A Pilot Model for Estimating Pedestrian Intersection Crossing Volumes
Robert J. Schneider, AICP*
University of California, Berkeley, Traffic Safety Center
2614 Dwight Way #7374
Berkeley, CA 94720
Telephone: 510-642-0566, Fax: (510) 643-9922
E-mail: rschneider@berkeley.edu
Lindsay S. Arnold, MPH
University of California, Berkeley, Traffic Safety Center
2614 Dwight Way #7374
Berkeley, CA 94720
Telephone: (510) 643-5659, Fax: (510) 643-9922
E-mail: larnold@berkeley.edu
David R. Ragland, PhD
University of California, Berkeley, Traffic Safety Center
2614 Dwight Way #7374
Berkeley, CA 94720
Telephone: (510) 642-0655, Fax: (510) 643-9922
E-mail: davidr@berkeley.edu
*(Corresponding Author)
November 2008
Word Count: 5,997 text + 6 figures and tables = 7,497 words
Key Words: Pedestrian, volume, exposure, intersection, crossing, model
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.
Schneider, Arnold, Ragland
1
ABSTRACT
Better data on pedestrian volumes are needed to improve the safety, comfort, and convenience of
pedestrian movement. This requires more carefully-developed methodologies for counting
pedestrians as well as improved methods of modeling pedestrian volumes. This paper describes
the methodology used to create a simple, pilot model of pedestrian intersection crossing volumes
in Alameda County, CA. The model is based on weekly pedestrian volumes at a sample of 50
intersections with a wide variety of surrounding land uses, transportation system attributes, and
neighborhood socioeconomic characteristics. Three alternative model structures were
considered, and the final recommended model has a good overall fit (adjusted-R2=0.897).
Statistically-significant factors in the model include the total population within a 0.5-mile radius,
employment within a 0.25-mile radius, number of commercial retail properties within a 0.25mile radius, and the presence of a regional transit station within a 0.1-mile radius of an
intersection. The model has a simple structure, and it can be implemented by practitioners using
geographic information systems and a basic spreadsheet program. Since the study is based on a
relatively small number of intersections in one urban area, additional research is needed to refine
the model and determine its applicability in other areas.
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.
2
Schneider, Arnold, Ragland
INTRODUCTION
Pedestrian street crossings are among the most common movements in urban transportation
systems. People walk and use wheelchairs to go to work and school, access bus stops and train
stations, patronize neighborhood stores, travel from automobile parking spaces to buildings, and
participate in many other activities. It is important for planners, engineers, designers, public
health professionals and others to have reliable estimates of pedestrian activity. Pedestrian
volumes can be used to:
• Quantify pedestrian exposure in safety analyses (express pedestrian risk as the number
of reported pedestrian crashes per pedestrian crossing)
• Set priorities for pedestrian engineering, education, enforcement, and encouragement
projects (in conjunction with public input, safety data, and other inputs)
• Guide the design of sidewalks, trails, median islands, and other facilities to serve
anticipated pedestrian volumes (beyond meeting minimum accessibility requirements)
• Predict the number of pedestrian crossings that will occur after a land use
development, roadway project, or transit service change is implemented
• Analyze whether or not a crosswalk will meet an engineering warrant for a pedestrian
crossing signal or other crossing treatment
• Predict the amount of pedestrian traffic near commercial businesses.
Study Purpose
The purpose of this paper is to present a preliminary model of pedestrian intersection crossing
volumes. While many variables for predicting pedestrian volumes were analyzed, the
recommended pilot model has a simple structure. The model uses inputs that are available from
common data sources (the US Census, local transit agencies, local roadway databases, aerial
photographs, etc.), so it can be implemented by practitioners using geographic information
systems and a simple spreadsheet program.
The results demonstrate that it is possible to use a relatively small sample of intersection
pedestrian counts to develop a basic, initial predictive model of pedestrian activity for other
intersections in an urban area. Since the analysis was conducted in one urban area (Alameda
County, CA), more research is needed to refine the model equation and determine the
applicability of the results for other communities.
PREVIOUS STUDIES
A variety of methods have been used to estimate or approximate pedestrian volumes. Overlay
mapping techniques, or sketch-plan methods, such as the Latent Demand Score, are useful for
planning and prioritization (1,2,3,4,5,6). However, they are not typically calibrated to actual
pedestrian volume counts. Instead, they rank locations of pedestrian activity on a relative scale.
Regional traffic volume models have also been modified to include walking as a mode choice
(7,8,9). Yet these regional models typically base trip generation on the characteristics of traffic
analysis zones (TAZs). Since many pedestrian trips are made within a TAZ, the geographic
scale of analysis is too large to capture fine-grained differences in pedestrian activity at
individual intersections.
Several pedestrian models have been developed using regression modeling techniques.
Cameron estimated a model of mid-day and evening pedestrian activity on sidewalk segments in
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.
Schneider, Arnold, Ragland
3
Manhattan. Variables in the model included square footage of office, retail, and restaurant uses
on block faces, sidewalk width, streets (east-west direction) versus avenues (north-south
direction), and distance to subway entrances (R2 = 0.23 to 0.61, depending on street orientation
and time of day) (10). Benham and Patel predicted noon-hour pedestrian activity in Milwaukee
based on the total square footage of commercial, office, cultural and entertainment, and storage
and maintenance uses on adjacent block faces. These land use factors explained approximately
60 percent of the variation in pedestrian volumes (11). These models were limited to very dense,
central business districts and specific times of day.
Space Syntax models also use regression modeling. They include street and pedestrian
network characteristics such as connectivity (number of street segment connections to a given
intersection node), mean depth (average number of street segments between a given node and
any other node in the network), visibility (the area visible by direct lines of sight from any
location in the street network), and relative asymmetry/integration (the number of turns that
would be required to travel from a given point in the street network to any other point in the
network). Applications of the Space Syntax model have predicted pedestrian flows with a high
level of accuracy on 7,000 street segments and 670 intersections in Oakland (R2 = 0.77)(12), 82
sample count locations predicting 468 street segments and intersections in Boston (R2 =
0.81)(13), and 237 sample sidewalk blocks predicting volumes on 7,526 blocks in Central
London (R2 = 0.82)(14). However, Space Syntax requires special software, such as Fathom
Visibility Graph Analysis Software (15).
The dependent variables used in previous models have included pedestrians crossing
intersections and pedestrians using sidewalks. Pedestrian counts have been taken during specific
observation periods throughout the day (10,11,12,13,14) (TABLE 1). These counts are often
converted to typical hourly volumes. One study extrapolated two-hour pedestrian counts to
annual volume estimates (12). However, few studies to date have used continuous counts to
account for daily, weekly, and seasonal variations in pedestrian activity or capture the effects of
weather and other factors on pedestrian volumes. Therefore, few models can produce accurate
estimates of full day, complete week, or total annual pedestrian volumes.
Independent variables associated with pedestrian volumes include land use factors,
transportation system attributes, and neighborhood socioeconomic characteristics (TABLE 1).
Many studies over the last 15 years have identified variables that are associated with different
levels of pedestrian activity (though not all of the variables have been included in pedestrian
models). The results of these studies are summarized in several comprehensive literature
reviews (16,17,18,19).
Several other factors have also been used to evaluate pedestrian activity in communities.
Time and cost variables are found in regional transportation demand modeling and elasticity
studies (7,8). Environmental factors, such as weather and topography, are also considered in
several studies (9,16,17,18,20).
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.
4
Schneider, Arnold, Ragland
TABLE 1 Previous Research on Factors Associated with Pedestrian Volume
DEPENDENT VARIABLES
Variable
Pedestrian volumes on sidewalk segments
(block faces) in Manhattan
Time Period
Instant captured in aerial photos
during the mid-day and evening
peak periods
6-minute cnts. during the noon
hr. extrapolated to the full hr.
Two-hr. morning and evening
peak counts extrapolated to
annual volume
Source (Year)
Cameron (1976)
Pedestrian volumes crossing intersections
and using sidewalk segments in Downtown
Boston
Pedestrian volumes at mid-points of
sidewalk segments in Central London
INDEPENDENT VARIABLES
Land Use Variable
Nearby population density
5-minute counts extrapolated to
morning, mid-day, evening
peak hrs.
Raford & Ragland (2005)
5-minute counts extrapolated to
a one-hr. volume
Desyllas, et al. (2003)
Relationship with Ped. Volume
Nearby housing unit density
Nearby employment density
Nearby land use mix
Proximity to mixed-use buildings
Proximity to multi-story buildings
Proximity to commercial buildings
Proximity to parks
Proximity to activity destinations
Proximity to vacant lots
Nearby building setback distances
Transportation System Variable
Sidewalk presence on nearby streets
Nearby sidewalk connectivity
Access to multi-use trails
Nearby multi-use trail connectivity
Access to transit
Nearby street network connectivity
Nearby intersection density
Nearby four-way intersections
Buffer between sidewalk and street on
nearby streets
Presence of street trees on nearby streets
Presence of street lighting on nearby streets
Nearby street block length
Amount of arterial roadways nearby
Auto speeds on nearby residential streets
Automobile parking spaces in nearby area
Difficulty of crossing nearby streets
Socioeconomic Variable
Student status
Larger household of unrelated individuals
Household automobile availability
Household income
Age
+
+
+
+
+
+
+
+
-
Source(s)
Ewing & Cervero (2001), Handy
(2005), Krizek (2003)
Krizek (2003)
Pedestrian volumes on sidewalk segments
(block faces) in Downtown Milwaukee
Pedestrian volumes crossing intersections
and using sidewalk segments in Oakland
TRB 2009 Annual Meeting CD-ROM
+
Relationship with Ped. Volume
+
+
+
+
+
+
+
+
+
+
+
Relationship with Ped. Volume
+
+
-
Benham & Patel (1977)
Raford & Ragland (2004)
Ewing & Cervero (2001), Handy (2005)
Handy (2005), Shriver (1997)
Ewing & Cervero (2001)
Ewing & Cervero (2001)
Ewing & Cervero (2001)
Ewing & Cervero (2001)
Handy (2005)
Ewing & Cervero (2001)
Shriver (1997)
Source(s)
Ewing & Cervero (2001), Handy (2005)
Ewing & Cervero (2001), Shriver (1997)
Handy (2005)
Shriver (1997)
Shriver (1997)
Handy (2005), Shriver (1997)
Ewing & Cervero (2001), Krizek (2003)
Ewing & Cervero (2001)
Ewing & Cervero (2001)
Ewing & Cervero (2001)
Ewing & Cervero (2001), Handy (2005)
Ewing & Cervero (2001), Krizek (2003)
Ewing & Cervero (2001)
Handy (2005)
Ewing & Cervero (2001)
Ewing & Cervero (2001)
Source(s)
Shriver (1997)
Shriver (1997)
Handy (2005), Shriver (1997)
Shriver (1997)
Shriver (1997)
Paper revised from original submittal.
5
Schneider, Arnold, Ragland
The literature indicates that there is a need for a pedestrian volume model that
incorporates land use, transportation system, socioeconomic, and other factors. Ideally, this
model should be able to predict the pedestrian volume at specific locations for an entire day,
week, or year. The approach used in this study addresses both of these needs. First, a thorough
method was used to develop accurate pedestrian counts. Second, a large number of possible
explanatory factors were documented and considered in the statistical modeling process. In
addition, the recommended pilot model has a simple structure that will make it straightforward
for practitioners to test.
METHODOLOGY
This section describes the study area and methodology used to develop the pedestrian crossing
model. The model is based on pedestrian counts taken at 50 intersections along arterial and
collector roadways in Alameda County, CA.
Study Area
Alameda County, CA is part of the San Francisco Bay Metropolitan Region. The county
(Census Bureau 2007 estimated population 1.46 million) is an excellent location for this pilot
study because it includes urban, suburban, and exurban communities—many of the built
environments in the county can be found throughout the United States. Oakland is the largest
city in the county (population 401,000). Alameda County includes the Oakland central business
district and other downtown commercial areas, clusters of neighborhood retail shops as well as
larger malls, mixed and single-use zoning, various street patterns, bus and rail transit systems,
and a population with a wide range of socioeconomic characteristics. According to Census 2000,
approximately 51 percent of residents referred to themselves as White, 27 percent Asian, 21
percent Hispanic or Latino, and 14 percent Black. Twenty-four percent of residents are under
age 18, 11 percent are over age 64, and 13 percent have some type of disability (21).
Intersection Selection
The selection of intersection sites for pedestrian volume counts is critical for developing an
unbiased model. Previous modeling efforts that have been based on existing counts may reflect
the bias in how those locations were chosen (e.g., communities often take counts at locations
with the highest pedestrian volumes or locations of interest to residents or advocacy groups, so a
model based on these locations may not represent other types of locations very well). In
contrast, this study used a deliberate process to select sites with a wide range of pedestrian
volumes, as well as locations surrounded by diverse land use, transportation system, and
neighborhood socioeconomic characteristics.
A strategic sampling process was used to select the 50 intersections for this study. First,
30 intersections were selected from all 528 intersections along state-maintained arterial
roadways. These 528 intersections were divided into high (highest third), moderate (middle
third), and low (lowest third) categories for three variables: population density, median income,
and proximity to commercial properties. This resulted in 27 different strata (e.g., Population
Density = High, Median Income = High, Commercial Retail = High; Population Density = High,
Median Income = Medium, and Commercial Retail = Low; etc.). One intersection was to be
chosen from each of these strata. Since two of the strata did not contain any intersections
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.
Schneider, Arnold, Ragland
6
(Population Density = High, Median Income = Low, Commercial Retail = Low and Population
Density = Medium, Median Income = Low, Commercial Retail = Low), one intersection was
chosen from each of the 25 strata that were represented. Five additional intersections were
chosen randomly from five different strata to complete the initial sample of 30 state-maintained
roadway intersections. Counts at these intersections were funded by the state department of
transportation.
The 20 remaining intersections for the study were selected from a total of 6,938
intersections (6,902 roadway intersections and 36 major roadway/multi-use trail intersections)
along other (non-state-maintained) arterial and collector roadways in Alameda County. Several
constraints were placed on the selection of these 20 intersections to ensure variation in land use
and transportation system characteristics surrounding the selected points:
• At least four intersections needed to be in the Oakland, Fremont, Hayward, or
Berkeley central business districts.
• At least two intersections needed to be where major multi-use trails crossed roadways.
• At least three intersections needed to be in each of the four county planning areas
(North, Central, South, and East).
Counts at these 20 intersections were funded by the county transportation authority.
Additionally, the following rules were used to select all 50 intersection points:
• No intersection could be located within 0.25-mile (402 m) of any other selected
intersection. This was done to limit spatial autocorrelation (observations at adjacent
intersections may not be independent because the same people have a high likelihood
of crossing both intersections).
• Intersections were required to have a population density of at least 50 residents per
square mile within a 0.25-mile (402 m) buffer to be considered for selection. Lowdensity areas are likely to have very sparse, variable pedestrian activity, which is
difficult to model. This ruled out 485 intersections.
• Intersections needed to be at least 0.25-mile (402 m) from an adjacent county border in
order to reduce the amount of data needed from surrounding counties. This ruled out
104 intersections.
• Offset intersections were considered to be a single intersection if the centerline of one
intersection was within 20 feet (6.10 m) of the center of the other intersection.
Otherwise, they were counted as two separate intersections.
• Intersections undergoing construction were not considered.
• Grade-separated intersections were not considered.
The 50 selected intersections had a wide variety of characteristics and were spread
throughout the county (FIGURE 1). While there was significant variation between sites, the
average values were similar to the county as a whole (descriptive statistics are provided in
TABLE 2). The selected intersections included:
• 9 intersections within 0.5-miles (805 m) of a Bay Area Rapid Transit (BART) station
• 20 intersections within 0.25-miles (402 m) of a elementary, middle, or high school
• 33 intersections with sidewalks on both sides of all roadways within 0.25-miles
(402 m)
• 4 trail/roadway intersections
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.
7
Schneider, Arnold, Ragland
• 6 central business district intersections
o Oakland (4)
o Hayward
o Fremont
• Intersections in neighborhoods with a wide range of socioeconomic characteristics
(within 0.25-miles (402 m))
o Between 45.5 and 56.0 percent male
o Between 7.42 and 36.4 percent under age 18
o Between 5.55 percent and 91.2 percent rental housing
o Median incomes between $14,600 and $114,000 per year (1999 dollars)
• A variety of site characteristics, including number of travel lanes, traffic volumes,
speed limits, median islands, curb radii, and types of traffic control.
Pedestrian Crossing Counts
One of the overall goals of the study was to evaluate intersection exposure to vehicle-pedestrian
collisions. Therefore, pedestrian counts were taken at intersection pedestrian crossings. While
pedestrian counts along sidewalk segments are important for planning and prioritization,
pedestrians are typically exposed to the risk of collision and injury when crossing the street.
Further, the study focused on pedestrians crossing arterial and collector roadway intersections
because these roadways often have the most activity destinations for pedestrians, highest motor
vehicle volumes, most popular bus lines, and most challenging pedestrian crossings.
Intersections of two local streets were not evaluated.
Any pedestrian who crossed within a crosswalk or within 50 feet (15.2 m) of either side
of any marked or unmarked crosswalk was counted (this included people walking their bicycle
across the street, babies being carried, and individual pedestrians crossing each intersection leg
multiple times, but it did not include skateboarders, in-line skaters, or people riding bicycles).
Each leg of the intersection was counted separately and summed to derive the total pedestrian
volume. Midblock crossings (more than 50 feet (15.2 m) from the intersection crosswalk) were
not observed during this analysis. Midblock counts require data collectors to focus
simultaneously at the intersection and further down all approaching roadways. This is difficult
to do accurately without additional data collectors at midblock locations.
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.
Schneider, Arnold, Ragland
TRB 2009 Annual Meeting CD-ROM
FIGURE 1 Locations of 50 Study Intersections in Alameda County
8
Paper revised from original submittal.
Schneider, Arnold, Ragland
9
A single pedestrian was counted multiple times at the same intersection if he or she
crossed more than one leg. Right-turning pedestrians on the sidewalk were not counted because
they did not cross the roadway. At “T-intersections” there are only three roadway crossings.
However, pedestrians using the sidewalk on the fourth side of the intersection were counted as
long as they walked at least half of the “crossing” distance on that leg. This made it possible to
make direct comparisons between the total intersection volumes at 3- and 4-way intersections.
Two separate manual counts were taken at each study intersection between April and
June 2008. One was a weekday count (Tuesday, Wednesday, or Thursday) and the other was a
Saturday count. All manual count periods were two hours long, from 9:00 a.m. to 11:00 a.m.,
12:00 p.m. to 2:00 p.m., or 3:00 p.m. to 5:00 p.m. These time periods were chosen because they
were expected to have the most consistent pedestrian travel patterns from week to week.
Daily, Weekly, and Seasonal Pedestrian Volume Patterns
It is critical to understand hourly, daily, weekly, and seasonal variations in pedestrian and bicycle
activity at different sites in order to interpret count data correctly (21,22,23,24). The peak hour
period for pedestrian travel may not be the same at all locations, even within the same
neighborhood. Near a school, the peak hour may be from 3:00 p.m. to 4:00 p.m.; in a restaurant
district, the peak hour may be from 7:00 p.m. to 8:00 p.m. Several studies have classified daily
distributions of pedestrian volumes into different categories based on land use type.
Classifications have been based on the type of pedestrian activity near a count location, including
shopper, employee, visitor, commuter, mixed, or special (21) and on the location of a count site
within a region, including central business district, residential, or fringe area (23). A study of 14
count sites in Washington, DC found six distinct pedestrian volume distribution patterns (22).
EcoCounter Dual Infrared Pyroelectric pedestrian sensors were used to extrapolate the
two-hour weekday and Saturday manual counts to average daily, weekly, and annual pedestrian
volumes. Between April and June 2008, these devices were rotated among a set of 13 of the 50
locations to identify different patterns of pedestrian activity (FIGURE 1).
The infrared counters were installed to count pedestrians on the sidewalk near the
intersection (i.e., within approximately 100 feet of the intersection). This is because the
technology is not suitable for counting pedestrians crossing the street. Therefore, it was assumed
that the daily pattern of sidewalk activity was nearly identical to the daily pattern of pedestrian
activity in the crosswalks. The infrared sensors tend to undercount pedestrians slightly, most
likely because they do not detect pedestrians walking exactly side-by-side.
To address undercounting, the sensor counts were tested against manual counts during
several different time periods. This comparison revealed undercounting between one and 20
percent, similar to a previous evaluation of the same technology (25). There was no evidence
that rate of undercounting was related to pedestrian volume. Therefore, undercounting was
assumed to have no effect on the daily pedestrian volume distributions used in the analysis (e.g.,
a given time period represents the same percentage of the total daily volume, regardless of how
much each hour is undercounted, since the rate of undercounting is the same throughout the day).
Reliable counts were gathered at 11 of the 13 infrared sensor locations. The analysis
identified differences in pedestrian volume patterns near employment centers, residential areas,
neighborhood commercial areas, and multi-use trails. Adjustments were made to account for
these differences at all 50 study intersections. The two-hour manual counts were also adjusted to
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.
Schneider, Arnold, Ragland
10
account for differences in time (e.g., time of day, day of week) and weather (e.g., cloud cover,
temperature).
Extrapolated weekly pedestrian crossing volume at each intersection was the dependent
variable used in the pilot model. Weekly pedestrian volumes ranged from 323 at an intersection
surrounded by open space, hotel, and industrial land uses near Oakland International Airport to
113,000 at an intersection in the Oakland central business district. For all 50 intersections, the
average weekly pedestrian volume was 9,260 (standard deviation = 18,000).
A more extensive description of the pedestrian counting methodology used for this study
is provided in the paper, “A Methodology for Counting Pedestrians at Intersections: Using
Automated Counters to Extrapolate Weekly Volumes from Short Manual Counts” (26).
ANALYSIS
This section describes the data analysis process used to estimate regression models and
recommend one of the alternatives as the final pilot model. After the pedestrian count data were
extrapolated to weekly volume estimates (described above), four main steps were followed: 1)
screen all of the potential explanatory variables and do not consider any variables that do not
have a relatively high correlation with pedestrian volume, 2) screen the remaining explanatory
variables for collinearity and avoid including highly-correlated variables in the same model, 3)
choose three alternative model structures that had a good overall fit and statistically-significant
explanatory variables to compare, and 4) examine the three alternatives and recommend one
preferred pilot model.
Regression Modeling
Ordinary least squares (OLS) regression was used to develop the pedestrian crossing volume
model. The modeling process tested the statistical relationship between weekly pedestrian
crossings at each intersection and a range of land use, transportation system, socioeconomic, and
intersection site characteristics. More than 50 variables were reviewed for inclusion in the model
(TABLE 2). The land use factors, transportation system attributes, and socioeconomic
characteristics were gathered from the area surrounding each intersection. Three different buffer
radii were used: 0.1 miles (161 m), 0.25 miles (402 m), and 0.5 miles (805 m) (most of the 0.5mile variables are omitted from Table 2). Few previous studies have examined the distances at
which specific variables influence pedestrian volumes. Using several different radii for each
variable makes it possible to explore this issue.
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.
11
Schneider, Arnold, Ragland
TRB 2009 Annual Meeting CD-ROM
TABLE 2 (Part 1) Land Use and Transportation System Variables Considered for the Pedestrian Volume Model
Variable Name
TOTPOP_T
TOTPOP_Q
TOTPOP_H
TOTEMP_T
TOTEMP_Q
PCTVAC_T
PCTVAC_Q
TOTVAC_T
TOTVAC_Q
PCTRENT_T
PCTRENT_Q
TOTRENT_T
TOTRENT_Q
NCOMPROP_T
NCOMPROP_Q
NESCH_T
NESCH_Q
NMSCH_T
NMSCH_Q
NHSCH_T
NHSCH_Q
NTSCH_T
NTSCH_Q
COLDUM_T
COLDUM_Q
Paper revised from original submittal.
Variable Name
NBARTSTA_T
NBARTSTA_Q
NBUSSTOP_T
NBUSSTOP_Q
TRAILMI_T
TRAILMI_Q
STREETMI_T
STREETMI_Q
BL_MI_T
BL_MI_Q
FWY_DUM_T
FWY_DUM_Q
SWCOV_Q
SWBUF_Q
LAND USE VARIABLES
50 Study Intersections (N = 50)
Description
Mean Std. Dev.
Min.
Max.
Total population within 1/10‐mile (161 m)
291
156
7.03
614
Total population within 1/4‐mile (402 m)
1880
869
254
3670
Total population within 1/2‐mile (805 m)
7500
3290
798
15100
Total employment within 1/10‐mile (161 m)
315
764
1.54
4170
Total employment within 1/4‐mile (402 m)
1660
3510
9.60
18900
Proportion of housing units within 1/10‐mile (161 m) that are vacant
0.0398
0.028 0.00673
0.122
Proportion of housing units within 1/4‐mile (402 m) that are vacant
0.0385
0.026 0.00849
0.106
Number of housing units within 1/10‐mile (161 m) that are vacant
5.22
5.66
0.127
29.7
Number of housing units within 1/4‐mile (402 m) that are vacant
32.1
28.5
1.58
124
Proportion of housing units within 1/10‐mile (161 m) that are rented
0.549
0.198
0.0560
0.923
Proportion of housing units within 1/4‐mile (402 m) that are rented
0.544
0.186
0.0555
0.912
Number of housing units within 1/10‐mile (161 m) that are rented
69.4
52.1
3.39
230
Number of housing units within 1/4‐mile (402 m) that are rented
453
310
22.0
1240
Number of commercial properties within 1/10‐mile (161 m)
6.66
8.11
0.00
40.0
Number of commercial properties within 1/4‐mile (402 m)
25.3
26.5
0.00
50.0
Number of elementary schools within 1/10‐mile (161 m)
0.0400
0.196
0.00
1.00
Number of elementary schools within 1/4‐mile (402 m)
0.320
0.508
0.00
2.00
Number of middle schools within 1/10‐mile (161 m)
0.00
0.00
0.00
0.00
Number of middle schools within 1/4‐mile (402 m)
0.08
0.337
0.00
2.00
Number of high schools within 1/10‐mile (161 m)
0.00
0.00
0.00
0.00
Number of high schools within 1/4‐mile (402 m)
0.0400
0.196
0.00
1.00
Number of elem., middle, high, and other schools within 1/10‐mile (161 m) 1
0.060
0.237
0.00
1.00
Number of elem., middle, high, and other schools within 1/4‐mile (402 m)1
0.480
0.700
0.00
3.00
Presence of college campus within 1/10‐mile (161 m) (Yes=1, No=0)
0.00
0.00
0.00
0.00
Presence of college campus within 1/4‐mile (402 m) (Yes=1, No=0)
0.00
0.00
0.00
0.00
TRANSPORTATION SYSTEM VARIABLES
50 Study Intersections (N = 50)
Description
Mean Std. Dev.
Min.
Max.
Number of regional rail transit stations within 1/10‐mile (161 m)
0.0200
0.140
0.00
1.00
Number of regional rail transit stations within 1/4‐mile (402 m)
0.0400
0.196
0.00
1.00
2
Number of bus route stops within 1/10‐mile (161 m)
12.9
17.1
0.00
118
2
Number of bus route stops within 1/4‐mile (402 m)
47.4
55.9
2.00
335
Total multi‐use trail centerline distance (miles) within 1/10‐mile (161 m)
0.0258
0.0765
0.00
0.365
Total multi‐use trail centerline distance (miles) within 1/4‐mile (402 m)
0.0916
0.262
0.00
1.32
Total street centerline distance (miles) within 1/10‐mile (161 m)
0.939
0.287
0.227
1.53
Total street centerline distance (miles) within 1/4‐mile (402 m)
5.64
1.43
2.23
9.40
Total centerline (miles) of streets with bicycle lanes within 1/10‐mile (161 m)
0.101
0.157
0.00
0.471
Total centerline (miles) of streets with bicycle lanes within 1/4‐mile (402 m)
0.321
0.365
0.00
1.10
Freeway presence within 1/10‐mile (161 m) (Yes = 1, No = 0)
0.120
0.325
0.00
1.00
Freeway presence within 1/4‐mile (402 m) (Yes = 1, No = 0)
0.180
0.384
0.00
1.00
3
Est. sidewalk coverage (0.00,0.25,0.50,0.75,1.00) within 1/4‐mile (402 m)
0.875
0.195
0.25
1.00
4
Est. prop. of sidewalks with buffer (0.00,0.25,0.50,0.75,1.00) within 1/4‐mile
0.525
0.288
0.00
1.00
All Major Street Intersections (N = 8055)
Mean Std. Dev.
265
208
1640
1180
6410
4140
151
350
930
1930
0.0373
0.0349
0.0366
0.0325
4.13
5.51
25.4
30.2
0.449
0.254
0.45
0.244
60.0
76.8
369
427
3.48
6.04
15.3
20.6
0.049
0.22
0.307
0.53
0.00857
0.0922
0.0567
0.233
0.00372
0.0609
0.0539
0.232
0.0683
0.260
0.458
0.669
0.0283
0.166
0.0622
0.242
Min.
0.0658
0.390
1.96
0.896
5.60
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
Max.
1700
7430
21700
4190
19600
0.371
0.290
75.9
316
1.00
1.00
970
2850
48.0
134
2.00
3.00
1.00
2.00
1.00
2.00
2.00
4.00
1.00
1.00
Data Source (Year)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
SF MTC6(2005)
SF MTC6(2005)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
Alameda Co. Assessor (2007)
Alameda Co. Assessor (2007)
Alameda Co. Assessor (2007)
Alameda Co. Assessor (2007)
Alameda Co. Assessor (2007)
Alameda Co. Assessor (2007)
Alameda Co. Assessor (2007)
Alameda Co. Assessor (2007)
Alameda Co. Assessor (2007)
Alameda Co. Assessor (2007)
Alameda Co. Assessor (2007)
Alameda Co. Assessor (2007)
All Major Street Intersections (N = 8055)
Mean Std. Dev.
Min.
0.00993
0.0992
0.00
0.0467
0.212
0.00
8.64
11.4
0.00
36.1
39.8
0.00
0.014
0.0558
0.00
0.0719
0.207
0.00
0.758
0.254
0.00
4.70
1.57
0.278
0.086
0.153
0.00
0.327
0.443
0.00
0.168
0.374
0.00
0.302
0.459
0.00
Not calculated5
Not calculated5
Max.
1.00
2.00
135
337
0.609
1.78
2.34
10.5
0.936
2.20
1.00
1.00
Data Source (Year)
SF MTC6(2007)
SF MTC6(2007)
SF MTC6(2007)
SF MTC6(2007)
SF MTC6(2007)
SF MTC6(2007)
SF MTC6(2007)
SF MTC6(2007)
SF MTC6(2007)
SF MTC6(2007)
SF MTC6(2007)
SF MTC6(2007)
Google Earth® (2008)
Google Earth® (2008)
1) Total schools does not include colleges. Colleges are included in a separate variable.
2) The number of "bus route stops" is the sum of the number of different bus routes servicing each bus stop within a given distance of the intersection (e.g., if 4 routes service a single bus stop, that particular bus stop will be counted 4 times).
3) Sidewalk coverage is estimated from aerial photography. 100% coverage (1.00) is sidewalks on both sides of all surface streets within 1/4‐mile of the intersection. Sidewalks on only one side of all streets would be considered 50% coverage (0.50).
4) Sidewalk buffer is estimated from aerial photography. 100% buffer (1.00) indicates that the sidewalks on both sides of all surface streets are separated from the edge of the roadway by a grass, tree, shrub, or other type of buffer. If
5) Detailed intersection characteristics were not gathered for all roadways in Alameda County. Because of cost, these characteristics would only be collected if they were significant in the final regression model.
6) SF MTC = San Francisco Bay Area Metropolitan Transportation Commission.
12
Schneider, Arnold, Ragland
TRB 2009 Annual Meeting CD-ROM
TABLE 2 (Part 2) Neighborhood Socioeconomic and Intersection Site Variables Considered for the Pedestrian Volume Model
Variable Name
PCTWHITE_T
PCTWHITE_Q
PCTMALE_T
PCTMALE_Q
PCT0VEH_T
PCT0VEH_Q
TOT0VEH_T
TOT0VEH_Q
MEDINC_T
MEDINC_Q
PCTU18_T
PCTU18_Q
PCTO64_T
PCTO64_Q
Variable Name
NEIGHBORHOOD SOCIOECONOMIC VARIABLES
50 Study Intersections (N = 50)
Description
Mean Std. Dev.
Min.
Max.
Proportion of population within 1/10‐mile (161 m) that is white
0.461
0.202
0.0401
0.822
Proportion of population within 1/4‐mile (402 m) that is white
0.463
0.196
0.0655
0.822
Proportion of population within 1/10‐mile (161 m) that is male
0.489
0.0286
0.421
0.556
Proportion of population within 1/4‐mile (402 m) that is male
0.492
0.0222
0.455
0.560
Proportion of households within 1/10‐mile (161 m) that have no automobile
0.168
0.168
0.0138
0.769
Proportion of households within 1/4‐mile (402 m) that have no automobile
0.159
0.150
0.0150
0.638
Total households within 1/10‐mile (161 m) that have no automobile
23.2
36.2
0.299
182
Total households within 1/4‐mile (402 m) that have no automobile
148
200
2.06
964
1,2
Median income (1999 dollars) of households within 1/10‐mile (161 m)
47800
21000
122 107500
1,2
Median income (1999 dollars) of households within 1/4‐mile (402 m)
49400
20300
14600 114000
Proportion of population within 1/10‐mile (161 m) that is under 18 years old
0.223
0.0675
0.0563
0.372
Proportion of population within 1/4‐mile (402 m) that is under 18 years old
0.223
0.0633
0.0742
0.364
Proportion of population within 1/10‐mile (161 m) that is over 64 years old
0.117
0.0776
0.0245
0.423
Proportion of population within 1/4‐mile (402 m) that is over 64 years old
0.114
0.0631
0.0245
0.340
INTERSECTION SITE VARIABLES
50 Study Intersections (N = 50)
Description
Mean Std. Dev.
Min.
Max.
All Major Street Intersections (N = 8055)
Mean Std. Dev.
0.495
0.234
0.492
0.232
0.492
0.0319
0.491
0.0271
0.124
0.13
0.126
0.127
18.3
31.3
112
172
59700
27900
59600
27200
0.234
0.0728
0.236
0.0694
0.108
0.0573
0.108
0.0508
Min.
0.0233
0.0364
0.231
0.356
0.00
0.00
0.00
0.00
122
1051
0.00872
0.0112
0.00
0.00
Max.
1.00
0.920
0.742
0.679
0.802
0.738
483
1600
167000
169400
0.626
0.625
0.502
0.394
Data Source (Year)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
U.S. Census (2000)
All Major Street Intersections (N = 8055)
Mean Std. Dev.
Min.
Max. Data Source (Year)
Paper revised from original submittal.
0.560
24000
79.2
4.44
0.496
12200
27.8
1.47
0.00
3000
29.0
2.00
1.00
55000
163
8.50
Not calculated7
Not calculated7
Not calculated7
Not calculated7
0.755
0.494
0.439
0.806
0.00
0.00
0.00
1.00
2.00
1.00
1.00
3.00
Not calculated7
Not calculated7
Not calculated7
Not calculated7
Field obs., Google Maps® (2008)
Curb radius category (<15 feet (<4.57 m)=1, 15‐25 feet=2, >25 feet (>7.62 m)=
1.30
0.580
0.260
1.90
Intersection is a "T" intersection (Yes=1, No=0)6
0.240
0.427
0.00
1.00
Not calculated7
Field obs., Google Maps® (2008)
CONTROLDUM
MAXADT
MAIN_WIDTH
MAIN_LANES
Either traffic signal or stop sign controling mainline roadway (Yes=1, No=0)3
MAIN_XW
MAIN_MED
MAIN_BL
CURBRADCAT
Number of marked crosswalks across the mainline roadway3
TINTER
Max. average daily traffic volume on a roadway passing through intersection
Average curb‐to‐curb length (feet) of the 2 crosswalks across the mainline roa
Average number of lanes on mainline approaches to the intersection3,4
Median refuge area present for at least one mainline roadway crosswalk3
Bicycle lanes on at least one mainline approach to intersection
3
Field observation (2008)
CA DOT (2007); local municipalities8
Field obs., Google Maps® (2008)
Field obs., Google Maps® (2008)
Field obs., Google Maps® (2008)
Field obs., Google Maps® (2008)
Field obs., Google Maps® (2008)
1) Median income is calculated as the weighted average of median incomes reported for the census block groups surrounding the intersection. Weights are assigned based on the proportion of the census block group within the specific buffer diatance from the intersection.
2) Several census block groups did not have data for median income. Intersections with a median income of 0 within the given buffer distance considered in this statistical summary.
3) Mainline roadway is the intersecting roadway with the higher traffic volume.
4) Average number of lanes on each mainline approach includes all through‐, left‐, and right‐turn lanes.
5) Curb radius category reflects the average estimated curb radius of all corners at the intersection.
6) "T" intersections are 3‐way intersections. Intersections were not considered to be "T" intersections if the fourth approach was a commercial driveway.
7) Detailed intersection characteristics were not gathered for all roadways in Alameda County. Because of cost, these characteristics would only be collected if they were significant in the final regression model.
8) Traffic volume data were gathered from Alameda (2004), Berkeley (2000‐2007), Dublin (2000‐2007), Fremont (2005), Hayward (2003‐2008), Livermore (2007), Pleasanton (2007), and Oakland (2007).
Schneider, Arnold, Ragland
13
With a sample size of 50 intersections, it is not possible to include all of the potential
independent variables in the model. Variables were selected for the model or removed from
consideration using several steps. First, variables with a low correlation (|ρ|<0.4) with the
dependent variable (e.g., number of weekly pedestrian crossings) were removed. Then,
correlations between the remaining independent variables were analyzed. Independent variables
with high correlation coefficients (|ρ|>0.7) were not included in the same model. Finally, several
OLS models were estimated. The three models with the best balance of statistically-significant
independent variables, overall model fit, and intuitive coefficients were selected as final
alternatives.
Model Alternatives
Three alternative model structures were considered when choosing the final recommended model
(TABLE 3). All three models have a good overall fit. The adjusted R2-values are between 0.87
and 0.91, and F-tests show that each model is significant at greater than the 99 percent
confidence level. Model A shows that weekly pedestrian volume at an intersection is higher
when there is more population within 0.5-miles (805 m), more employment within 0.25-miles
(402 m), more commercial properties within 0.25-miles (402 m), a regional transit station within
0.10-miles (161 m), and a lower percentage of the population under age 18 within 0.25-miles
(402 m). For example, the NCOMPROP_Q variable coefficient means that for each additional
commercial property within 0.25-miles (402 m) of an intersection, it will, on average, have
approximately 106 more pedestrian crossings per week. Most of these factors have also been
identified in previous studies. Neighborhoods with a greater percentage of children under age 18
may have lower pedestrian volumes at arterial and collector roadway intersections because
parents may be cautious about letting their children cross these types of busy streets.
Model B is similar to Model A, but it does not include the variable for percentage of the
surrounding population that is under age 18. As a result, Model B may be easier for practitioners
to use than Model A since it does not require gathering neighborhood-level age data from census
files.
Model C uses many of the same types of variables as the other two models, but the
distance thresholds are different. Population and commercial properties are evaluated at a
distance of 0.10-miles (161 m), and a 0.25-mile (402 m)-distance is used for proximity to a
regional transit station. The total number of bus stops within 0.10-miles is also included in this
model. Model C has several disadvantages. While the smaller buffer areas provide more
localized population and commercial property information, these variables may not work well in
larger-scale suburban areas. For example, commercial retail properties in malls may be more
than 0.10 miles (161 m) from adjacent intersections. In addition, data on the total number of bus
route stops (each bus line is counted separately) may be difficult to gather.
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.
14
Schneider, Arnold, Ragland
TABLE 3 Alternative Pedestrian Volume Model Specifications
Dependent Variable = Total Weekly Pedestrian Intersection Crossings
Model A
Model B
Model Variables
Coeff. (Std. Err.)1
Coeff. (Std. Err.)1
CONSTANT
4170 (4270)
‐4910 (2050)**
TOTPOP_T
TOTPOP_H
0.884 (0.254)***
0.928 (0.266)***
TOTEMP_Q
1.72 (0.400)***
2.19 (0.367)***
NCOMPROP_T
NCOMPROP_Q
106 (39.0)***
98.4 (40.8)**
NBARTSTA_T
56,800 (7810)***
54,600 (8160)***
NBARTSTA_Q
NBUSSTOP_T
PCTU18_Q
‐36,400 (15,200)**
Overall Model
Sample Size (N)
50
50
2
Adjusted R
0.907
0.897
96.6***
108***
F‐Test
Model C
Coeff. (Std. Err.)1
‐5790 (1990)***
14.5 (7.19)**
456 (118)***
44,800 (6280)***
465 (81.4)***
50
0.870
83.2***
1) Significance is indicated by asterisks: *** indicates significant at 99% (ρ<0.01), ** indicates significant at 95% (ρ<0.05), * indicates significant at 90% (ρ<0.10)
The model residuals (difference between predicted volume and observed volume at each
intersection) were also graphed for all three models. Differences in predicted and observed
values ranged between almost zero to less than 20,000 pedestrian crossings per week (FIGURE
2). None of the three residual patterns suggested that one model was better or worse than the
others.
While all three models could be used, Model B was recommended because it is has a
good overall model fit, includes statistically significant and logical independent variables, and
can be estimated using readily-available data.
RESULTS
The recommended pilot pedestrian intersection crossing model has the following form:
Total pedestrian intersection crossings per week =
0.928 * Total population within 0.5-miles of the intersection
+ 2.19 * Total employment within 0.25-miles of the intersection
+ 98.4 * Number of commercial retail properties within 0.25-miles of the intersection
+ 54,600 * Number of regional transit stations within 0.10-miles of the intersection
- 4910
(1)
The adjusted R2-value of the model is 0.897, and the F-test is significant at a 99.9 percent
confidence level. All of the independent variables in the recommended model are statistically
significant at the 95 percent confidence level and have a logical interpretation.
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.
15
Schneider, Arnold, Ragland
20000
FIGURE 2 Residual Analysis for Three Alternative Pedestrian Volume Models
0
50000
Weekly Pedestrian Crossing Volume
100000
20000
-20000
Residual (Expected - Actual)
-10000
0
10000
Model A
0
50000
Weekly Pedestrian Crossing Volume
100000
20000
-20000
Residual (Expected - Actual)
-10000
0
10000
Model B
-20000
Residual (Expected - Actual)
-10000
0
10000
Model C
0
TRB 2009 Annual Meeting CD-ROM
50000
Weekly Pedestrian Crossing Volume
100000
Paper revised from original submittal.
Schneider, Arnold, Ragland
16
Validation
An important next step in the modeling process will be to test the predictive accuracy of the
model at other intersections in Alameda County. This will require collecting comparison counts
using a methodology identical to the approach used in this study. While new counts are needed,
several counts from the past six years were available in different parts of Alameda County. Of
hundreds of previous counts, only 46 included a clear description of the number of crossings at
each intersection and the time of day and day of week when the count was taken. Important
information about the specific data collection methods (e.g., how close to the intersection
pedestrians needed to cross to be counted) and data collector accuracy were not available.
Nonetheless, the pilot model pedestrian volume estimates were within 50 percent of the historic
manual counts at 30 of the 46 comparison intersections. In the future, more consistent counts are
needed to compare with the predicted model volumes. However, these historic count
comparisons suggest that the pilot model and future refinements will be useful for estimating
pedestrian volumes.
CONSIDERATIONS FOR FUTURE RESEARCH
The pilot model described in this paper is based on observations at 50 intersections in
Alameda County, CA. The pedestrian volume estimates produced by the model are intended for
planning, prioritization, and safety analysis at the community, neighborhood, and corridor levels.
Since the model provides rough estimates of pedestrian activity, actual pedestrian counts should
be used for site-level safety, design, and engineering analyses.
Although a systematic process was used to select the study intersections and gather count
data, there are many variables that cannot be captured in a model with a sample size of 50.
Variables that were not included in the model but may have a significant relationship with
pedestrian volumes when more counts are taken include:
• Sidewalk coverage and buffer between roadway and sidewalk
• Roadway width and number of motor vehicle lanes
• Street network density
• Percentage of households with no vehicles available
While this pilot study suggests that pedestrian activity levels are influenced more by land
use patterns than pedestrian facility and roadway design characteristics, accessible sidewalks and
well-designed pedestrian street crossings are critical for pedestrian safety. In addition, many
community surveys indicate that a safe and comfortable walking environment will increase
pedestrian activity. Future research should continue to examine pedestrian facility quality
variables.
Different types of statistical models should also be tested. This may include categorical
models, model specifications that constrain predicted values to positive numbers, and models
with interaction terms. Testing the predictive ability of the pilot model against future pedestrian
counts will help suggest the types of refinements that should be made.
Due to restrictions used in the intersection selection process, the model may not be
appropriate for predicting volumes at intersections adjacent to freeways and other grade
separated roadways or in rural areas with less than 50 residents per square mile. In addition, the
model may not perform well in locations close to special attractors, such as amusement parks,
waterfronts, sports arenas, and regional recreation areas. Pedestrian volumes in these areas tend
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.
Schneider, Arnold, Ragland
17
to be highly variable, with high volumes during certain seasons or during nice weather. Bridges
and underpasses may also channel pedestrian activity, so more research may be necessary to
adjust volume estimates near these features.
Daily, weekly, and seasonal pedestrian volume patterns were captured by infrared sensors
at 11 of the 50 locations. While the temporal patterns from these intersections were applied to
other locations with similar land use characteristics, they may not be precise matches. For
example, a particular attractor, such as a gym or a school with certain hours of operation may
exert a significant influence over the distribution of pedestrian activity at nearby intersections.
This effect may not be captured in more general daily patterns of activity from other locations.
While people often walk along street segments and other pathways, straight-line distances
were used in this study. Therefore, a commercial property that is within a 0.25-mile (402 m)
radius of a person’s home may actually require walking further than 0.25 miles (402 m).
Network distances were not used because micro-scale data on pedestrian-only pathways, internal
property circulation patterns, and informal pedestrian cut-throughs were not available.
The study used a stratified random sampling process. This was chosen over other
methods in order to provide the greatest variation in the characteristics of study intersections
while maintaining random selection for each individual intersection. If a simple random method
had been used, intersections in the most common types of areas (such as low-density residential
neighborhoods with middle-range incomes) would likely have made up most of the sample.
Alternatively, selecting study intersections by convenience (such as locations suggested by local
experts or community members) would introduce bias into the method.
Pedestrian trip attractors, such as commercial properties, regional transit stations, and
schools were treated with equal weight in the modeling process. Future analyses could be done
to incorporate weighting factors based on retail square footage, transit station access/egress
counts, or school enrollment to capture these differences. However, an advantage of treating
pedestrian attractors equally is that the pilot model remains relatively easy for practitioners to
use.
The number of people walking in a particular community may also vary due to the
overall condition of pedestrian facilities and attitudes towards walking in the community. These
broader characteristics may change over time. Additional analysis in multiple communities is
needed to identify these broader geographic and cultural influences on pedestrian volumes.
Further research is needed to increase the number of intersection count locations to
increase the predictive capability of the model. These additional counts could be taken in
Alameda County as well as other locations throughout the United States and other countries.
Communities that are good candidates for refining the model would have a variety of pedestrian
environments and have access to all necessary land use, transportation system, and
socioeconomic data in GIS. It may also be possible to apply a similar methodology to develop
models of midblock pedestrian crossing volumes and pedestrian volumes on sidewalk segments.
Finally, it would be beneficial to do a study to compare this study with other pedestrian
models, such as Space Syntax. The models could be compared based on overall predictive
ability, applicability to different geographic areas, cost, ease of implementation, and other
factors.
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.
Schneider, Arnold, Ragland
18
CONCLUSION
The recommended pilot model is a simple tool that can be used to develop rough estimates of
pedestrian intersection crossing volumes. Additional research and testing is needed to refine the
model and determine its applicability in other communities. However, the model has a good
overall fit, and the variables are statistically-significant and have a logical relationship with
pedestrian volume. Practitioners can use this initial model and future refinements to estimate
pedestrian exposure for safety analyses, prioritize locations for pedestrian projects, predict
pedestrian volumes at intersections in new developments, and for many other purposes. Better
pedestrian volume estimates will help planners, designers, engineers, public health professionals,
and others improve the safety and convenience of pedestrian transportation.
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.
Schneider, Arnold, Ragland
19
ACKNOWLEDGEMENTS
This study was funded by the Alameda County Transportation Improvement Authority and
California Department of Transportation. The authors would like to thank Ryan Greene-Roesel,
Mara Chagas Diogenes, and Noah Raford for their previous contributions to this study. We
would also like to thank Population Research Systems for assisting with the manual pedestrian
counts and EcoCounter for providing the infrared pedestrian sensors.
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.
20
Schneider, Arnold, Ragland
REFERENCES
1. Turner, S., G. Shunk, and A. Hottenstein. Development of a Methodology to Estimate Bicycle
and Pedestrian Travel Demand. Texas Transportation Institute, Research Project Number 01723, Report 1723-S, Texas Department of Transportation, Federal Highway Administration,
September 1998.
2. Ercolano, J., J. Olson, D. Spring. “Sketch-Plan Method for Estimating Pedestrian
Traffic for Central Business Districts and Suburban Growth Corridors,” Transportation Research
Record 1578, Transportation Research Board, Washington D.C., 1997, pp. 38-47.
3. Landis, B.W. “The Bicycle System Performance Measures: The Intersection Hazard and
Latent Demand Score Models,” ITE Journal, Vol. 66, No. 2, 1996, pp. 18-26.
4. City of Portland, OR. Portland Pedestrian Master Plan. Office of Transportation Engineering
and Development, June 1998.
5. District of Columbia Department of Transportation. District of Columbia Pedestrian Master
Plan, Forthcoming.
6. City of Alexandria, VA. Alexandria Pedestrian and Bicycle Mobility Plan. Available online:
http://alexandriava.gov/localmotion/info/default.aspx?id=11418, June 2008.
7. Purvis, C.. Incorporating Land Use and Accessibility Variables in Travel Demand Models.
Available online: http://www.mtc.ca.gov/maps_and_data/datamart/research/ascepurv.htm, 1998.
8. Purvis, C. Incorporating Effects of Smart Growth and TOD in San Francisco Bay Area Travel
Demand Models: Current and Future Strategies. Available online:
http://www.mtc.ca.gov/maps_and_data/datamart/research/Incorporating_Smart_Growth_MTC_
models.pdf, 2003.
9. Parsons Brinckerhoff Quade and Douglas, Inc. with Cambridge Systematics, Inc. and
Calthorpe Associates. LUTRAQ: Making the Land Use Transportation Air Quality Connection,
Volume 4A: The Pedestrian Environment. Prepared for 1000 Friends of Oregon, Available
online: http://ntl.bts.gov/DOCS/tped.html, December 1993.
10. Pushkarev, B. and J. Zupan. “Pedestrian Travel Demand,” Highway Research Record 355,
Washington, D.C., 1971.
11. Benham, J. and B. G. Patel. “A Method for Estimating Pedestrian Volume in a Central
Business District,” Transportation Research Record 629, Transportation Research Board,
Washington D.C., 1977, pp. 22-26.
20
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.
21
Schneider, Arnold, Ragland
12. Raford, N. and D. Ragland. “Space Syntax: Innovative Pedestrian Volume Modeling Tool
for Pedestrian Safety,” Transportation Research Record 1878, Transportation Research Board,
Washington D.C., 2004, pp. 66-74.
13. Raford, N. and D. Ragland. Pedestrian Volume Modeling for Traffic Safety and Exposure
Analysis. University of California Traffic Safety Center white paper, Available online,
http://repositories.cdlib.org/its/tsc/UCB-TSC-RR-2005-TRB2/. December 2005.
14. Desyllas, J., E. Duxbury, J. Ward, and A. Smith. Pedestrian Demand Modelling of Large
Cities: An Applied Example from London. Center for Advanced Spatial Analysis, University
College London, Available online, http://www.casa.ucl.ac.uk/working_papers/paper62.pdf. June
2003.
15. Intelligent Space Partnership. Fathom: Visibility Graph Analysis Software, Available online,
http://www.intelligentspace.com/tech/fathom.htm/. Accessed July 15, 2008.
16. Ewing, R. and R. Cervero. “Travel and the Built Environment: A Synthesis,” Transportation
Research Record 1780, Transportation Research Board, 2001, pp. 87-113.
17. Handy, S. Critical Assessment of the Literature on the Relationships Among Transportation,
Land Use, and Physical Activity, Transportation Research Board Special Report 282, Available
online: http://trb.org/downloads/sr282papers/sr282Handy.pdf, 2005.
18. Krizek, K. “Operationalizing Neighborhood Accessibility for Land Use-Travel Behavior
Research and Regional Modeling,” Journal of Planning Education and Research, Volume 22,
2003, pp. 270-287.
19. Shriver, K. “Influence of Environmental Design on Pedestrian Travel Behavior in Four
Austin Neighborhoods,” Transportation Research Record 1578, Transportation Research Board,
Washington D.C., 1997, pp. 64-75.
20. Cameron, R.M. Pedestrian Volume Characteristics. Institute of Transportation Engineers
Compendium of Technical Papers, 1976. Cited in Greene-Roesel, R., M.C. Diogenes, and D.R.
Ragland. Estimating Pedestrian Accident Exposure: Protocol Report. University of California at
Berkeley Traffic Safety Center, Available online: http://repositories.cdlib.org/its/tsc/UCB-TSCRR-2007-5/, March 2007.
21. U.S. Census Bureau. 2007 American Community Survey 1-Year Estimates, 2007.
22. Davis, S.E., L.E. King, and D.H. Robertson. “Predicting Pedestrian Crosswalk Volumes,”
Transportation Research Record 1168, Transportation Research Board, Washington D.C., 1988,
pp 25-30.
21
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.
22
Schneider, Arnold, Ragland
23. Zeeger, C.V., R. Stewart, H. Huang, P.A. Lagerwey, J. Feaganes, and B.J. Campbell. Safety
Effects of Marked versus Unmarked Crosswalks at Uncontrolled Locations: Final Report and
Recommended Guidelines, Federal Highway Administration, , FHWA–HRT–04–100, 2005.
24. Hocherman, I., A.S. Hakkert, and J. Bar-Ziv. “Estimating the Daily Volume of Crossing
Pedestrians from Short-Counts,” Transportation Research Record 1168, Transportation
Research Board, Washington D.C., 1988, pp. 31-38.
25. Greene-Roesel, R., M.C. Diógenes, D.R. Ragland & L.A. Lindau. Effectiveness of a
Commercially Available Automated Pedestrian Counting Device in Urban Environments:
Comparison with Manual Counts. UC-Berkeley Traffic Safety Center, Available online,
http://www.tsc.berkeley.edu/news/08-0503session240ryanposter.pdf, 2007.
26. Schneider, R.J., L.S. Arnold, and D.R. Ragland. "Extrapolating Weekly Pedestrian
Intersection Crossing Volumes from 2-Hour Manual Counts," UC-Berkeley Traffic Safety
Center, Submitted for Presentation at Transportation Research Board Annual Meeting, 2009.
22
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.
1
Schneider, Arnold, Ragland
LIST OF TABLES AND FIGURES
TABLES
1. Previous Research on Factors Associated with Pedestrian Volume
2. (Part 1) Land Use and Transportation System Variables Considered for the Pedestrian Volume
Model
2. (Part 2) Neighborhood Socioeconomic and Intersection Site Variables Considered for the
Pedestrian Volume Model
3. Alternative Pedestrian Volume Model Specifications
FIGURES
1. Locations of 50 Study Intersections in Alameda County
2. Residual Analysis for Three Alternative Pedestrian Volume Models
TRB 2009 Annual Meeting CD-ROM
Paper revised from original submittal.