conducted annually. It
includes questions on physical activity, but does not distinguish between walking and
other forms of physical activity ( BRFSS, 2006). The state of California could choose
to add additional questions to the BRFSS in order to gain information about the
prevalence of walking in the state.
3.2.2. Travel Surveys
Travel surveys are conducted at the metropolitan, state, and national level for
transportation planning purposes. Most rely on travel diaries, in which respondents
record detailed information about trips taken during a designated travel period. The
detail provided by travel diaries is valuable in estimating pedestrian volume, because
it allows volume to be expressed in terms of the amount of time walked, the distance
walked, or the number of walking trips made.
The largest travel survey conducted nationally is the National Household Travel
Survey ( NHTS). The survey is conducted about every six years by the Federal
Highway Administration, and records the travel patterns of about 20,000 randomly
selected U. S. households. The NHTS reports the number of trips by mode that
respondents took in the week the survey was administered. It can be used to
quantify pedestrian trips as a share of all trips taken nationally or by major Census
division ( e. g. Mountain; Pacific, West South Central, etc.). The NHTS is not intended
for use at the state or sub- state levels, but states or metropolitan areas can purchase
add- ons ( NHTS, 2006).
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 34
Several states and metropolitan areas also conduct travel surveys to serve local
needs ( TRB, 2006). In the state of California, travel surveys are conducted in several
metropolitan areas and on at the state level. The California Statewide Household
Travel Survey ( CSTS), a travel survey of 17,040 California households, was
conducted between 2000- 2001 by the California Department of Transportation
( Caltrans). The CSTS quantifies the number, duration, and approximate distance of
trips taken by survey respondents on an average weekday for each mode of
transportation. It also captures household demographic and economic
characteristics.
The CSTS provides a robust estimate of the amount of pedestrian activity in the
state of California, and for 17 sub- state regions, for the year 2000. The survey must
be used cautiously or not at all for small geographic areas such as cities or counties
( Caltrans, 2002). In addition, the CSTS cannot be used to track short- term trends in
pedestrian activity because it is not conducted on a regular basis.
Several metropolitan areas in California also collect travel surveys similar to the
CSTS and the NHTS. For example, the Metropolitan Transportation Commission
conducts the Bay Area Travel Survey ( BATS) a study of the travel patterns of
approximately 15,000 Households in the 9- county Bay Area. The BATS was
conducted in 2000, 1996, 1990, 1981, and 1965. The Sacramento Area Council of
Governments and the Southern California Association of Governments also conduct
travel surveys about once a decade.
3.2.3. U. S. Census Journey- to- Work and the American Community Survey
The Journey- to- Work component of the U. S. Decennial Census long form contains
detailed information about the work- trip characteristics of one in six U. S. households.
Respondents are asked about the location of their workplace; their usual means of
transportation to work; and the amount of time it usually took them to get to work.
The data is free to the public, available online, and covers large and small
geographies throughout the nation.
However, Journey- to- Work data has some limitations. The survey questionnaire asks
only about which mode of transport the respondent used most frequently to commute
to work in the previous week. By doing so, it accounts only for work trips, which
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 35
make up a minority of all walking trips ( Komanoff and Roelofs, 1993), and for
employed adults, who make up less than half of the population ( U. S. Census
Bureau, 2004). Moreover, the form asks how the respondent “ usually” got to work,
and thus does not capture occasional trips to work made by another mode. Neither
does it account for walking trips made as a component of the work trip, such as trips
to and from a bus stop. This is because the survey questionnaire asks the
respondent to name only the mode they used for the majority of the distance of their
trip ( U. S. Census Bureau, 2005).
In spite of these weaknesses, Census Journey- to- Work data has been used as proxy
for pedestrian exposure because it provides some information about how much
people are walking in an area, and is often the only data on walking available at the
level of the city. One widely- known report on pedestrian safety, which was published
by the Surface Transportation Policy Project, used the percentage of people walking
to work and population data from the Census to compare pedestrian risk in
metropolitan areas across the nation ( STPP, 2002, 2004).
The Census long form that provides Journey- to- Work data is currently being
replaced by a new product called the American Community Survey ( ACS). Although
the information being collected in the ACS is the same as what was collected in the
Census long form, the two surveys differ in important ways. The most important
difference is that Journey- to- Work data will be available every year through the ACS,
rather than once a decade. Another important difference lies in the sample design.
Whereas the Census long form data was collected during a specific week in April,
the ACS samples households on a rolling basis during each month of the year. This
means that ACS data will reflect traveler behavior throughout the year rather than for
a specific season. When fully implemented, the ACS will sample about 3 million, or 1
in 10, U. S. households annually.
ACS data are currently available for communities of 65,000 or more on a yearly
basis. For smaller communities, it will take between several years to accumulate
enough samples to provide data. Beginning in 2008, yearly estimates based on three
year averages will be available for communities of 20,000 or more, and beginning in
2010, yearly estimates based on five- year averages will be available at the Census
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 36
tract and block group level A summary of ACS data availability is displayed in Table
3.1.
Table 3.1: Block Group Level A Summary of ACS Data Availability
Data for the Previous Year Released in the Summer
of:
Type of Data
Population
Size of Area 2003 2004 2005 2006 2007 2008 2009 2010+
Annual
estimates ≥ 250,000
Annual
estimates ≥ 65,000
3- year averages ≥ 20,000
5- year averages
Census Tract
and Block
Group*
Data reflect American Community Survey testing through 2004
* Census tracts are small, relatively permanent statistical subdivisions of a country averaging about 4,000
inhabitants. Census block groups generally contain between 600 and 3,000 people. The smallest
geographic level for which data will be produced is the block group; the Census Bureau will not publish
estimates for small numbers of people or areas if there is a probability that an individual can be identified.
Source: U. S. Census Bureau, 2006
Table 3.2: Characteristics of Existing Pedestrian Related Surveys
Survey Walking Question Geographies Years available
Decennial Census Usual mode to work Census tract 􀃆 nation 1980, 1990, 2000
American Community
Survey Usual mode to work Census tract 􀃆 nation Every year after
2003*
Behavioral Risk Factor
Surveillance System None- possible add on States, nation Every year
National Household Travel
Survey
Number, length,
duration of walk trips
Census divisions,
nation
Every 6 years: 1969,
1997, 1983, 1990,
1995, 2001
California State Travel
Survey
Number, length,
duration of walk trips
Caltrans Districts,
state of California Every 10 years
Metro Area Surveys Number, length,
duration of walk trips
SF, La & Sac metro
area
Varies – about every
6- 10 years
* ACS release schedule varies by geography; data at the census tract level not available until 2010
3.3. Modeling Methods
Mathematical models can be used to estimate pedestrian volumes by combining key
assumptions with existing data. If properly calibrated and tested, models can be
powerful tools in estimating pedestrian volumes when direct measurement is not
feasible. The advantages and disadvantages of modeling depend to some degree on
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 37
the model itself, but in general, models have the potential to save time and resources
without overly compromising accuracy.
Radford and Ragland ( 2006) identified three main types of models: sketch plan
models, network analysis models, and microsimulation models. The strengths and
weakness of each for measuring pedestrian exposure are presented below.
3.3.1. Sketch plan models
Sketch plan models use available data to estimate pedestrian volumes for regional
or city- wide planning purposes. These models rely on known or estimated
correlations between pedestrian activity and adjacent land uses, such as square feet
of office or retail space, and/ or indicators of transportation trip generation such as
parking capacity, transit volumes, or traffic movements ( Schwartz et al., 1999). Some
of these models are not capable of producing pedestrian volumes, but rather
produce a dimensionless indicator of pedestrian activity.
The city of Sacramento, California, recently used a sketch plan method developed by
Fehr and Peers Transportation Consultants ( 2005) as part of its pedestrian master
plan. The method inputs demographic, economic and land use variables associated
with walking into Geographic Information Systems software to produce a
dimensionless “ pedestrian demand index” for each street segment in the city.
3.3.2. Network analysis models
Network analysis models are more complex than sketch plan models because they
rely on a map or model of the pedestrian network. As a result, they are capable of
estimating volumes for specific street segments and intersections over an entire city
or neighborhood. Although the models vary in technique, most use a variation on the
four- step modeling approach to generate and distribute trips based upon
assumptions about the amount of walking trips in a study area and various route
choice algorithms ( Senevarante and Morall, 1986; Ben- Akiva and Lerman, 1985;
McNally, 2000).
Radford and Ragland ( 2004) used a network analysis model, Space Syntax, to
estimate pedestrian volumes on streets and intersections throughout Oakland,
California. The model required input of a pedestrian route map derived from publicly
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 38
available Census TIGER/ line GIS centerline road maps; population and employment
data from the U. S. Census and the California Economic Census; and raw pedestrian
count data needed to calibrate the model. The model produced reasonable
estimates of city- wide pedestrian volume.
The Space Syntax model is also useful for estimating pedestrian flow along
corridors. This is very helpful because direct measurement of flow along corridors is
difficult. It may be achieved by dividing the road network into small segments, such
as a block length, and assuming that flow along the segment is constant. This is not
always a fair assumption because of the complexity of pedestrian movement. For
example, if a pedestrian is counted at the end of a block, it is uncertain whether she
has been traveling for the entire block or if she just exited a building. With vehicle
volumes, by contrast, it is often assumed that any vehicle passing through a point
has been traveling along the length of the segment ( FHWA, 2001). Space Syntax
provides an alternative method of estimating flow along many corridors with a small
set of samples as input.
3.3.3. Microsimulation models
Microsimulation models use flow principles from physical science to model
pedestrian behavior in confined spaces such as the interior of shopping malls or
subway stations, on a single or small number of streets, or within building interiors.
Microsimulation models provide highly accurate, detailed information about
pedestrian movement, but require specialized software, knowledge and extensive
data inputs ( Radford and Ragland, 2006).
3.3.4. Comparison of modeling techniques
Table 3.3 presents a comparison of these approaches, highlighting their advantages
and disadvantages for estimation of wide- area pedestrian volumes. This table was
adapted from Radford and Ragland ( 2006). Each of the modeling approaches
discussed in this paper is suited to a different scale of geographic analysis. Sketch
plan models are best for broad regional or statewide analysis; network analysis
models are appropriate for corridor, neighborhood, or urban area analysis; and
microsimulation models are best for a single street or smaller area.
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 39
Relevant literature indicates that sketch plans have the most potential to be put into
standard use for estimating pedestrian volume throughout the state. While less
accurate than other types of models, sketch plans are relatively simple to use and
make the most out of existing data sources. A simple, standardized sketch plan
method would be an improvement over the current absence of volume estimation
methods in many areas.
Microsimulation models are much too complex and costly to be practical beyond the
level of the street or intersection. Network analysis models have been successfully
used to estimate pedestrian volumes in most large urban areas, but may be
impractical in many small cities and rural areas that lack staffing and resources to
perform the GIS analysis and calibration necessary to complete the model.
Table 3.3: Comparison of Modeling Methods
Scale of Application Advantages Disadvantages
Sketch Plan Large scale ( city,
region, state)
Little data collection
required;
No specialized
expertise needed;
Quick estimations.
Aggregate level;
Low accuracy.
Network Analysis Urban and
neighborhood level
Good detail;
Reasonable accuracy;
Limited data
requirements;
Useful for estimating
pedestrian flows along
corridors;
Appropriate to urban
volume analysis.
Model must be calibrated
with pedestrian counts;
Requires existing GIS
data;
Must be submitted to
sensitivity test.
Microsimulation Individual Streets or
intersections
Highly accurate;
Detailed;
Allows visualization of
pedestrian flow.
Complex;
Steep learning curve;
Significant initial data
requirements.
3.4. Comparison of Methods
This chapter reviewed and evaluated three possible systematic approaches to
measurement of pedestrian volumes over wide areas. The choice of area wide
counting methods depends on budget constraints and data needs, and the
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 40
availability of existing data. No single approach is best, but each has strengths and
weakness. These are summarized in Table 3.4.
Table 3.4: Comparison of Approaches to Pedestrian Volume Estimation
Approaches Advantages Disadvantages
Direct sampling
methods
Based on real, not reported pedestrian
activity;
All pedestrians at each site are
sampled;
Pedestrian volumes linked to specific
sites;
If designed appropriately, data could be
aggregated from small to large
geographies.
Difficult to devise a sampling scheme;
Need a good inventory of the pedestrian
network;
Would require significant manpower;
No demographic or attitudinal
information captured;
No information on distance, length, or
time walked.
Survey methods
Can capture demographic and
household data;
Can capture distance, length, and time
walked;
Existing surveys could be adapted /
expanded.
Walk trips are consistently
underreported in surveys;
Difficult to link walking to specific
infrastructure;
Difficult to determine whether walking
occurred in areas exposed to vehicle
traffic.
Modeling
methods
Make the most of available data;
Dynamic and flexible;
Potential for lowest cost.
Different models may be needed for
different geographic areas;
Output may be limited to dimensionless
measure of pedestrian demand.
4. SITE SPECIFIC METHODS
The previous chapter discussed approaches to measuring pedestrian exposure over
wide areas such as cities or states. In many cases it is necessary to collect
pedestrian exposure data at specific sites such as intersections, pedestrian
crossings, or along a city block. Site- specific measurement of pedestrian exposure is
used to identify high collision locations; to evaluate how infrastructure influences
pedestrian risk; or to track changes in risk over time at a specific site or sites.
There are three main methods of counting pedestrians at specific sites: ( i) field
observation ( ii) video observation with manual review and ( iii) automated methods.
This chapter describes these methods and evaluates the strengths and weakness of
each.
4.1. Pedestrian Counts at Specific Sites
Pedestrian volumes at specific sites are usually collected directly using either ( i)
manual counts taken by collectors in the field or through video observation, or ( ii)
automated counts using specialized equipment. Push button counters are also used
to count pedestrians. However, because of their lack of accuracy relative to the other
counting methods, push button counters were not reviewed in this protocol. It has
been determined that only 35 percent of all pedestrians use push button devices
when they are available ( Zeeger et al., 1982).
Pedestrian counting methods differ in their cost, convenience, level of data detail,
and accuracy. In order to select the most appropriate method for different conditions
and study purposes, it is important to understand the strengths and weaknesses of
each method.
4.1.1. Manual counting methods
Manual counting methods are frequently used to quantify all types of transportation
activity, including vehicle, bicycle, and pedestrian volumes. Manual methods are the
most frequently used method of counting pedestrians, particularly for studies that
require small samples of data at specific locations, such as pedestrian crossings.
The two most common manual counting methods used to measure pedestrian flows
at crossings are:
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 42
􀀹 Field observations: in which pedestrians are observed in the field and counted by
hand.
􀀹 Video- recordings: in which camera recordings of pedestrian crossings are taken
and then processed through playback and manual recording.
Field observations are typically used for periods of less than a day. In this case,
the normal intervals for counting are 5, 10, or 15 minutes. The counts are
recorded with tally sheets, hand- held computers, or clickers. Tally sheets can
include an individual line for each pedestrian and his or her characteristics and/ or
behavior can be recorded, although not all tally sheets are designed this way.
Some include only boxes in which the number of pedestrians crossing within a
certain time are recorded. An example of a field sheet used to count pedestrians
and make inferences about their characteristics is provided in Appendix A. Hand-held
computers ( PDAs) are more frequently used to count and classify vehicle
movements, but can also be used to collect information about pedestrian flow
and movement directions.
Clickers, Figure 4.1, are appropriate in situations where there is no need to
record individual pedestrian characteristics. They are also helpful in areas of high
volume, where it is important that the observer have his or her eyes focused on
the street. Schweizer ( 2005) found that a person can count about 2,000 to 4,000
pedestrians in an hour using clickers, and only half that amount without them.
Using more than one clicker, the field observer can factor in the difference
between males and females or the direction of movement. However, recording
these characteristics would decrease the capacity of the field observer.
Manual- video recording uses cameras to record images of pedestrians which are
later reviewed by an observer. The observer records the number of pedestrians
as well as pedestrian characteristics and behavior, if needed. Detailed review of
behaviors, or crowded pedestrian conditions, may require that the observer
review the video in variable time ( e. g. slowing and speeding the video as
needed). Specialized video- playback tools may be used to facilitate review of
the videos. One such tool was developed by the Partners for Advanced Transit
and Highway Research ( PATH), and is depicted in Figure 4.3.
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 43
The central issues with the manual- video method of counting pedestrians are the
need for a good camera angle and resolution ( Figure 4.2) and the long time
required to review the video tapes, estimated to be three times the tape length
( Diogenes et al., 2007).
Figure 4.1: Field Observation using clickers ( Schweizer, 2005)
Figure 4.2: Video- image camera angle and resolution
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 44
Figure 4.3: Path QuickTime Playback Tool
4.1.1.1. Cost of manual methods
The relative cost of field observations and video- recording counts varies based on
the source of labor, the volume of the intersection, and the amount and type of data
being collected. Costs can be broken into labor and equipment costs. In general,
field observations are labor intensive but have low equipment costs. Video methods
have higher equipment costs, and may have equally high if not higher labor costs
depending on the amount of staff time taken reviewing video, and on whether the
video camera can be left unattended in the field.
Cost of Manual Field Observations:
􀀹 Few equipment costs, though they may be increased if electronic hand- held
devices ( PDAs) are used to record pedestrian activity. However, use of these
devices reduces the labor costs associated with data entry
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 45
􀀹 High labor costs. Staff are needed to observe pedestrians in the field and to
perform data entry. More staff are needed at high- volume intersections, when
several data points are being collected ( e. g. pedestrian characteristics), or
when detailed pedestrian behaviors are being investigated
􀀹 Training costs vary. The cost of training relates to whether consultant
observers are used or whether observers are on staff, and to the need for
data quality. Generally, more training can be expected to produce better
quality data
􀀹 Costs can be reduced if counts performed by volunteers/ students; if counts
are integrated in to regularly scheduled vehicle counts ( Schneider et al.,
2005); and if counts are scheduled efficiently to maximize the use of available
labor
􀀹 Example: the District Department of Transportation performs 10- hour counts
at intersections across the city. The Department estimated in 2005 that each
intersection counted cost between $ 400 - $ 500, including the cost of labor for
pedestrian and motor vehicle counts and the cost of entering the field data
into spreadsheets ( Schneider et al., 2005)
Cost of Manual Video Recording:
􀀹 Equipment costs include the price of camcorders, tapes, and recording
accessories. Camcorders vary in price depending on the quality required, but
range from hundreds to a few thousand dollars. The cost of video tapes varies
by number of hours recorded
􀀹 High- resolution or time- lapse video equipment may be required to record
detailed pedestrian characteristics, or to monitor more than one crossing at a
time. For example, the City of Davis, California, purchased a time lapse video
system ( including camera, playback system and videotapes) for $ 7,000 in
1998/ 99 ( Schneider et al., 2005). The cost- burden of video equipment should
be assessed over the life of the equipment
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 46
􀀹 Costs can be reduced if video counts are combined with other purposes, such
as security
􀀹 Staff are needed for initial setup of camera and camera maintenance
􀀹 One staff person per video camera is typically required in the field to prevent
vandalism and theft, unless the camera is concealed or made inaccessible
􀀹 Only one staff person is needed to review the video, regardless of the
intersection volume, because video can be slowed down and rewound.
However, staff may take many hours to review the video if detailed
information or a high level of accuracy is required
􀀹 Transportation costs must be paid for staff and video camera. In some cases,
a flat- bed truck may be required for set up of the video camera
4.1.1.2. Convenience and data detail of manual methods
Field observations and video- recording differ in their relative convenience and in the
data detail that can be collected. Generally, field observers can capture a broad
array of pedestrian characteristics and behaviors. Video- recordings are sometimes
capable of capturing these details, but not without careful camera positioning and/ or
high resolution film. Video cameras may be able to record at times inconvenient for
field observers, such as night time or weekends; however, this is only possible if the
video is positioned or disguised such that it can be left alone without protection from
vandalism or theft, and if the video image is unobscured by poor lighting or weather.
Convenience and Data Detail of Manual Field Observations:
􀀹 Staff schedules must be coordinated
􀀹 Inconvenient to collect data during inclement weather or during night/ weekend
hours
􀀹 Can waste labor time in areas of low volume
􀀹 Possible to capture detailed pedestrian characteristics like age, race, and
specific behaviors ( Mitman and Ragland, 2007; Diogenes et al., 2007)
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 47
􀀹 Difficult to record extra details if pedestrian volumes are high, unless
additional staff are used
􀀹 Possible to capture mid- block crossings if observers trained properly
􀀹 Possible for a single staff person to observe multiple crossings if pedestrian
volume is low
􀀹 Difficult to record the amount of time it takes pedestrians to cross
􀀹 Possible to record detailed information about the setting or nearby events that
are not captured within a camera’s field of view
Convenience and Data Detail of Manual Video Observations:
􀀹 If camera is positioned securely and disguised such that no on- site
videographers are required to protect it, data can be collected at inconvenient
times such as nights and weekends, assuming there is adequate lighting at
the site
􀀹 If camera is rain- proof, it is possible to collect data during inclement weather
􀀹 Difficult to find a suitable place for video camera. Installation and use of
cameras requires permits as well as security and safety procedures to protect
the camera and those around it. For example, permits are typically needed to
park a flat- bed truck near an intersection, and police must be notified so they
do not suspect illegal activity.
􀀹 Difficult to capture pedestrian characteristics such as age or behavior without
expensive cameras or precise positioning
􀀹 Presence of camera may influence pedestrian behaviors
􀀹 Cannot capture crossings from multiple directions unless multiple cameras
are used or camera positioned at a very wide angle, which may compromise
the image quality
􀀹 Cannot capture pedestrian behavior outside of the camera’s field of view
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 48
􀀹 Possible to capture time and speed
􀀹 Cannot capture detailed information about the setting
4.1.1.3. Accuracy of manual methods
It is important to understand the accuracy of each counting method in order to make
adjustments to counts or to choose the method with the desired level of accuracy.
Although there are few empirical studies of the error of pedestrian counting methods,
it is possible to identify and discuss the sources of error in each. In general, the
accuracy of manual counts is affected by the level of observer training and attention,
and whether the observer is in the field or reviewing video recordings. Mitman and
Ragland ( 2007) compared the inter- reliability between different field observers and
found there is a significant and measurable difference in the data quality produced
by observers with different levels of motivation.
In both methods, error can be avoided by choosing observers carefully, conducting
adequate training, and matching the collection method with location scenarios
( Mitman and Ragland, 2007). However, video- recordings provide additional
insurance against lack of observer motivation because they can be reviewed multiple
times by different observers to check data quality.
Sources of error in manual field observations:
􀀹 Lack of attention. The motivation and training of field observers may affect
their attention in the field.
􀀹 Differences in judgment. The unique personality attributes of field observers
may affect their ability to judge pedestrian characteristics and behaviors, such
as age and gender.
􀀹 Level of pedestrian activity. The amount of pedestrian activity may impact the
accuracy of the count in a variety of ways. Very low or high volumes can
impact the observer’s attention and their ability to record all data points. More
research is needed to determine the relationship between pedestrian volume
and the accuracy of field observations.
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 49
􀀹 Amount of data needed. If it is necessary to record several data points for
each pedestrian ( e. g. gender, direction, age), the quality of the data recorded
may decrease if the capacity of the observer is exceeded, or if recording the
data requires the observer to take his or her eyes off the street.
􀀹 Length of time collecting data. If the collection period is long, the observer
may take unscheduled breaks or get distracted.
Sources of error in manual video recordings:
􀀹 Lack of quality images. The camera angle, positioning, and image resolution
affect the quality of the image and therefore the ability of the video observer to
discern individual pedestrians and their characteristics.
􀀹 Differences in judgment. As with field observation, the attributes of video
observers may affect their judgment of pedestrian characteristics.
􀀹 Lack of attention. As with field observation, the motivation and training of
video observers may affect their attention. However, video recordings can be
reviewed multiple times to ensure data quality.
􀀹 Traffic composition. Large vehicles may block the view of the crossings and
render the video unusable in some instances. In contrast, field observers can
adjust their viewing angle in real time to continue the observations and
therefore eliminate this issue ( Mitman and Ragland, 2007).
􀀹 Level of pedestrian activity. The level of pedestrian activity does not much
affect the quality of counts because video can be reviewed in variable time to
ensure all pedestrian are counted. However, the level of pedestrian activity
may increase the time required to review the video, which may negatively
impact the motivation of the video observer.
􀀹 Gaps in data collection. Data may be lost, and accuracy affected, when
recording is stalled to change tapes, and if the camera malfunctions or is
vandalized during counting.
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 50
4.1.2. Automated methods
In general, automated counting of pedestrians is advantageous because it can
reduce the labor costs associated with manual methods. It also has the potential to
record pedestrian activity for long periods of time that are currently difficult to capture
through traditional methods.
Automated methods are commonly used to count motorized vehicles, but are not
frequently used to count pedestrians at this time. This is because the automated
technologies available to count pedestrians are not very developed, and their
effectiveness has not been widely researched. Moreover, most automated methods
are used primarily for the purpose of detecting, rather than counting, pedestrians
( Dharmaraju et al., 2001; Noyce and Dharmaraju, 2002; Noyce et al., 2006).
A review of pedestrian detection technologies was performed by and Noyce and
Dharmaraju ( 2002) and by Chan et al. ( 2006). Technologies include piezoelectric
sensors, acoustic, active and passive infrared, ultrasonic sensors, microwave radar,
laser scanners, video imaging ( computer vision). A detailed review of these
technologies and their potential for counting, not merely detecting pedestrians is
being conducted for this project, and will be presented in the final report.
Of the technologies listed above, those most adaptable to the purpose of pedestrian
counting are video imaging ( computer vision) and passive infrared devices. Video
imaging utilizes intelligent processing of digital images of pedestrians captured with a
video camera ( Figure 4.4) that is mounted above the area of pedestrian movement.
The processor subtracts the static background from the image and then tracks the
remaining objects to determine whether they are pedestrians ( CLP, 2005).
Passive infra- red devices count pedestrians by tracking the heat emitted by moving
objects. The company “ Irysis”, based in Great Britain, has developed infrared
pedestrian counting devices that can be located either in or outdoors, and are
mounted directly above the area of pedestrian activity ( Figure 4.5). These sensors
have the advantage of being relatively easy to install and configure, and are not
affected by lighting conditions since they rely on heat to produce the images ( CLP,
2005).
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 51
Figure 4.4: Video Imaging for Counting Pedestrian ( CLP, 2005)
Figure 4.5: Irysis Infrared Pedestrian Counting Device ( CLP, 2005)
4.2. Comparison Between Methods
The choice of pedestrian counting method should be based on the accuracy level
desired, budget constraints, and the project data needs. For example, manual counts
must be used when the effort and expense of automated equipment are not justified
or when information about pedestrian characteristics or behavior is required.
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 52
To guide the selection of a method, it is important to review the advantages and
disadvantages of each in collecting pedestrian exposure data at specific sites ( Table
4.1). As specific advantages and disadvantages of the automated methods depend
on the particular technology, only general aspects of these methods are highlighted.
It is important to emphasize that little is known about the relative accuracy and
reliability of these methods. Field tests were performed within the context of this
project to compare the particularities of the manual methods and a summary paper
was submitted to the Transportation Research Board Conference ( Appendix B).
However, further work is needed to draw more specific conclusions about these
methodologies.
Table 4.1: Comparison of Methods to Count Pedestrian at Crossings
Method Advantages Disadvantages
Field
Observations
Relatively low cost;
Observer can record detailed
pedestrian characteristics and
behaviors ( Tally sheet)
Labor- intensive;
Difficult to control the counting process;
Problems at night, in unsafe locations,
and during rainy weather;
Cannot check accuracy of counts after
they occurred;
Video
Observations
Small error rate;
Can replace several counters;
Evaluation can be repeated several
times;
Possible to observe characteristics of
road environment.
Difficult to find suitable place for video
camera;
May be gaps in the counting process
( battery and tape change);
Labor intensive ( long analysis time) if
good data quality is required;
Can be hard to identify pedestrian
characteristics and behaviors.
Automated
Methods
Can collect data for long periods;
Data storage is less time consuming.
Capital cost may be high;
Specialized training may be required;
Can not collect pedestrian
characteristics / behavior.
5. DATA COLLECTION PLANNING AT INTERSECTIONS
Another aspect of site- specific measurement of pedestrian volume is the issue of
where to collect data. The ideal would be to collect pedestrian volumes at all
intersections of a city, but most projects have both budget and time constraints. In
this case, a sample of the target population of sites must be selected for study.
Nassirpour ( 2004) points out that there is no uniform standard of quality that must be
reached by every sample and that the quality of the sample depends entirely on the
stage of the research and how the information will be used. So, the development of a
sample design that satisfies the project goals is crucial to obtain the necessary data
efficiently.
This chapter describes a simplified set of statistical issues that should be considered
when designing a methodology for collecting pedestrian volumes at intersections for
different purposes. The proposed methodology is based on the recommendations of
the Bureau of Transportation Statistics ( BTS, 2003, 2005).
5.1. Sample Design Issues
Sample design is composed of three critical tasks: ( i) definition of the target
population; ( ii) selection of sample technique; and ( iii) determination of sample size.
All these tasks have as constraints the objectives of the research, the type of the
study and the resources available for the study, as shown in the Sampling Strategy
Scheme of a Sampling Strategy ( Figure 5.1). These constraints will play an important
role when selecting the sample technique and determining the sample size.
Figure 5.1: Generalized Model of Sampling ( Adapted from Aggarwal, 1988 and Nassirpour, 2004)
SAMPLING STRATEGY
Define the Objetives
Define the Target Population
Select the Sampling Technique
Determine the Sample Size
Cost
Time
Manpower
Equipment
Resources
Survey
Direct Observation
Historical
Type of Study
Experimental
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 54
5.1.1. Definition of target population
The target population can be defined as the complete set of sites from which you
need to collect information ( Nassirpour, 2004). Determining the population targeted
is the first step in the sampling strategy and it is dependent on the study objective.
For example, if you want to quantify pedestrian volume in the downtown’s
intersections, your target population is all the intersections in the downtown area. If
you are interested in determining the average pedestrian volume in signalized
intersections in California, so all signalized intersections within the state of California
is your target population.
When defining the target population you must define the project objectives and
specifications clearly to avoid collecting unnecessary data or generating bias. For
example, if you want collect pedestrian volumes at marked and unmarked
crosswalks you must define how to identify and distinguish between these
intersections and define the geography of the study area.
After defining the target population, the operational sampling frame must be
constructed. The sampling frame is a list of sampling units from which the sample
can be selected at each sampling stage ( Aggarwal, 1988). For example, in a study of
intersection in the central business district, the sampling frame would be a database
of all the intersections within the area. Ideally the target population must be
coincident with the available list of sampling units. In situations where a complete
database of the sampling units is unavailable, it is necessary to adjust the sample
from the frame population to the target population.
In traffic observation studies, the Geographic Information Systems ( GIS) and digital
road databases are commonly used to develop the sampling frame ( Shapiro et al.,
2001). GIS can be very useful in defining the sets of intersections that are eligible for
sampling, and can also provide additional information about the site, such as the
number of pedestrian collisions.
5.1.2. Selection of sampling technique
After selecting the target population it is necessary to choose a sampling technique
( Figure 5.2). The first step in selecting this technique is to decide whether to use
non- probabilistic or probabilistic sampling.
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 55
SAMPLING TECHNIQUE
Non Probability Sampling Probability Sampling
Simple Random
Stratified
Cluster
Multi Stage Random
Systematic Random
Convenience
Quota
Snowball
Judgment
Figure 5.2: Classification of Sampling Techniques ( Adapted from Aggarwal, 1988)
The non- probabilistic samples are selected through non- random methods, where the
researcher has a lack of control over the sampling error. This type of sampling is
most often used in experimental studies or case studies, when the researcher is
interested in specific units or individuals and not in making conclusions about an
entire population.
Non- probabilistic samples do not require the determination of sample size. Instead,
the researcher will typically select a small number of samples based on subjective
criteria. Table 5.1 describes in few words some of the existing non- probabilistic
sampling techniques, pointing out the advantages and disadvantages of each
method.
In contrast to non- probabilistic sampling, probabilistic sampling involves the use of
statistical principles to select units or individuals randomly. This allows the
researcher to calculate the sampling error and to make inferences about the target
population. Probabilistic sampling requires more time and money to design the
sample and to calculate the sample size necessary to obtain a representative
sample. Table 5.2 describes the most frequently used probabilistic sampling
techniques.
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 56
It is important to keep in mind that the selection of a sampling technique must be
based on the research objectives and on the type of study.
Table 5.1: Non- Probabilistic Sampling Techniques
Non-probabilistic
method
Definition Example Advantage Disadvantage
Convenience Obtaining a sample
of people or units
that are most
convenient to study.
Selecting
intersections with
available collision
data
Low Cost;
Easy method of
sample design.
No representative
sample;
Not recommended
for descriptive or
casual studies.
Judgment Selecting a sample
based on individual
judgment about the
desirable
characteristics
required of the
sampling units.
Selecting
signalized
intersections
because of
experience or
intuition that they
have higher
pedestrian flow.
Low cost;
Allow to draw
some conclusions
about the
characteristics of
the selected
sample.
Does not allow
drawing general
conclusions about
the entire
population.
Quota It is similar to the
judgment sample,
but requires that the
various subgroups
in a population are
represented.
Making sure to
select some
signalized and
some
unsignalized
intersections in a
sample.
Low cost;
Allow to draw
some conclusions
about the
characteristics of
the selected
sample.
Does not allow
drawing general
conclusions about
the entire
population, or
sample subgroups.
Snowball Additional survey
respondents are
obtained from
information provided
by the initial sample
of respondents.
Used when
surveying
individuals about
their behaviors
( e. g. how much
they walk in
specific areas)
Some
characteristics
about the target
population can be
known
Requires a lot of
time and
resources;
Used only for
surveys.
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 57
Table 5.2: Probabilistic Sampling Techniques
Probabilistic
method Definition Example Advantage Disadvantage
Simple
Random
A sampling procedure
that ensures each
element in the
population will have an
equal chance of being
included in the sample
Subgroups
within the target
population may
not be
represented in
the sample;
Larger samples
are necessary.
Systematic
Random
Samples are randomly
selected from a list in
order, but not every one
has an equal chance of
being selected.
When there are
enough
resources; to
inquire about the
characteristics of
the entire
population
Simple;
Conclusions
about the
population can be
drawn.
The sample may
not be
representative
because of the
ordering of the
original list.
Stratified Sub- samples are drawn
within different strata.
Each stratum is
composed of samples
with similar
characteristics.
When
representation of
all subgroups
within a particular
sample is
necessary.
More efficient
sample ( variance
differs between
the strata);
Small sampling
error between
strata;
Smaller samples.
May be difficult
to determine
characteristics of
individuals to
appropriate
classify them in
specific strata.
Cluster Entire groups, not
individuals, are selected
to participate in the data
collection;
Simple random sampling
is applied to the
representative “ clusters”
to select the clusters in
which all members will
participate.
Sample may not
be as
representative
as desired;
Error may be
greater than with
other
techniques;
Pilot studies
may be
necessary to
identify the
clusters.
Multi Stage
Random
Stratification techniques
within the clusters used
to refine and improve
the sample. Examples of
this kind of sampling:
National Safety Belt
Survey.
When the
population is too
big or when there
is a lack of
information about
individual
sampling units
( e. g. all vehicle
occupants in the
United States)
Efficient for large
numbers.
Do not need to
identify all units.
Smaller samples;
Less expensive
relative to the
population size. Like cluster
sampling but
more
representative
within clusters.
* Based on Nassirpour, 2004 and MRUTC, 2005
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 58
5.1.3. Determination of sample size
There are many considerations that come into play when determining the sample
size, such the level of precision to be achieved, operational constraints, available
resources and the chosen sampling technique. The more accurate the desired
results, the greater the sample size required. In order to achieve a certain level of
precision, the sample size will depend, among other things, on the following factors
( Statistics Canada, 2006):
􀀹 The variability of the characteristics being observed: If all intersections have the
same pedestrian flow, then a volume count in one would be sufficient to estimate
the average pedestrian flow for all the intersections. If intersections have very
different flows, then a bigger sample is needed to produce a reliable estimate.
􀀹 The sampling and estimation methods: Not all sampling and estimation methods
have the same level of efficiency. Operational constraints and the unavailability of
an adequate frame sometimes mean that the most efficient technique cannot be
used. A larger sample size is needed if the method used is inefficient.
Som ( 1996) points out other important observations about sample size:
􀀹 Estimates of sample size required to obtain measures with a given precision will
often be found to be quite large, when derived on the basis of unrestricted simple
random sampling;
􀀹 Small samples have proved useful, not only as pilot studies to full- scale surveys,
but also providing interim estimates;
􀀹 An organizations with inadequate resources can start from a small sample and
with increasing resources build up a fully adequate sample; the Current
Population Survey of the U. S. A., for example, started in 1943 with 68 primary
areas which were enlarged to the present 449.
􀀹 It is possible to combine smaller monthly or quarterly estimates into yearly
estimates, and the yearly estimates into estimates covering longer periods, to
provide estimates with acceptable precision.
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 59
􀀹 In the interest of true accuracy, it may sometimes be better to conduct a smaller
sample with adequate control than try to canvass a much larger sample but with
poor quality data.
In this protocol, examples are given on how to estimate the sample size for collecting
pedestrian volumes at intersections for different purposes. However, these examples
are based on specific scenarios, and if any variable of the scenario is changed the
sample size must be recalculated.
5.2. Sampling Intersections in a City
As presented above, the sample design must be based on the research objective,
the type of study and the available resources. Therefore, when planning to collect
data about pedestrian exposure at intersections, the data needs and goals must be
clearly defined. These considerations include: ( i) what data items are needed and
how they will be used; ( ii) the precision level required for estimates; ( iii) the format,
level of detail, and types of tabulations and outputs; and ( iv) when and how
frequently users need the data ( BTS, 2005).
Once data needs are defined, the existing data collection systems must be reviewed
in order to determine whether all or part of the required data are already available, or
could be more easily obtained by adding or modifying other data collection systems
( BTS, 2005). Sometimes, manual pedestrian counts can be combined with existing
motor vehicle counts at little or no additional cost. This has already been achieved
with good results in some U. S. communities such as Albuquerque, NM, Baltimore,
MD, and Washington, DC ( Schneider et al., 2005). Pedestrian counts can also be
combined with other initiatives such as general plans, pedestrian plans, or studies
( e. g. the National Seat Belt Survey). When it is not possible to obtain the necessary
pedestrian exposure data by adding or modifying the existing data collection system,
a sample design is needed.
Data collection and analysis occurs after the data collection methodology has been
defined. However, in systematic studies where data collection is performed
repeatedly, it is necessary to reevaluate the study objectives and methodology each
time data is collected, creating a loop in the data collection planning process. This
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 60
loop ensures changing conditions are reflected in the study design. Figure 5.1
illustrates this process.
Define Goals
Determine Data Needs
Could data be obtained by
adding or modifying other
data collections systems?
YES NO
Define approches
to modify or
combine existing
data collection
systems
Develop a new
data collection
system
Collect data
Review the
Initial Objectives
Figure 5.3: Methodology for Planning Pedestrian Exposure Data Collection at Intersections
This chapter focuses on the development of new data collection systems. Three
hypothetical scenarios involving the collection of pedestrian exposure data were
constructed to illustrate the necessary procedures. These scenarios are intended to
be brief sketches of data collection planning. Not all methods and purposes are
explored in the scenarios.
To simplify the analysis of the scenarios, we have organized the sampling design in
4 steps, as shown in the Figure 5.4.
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 61
Define Goals and Data Needs
Determine the Sampling Technique
Is it necessary to draw conclusions about the target population?
YES NO
Select a
Probabilistic
Method
Select a Non-
Probabilistic
Method
Determine sample
size and error
Figure 5.4: Sampling Design Steps for Pedestrian Exposure Data Collection at Intersections
5.2.1. Scenario 1: Evaluate change over time
One of the uses of pedestrian exposure data is to evaluate change over time, such
as the change in pedestrian risk in an area or a countermeasure’s effectiveness
( before- and- after studies, such as Banerjee and Ragland, 2007). In such
circumstances, it is common that the researcher is more interested in studying
specific sites using non- probabilistic methods to choose where to collect data.
In the first scenario the research goal is to evaluate pedestrian risk among 10
specific intersections before and after signalization. In this case, there is no need to
make general inferences about the sample population, and the sites are already
chosen using the judgment method ( i. e. the intersections that will be signalized).
However, the researcher must be aware that when evaluating a temporal series of
data it is important to use the same methodologies through time, thus avoiding
seasonal influence ( Cameron, 1976; Hocherman et al., 1988; Hottenstein et al.,
1997).
Estimating Pedestrian Accident Exposure: Protocol Report, March 23, 2007 62
5.2.2. Scenario 2: Evaluate risk related to infrastructure type
Pedestrian exposure can also be used to compare the safety associated with
infrastructure. For example, Zeeger et al. ( 2005) compared pedestrian risk among
marked and unmarked crosswalks. For this purpose, judgment samples or random
samples can be used.
The research goal of the second scenario is to determine if pedestrian collision rates
at marked mid- block crossings are higher than at unsignalized intersections. The
available annual numbers of collisions are aggregated by type of crosswalk in
business area of San Francisco. Therefore, the sample frame is marked mid- block
crossings and unsignalized intersections in the San Francisco central business
district.
To perform the analysis, the annual volume of pedestrians at each type of crossing
must be determined. Since the study goal is to understand target population
characteristics, a representative sample is needed.
Two random sample sites must be