Content uploaded by Ahmed M. El-Geneidy
Author content
All content in this area was uploaded by Ahmed M. El-Geneidy on Sep 20, 2018
Content may be subject to copyright.
103
Transportation Research Record: Journal of the Transportation Research Board,
No. 2661, 2017, pp. 103–110.
http://dx.doi.org/10.3141/2661-12
Many fields of study recognize the interdependent health, environ-
mental, and economic benefits of walking. To promote walking in
entire populations, measures such as Walk Score have been developed
to classify the walking friendliness or walkability of places. Yet high
walkability is not always equated with increased walking. This paper
investigates this discrepancy with the use of survey data on pedestrian
behavior; a variety of geographic information system–derived land
use and built environment measures of neighborhoods in Montreal,
Quebec, Canada; and socioeconomic characteristics obtained from
the 2011 National Household Survey. A descriptive analysis of walking
behavior and neighborhood characteristics reveals that some neighbor-
hoods with higher walking rates are characterized by a lower presence
of parking lots and setbacks and a greater proportion of on-street tree
canopy. Linear regressions predicting walking rates confirm these associ-
ations after adjusting for Walk Score and neighborhood socioeconomic
characteristics. These findings suggest that more work is needed for
nuancing walkability measures and offer particular insight for health
professionals, planners, and engineers looking to promote walking as
an alternative and healthier mode of transport. Reducing open space,
such as parking lots and setbacks, and increasing street-level tree
canopy are two ways that the urban built environment can be modified
to support walking, especially in areas with high Walk Score and low
walking rates.
Increasing awareness of the environmental, health, and economic
benefits of walking has generated interest across disciplines in iden-
tifying built environment factors that facilitate walking. These factors
are often referred to collectively as “walkability” and include ele-
ments related to amenity density (e.g., number of shops or jobs in
an area), land use, and street connectivity. Because of the number of
factors considered when one attempts to quantify walkability, many
researchers have begun developing indexes to capture these elements
in a single measure (1). Most existing walkability indexes consider
mixed land use, accessibility (i.e., number of destinations reachable
on foot), street connectivity, and presence of pedestrian infrastruc-
ture as indicators of higher walkability (2). One such index is Walk
Score (www.walkscore.com), a proprietary web-based algorithm that
assigns a score from 0 (low walkability) to 100 (high walkability) for
any address. Walk Score measures street connectivity, population
density, and block length, as well as proximity to 13 types of ameni-
ties (e.g., grocery stores, restaurants, bars, schools, parks, etc.). A
distance decay function assigns weights to these amenities, in which
destinations within 0.25 mi (0.40 km) are assigned full weight and
less weight is given to more distant amenities up to 1.5 mi (2.4 km)
(3–5). Walk Score data are easily accessible and thus widely used
among researchers (6, 7). Although Walk Score makes much of its
data available for free on its website, the exact parameters of its
index are not public, given the proprietary nature of some of Walk
Score’s services.
While Walk Score provides an accessible and mostly free resource
for researchers to use, it might not fully capture every element of
walkability. In Montreal, Quebec, Canada, many areas are found with
similar Walk Scores that vary substantially with respect to walking
rates. Figure 1 shows the distribution of census tracts’ walking rates
calculated from survey data, within the same range of Walk Scores
using data from Montreal. The distributions reveal that while Walk
Score is generally associated positively with walking rates, there is
still a high degree of variation in recorded pedestrian activity between
census tracts within the Walk Score categories. Given the increas-
ing interest from researchers and policy makers in walkability and
walkability indexes, and the discrepancies found between Walk
Score and walking rates, the aim of this paper is to identify land
use characteristics that are associated with higher rates of walking
at the neighborhood level. The paper focuses on identifying walk-
ability factors not included in the Walk Score that might explain the
discrepancy between walking rates and Walk Scores. The hypoth-
eses are that (a) parking lots and setbacks will be associated with
lower walking rates, while on-street tree canopy cover will be
associated with higher walking rates; and (b) Walk Score does not
fully account for these neighborhood characteristics. To achieve the
research objective, census tracts are identified with a Walk Score
and walking rate divergence using a cluster analysis and then the
discrepancies in parking lots and setbacks and tree canopy cover-
age are assessed. Linear regression models then are estimated that
predict walking rates as a function of Walk Score, parking lots and
setbacks, on-street tree canopy cover, and control variables. Because
this study focuses on the influence of the built environment on walk-
ing, the authors perform a neighborhood-level analysis as opposed to
an individual-level analysis. Accordingly, the findings of this study
may be used to improve neighborhood walkability metrics and sug-
gest built environment improvements for increasing walking rate for
entire populations.
The Missing Middle
Filling the Gap Between Walkability
and Observed Walking Behavior
Thomas Herrmann, Geneviève Boisjoly, Nancy A. Ross,
and Ahmed M. El-Geneidy
T. Herrmann and N. A. Ross, Department of Geography, and G. Boisjoly and A. M.
El-Geneidy, School of Urban Planning, McGill University, Suite 400, 815 Sherbrooke
Street West, Montréal, Québec H3A 0C2, Canada. Corresponding author: A. M.
El-Geneidy, ahmed.elgeneidy@mcgill.ca.
104 Transportation Research Record 2661
LITERATURE REVIEW AND RESEARCH CONTEXT
Walking is recognized across several disciplines for its interdependent
health, environmental, social, and economic benefits. Starting in the
1990s, health officials in the United States began recommending
walking as a form of exercise to combat the onset of chronic illness
related to physical inactivity (8–10). While personal characteristics
such as age and motivation affect walking frequency and intensity
(11), research indicates that certain elements of the built environment
influence walking behavior and improve physical health (12). In light
of these findings, urban planners now advocate for more compact
and less auto-dependent development patterns that facilitate walking
(13). Planners cite these walkable compact environments as benefi-
cial not only for health but also for more cost-efficient allocation
of transportation infrastructure investments and lower ecological
footprints (14, 15).
Walk Score is easily accessible and its use in research is now
widespread. In health research, for instance, Walk Score has been
used to analyze association between obesity (body mass index) and
neighborhood built environment (16, 17) and to measure walking
rates and physical activity levels in socially disadvantaged popula-
tions (18). In the field of urban planning, research using Walk Score
has been used for policy goals, such as integrating higher-density land
uses near transit (19) and simulating travel behavior and carbon foot-
print impacts of proposed developments (20). While Walk Score has
been validated as a general measure of neighborhood walkability
(6, 21), other research has found gaps in its predictability of walk-
ing rates. Koschinsky et al. found that the association between Walk
Score and walking is less strong in low-income than in high-income
neighborhoods (22). Weinberger and Sweet validate Walk Score as a
predictor of walking, but they point out that sensitivity between the
Walk Score and walking rates differs by trip type (23). Accordingly,
these discrepancies deserve attention and require further investigation
in research.
DATA AND METHODOLOGY
Data
This study examined the relationship between features of the
built environment and walking rates for 466 of 477 census tracts
(11 census tracts lacked data) in the city of Montreal. Three types
of variables were collected in addition to the Walk Score data:
(a) measures of walking behavior, (b) neighborhood characteristics,
and (c) neighborhood socioeconomic characteristics. A summarized
description of these variables is provided in Table 1.
Walking Rates and Walk Score
Walking rates were measured as the “pedestrian modal share.” These
rates are determined from the 2013 Montreal Origin–Destination
(O-D) Survey (Table 1). The O-D survey is a phone-based survey
conducted once every 5 years during the fall by the Agence Métro-
politaine de Transport (AMT) (24). Respondents provide demographic
information about their household, characteristics of individual per-
sons taking trips, and disaggregate characteristics of each trip taken
Walking Rates
Frequency
80
40
20
60
0
0.0 0.1 0.2 0.3 0.4
0.5
(b)
(d)
Walking Rates
Frequency
20
10
5
15
0
0.0 0.1 0.2 0.3 0.4 0.5
(a)
Walking Rates
Frequency
60
50
40
30
20
10
0
0.0 0.1 0.2 0.3 0.4 0.5
(c)
Walking Rates
Frequency
60
50
40
30
20
10
0
0.0 0.1 0.2 0.3 0.4 0.5
FIGURE 1 Distribution of walking rates by Walk Score interval in Montreal.
Herrmann, Boisjoly, Ross, and El-Geneidy 105
by individuals during the previous day. Trip information, including
trip mode, purpose, origin, and destination, is collected for 5% of
households in the metropolitan area and this study’s analyses are
based on home-based trips for 51,547 adults (ages 18–65) (25). For
this analysis, any trip where both the origin and destination were
reached by walking was considered a pedestrian trip. The walking rate,
therefore, reflects the percentage of all trips that were pedestrian trips.
A walk-to-work rate was also calculated by finding the percentage of
all trips made for commuting to work that were pedestrian trips. The
shortest path walking distance for all pedestrian trips was calculated
in a geographic information system using a pedestrian street network
(i.e., one with highways removed and on-street paths included). Using
expansion factors provided by the AMT, the data are aggregated to the
census tract level to calculate the walking rates. While many studies
examine walkability by using trip origin or destination coordinates
(4, 26), other studies have used Walk Score at a similar neighborhood
level of analysis (22, 27).
The Walk Score is a continuous variable between 0 (lowest pos-
sible walkability) and 100 (highest possible walkability). Walk Scores
were downloaded at the postal code level. In Canada, postal codes are
smaller than census tracts: approximately the size of one side of a city
block. The 39,648 Walk Scores for each postal code were aggregated
to the 466 census tracts by determining the centroid of each postal
code and averaging Walk Score values for postal code centroids
within each census tract. The average number of points aggregated
to the census tract level was 78. The average standard deviation for
Walk Score points within census tracts was 4.90.
Land Use Measures: Parking Lots, Setbacks,
and Tree Canopy
Parking lots and setbacks were determined with clutter data from
DMTI Spatial, Inc. (28). The clutter data consist of raster data sets
at a 30-m resolution and 10 values representing different land
use classifications. Most of the land uses are based on data from
the National Topographic Database, which is itself a database of
delineations of different terrains, forest cover, populated places, and
industrial infrastructure collected by Natural Resources Canada at
the 1:50,000 scale. “Open land” is one of 10 land use classifications
included in the clutter data. Open land refers to areas where the
National Topographic Database has no mapped features. These areas
contain neither natural terrains (e.g., rivers, lakes, forest, wetlands,
etc.) nor built features (e.g., buildings, pipelines, dams, etc.). Satellite
imagery reveals that most open land uses in urban settings are human-
manipulated areas, such as parking lots and other forms of setbacks,
such as driveways and lawns. The clutter data do not classify parks
as open space. A neighborhood with a large park, therefore, will not
necessarily have a large proportion of open space, but a neighbor-
hood with more space between buildings will have more open space.
The percentage of each census tract’s parking lots and setbacks is
calculated by converting the raster pixels to vector centroids in a
geographic information system and calculating the percentage of open
land centroids as a total of all land class centroids for each census
tract. It is expected that open land use (parking lots and setbacks)
will depress walking rates, because parking lots and setbacks reduce
dwelling densities.
On-street tree canopy cover is assessed from a shapefile contain-
ing polygon features of each tree in the city of Montreal, which was
downloaded from the city’s website (29). The tree canopy shapefile
was created by the city in 2007 and made available online in 2013.
A central assumption of using the tree canopy data is that only trees
near streets will correlate positively with walking rates. Only trees
with a majority of their area within 10 m of the street centerline were
included. This assumption is consistent with research that suggests
that on-street trees provide a more favorable walking environment
(30). This on-street tree canopy variable represents the percentage of
the area of the 10-m street buffer that is covered by the tree canopy
TABLE 1 Description of Variables
Variable Description Source
Walking Behavior Measures
Walking rates Proportion of all trips where the primary reported mode is walking Montreal O-D Survey (24)
Mean walk trip distance Mean distance in meters of all walk trips originating in the census tract assuming
shortest path calculations
Montreal O-D Survey (24)
Walk-to-work rate Percentage of all work trips where the primary reported mode is walking Montreal O-D Survey (24)
Walk Score Average of all Walk Scores in the census tract (0–100), with 0 indicating
lowest possible walkability and 100 indicating highest possible walkability
Walk Score
Land Use Measures
Parking lots and setbacks Proportion of total area characterized by the absence of buildings, water,
or natural environments (e.g., lawns, parking lots, cropland)
DMTI Spatial
On-street tree canopy cover Proportion of the total area of trees within 10 m of the street and the total
area of a buffer of 10 m within the street centerline
City of Montreal
Controls: Auto Dependency
Paid parking Binary variable indicating the absence (0) or presence (1) of metered parking Stationnement de Montréal
Distance from highway Distance in meters from the nearest (grade-separated, limited access) highway DMTI Spatial
Controls: Sociodemographic
Median household income Median household income reported in 2010 Statistics Canada (National
Household Survey 2011)
Percentage immigrant Percentage of neighborhood population that is an immigrant to Canada
(i.e., not Canadian born)
Statistics Canada (National
Household Survey 2011)
106 Transportation Research Record 2661
in each census tract. The average census tract in Montreal has an
18.98% tree cover surrounding its streets, with a standard deviation
of 11.06 across the sample.
Auto Dependency Controls: Paid Parking
and Distance from Highway
Two variables measured auto dependency: the presence of on-street
metered parking (paid parking) in the census tract and distance
from the centroid of the census tract to the nearest limited-access
highway. The geocoded location of parking meters was taken from
the Stationnement de Montréal (local parking authority) website.
It was expected that paid parking would be positively associated
with walking rates because paid parking is generally implemented in
areas with lower car ownership (3), whereas free parking is associ-
ated with higher rates of automobile-oriented investment and use
(31). The distance from a highway variable is calculated by measuring
the distance in meters between the centroid of the census tract and
the nearest highway segment using shapefiles provided by DMTI
Spatial, Inc. Distance from highway is expected to be positively
associated with walking rates, because highways often represent
physical barriers in the environment that correlate with lower walking
rates (32).
Sociodemographic Controls: Median Household
Income and Percentage of Immigrants
The sociodemographic profile of the census tracts is thought to influ-
ence walking behavior (33). Median household income and the per-
centage of immigrants (i.e., people who immigrated to Canada at
any time) were calculated for each census tract from 2011 National
Household Survey data. In Canada, major urban centers host substan-
tial immigrant populations whose travel behaviors may differ from
those of Canadian-born individuals. In the Montreal Census Metropol-
itan Area (CMA), for instance, 48.6% of recent immigrants commuted
to work by public transit compared with 20.9% of Canadian-born
commuters (34).
Methodology
A cluster analysis (k-means approach) was performed to identify census
tracts with a Walk Score and walking rate divergence. A descriptive
analysis then compared walking behavior (walking rate, walk-to-
work rate, and average walk trip distance) and neighborhood char-
acteristics (Walk Score, median household income, parking lots and
setbacks, on-street canopy cover, and population density) between
the different clusters.
Three linear regression models of walking rates were estimated.
Each of the three models uses walking rate as the independent vari-
able, four control variables (paid parking, distance from highway,
median household income, and percentage of immigrant population),
and different sets of predictor variables. The first model considers
the relationship between walking rate and the land use measures of
interest (parking lots and setbacks and the on-street tree canopy),
while the second model examines the relationship between the pedes-
trian modal share and Walk Score. The third model uses both Walk
Score and the land use measures of interest (parking lots and setbacks
and the on-street tree canopy) as independent variables. The work-
ing assumption is that these two variables will remain significant
when one adjusts for Walk Score.
Independent variables were tested for multicollinearity in each
regression model. The multicollinearity between each independent
variable and the dependent variable was determined by finding
the variable inflation factor (VIF). The highest VIF found of any
independent variable in any model was 2.290 (Walk Score). Walk
Score was the only variable that had a VIF above 2. Often, variables
are eliminated on the basis of multicollinearity if the VIF exceeds 10
(35). The VIFs of this study’s independent variables were well below
this common threshold.
RESULTS AND ANALYSIS
Cluster Analysis
Six clusters displayed the least redundancy and most variation across
groups (Figure 2). Clusters 1 and 2 had above-average (high) Walk
Scores (HWS), Clusters 3 and 4 had average Walk Scores (AWS),
and Clusters 5 and 6 had below-average (low) Walk Scores (LWS),
when compared with the citywide sample. Clusters 1 and 2 had above-
average Walk Scores but Cluster 2 exhibited a much lower walking
rate, only slightly above the average. With respect to Clusters 3
and 4, with average Walk Scores, Cluster 3 had an above-average
walking rate, whereas Cluster 4 had a below-average rate.
Walk Scores and walking rates gradually decrease away from
the central area of the city (Figure 3). The distinct colors represent
different Walk Score pairs: Clusters 1 and 2 (blue) with high Walk
Scores, Clusters 3 and 4 (orange) with average Walk Scores, and
Clusters 5 and 6 (gray) with low Walk Scores. Each shade signifies
where the walking rates are higher or lower than the other clus-
ter within that pair, revealing the variation between census tracts
within the same Walk Score classification and in similar geographic
locations.
Despite their comparable Walk Scores (75 and 76), the walking
rate of Cluster 3 (23%) was nearly double that of Cluster 4 (12%)
(Table 2). The walk-to-work rate of both clusters was the same (6%),
suggesting that noncommuting trips (such as shopping trips, trips
to visit a friend, etc.) are driving higher walking rates in Cluster 3.
These groups are very similar socioeconomically, suggesting that
elements of the built environment, and not personal or socioeconomic
characteristics, are influencing the difference in walking behavior
between these clusters. Similar patterns were observed in the HWS
pair (Clusters 1 and 2). Despite similar Walk Scores (90 and 91),
walking rates were much higher in Cluster 1 (35%) than in Cluster 2
(20%). As in the AWS cluster, the socioeconomics and walk-to-work
rates of Clusters 1 and 2 were similar.
Descriptive statistics reveal that the land use characteristics of
interest in higher walking rate clusters follow the expected direction.
For HWS and AWS cluster pairs, the cluster with the lowest walking
rate also has the highest proportion of parking lots and setbacks.
In Cluster 3, the proportion of land used devoted to parking lots and
setbacks (14%) is lower than in Cluster 4 (26%). Conversely, in the
same pair, the cluster with the highest walking rate has the highest
on-street canopy cover: Cluster 3 has a larger proportion of land
as on-street tree canopy (26%) than Cluster 4 (20%). In the HWS
pair, the cluster group with more walking also follows the expected
pattern: there are fewer parking lots and setbacks (6% versus 14%)
and more trees (28% versus 22%) in Cluster 1, where walking rates
are 35%, compared with Cluster 2, where walking rates are 20%.
Herrmann, Boisjoly, Ross, and El-Geneidy 107
Mean Z-Score
High Walk Score (HWS) Average Walk Score (AWS) Low Walk Score (LWS)
FIGURE 2 Cluster analysis of Montreal census tracts (n = 466) with the use of Walk Score and walking rate.
1. High Walk Score–more walking
2. High Walk Score–less walking
5. Low Walk Score–more walking
6. Low Walk Score–less walking
No data
Metro
3. Average Walk Score–more walking
4. Average Walk Score–less walking
FIGURE 3 Geographic distribution of clusters of Walk Score and walking, Montreal census tracts (24, 28).
108 Transportation Research Record 2661
Regression Models
The results of the regression models are reported in Table 3. The first
model explains 46.1% of the variation in walking rate, with parking
lots and setbacks driving the model strongly; for each 10% increase
in the proportion of parking lots and setbacks, a 1.7% decrease in
walking rate is expected (Table 3). Conversely, the presence of a
larger on-street tree canopy is shown to be positively associated
with walking rates; for each 10% increase in tree canopy, a 0.7%
increase in the walking rate is predicted. Model 2 examines the
relationship between walking rates and Walk Scores. The Walk Score
term is significant and positively associated with walking rates, and
the model explains a similar amount of variation in walking rates
(R2 = 44.1%) as Model 1. When Walk Score, parking lots and set-
backs, and tree canopy are modeled together in Model 3, the model
fit (R2 = 50.4%) is better compared with Models 1 and 2. The com-
parative assessment of Model 1 and Model 3 reveals that the pro-
portion of parking lots and setbacks and on-street canopy remains
significant when Walk Score is included in the model. Also, including
the new land use variables (parking lots and setbacks and on-street
tree canopy), in addition to the Walk Score variable, increases the
explanatory power of the model.
With respect to the other explanatory variables, presence of paid
parking and distance to highway are positively associated with
walking rates in all models, as expected. The paid parking variable,
in particular, is highly predictive of walking rates. For example, in
Model 1, walking rates are found to be 6% higher in the census tracts
with paid parking versus those without it. Finally, census tracts with
higher proportions of immigrants and higher median incomes are
negatively associated with walking rates in all three models.
DISCUSSION OF RESULTS
Neighborhoods with similar levels of walkability may “produce”
different levels of walking rates in their populations. The cluster
analysis confirms the preliminary findings that there is considerable
variation in walking rates among census tracts with similar Walk
Scores. The study found three pairs of cluster groups with similar
Walk Scores and similar socioeconomics, but substantial differences
between the groups within the pairs in walking rates and land uses.
The descriptive statistics associated with each cluster group high-
light that in the higher walking rate clusters, the presence of parking
lots and setbacks is lower, whereas the proportion of on-street tree
TABLE 2 Descriptive Statistics of Six Clusters
High Walk Score Average Walk Score Low Walk Score
Variable
High Walking
Cluster 1
Low Walking
Cluster 2
High Walking
Cluster 3
Low Walking
Cluster 4
High Walking
Cluster 5
Low Walking
Cluster 6
Sample size 63 96 79 110 91 27
Walk Score (0–100) 90.02 91.34 74.84 76.45 58.88 37.06
Median household income ($) 43,493 41,068 42,245 42,107 54,586 67,763
Parking lots and setbacks, average (%) 6 14 14 26 28 41
On-street canopy cover, average (%) 28 22 26 20 15 15
Walking rate (%) 35 20 23 12 10 5
Walk-to-work rate (%) 17 15 6 6 4 2
Population density (per km2) 14,020 10,616 9,895 7,751 5,319 2,548
Average walk trip distance (m) 856 942 813 1,019 975 1116
TABLE 3 Regression Coefficients Predicting Walking Rate at Census Tract Level
Model 1 Model 2 Model 3
CI CI CI
Variable Coefficient Lower Upper Coefficient Lower Upper Coefficient Lower Upper
Constant 0.250** 0.219 0.280 –0.015 –0.071 0.042 0.083** 0.023 0.142
Walk Score — — — 0.003** 0.002 0.003 0.002** 0.001 0.002
Parking lots and setbacks –0.169** –0.209 –0.129 — — — –0.130** –0.171 –0.090
On-street tree canopy 0.073** 0.032 0.114 — — — 0.057** 0.017 0.096
Paid parking 0.062** 0.047 0.076 0.035** 0.017 0.052 0.031** 0.015 0.048
Distance from highway 0.010** 0.004 0.017 0.016** 0.009 0.023 0.011** 0.005 0.018
Median household income
(tens of thousands)
–0.014** –0.018 –0.010 –0.005* –0.010 –0.001 –0.008** –0.012 0.003
Immigrant percentage –0.100** –0.148 –0.052 0.092** –0.141 –0.044 –0.098** –0.144 –0.052
Note: CI = confidence interval; — = not used in model; n = 466.
R2: Model 1 = .461; Model 2 = .441; Model 3 = .504.
*p < .05; **p < .01; ***p < .001.
Herrmann, Boisjoly, Ross, and El-Geneidy 109
canopy is greater, as hypothesized. These associations follow both
within and between each cluster pair.
The regression models validate the trends observed in this cluster
and descriptive statistics analysis after adjustment for neighborhood
socioeconomic characteristics. The presence of parking lots and set-
backs was found to be negatively associated with walking rates,
even when controlling with Walk Scores. Conversely, the presence
of an abundant on-street tree canopy was found to be associated
favorably with walking rates in neighborhoods, over and above the
influence of the overall walkability of a neighborhood as measured
by the Walk Score.
The finding on the relationship between walking rates and the
presence of parking lots and setbacks is complementary with many
conventional measures of walkability. Since the presence of parking
lots and setbacks denotes the absence of buildings, it is reasonable to
assume that areas with higher amounts of parking lots and setbacks
will also have lower dwelling densities. As low dwelling densities
decrease local accessibility of destinations and the presence of park-
ing lots increases access to destinations by car (36), areas with more
parking lots and setbacks simultaneously discourage walking trips
and incentivize car travel. Parking lots and setbacks may also impede
street connectivity, especially in urban areas where they are associ-
ated with larger industrial footprints (37). Areas with large industrial
setbacks have also been found to increase pedestrian crash frequency
(38), which could discourage walking. While some of these effects
are accounted for by Walk Score or similar walkability measures, this
study highlights that they are not fully captured by Walk Score.
It was also found that census tracts with a larger street tree
canopy had higher rates of walking, independent of Walk Score. This
finding contradicts a similar study conducted in the Twin Cities
(Minneapolis and Saint Paul, Minnesota), which did not find a
significant difference between the presence of trees and walking rates
for transport (39). Nevertheless, the positive influence of the street
tree canopy on walking can be attributed to a more aesthetically
pleasing walking experience (40, 41). The street tree canopy also
influences the comfort of walking in warmer weather by providing
shade and cooling from evapotranspiration (42, 43). These effects
may encourage walking, as opposed to areas with less trees but more
parking lots and setbacks.
STUDY LIMITATIONS
First, while the Walk Score values used in this analysis are rela-
tively recent (2013), Walk Score has since changed its methodology.
Beginning in 2014, Walk Score began using street network buffers
to derive neighborhood walkability measures; the older Walk Score
values used for this analysis used Euclidean distance buffers instead
(4). Accordingly, this study’s models may slightly underestimate
the association between Walk Score and walking rates. Within the
scope of this study, it was not feasible to obtain more recent Walk
Score values at the level of analysis desired. Nonetheless, the findings
identify environmental and land use factors with strong associations
to walking rates that have never been used by Walk Score.
Second, given the methodology used by the AMT for the 2013
Montreal Origin–Destination Survey, the authors cannot guarantee
that all walking trips that occur in the Montreal CMA were surveyed
evenly at the level of analysis (census tracts). Nonetheless, each
census tract contains at least 30 trips of any travel mode, and the
sample size of the survey (5% of households in the Montreal CMA)
is substantial.
Third, because of the neighborhood level of this analysis, the
results of this analysis are subject to some generalization and smooth-
ing, especially as some data were aggregated at the census tract
level. The effects of these externalities are mitigated by using the
smallest aggregation geography available (census tract). The analy-
sis is intended to understand broad patterns of walking behavior at
the population level and not determinants of an individual’s walking
behavior.
CONCLUSION
This paper examined areas where the strength of association between
Walk Score and walking rates was less strong than expected. The
cluster analysis reveals substantially different walking rates between
clusters with similar Walk Score means. On the basis of the findings in
the regression analysis, the authors can confirm that consideration of
parking lots and setbacks and on-street tree canopies would improve
the predictive power of Walk Score. In areas where the association
between Walk Score and walking rates diverged, this finding is often
explained by the presence of parking lots and setbacks and absence of
trees, as these areas neither increase amenity density (as buildings do)
nor improve the aesthetics or comfort of the pedestrian environment
(as trees do). Therefore, it is important for researchers to consider this
balance between neighborhood land use and greenness when using,
interpreting, and designing walkability metrics.
ACKNOWLEDGMENTS
The authors thank the anonymous reviewers for their feedback on the
earlier version of the manuscript. The authors gratefully acknowledge
the Agence Métropolitaine de Transport (AMT) for providing the
detailed 2013 Origin–Destination Survey data that enabled this study
and particularly Gabriel Sicotte for his feedback. This research was
funded by the Natural Science and Engineering Research Council
of Canada and the Social Sciences and Humanities Research Coun-
cil of Canada. Last but not least, the authors thank Lesley Fordham,
Emily Grisé, Dea van Lierop, and Derrick Swallow of Transportation
Research at McGill University for their support and feedback.
REFERENCES
1. Frank, L. D., J. F. Sallis, B. E. Saelens, L. Leary, K. Cain, T. L. Conway,
and P. M. Hess. The Development of a Walkability Index: Application
to the Neighborhood Quality of Life Study. British Journal of Sports
Medicine, Vol. 44, No. 13, 2010, pp. 924–933. https://doi.org/10.1136
/bjsm.2009.058701.
2. Shay, E., S. C. Spoon, and A. J. Khattak. Walkable Environments and
Walking Activity. Final report. Southeastern Transportation Center,
University of Tennessee, Knoxville, 2003.
3. Marsden, G. The Evidence Base for Parking Policies—A Review. Trans-
port Policy, Vol. 13, No. 6, 2006, pp. 447–457. https://doi.org/10.1016
/j.tranpol.2006.05.009.
4. Duncan, D., J. Aldstadt, J. Whalen, and S. Melly. Validation of Walk
Scores and Transit Scores for Estimating Neighborhood Walkability and
Transit Availability: A Small-Area Analysis. GeoJournal, Vol. 78, No. 2,
2013, pp. 407–416. https://doi.org/10.1007/s10708-011-9444-4.
5. Walk Score Methodology. https://www.walkscore.com/methodology
.shtml.
6. Carr, L., S. Dunsiger, and B. Marcus. Walk Score™ as a Global Esti-
mate of Neighborhood Walkability. American Journal of Preventive
Medicine, Vol. 39, No. 5, 2010, pp. 460–463. https://doi.org/10.1016
/j.amepre.2010.07.007.
110 Transportation Research Record 2661
7. Lwin, K., and Y. Murayama. Modelling of Urban Green Space Walk-
ability: Eco-Friendly Walk Score Calculator. Computers, Environment
and Urban Systems, Vol. 35, No. 5, 2011, pp. 408–420. https://doi.org
/10.1016/j.compenvurbsys.2011.05.002.
8. Lee, I., and D. Buchner. The Importance of Walking to Public Health.
Medicine and Science in Sports and Exercise, Vol. 40, No. 7 (Suppl.),
2008, pp. S512–S518.
9. Heath, G., R. Brownson, J. Kruger, R. Miles, K. Powell, and L. Ramsey.
The Effectiveness of Urban Design and Land Use and Transport Policies
and Practices to Increase Physical Activity: A Systematic Review. Jour-
nal of Physical Activity and Health, Vol. 3, Suppl. 1, 2006, pp. S55–S76.
https://doi.org/10.1123/jpah.3.s1.s55.
10. Powell, K., and S. Blair. The Public Health Burdens of Sedentary Living
Habits: Theoretical but Realistic Estimates. Medicine and Science in
Sports and Exercise, Vol. 26, No. 7, 1994, pp. 851–856. https://doi.org
/10.1249/00005768-199407000-00007.
11. Chastin, S., N. Fitzpatrick, M. Andrews, and N. DiCroce. Determinants
of Sedentary Behavior, Motivation, Barriers, and Strategies to Reduce
Sitting Time in Older Women: A Qualitative Investigation. Inter-
national Journal of Environmental Research and Public Health, Vol. 11,
No. 1, 2014, pp. 773–791. https://doi.org/10.3390/ijerph110100773.
12. Frank, L., J. Sallis, T. Conway, J. Chapman, B. Saelens, and W. Bachman.
Many Pathways from Land Use to Health: Associations Between Neigh-
borhood Walkability and Active Transportation, Body Mass Index, and
Air Quality. Journal of the American Planning Association, Vol. 72,
No. 1, 2006, pp. 75–87. https://doi.org/10.1080/01944360608976725.
13. Filion, P. Metropolitan Planning Objectives and Implementation Con-
straints: Planning in a Post-Fordist and Postmodern Age. Environment
and Planning A, Vol. 28, No. 9, 1996, pp. 1637–1660. https://doi.org
/10.1068/a281637.
14. Norman, J., H. MacLean, and C. Kennedy. Comparing High and Low
Residential Density: Life-Cycle Analysis of Energy Use and Greenhouse
Gas Emissions. Journal of Urban Planning and Development, Vol. 132,
No. 1, 2006, pp. 10–21. https://doi.org/10.1061/(ASCE)0733-9488
(2006)132:1(10).
15. Burchell, R. W., N. A. Shad, D. Listokin, H. Phillips, A. Downs, S. Seskin,
J. S. Davis, T. Moore, D. Helton, and M. Gall. TCRP Report 39: The Costs
of Sprawl—Revisited. TRB, National Research Council, Washington,
D.C., 1998.
16. Jilcott Pitts, S., J. McGuirt, L. Carr, Q. Wu, and T. Keyserling. Associa-
tions Between Body Mass Index, Shopping Behaviors, Amenity Density,
and Characteristics of the Neighborhood Food Environment Among
Female Adult Supplemental Nutrition Assistance Program (SNAP)
Participants in Eastern North Carolina. Ecology of Food and Nutrition,
Vol. 51, No. 6, 2012, pp. 526–541. https://doi.org/10.1080/03670244
.2012.705749.
17. Chiu, M., B. Shah, L. Maclagan, M. Rezai, P. Austin, and J. Tu. Walk
Score® and the Prevalence of Utilitarian Walking and Obesity Among
Ontario Adults: A Cross-Sectional Study. Health Reports, Vol. 26, No. 7,
2015, pp. 3–10.
18. Brown, S., H. Pantin, J. Lombard, M. Toro, S. Huang, E. Plater-Zyberk,
T. Perrino, G. Perez-Gomez, L. Barrera-Allen, and J. Szapocznik. Walk
Score®: Associations with Purposive Walking in Recent Cuban Immi-
grants. American Journal of Preventive Medicine, Vol. 45, No. 2, 2013,
pp. 202–206. https://doi.org/10.1016/j.amepre.2013.03.021.
19. Liu, C., S. Erdogan, T. Ma, and F. W. Ducca. How to Increase Rail
Ridership in Maryland? Direct Ridership Models (DRM) for Policy
Guidance. Presented at 93th Annual Meeting of the Transportation
Research Board, Washington, D.C., 2014.
20. Rakha, T., and C. Reinhart. A Carbon Impact Simulation-Based Frame-
work for Land Use Planning and Non-Motorized Travel Behavior
Interactions. In Proceedings of the 13th Conference of the International
Building Performance Simulation Association, 2013, pp. 1248–1255.
21. Duncan, D., J. Aldstadt, J. Whalen, S. Melly, and S. Gortmaker. Vali-
dation of Walk Score® for Estimating Neighborhood Walkability:
An Analysis of Four US Metropolitan Areas. International Journal
of Environmental Research and Public Health, Vol. 8, No. 12, 2011,
pp. 4160–4179. https://doi.org/10.3390/ijerph8114160.
22. Koschinsky, J., E. Talen, M. Alfonzo, and S. Lee. How Walkable Is
Walker’s Paradise? Environment and Planning B: Urban Analytics
and City Science, Vol. 44, No. 2, 2017, pp. 343–363.
23. Weinberger, R., and M. N. Sweet. Integrating Walkability into Planning
Practice. Transportation Research Record: Journal of the Transporta-
tion Research Board, No. 2322, 2012, pp. 20–30. https://doi.org/10.3141
/2322-03.
24. Fichier de Déplacements des Personnes dans la Région de Montréal
Enquête Origine-Destination 2013. Agence Metropolitaine de Transport,
Montreal, Quebec, Canada, 2013.
25. Chapleau, R. Conducting Telephone Origin-Destination Household
Surveys with an Integrated Informational Approach. In Proceedings of
the International Conference on Transport Survey Quality and Inno-
vation, Transportation Research Board, Washington, D.C., 1997.
26. Manaugh, K., and A. El-Geneidy. What Makes Travel ‘Local’: Defining
and Understanding Local Travel Behaviour. Journal of Transport and
Land Use, Vol. 5, No. 3, 2012, pp. 15–27.
27. Winters, M., C. Voss, M. C. Ashe, K. Gutteridge, H. McKay, and
J. Sims-Gould. Where Do They Go and How Do They Get There? Older
Adults’ Travel Behaviour in a Highly Walkable Environment. Social
Science and Medicine, Vol. 133, 2015, pp. 304–312. https://doi.org
/10.1016/j.socscimed.2014.07.006.
28. Clutter Data. DMTI Spatial, Inc., Markham, Ontario, Canada, 2005.
29. Canopée. Ville de Montréal, Montréal, Quebec, Canada, 2007.
30. Onishi, A., X. Cao, T. Ito, F. Shi, and H. Imura. Evaluating the Potential
for Urban Heat-Island Mitigation by Greening Parking Lots. Urban
Forestry and Urban Greening, Vol. 9, No. 4, 2010, pp. 323–332. https://
doi.org/10.1016/j.ufug.2010.06.002.
31. Shoup, D. The High Cost of Free Parking. Journal of Planning Edu-
cation and Research, Vol. 17, No. 1, 1997, pp. 3–20. https://doi.org
/10.1177/0739456X9701700102.
32. Zhu, X., and C. Lee. Correlates of Walking to School and Implications
for Public Policies: Survey Results from Parents of Elementary School
Children in Austin, Texas. Journal of Public Health Policy, Vol. 30,
No. S1, 2009, pp. S177–S202. https://doi.org/10.1057/jphp.2008.51.
33. Manaugh, K., and A. El-Geneidy. Validating Walkability Indices: How
Do Different Households Respond to the Walkability of Their Neigh-
borhood? Transportation Research Part D: Transport and Environment,
Vol. 16, No. 4, 2011, pp. 309–315. https://doi.org/10.1016/j.trd.2011
.01.009.
34. Heisz, A., and G. Schellenberg. Public Transit Use Among Immigrants.
Canadian Journal of Urban Research, Vol. 13, No. 1, 2004, pp. 170–191.
35. O’Brien, R. A Caution Regarding Rules of Thumb for Variance Infla-
tion Factors. Quality and Quantity, Vol. 41, No. 5, 2007, pp. 673–690.
https://doi.org/10.1007/s11135-006-9018-6.
36. Leslie, E., N. Coffee, L. Frank, N. Owen, A. Bauman, and G. Hugo.
Walkability of Local Communities: Using Geographic Information
Systems to Objectively Assess Relevant Environmental Attributes.
Health and Place, Vol. 13, No. 1, 2007, pp. 111–122. https://doi.org
/10.1016/j.healthplace.2005.11.001.
37. Leigh, N., and N. Hoelzel. Smart Growth’s Blind Side: Sustainable
Cities Need Productive Urban Industrial Land. Journal of the American
Planning Association, Vol. 78, No. 1, 2012, pp. 87–103. https://doi.org
/10.1080/01944363.2011.645274.
38. Ukkusuri, S., L. Miranda-Moreno, G. Ramadurai, and J. Isa-Tavarez.
The Role of Built Environment on Pedestrian Crash Frequency. Safety
Science, Vol. 50, No. 4, 2012, pp. 1141–1151. https://doi.org/10.1016
/j.ssci.2011.09.012.
39. Forsyth, A., M. Hearst, J. M. Oakes, and K. Schmitz. Design and Desti-
nations: Factors Influencing Walking and Total Physical Activity. Urban
Studies (Edinburgh, Scotland), Vol. 45, No. 9, 2008, pp. 1973–1996.
https://doi.org/10.1177/0042098008093386.
40. Litman, T. A. Economic Value of Walkability. Transportation Research
Record: Journal of the Transportation Research Board, No. 1828, 2003,
pp. 3–11. http://dx.doi.org/10.3141/1828-01.
41. Ewing, R., and S. Handy. Measuring the Unmeasurable: Urban Design
Qualities Related to Walkability. Journal of Urban Design, Vol. 14,
No. 1, 2009, pp. 65–84. https://doi.org/10.1080/13574800802451155.
42. Hart, M., and D. Sailor. Quantifying the Influence of Land-Use and
Surface Characteristics on Spatial Variability in the Urban Heat
Island. Theoretical and Applied Climatology, Vol. 95, No. 3–4, 2009,
pp. 397–406. https://doi.org/10.1007/s00704-008-0017-5.
43. Solecki, W., C. Rosenzweig, L. Parshall, G. Pope, M. Clark, J. Cox, and
M. Wiencke. Mitigation of the Heat Island Effect in Urban New Jersey.
Global Environmental Change Part B: Environmental Hazards, Vol. 6,
No. 1, 2005, pp. 39–49.
The Standing Committee on Pedestrians peer-reviewed this paper.
