Gendered walkability: Building a daytime walkability index for women

Abstract

Urban walkability is influenced both by built environment features and by pedestrian demographics. Research has shown that fac- tors influencing women’s walking differ from those affecting men’s. Us- ing a mixed-method approach, this study creates a new women-specific, GIS-based walkability index using San Francisco as a case study, and an- swers two questions: Which variables most influence women’s propensity to walk? And Does the leading walkability index, Walk Score, reflect women’s walkability? Focus group participants (n=17) ranked crime, homeless- ness and street/sidewalk cleanliness as the three most influencing factors on women’s walkability, accounting for 58% to 67% of the Women’s Walkability Index’s total score. The least walkable areas in San Francisco, according to this index, are rated as some of the most walkable neigh- borhoods in the city by Walk Score, despite high crime and homeless- ness density. Walk Score is negatively correlated with the new Women’s Walkability Index (Spearman’s rho = -0.585) and inaccurately represents women’s walkability. If the new index accurately captures the reality of women’s walking, then some of the most widely accepted conventions about what kind of areas promote walking could be inaccurate when it comes to women.

Full text

Abstract: Urban walkability is influenced both by built environment
features and by pedestrian demographics. Research has shown that fac-
tors influencing women’s walking differ from those affecting men’s. Us-
ing a mixed-method approach, this study creates a new women-specific,
GIS-based walkability index using San Francisco as a case study, and an-
swers two questions: Which variables most influence women’s propensity to
walk? And Does the leading walkability index, Walk Score, reflect women’s
walkability? Focus group participants (n=17) ranked crime, homeless-
ness and street/sidewalk cleanliness as the three most influencing factors
on women’s walkability, accounting for 58% to 67% of the Women’s
Walkability Index’s total score. The least walkable areas in San Francisco,
according to this index, are rated as some of the most walkable neigh-
borhoods in the city by Walk Score, despite high crime and homeless-
ness density. Walk Score is negatively correlated with the new Women’s
Walkability Index (Spearman’s rho = -0.585) and inaccurately represents
women’s walkability. If the new index accurately captures the reality of
women’s walking, then some of the most widely accepted conventions
about what kind of areas promote walking could be inaccurate when it
comes to women.
Gendered walkability: Building a daytime walkability index for
women
Article history:
Received: October 12, 2018
Received in revised form:
April 7, 2019
Accepted: May 6, 2019
Available online: June 25, 2019
Data availability: https://datasf
.org/opendata/
Copyright 2019 Yael Golan , Nancy Lee Wilkinson, Jason Henderson, & Aiko Weverka
http://dx.doi.org/10.5198/jtlu.2019.1472
ISSN: 1938-7849 | Licensed under the Creative Commons Attribution – Noncommercial License 4.0
e Journal of Transport and Land Use is the ocial journal of the World Society for Transport and Land Use (WSTLUR)
and is published and sponsored by the University of Minnesota Center for Transportation Studies.
T J  T  L U http://jtlu.org
V. 12 N. 1 [2019] pp. 501–526
Yael Golan
San Francisco State University
yaelg@mail.sfsu.edu
Jason Henderson
San Francisco State University
jhenders@sfsu.edu
Nancy Lee Wilkinson
San Francisco State University
nancyw@sfsu.edu
Aiko Weverka
San Francisco State University
aweverka@gmail.com
1 Introduction
Urban walkability has been linked to many benets, including increased levels of physical activity
(Calise, Heeren, DeJong, Dumith, & Kohl, 2013; Durand, Andalib, Dunton, Wolch, & MA, 2011;
Frank et al., 2006), improved economic performance (Cortright, 2009; Leinberger & Alfonzo, 2012);
and some environmental benets, including a reduction in vehicle miles traveled (Diao & Ferreira,
2014; Frank et al., 2006; Talen & Koschinsky, 2014) and lower levels of hazardous air pollution (Frank
et al., 2006; Zahran, Brody, Maghelal, Prelog, & Lacy, 2008).
An area’s walkability is largely inuenced by built environment features like pedestrian amenities,
land-use mix and proximity to destinations, and also by individual characteristics like the level of famil-
iarity with the area, age and gender. With the aid of Geographic Information Systems (GIS) and other
advanced technologies, researchers can now measure and model the physical and built-environment

502 JOURNAL OF TRANSPORT AND LAND USE 12.1
features that inuence walkability better than ever before (Leslie et al., 2007), though the question of
how specic personal characteristics inuence walking remains largely unanswered.
One prominent online tool applying modern technology to analyze walkability is Walk Score,
which uses a distance decay function to rank the walkability of an address based on its proximity to
nearby destinations such as restaurants and retail (Walk Score, n.d.). Walk Score is considered by some
academics a reliable proxy for neighborhood walkability and has been validated as such by several stud-
ies. Manaugh & El-Geneidy (2011) compared four existing walkability indices at multiple geographic
scales with actual observed travel behavior for over 44,200 home-based trips in Montreal and found
that all four walkability indices performed well in describing pedestrian behavior, though Walk Score
proved to be slightly superior to the other indices at predicting whether a home-based shopping trip
would be made on foot. Duncan, Aldstadt, Whalen, & Melly (2012) tried to validate Walk Score as a
measure of neighborhood walkability and found that while it accurately captures features like retail and
open space density, intersection density and residential density, it does not account for other aspects of
walkability, like sidewalk completeness and average speed limit (Park et al., 2012). Walk Score has been
criticized for focusing solely on exible, leisure-time activities, and ignoring xed activities like work
and childcare (Vale, Saraiva, & Pereira, 2015). Carr, Dunsiger, & Marcus (2010) explored Walk Score’s
validity as a walkability index and found strong and signifıcant correlations between Walk Score and all
objective measures of the physical environment assessed in their study, but also found positive correla-
tions between Walk Score and reported crime, suggesting that walkability-inhibiting factors may not be
reected by the index (Koschinsky & Talen, 2015).
Despite the richness of available data and the large volume of current research on walkability, the
study of gender dierences in walking remains limited, especially in the context of built environment
attributes, (Owen, Humpel, Leslie, Bauman, & Sallis, 2004), with the exception of studies focusing on
walking for exercise (Durand et al., 2011; Kerr et al., 2014; Van Dyck et al., 2015) or on the associations
between walkability and health-related measures in women (Sugiyama, Salmon, Dunstan, Bauman, &
Owen, 2007). Studies that do focus on gender in the context of built environment features generally
point to substantial dierences between men and women in regards to their sensitivity to such aspects
as trac (Clifton & Livi, 2005), sidewalks (Brookeld & Tilley, 2016; Clifton & Dill, 2005), neigh-
borhood aesthetics (Pelclová, Frömel, & Cuberek, 2013) and other built environments and streetscape
features (Owen et al., 2004; Park & Calvert, 2008). Fear of violent crime has been a dominant theme
in research on women’s relationship with public spaces, built environment characteristics and walking
(Koskela & Pain, 2000; Loukaitou-Sideris, 2006; Pain, 2001; Valentine, 1989), but little attention has
been given to the gendered patterns of walking outside of women’s fear of crime. e limited academic
attention to women’s pedestrian experiences points to a gap in our understanding of both walkability
and women’s mobility.
is study aims to address this gap by exploring women’s subjective pedestrian experiences and
creating a women-specic walkability index focusing on daytime walking, when fear of crime is assumed
to be less pronounced. By employing a mixed-method approach comprised of focus groups, GIS and
statistical analysis, two research questions are explored: Which variables most inuence women’s propensity
to walk? And Does the leading walkability index, Walk Score, reect women’s walkability?
2 Methods
2.1 Study area
San Francisco is consistently ranked the second most walkable large city in the U.S. (Walk Score,
2016)—and has a culture of walking, with almost a quarter of daily trips made on foot (SFMTA,

503Gendered walkability: Building a daytime walkability index for women
2014), a much larger share than the national average of about 10% (U.S. Department of Transporta-
tion, 2011). It is a stated goal of the city to further improve walking conditions (San Francisco Planning
Department, 2011).
Understanding the mechanisms and built-environment features that inuence women’s propensity
to walk in a walkable city like San Francisco can have important implications for the ability of urban
planners and policy makers—locally, regionally and nationally—to improve female residents’ pedestrian
experience, increase overall walkability, and create better, more pedestrian-friendly environments that
encourage walking over driving, for both genders.
2.2 Focus groups
Two focus groups were held in fall 2016. Each meeting took about 90 minutes, and a total of seventeen
women participated. Participants were recruited using snowball sampling, with a recruitment message
sent via email and posted on the Nextdoor neighborhood social network in 31 San Francisco neigh-
borhoods. Participants were female residents of San Francisco over the age of eighteen who had been
San Francisco residents for at least three months and did not reside in on-campus housing in a school,
hospital, or military base. Participants were not paid for their participation.
Focus group participants were encouraged to discuss daytime walking and ignore nighttime walk-
ing, which is both less prevalent (Cervero & Duncan, 2003; Clifton & Livi, 2005) and more heavily
inuenced by fear of crime (Loukaitou-Sideris, 2006). An emphasis was placed on participants’ day-to-
day walking for transit, leisure and errands, as opposed to walking for exercise. Participants were asked
to describe and discuss positive and negative walking experiences, favorite and least favorite routes,
and what aects them the most when walking or making walking-related decisions, like route choice.
Participants were also asked to discuss the importance of ten index variables selected by the researcher
a priori based on an extensive literature review pointing to fear of crime, safety from trac, and safe
sidewalk conditions as key components of women’s walkability. e majority of variables included in the
index and discussed by participants represent these three main concerns, with the addition of the types
of businesses on the street. e ten variables discussed included sidewalk quality, sidewalk cleanliness,
crime, presence of parks, presence of homeless people or encampments, vehicular trac speed, grati,
o-street parking, Americans with Disabilities Act (ADA) curb ramps, and types of businesses that de-
terred or promoted women’s walking.
At the end of each focus group meeting, participants were asked to ll out a prioritizing grid based
on Pairwise Comparison (Figure 1), a method for comparing alternatives in pairs to judge which of each
pair is preferred. In this case, the alternatives compared were the ten variables discussed at focus group
meetings, and participants were asked to choose, for each pair of variables, the one that had a stronger
eect on their walking. e number of times each variable was chosen as the more important one was
summarized across all participants and the variables were ranked accordingly, resulting in an aggregate
ranking of all ten variables based on focus group participants’ preferences.

504 JOURNAL OF TRANSPORT AND LAND USE 12.1
A:Sidewalk
quality
B:Street&
sidewalk
cleanliness
C:Fearof
crime
D:Presence
ofparks E:Curbramps
F:Homeless
people/enca
mpments
G:Volumeof
vehicular
traffic
H:Parking
garages I:Grafitti
J:Typesof
Businesseson
thestreet
A:SidewalkQuality
(width,completeness)
B:Street&sidewalk
cleanliness
C:Fearofcrime
D:Presenceofparks
E:Curbramps
F:Presenceofhomeless
people/encampments
G:Volumeofvehicular
traffic
H:Parkinggarages
I:Grafitti
J:TypesofBusinesseson
thestreet
Figure 1. Pairwise Comparison prioritizing grid for the ten variables

505Gendered walkability: Building a daytime walkability index for women
e Analytical Hierarchy Process (AHP) (Saaty, 2008) was then used to transform the Pairwise
Comparison-based ordinal rankings into mathematical weights that could be used to assign relative
importance of spatial data layers in GIS. AHP measures the relative importance of each pair of variables
on a scale of 1 to 9 (Figure 2). Here, the AHP scale was used to measure the relative importance of each
pair of variables across all participants:
Figure 2. e scale of relative importance (based on Saaty, 2008)
Once the relative importance of each variable was calculated (Figure 3), variable weights were de-
rived by dividing the value assigned to the choice A vs. B by the sum of values for all choices regarding
A (A vs. C, A vs. D, etc.), and making equal to 1 the sum of entries on each column. Variable weights
were then calculated by averaging the entries on each row, with the sum of all weights totaling 100%.
Intensity of
Importance
Definition
Explanation
1 Equal importance
The two variables A, B, equally influence walking
(half of the participants chose A over B)
3
Weak importance
of one over the
other
Variable A is slightly more influential than B
(11 of 17 participants chose A over B)
5
Essential or
strong importance
Variable A is more influential than B
(13 of 17 participants chose A over B)
7
Demonstrated
importance
Variable A is much more influential than B
(15 of 17 participants chose A over B)
9
Absolute
importance
Variable A is absolutely more influential than B
(all 17 participants chose A over B)
2, 4, 6, 8 Intermediate values between the two adjacent statements
Reciprocals of
above
If variable A has one of the above numbers assigned to it when
compared with variable B, then B has the reciprocal value when
compared with A


506 JOURNAL OF TRANSPORT AND LAND USE 12.1
A:Sidewalk
quality
B:Street&
sidewalk
cleanliness
C:Fearof
crime
D:Presence
ofparks
E:Curb
ramps
F:Homeless
people/enc
ampments
G:Volume
ofvehicular
traffic
H:Parking
garages I:Grafitti
J:Typesof
Businesses
onthe
street
A:SidewalkQuality
(width,completeness)
1.00 0.17 0.11 0.25 8.00 0.17 0.20 3.00 5.00 0.33
B:Street&sidewalk
cleanliness 6.00 1.00 0.14 1.00 9.00 0.25 2.00 7.00 9.00 1.00
C:Fearofcrime 9.00 7.00 1.00 6.00 8.00 5.00 6.00 8.00 9.00 7.00
D:Presenceofparks 4.00 1.00 0.17 1.00 9.00 0.33 1.00 5.00 8.00 1.00
E:Curbramps 0.13 0.11 0.13 0.11 1.00 0.11 0.13 0.20 0.20 0.13
F:Presenceofhomeless
people/encampments 6.00 4.00 0.20 3.00 9.00 1.00 4.00 8.00 9.00 6.00
G:Volumeofvehicular
traffic 5.00 0.50 0.17 1.00 8.00 0.25 1.00 6.00 8.00 2.00
H:Parkinggarages 0.33 0.14 0.13 0.20 5.00 0.13 0.17 1.00 2.00 0.17
I:Grafitti 0.20 0.11 0.11 0.13 5.00 0.11 0.13 0.50 1.00 0.17
J:TypesofBusinesses
onthestreet
3.00 1.00 0.14 1.00 8.00 0.17 0.50 6.00 6.00 1.00
Figure 3. e relative importance matrix

507Gendered walkability: Building a daytime walkability index for women
2.3 Data and variables
e Women’s Walkability Index (WWI) was created using ArcGIS software version 10.4.1 (ESRI, Red-
lands, CA). In addition to the ten variables ranked by focus group participants, slope was added as an
eleventh variable to account for San Francisco’s hilly terrain. Participants in both groups noted that
slope was an inuential factor in their walking behavior. However, since it was not part of the original
prioritizing grid, slope was not given a rank by participants and a corresponding AHP weight. To solve
this, slope was added to the AHP analysis as an eleventh variable and given a median rank, as if half
of the women found it more important than other variables, and half of the women did not. is was
done after all other variable rankings and weights had been calculated, to allow for an analysis both with
and without slope. is provides an opportunity to see what the rankings for San Francisco would look
like if the city were at and allows for a replication of the index in other geographies that are not heavily
inuenced by slope without having to change variable rankings and weights.
e Sidewalk Quality variable was ultimately excluded from the GIS analysis, despite discussion by
focus group participants, since sidewalk condition data for San Francisco was not available. erefore,
AHP weights had to be recalculated excluding the Sidewalk Quality variable.
All data sets were obtained from SF OpenData (https://data.sfgov.org), the open-access online por-
tal for data published by the City & County of San Francisco. For parking garages, parks, and speed lim-
its, GIS shapeles were available on the SF Open Data platform. e other data sets were downloaded
as comma-separated value (CSV) les with location information. Latitude and longitude measurements
were converted to spatial data and address data were geocoded using an ESRI 2013 US street address
locator. Slope was derived from the 10-meter digital elevation model (DEM) of San Francisco obtained
from the United States Geological Survey (USGS) website (https://viewer.nationalmap.gov/basic/). All
datasets were re-projected and analyzed in the North American Datum (NAD) 1983 State Plane coor-
dinate system. Since 266 city blocks did not have speed limit data, these blocks were excluded from the
analysis altogether so that 14,507 of 14,773 city blocks were eventually scored by the index (98.2%).
For the crime variable, only daytime crimes relevant to pedestrians and street life were included:
assault, burglary, disorderly conduct, drug/narcotic, drunkenness, gang activity, liquor laws, loitering,
prostitution, robbery, sex oenses, vandalism and vehicle theft. Other types of crimes, like fraud, arson,
bribery etc. were excluded from analysis because it is assumed that they do not aect pedestrians. Ad-
ditionally, crimes occurring between 8pm and 6am were excluded, due to this index’s focus on daytime
walkability.
For the “types of businesses” variable, registered business locations were divided into three catego-
ries based on the literature review and focus group discussions: 1) Walkability-promoting businesses
such as retail, restaurants, coee shops, grocery stores, physical tness facilities, beauty salons, schools
and childcare facilities; 2) Walkability-inhibiting businesses such as liquor stores, auto repair shops, gas
stations, drinking places, warehouses and industrial activities; and 3) “Neutral” businesses that neither
promote nor inhibit walkability, including accountants, banks, law oces and such.
Figure 4 describes the variables, measures, data sources and direction of inuence of each measure
on women’s walking: negative (-) or positive (+).

508 JOURNAL OF TRANSPORT AND LAND USE 12.1
Figure 4. Variables, measures, direction of inuence and data sources
Figure 5 provides a visual step-by-step roadmap of the methodological steps taken.
Variable Measure Direction of
Influence
Data Source (City
Department) and Year
Crime
Number of daytime
“pedestrian affecting”
crimes per block
(-) SF Police Department,
2016 data
Presence of
homeless people
or encampments
Number of requests for
cleanup of encampments,
carts, needles or human
waste per block
(-) SF 311 (customer
service), 2016 data
Street and
sidewalk
cleanliness
Number of requests for
cleanup of street, garbage
cans, bulky items and
other waste per block
(-) SF 311 (customer
service), 2016 data
Vehicular traffic Maximum speed limit per
block (MPH) (-)
SF Municipal
Transportation Agency,
2016 data
Parks & open
space
Presence of parks and
open spaces on or adjacent
to a block
(+) SF Recreation & Parks
Department, 2016 data
Type of
businesses on the
street
Number of walkability-
promoting businesses per
block
(+)
SF Treasurer – Tax
Collector, 2015 data
Number of walkability-
inhibiting businesses per
block
(-)
Number of walkability-
neutral
businesses per block (+)
Off-street
parking lots and
parking garages
Number of parking spaces
in off-street parking
garages and parking lots
per block
(-)
SF Municipal
Transportation Agency,
2016 data
Graffiti incidents
Number of reported
graffiti incidences per
block in last 30 days
(-) SF 311 (customer
service), 2016 data
Curb ramps
(ADA)
Number of curb ramps per
block (+) SF Department of Public
Works, 2016 data
Slope
Block slope (difference in
elevation between two
ends of the block)
(-) United States Geological
Survey (USGS), 2017


509Gendered walkability: Building a daytime walkability index for women
Figure 5. Women’s Walkability Index step-by-step methodology
2.4 GIS analysis
A vector-based approach was used to analyze most variables, in order to capture their density at the city
block level and provide a per-block index score. e rationale for this approach was twofold: rst, the
City of San Francisco reports much of its data at the block level rather than providing exact addresses;
second, for many of the examined measures, like street cleaning requests, the intensity (i.e., density) of
incidents for any given city block is presumably more meaningful to women pedestrians than the exact
location of each incident.
First, a 10-meter buer was added to the streets layer to capture only those incidents that occurred
10 meters from the center of the road in either side. e buer distance was based on San Francisco’s
recommended street and sidewalk width of 17-18.3 meters (56-60 feet) (San Francisco Fire Depart-
ment, n.d.; San Francisco Planning Department, 2010). Next, the number of incidents per block was
calculated using a spatial join for each of the following variables: walkability-promoting businesses,
walkability-inhibiting businesses, walkability-neutral businesses, parking spaces in o-street lots and
garages, requests for street and sidewalk cleaning, reports of grati, and intersections with curb ramps.
Parks were given a score of 1 if the block was immediately adjacent to a park, and a score of 0 otherwise,
since the walkability-related benets of parks appeared to only occur immediately adjacent to them,
based on focus group discussions. e dataset for maximum speed limits was already in a per-block
format so no further manipulation was needed.
For businesses, a new dataset was calculated based on the three categories of businesses (walkability-
promoting, walkability-inhibiting, and neutral). e density per block of each of these three categories

AHP–Analyt ical
HierarchyProcess
Pairwise
Compari son
Focus groups
discuss & rank
variables
Transform
rankings into
weights totaling
100%
Re-calculate
weights with slope
added & sidewalk
condition excluded
Compare WWI
scores to Walk
Score
WWIScore
GISDataPrep
GISAnaly sis
Index Weights
SFOpenData
Exclude 266 blocks with missing data
Exclude nighttime & irrelevant crimes
Categorize businesses based on type
Transform all datasets to per-block measure
Normalize all datasets by block length
Use LineSlope to calculate block slope
10-meter buffer around street in each side
Assign 1 if block adjacent to park, 0 otherwise
0.25"
Calculate each block’s business score
Vector: per-block incidents for all datasets
Raster: Kernel Density (homel essness & crime)
Rescale all vari ables to a 1-10 scale
Researchers
SelectVariabl es
(Basedonlit.
review)

510 JOURNAL OF TRANSPORT AND LAND USE 12.1
was multiplied by a weight selected to reect each category’s inuence on walkability, under the assump-
tions that promoting and inhibiting businesses have the same magnitude of inuence (in opposite direc-
tions), and that neutral businesses have half as much the inuence as promoting or inhibiting businesses,
because even if they are not themselves attractants, neutral businesses still contribute to a livelier street,
which in turn contributes to walkability. e weights selected were therefore 40% each for promoting
and inhibiting businesses, and 20% for neutral businesses.
Once the density per block was calculated for each of the datasets, attributes from all the dierent
layers were joined into one master table, and the density-per-block measure for each variable was nor-
malized by block length. is was done to neutralize the inuence of exceptionally long blocks (i.e., a
very long block may have more crimes reported on it simply because it has greater area in which crimes
can occur rather than actually representing an unusually crime-dense block). To allow for comparison
of metrics with dierent units and scales, data were then rescaled to a continuous 1-10 scale, with 10
representing the “best-case” scenario for the variable (supporting walkability) and 1 representing the
“worst-case” scenario (impeding walkability). Most variables were rescaled using the Natural Breaks
method (Jenks, 1967), based on ten natural breaks. For the nonlinear variables speed limit, parking
garages, and slope rescaling was done manually (Figure 6).
Figure 6. Manually classied categories for non-linear variables
As mentioned, slope analysis was done separately since the slope was not ranked in the prioritiz-
ing grid and was therefore not assigned a weight. e slope for each city block was calculated using the
LineSlope Tool (Davis, 2014), which calculates the dierence in elevation between the two edges of the
block. is was done because some streets in San Francisco are cut into the side of hills, so while their
slope based strictly on a DEM may seem steep, they can be relatively at. Slopes were then reclassied
manually into ve categories informed by San Francisco’s Walk First methodology for determining pe-
destrian activity (San Francisco Planning Department, 2011) and assigned scores of 0, 3, 5, 8 or 10 to
allow for easy comparison with other datasets, with at blocks receiving the highest score.
With the two remaining variables, crime and presence of homeless people or encampments, a raster
approach was taken to account for the spillover eects of these variables, which may cause crime and
homelessness in one block to “spill” outwards and aect women’s walking in adjacent blocks, an eect
that cannot be captured with vector analysis counts of incidents on a block-by-block basis. For example,
small alleyways that have little or no crime or homelessness may be perceived to have similar charac-
teristics to adjacent main streets that have high levels of crime or homelessness. For the raster analysis,
the Kernel Density tool was used to calculate the density of crime reports and of requests for cleanup
of encampments, carts, needles or human waste per 3-meter cell, at a 100-meter kernel search radius.
Score Speed Limit (MPH) Off-Street Parking Spaces
Per Block Slope
1
3
5
8
10
> 45
> 350
> 0.11
36-45
36-350
0.09-0.11
25-35
11-35
0.06-0.08
16-25 1-10 0.03-0.05
=<15
0
=< 0.02


511Gendered walkability: Building a daytime walkability index for women
e values in each raster cell centroid were extracted and clipped to the buered streets layer, and then
aggregated by block using a Spatial Join and Summary Statistics combination. ese measures did not
need to be normalized by block length, so they were rescaled into ten natural breaks and joined back to
the master dataset.
Once the vector and raster sections of the analysis were combined, each variable’s normalized and
rescaled per-block density was multiplied by the weight assigned to it by the Analytic Hierarchy Process,
and the complete index was added together resulting in a Women’s Walkability Index (WWI) score
between 1 and 10 for each block in San Francisco.
2.5 Walk Score
Once the index was completed and scores were available for every city block, the scores were compared
to those generated by Walk Score. Since Walk Score provides point-specic scores (by exact address or
latitude-longitude coordinates) and WWI provides a block-level score, block-midpoint latitude and lon-
gitude coordinates served as the basis for comparison. A basic JavaScript tool (Salzman, 2017) was built
to use these midpoint coordinates to retrieve the corresponding walkability scores from Walk Score’s
API. Walk Scores were retrieved in March 2016. Spearman’s Correlation test was used to examine the
correlation between WWI scores and Walk Scores. is same test was also used to evaluate the correla-
tion between crime and Walk Scores given previous study’s observations about the relationship between
Walk Score and crime levels as well as focus group feedback on the relative importance of this factor.
3 Results
3.1 Focus group discussion
Seventeen women from 13 dierent neighborhoods participated in focus group meetings. None of the
women lived in the urban core areas of the city but many of them worked there and traveled there daily.
Participant ages ranged from early twenties to late sixties with an average age of 37.1 and a median age
of 36. Most participants had been long-time residents of the City and were very familiar with its layout
and geography. According to focus group participants, crime and personal safety proved the most inu-
ential variables on women’s walking, even during the daytime. Most participants stated that they would
prefer walking on streets where other people are present, rather than on small side streets and alleys that
felt deserted. Several participants noted that quiet streets are “eerie” or “unsettling.” Some participants
preferred quieter streets with fewer people but would only walk on such streets if they were familiar with
them and had walked there before. Most participants were highly aware of their surroundings while
walking and had adopted personal strategies to keep themselves safe from crime and threatening situa-
tions. Examples included avoiding walking under bridges or in alleys and small side streets, walking on
the trac side of the sidewalk so that they cannot be grabbed from within buildings or crevices between
buildings, and crossing the street when approaching a group of men.
Homelessness was also very inuential. Most participants only felt comfortable walking by home-
less people when there were not many of them, or if they were familiar “neighborhood regulars.” When
this was the case, participants tended to view homeless individuals as neighbors and were not intimidat-
ed by their presence. However, there was a consensus that groups of homeless people, as well as homeless
encampments, should be avoided entirely, partly because of personal safety issues and worrying about
erratic, unpredictable behavior by the homeless people (and their dogs), and partly because some par-
ticipants felt that they were intruding into the homeless individuals’ personal space and disturbing their
privacy. Most participants agreed that homeless encampments also aected their walking due to cleanli-
ness and smell issues associated with encampments.

512 JOURNAL OF TRANSPORT AND LAND USE 12.1
Sidewalk cleanliness seemed to highly aect participants’ walking. ey dierentiated between
“normal city dirtiness”—some trash, animal feces, etc.—and “nasty” streets where the location and
amount of garbage and trash suggest that the street is neglected and there are no “eyes on the street,”
indicating a possible threat to personal safety. Several participants also mentioned bad smells (mostly
of human feces or urine) as major deterrents to walking. San Francisco has a relatively large homeless
population (795 homeless individuals per 100,000 residents compared to an average of 479 among peer
cities) (City and County of San Francisco Oce of the Controller, 2017), which has drawn media and
public attention as the city tries to deal with an increase in complaints of human feces, syringes and
homeless encampments on city sidewalks (Smith, 2016), particularly in some areas usually considered
highly walkable.
Most participants agreed that urban parks are an attractant in terms of walking and that they gener-
ally promote walkability, but there was a consensus around the need to avoid walking in or near parks at
night or early in the morning. Even during the daytime, most women said they would not walk alone
in larger parks with lots of hidden spots and remote trails, echoing previous studies on women’s fear of
crime in urban green spaces (Sreetheran & Van Den Bosch, 2014).
Most women stated that they liked walking on “visually interesting” streets with businesses, par-
ticularly retail with interesting shop fronts, and food and beverage businesses, and would prefer to avoid
blocks with “walkability-inhibiting” types of businesses such as liquor stores, auto shops and industrial
uses. Two participants mentioned that when walking their dogs, they tend to avoid streets with multiple
retail businesses due to high density of people, which makes walking their dogs more challenging. One
participant noted that she tends to avoid commercial streets when walking with her toddler for the same
reason, though she does prefer commercial streets when walking alone.
Vehicular trac was a major cause for concern, and several participants tended to avoid major
streets with many trac lanes and high trac volumes or high speed limits. Strategies for avoiding or
minimizing the risks of vehicular trac included changing routes and walking on smaller streets, walk-
ing facing oncoming vehicular trac, and opting for streets with stop lights rather than stop signs. Some
participants also noted that trac noise and exhaust bothered them when walking on major streets. A
few participants, particularly older ones, were also troubled by cyclists and preferred to refrain from
walking on streets they knew to have a large volume of bicycle trac, even when there were designated
bike lanes present.
e four remaining variables—grati, o-street parking, curb ramps and sidewalk quality - were
only discussed briey as there was a consensus that these were relatively inconsequential in terms of
their impact on women’s daytime walking. Slope was also discussed; a few participants mentioned that
hilly streets, while challenging to climb, might be safer or more pleasant to walk on since there are fewer
crimes and fewer homeless people, but on the other hand are less appealing since fewer people walk on
them, which makes them feel deserted and unsettling.
3.2 Focus groups rankings and weights
Participant rankings were quite consistent, with a majority of participants ranking crime and homeless-
ness as the two most inuential variables, followed, for the most part, by vehicular trac and sidewalk
cleanliness (not necessarily in that order). All but one participant ranked at least one of the two vari-
ables—crime and homelessness—in the top three. Grati and curb ramps were consistently ranked as
the least important variables. Variation across participants’ rankings was largest with regards to busi-
nesses, parks, and sidewalk quality, but these variables, too, were generally consistently ranked as of
medium importance (Figure 7).

513Gendered walkability: Building a daytime walkability index for women
Figure 7. Focus groups rankings of the ten variables
As mentioned in the Methods section, the sidewalk quality variable had to be excluded from the
index since data for this variable was not available. And since participants did not rank slope, the index
needed to be computed both with and without slope. erefore, the AHP process used to transform
the rankings into weights for the GIS analysis was repeated twice, to create two models. e rst model
excluded both sidewalk quality and slope, and the second model excluded sidewalk quality, but included
slope. Variable weights do not vary signicantly between the two models, nor do they dier signicantly
when all variables are taken into account (Figure 8). Under both models, crime and homelessness exert
the strongest eects on women’s walkability (about 31%-37% and 17%-20% of the total index score,
respectively).
Variable
Rank
Fear of crime
1
Presence of homeless people or encampments
2
Street and sidewalk cleanliness
3
Vehicular traffic volume
4
Parks & open space
5
Type of businesses on the street
6
Sidewalk quality*
7
Off-street parking garages and parking lots
8
Graffiti incidents
9
Curb ramps
10
Slope**
-
* Sidewalk quality had to be excluded from the GIS analysis due to data limitations
** Slope was not part of the prioritizing grid and was no ranked by focus group participants


514 JOURNAL OF TRANSPORT AND LAND USE 12.1
Figure 8. Variable weights under the two models
Street and sidewalk cleanliness also proved highly important to women (around 10% of the index
score), and these three variables combined (crime, homelessness and street cleanliness) accounted for
about 58%-67% of the nal index score. e presence of parks (around 8.5%), the volume of vehicular
trac (about 9%) and the type of businesses found on the street (≈7.5%) were also important. e four
remaining variables - grati, o-street parking, curb ramps and sidewalk quality - together accounted
for only about 12% of the nal index score when all variables are taken into account, or about 7%-8%
when excluding sidewalk quality (Models 1 and 2).
Model 1
Model 2
All
Variables
Excluding
Slope &
Sidewalk
Quality
Excluding
Sidewalk
Quality
A: Sidewalk Quality
B: Street & Sidewalk Cleanliness
C: Fear of Crime
D: Presence of Parks
E: Presence of Curb Ramps
F: Homelessness
G: Vehicular Traffic Volume
H: Off-Street Parking
I: Graffiti Incidents
J: Types of Businesses
K: Slope
4.5%
0.0%
0.0%
9.7%
10.2%
9.7%
29.5%
37.0%
31.1%
8.3%
8.9%
8.5%
1.8%
1.4%
2.1%
17.3%
20.3%
17.5%
8.7%
9.1%
8.8%
2.9%
2.9%
3.4%
2.5%
2.4%
2.9%
7.2%
7.8%
7.6%
7.6%
0.0%
8.4%
Total
100.0%
100.0%
100.0%


515Gendered walkability: Building a daytime walkability index for women
Figure 9. Non-parametric correlations between variables—Spearman’s rho
e correlations between the variables used in the models were also examined. While many of the
variables showed statistically signicant correlations, these relationships varied in strength, with most
correlations being of weak to moderate strength. e two strongest correlations, between crime and
homelessness and between cleaning requests and homelessness, had Spearman’s rho coecient values of
0.629 and 0.628, respectively (Figure 9).
Index results
Both models had similar outcomes with few substantial dierences in minimum, maximum or
mean WWI score for San Francisco city blocks (Figure 10).
Parks
Off-Street
Parking
Graffiti
Incidents
Curb Ramps
Cleaning
Requests
Number of
Businesses
Crime
Density
Homeless
Density
Slope
Speed Limit
Parks
Coefficient 1.000 -.048** -.036** -.014 -.066** -.116** -.047** -.087** .030** -.015
Sig. (2-tailed) . .000 .000 .079 .000 .000 .000 .000 .000 .069
Off-Street
Parking
Coefficient -.048** 1.000 .195** .022** .195** .215** .316** .249** -.199** .066**
Sig. (2-tailed) .000 . .000 .009 .000 .000 .000 .000 .000 .000
Graffiti
Incidents
Coefficient -.036** .195** 1.000 .134** .429** .295** .413** .420** -.138** .084**
Sig. (2-tailed) .000 .000 . .000 .000 .000 .000 .000 .000 .000
Curb
Ramps
Coefficient -.014 .022** .134** 1.000 .328** .263** .168** .117** -.024** .088**
Sig. (2-tailed) .079 .009 .000 . .000 .000 .000 .000 .003 .000
Cleaning
Requests
Coefficient -.066** .195** .429** .328** 1.000 .448** .522** .628** -.183** .141**
Sig. (2-tailed) .000 .000 .000 .000 . .000 .000 .000 .000 .000
No, of
Businesses
Coefficient -.116** .215** .295** .263** .448** 1.000 .425** .371** -.129** .019*
Sig. (2-tailed) .000 .000 .000 .000 .000 . .000 .000 .000 .023
Crime
Density
Coefficient -.047** .316** .413** .168** .522** .425** 1.000 .629** -.199** .075**
Sig. (2-tailed) .000 .000 .000 .000 .000 .000 . .000 .000 .000
Homeless
Density
Coefficient -.087** .249** .420** .117** .628** .371** .629** 1.000 -.245** .112**
Sig. (2-tailed) .000 .000 .000 .000 .000 .000 .000 . .000 .000
Slope
Coefficient .030** -.199** -.138** -.024** -.183** -.129** -.199** -.245** 1.000 -.112**
Sig. (2-tailed) .000 .000 .000 .003 .000 .000 .000 .000 . .000
Speed
Limit
Coefficient -.015 .066** .084** .088** .141** .019* .075** .112** -.112** 1.000
Sig. (2-tailed) .069 .000 .000 .000 .000 .023 .000 .000 .000 .
**Correlation is significant at the 0.01 level (2-tailed)
* Correlation is significant at the 0.05 level (2-tailed)


516 JOURNAL OF TRANSPORT AND LAND USE 12.1
Figure 10. Descriptive statistics of the two models
Map 1 shows the aggregated WWI score for each city block based on Model 2 (the more extensive
model, which excludes sidewalk quality but includes slope). High-walkability areas are shown in blue
and low-walkability areas are shown in red. e city’s Downtown, Tenderloin, and South of Market
(SoMa) neighborhoods, all with relatively high prevalence of homelessness and crime, have the lowest
overall WWI scores and create a large, continuous area of low walkability for women in the heart of San
Francisco. Additional pockets of low walkability appear in other parts of the city, mostly in the eastern
side, in areas with light-industrial activity, near freeway underpasses, and along large commercial corri-
dors like Mission Street and Market Street, which are generally considered very walkable. Map 2 shows a
hot and cold spot analysis (Getis & Ord, 1992) of Model 2 WWI scores, where cold spots (marked blue)
are statistically signicant clusters of very low women’s walkability scores and hot spots (marked red)
are statistically signicant clusters of very high walkability scores. is map attests to the positive eects
of parks, open space and greenery on women’s walkability, as hot spots of high walkability closely align
with some of the city’s open spaces, like Lake Merced in the southwest, Ocean Beach and Park Presidio
Blvd going north from Lake Merced, and around the perimeter of Golden Gate Park in the northwest.
Model 1 Model 2
N (blocks scored)
Min WWI Score
Max WWI Score
Mean WWI Score
Standard Deviation
14,507
14,507
3.44
4.07
9.63
9.58
8.15
8.01
0.96
0.83


517Gendered walkability: Building a daytime walkability index for women
Map 1. Women’s Walkability Index block scores—Model 2
Women's Walkability Index
Model 2 Score*
4.23 - 6.38
6.39 - 7.41
7.42 - 8.14
8.15 - 8.81
8.82 - 9.58
* Model 2 is the model wi th 10 variables, excluding sidewalk condition but i ncluding slope
Classification: Natural Breaks
Yael Golan, March 2019
[
0 1 20.5
Kilometers

518 JOURNAL OF TRANSPORT AND LAND USE 12.1
Map 2. Hot and cold spot analysis—Model 2
Walk Score Comparison
Women’s Walkability Index scores were examined against Walk Score for each of the 14,507 city
blocks analyzed and for both models, by comparing the mid-block Walk Score to the corresponding
block’s WWI score. Both models showed a signicant (p < 0.01) moderate negative correlation with
Walk Score, with Spearman’s rho coecients of -0.585 (model 1) and -0.525 (model 2) (Figure 11).
Walk Score was also moderately positively correlated with crime (Spearman’s rho value of 0.678; p <
0.01).
Figure 11. Correlations between Walk Score and Women’s Walkability Index
Hot and Cold Spot Analysis - WWI Model 2
* Model 2 is the model wi th 10 variables, excluding sidewalk condition but i ncluding slope
Classification: Natural Breaks
Yael Golan, March 2019
[
0 1 20.5
Kilometers
Model 2* Hot Spot Analysis
Cold Spot - 99% Confidence
Cold Spot - 95% Confidence
Cold Spot - 90% Confidence
Not Significant
Hot Spot - 90% Confidence
Hot Spot - 95% Confidence
Hot Spot - 99% Confidence
Model 1 Model 2
Walk Score
Correlation Coefficient
-.585**
-.525**
Sig. (2-tailed)
.000
.000
N
14,507
14,507
** Correlation is significant at the 0.01 level (2-tailed)


519Gendered walkability: Building a daytime walkability index for women
WWI scores were transformed from their original 1-10 scale to a 1-100 scale to allow for easier
comparison with Walk Score. Map 3 shows the absolute value dierences between Walk Score and
WWI scores, computed by subtracting the rescaled WWI score from the corresponding Walk Score
for each block. Blue areas on this map indicate where indices agree (have similar scores). Red areas in-
dicate large dierences (in absolute values) between the two scores. Such areas include the Downtown,
SoMa and Tenderloin neighborhoods in the southeast, which are considered extremely walkable by
Walk Score, but have low WWI scores because of high crime and homelessness. Other areas where the
indices disagree include Lake Merced and Ocean Beach on the western waterfront and the hilly neigh-
borhoods around Mount Sutro, where lack of walking destinations drives Walk Scores down while low
crime and homelessness prevalence and nearby open space drive WWI scores up; and areas with few
roads, where lower intersection density and longer blocks are likely the reason for Walk Scores being
lower than WWI scores.
Map 3. Dierences between WWI score and Walk Score
4 Discussion
is study aimed to answer two research questions: Which variables most inuence women’s propensity to
walk? And Does the leading walkability index, Walk Score®, reect women’s walkability?
In regard to the rst research question, the results of this study rearm the importance of fear of
crime as an inuence, if not the inuence, on women’s walkability, even during the daytime. By focusing
specically on daytime walkability, this study attempted to capture some of the additional factors and
Differences Between Walk Score and Women's Walkability Index (Absolute Values)
* Calculated as midblock Walk Score mi nus WWI Model 2 block score multiplie d by 10 (absolute values).
Red indicated larger differences between the two i ndices. Blue indicates areas where the indices a gree. Classification: Natural Breaks.
Yael Golan, March 2019
[
0 1 20.5
Kilometers
Walk Score - WWI Score*
0 - 8
9 - 17
18 - 27
28 - 41
42 - 89

520 JOURNAL OF TRANSPORT AND LAND USE 12.1
considerations inuencing women’s walking behavior. However, focus group discussions and variable
rankings by focus group participants suggested that fear of crime remains the number one factor among
many factors aecting women when they decide where, when and how to walk, regardless of time of
day. e idea that fear of crime is signicantly less inuential on women’s daytime walking than it is at
night seems to be questionable, at least in San Francisco. While women may dier in the types of situa-
tions they nd threatening, there was a consensus among participants that a sense of personal safety is a
prerequisite for women’s walking. It is not surprising, then, that three of the four top-ranking variables
were related to sense of safety: crime, homelessness and vehicular trac (ranked numbers 1, 2, and 4,
respectively).
Studies exploring the relationship between crime and walking often encounter a disassociation be-
tween respondents’ sense of neighborhood safety and actual crime incidents: fear of crime is many times
only weakly associated with actual crime (Foster & Giles-Corti, 2008). By using both subjective (focus
group discussions) and objective (reported crime incidents) data, this study was able to minimize this
limitation. Focus group participants discussed their perceived safety in a more general way and were not
asked to rank specic locations in the city by how safe they perceived them to be. Instead, participants
were asked to choose the elements in the built environment that most inuenced them when walking
during the daytime, and their choices were used to create a ranking and weighting system which later
served as the basis for analysis of objective crime data. is minimized the potential for perception-
related biases.
Interestingly, some of the participants said they felt safer and much less vulnerable when running,
or when cycling, than when walking. e perception that walking was the most vulnerable mode of
transportation in the eyes of some women was intriguing and should be further investigated.
e presence of homeless people on a street block does not automatically deter women from walk-
ing there. Rather, as focus group participants pointed out, it is a matter of familiarity and magnitude.
While the “familiarity” eect of homeless people could not be accounted for in this index, the issue of
magnitude was addressed in the GIS analysis, by using a density measure for homelessness. Many of the
participants explained their inclination to avoid walking near homeless people as an attempt to avoid
unpredictable situations, echoing Valentine’s (1989) claim that women’s fear in public spaces is basically
a fear of unpredictable and uncontrollable behaviors by strangers. More importantly, Valentine’s asser-
tion that women perceive only men as strangers is also echoed in this study. As the women in one of the
focus groups agreed - after a long discussion of fear of crime and homelessness, objectication of women
in the public sphere, and catcalling by construction workers—that what they are really trying to avoid
are groups of men:
Participant A: “I would avoid construction sites because of catcalling… I’m not afraid of them,
just disgusted by male objectication of women”
Participant B: “Yeah, just leave me alone, just let me walk down the street”
Participant C: “Depends on the size of construction. Small contractors who live in the city and
raise daughters here are more respectful”
Participant D: “I get catcalled by homeless people and would prefer construction workers over
homeless (people)”
Participant E: “So I think the point is that it’s not necessarily construction workers, just groups of
men. Even a group of men in suits would bother me”
Participant A: “I agree, I’d avoid groups of men” (Most other participants nodded or mumbled in
agreement)
Street and sidewalk cleanliness (ranked third) and the presence of parks or open spaces (ranked
fth) both point to the importance of aesthetics to women when walking, in line with previous stud-
ies that found women’s walking to be strongly associated with neighborhood aesthetics (Pelclová et al.,

521Gendered walkability: Building a daytime walkability index for women
2013), more so than men’s walking (Van Dyck et al., 2013). However, the relationship between parks,
greenery and open spaces and women’s walking is convoluted. Previous studies have shown that parks
and badly placed bushes and greenery (Koskela & Pain, 2000) and deserted open spaces (Valentine,
1989) can deter women’s walking, but this study found small urban parks to be a major attractant to
women’s walking, at least during the daytime. Most women in this study would not avoid parks and
greenery, and in fact would actively seek green areas to walk in or near during the daytime, with the
exception of very large parks with remote trails, and large wilderness-like open spaces.
As for vehicular trac, previous research has shown that women are more likely than men to view
the presence of trac as an important factor when walking (Weinstein Agrawal, Schlossberg, & Irwin,
2008) and are more likely than men to list reasons related to the amount of trac as deterrents to
walking (Clifton & Dill, 2005). Indeed, focus group participants ranked vehicular trac as the fourth
most inuential factor on walking, and extensively discussed their fear of vehicular trac and major
thoroughfares, noise from trac and exhaust fumes. Interestingly, some participants were almost as
concerned about bicycle trac. Future research into the relationship between walkability and bicycling
may shed some light on this issue.
While mixed-use streets with retail businesses are generally shown to promote walkability (Pelclová
et al., 2013; Talen & Koschinsky, 2013), the presence of businesses on the street seems to have a mixed
eect on focus group participants. Most participants did state that having “eyes on the street” (Jacobs,
1961) and seeing other people on the street when walking were important to them, in accordance with
previous research by Weinstein Agrawal et al. (2008). However, at least during the daytime, this seems to
be an issue of personal preference, with a small minority of focus group participants saying they preferred
walking on quieter streets and avoiding the hustle and the bustle of busy commercial streets. e choice
of busy vs. quiet streets also depends, according to focus group discussions, on who is walking (walking
with dogs or small children seems to favor quieter streets, while walking alone favors busy commercial
streets), the time of day (rush hour vs. mid-day) and mood. In terms of the types of businesses that
promote or deter walking, participants mentioned liquor stores, mini-marts that sell liquor and snacks,
auto shops and large industrial/manufacturing businesses as deterrents to walking, and restaurants, cof-
fee shops and small boutique shops as attractants. Other types of establishments believed to promote
walkability for women, like churches and shopping malls (Clifton & Livi, 2005), were not discussed.
Other variables examined in this study, which were found in previous studies to inuence women’s
walking, like parking lots (Valentine, 1989; Clifton & Livi, 2005), grati (Craig, Brownson, Cragg,
& Dunn, 2002), sidewalk conditions (Clifton & Dill, 2005; (Handy, 2007) and curb ramps (Clifton
& Livi, 2005) turned out to be less important to focus group participants and had a smaller eect on
index score. ese four variables combined accounted for only 10.1% - 11.7% of a block’s nal score,
or 6.7% - 8.3% when excluding sidewalk quality.
Based on focus group rankings, the least walkable areas for women in San Francisco are in the
northeastern part of the city. Additional pockets of low women’s walkability are found along large com-
mercial streets and around highway intersections and underpasses. ese areas received low walkability
rankings because they are high-crime, high-homelessness and relatively dirty (i.e., have large numbers
of street and sidewalk cleaning requests), three variables that together account for 58.3% (Model 2) to
67.5% (Model 1) of the entire walkability score. Interestingly, the lowest-walkability areas according to
the WWI are also some of the most densely populated, mixed-use, business-rich areas of San Francisco,
most popular with tourists and locals. If WWI scores accurately represent the reality of women’s walking
then some of the most widely accepted conventions about what kind of areas promote walking may be
inaccurate when it comes to women.
e most consistently walkable areas for women are concentrated on the western side of the city,
mostly as a result of relatively low crime and low homeless densities in these areas. roughout the city,
low-crime and low-homelessness areas adjacent to parks receive very high WWI scores.

522 JOURNAL OF TRANSPORT AND LAND USE 12.1
It seems that the answer to the second research question, Does the leading walkability index, Walk
Score®, reect women’s walkability? is no. Based on focus group discussions and variable rankings, the
most inuential factor on women’s walking is fear of crime, accounting for 31.1% - 37% of the total
WWI score. Since Walk Score’s algorithm does not take crime into account, it is hard to claim that Walk
Score accurately reects women’s walkability. Although Walk Score has been validated by academics as
a reliable walkability measure (Duncan, et al., 2012; Manaugh & El-Geneidy, 2011), it has also been
criticized for ignoring crime, and one study found it to be positively correlated with crime (Carr et al.,
2010). Like Carr et al. (2010), this study found a statistically signicant positive correlation of moderate
strength (Spearman’s rho value of 0.678; p < 0.01) between Walk Score and crime reports in San Fran-
cisco. Walk Score also ignores other built environment variables that were part of WWI, such as trac,
topography and neighborhood aesthetics (Duncan et al., 2012). Additionally, Walk Score was found to
best capture walkability at a large spatial scale of 1,600 meters (1 mile) (Duncan et al., 2012), which is
much larger than the average city block length in San Francisco (about 120 meters, or just under 400
feet), the spatial scale analyzed by WWI. As a large-scale, nationally applied, “one size ts all” walkability
index, Walk Score may not be nuanced enough to detect relative walkability within a generally walkable
place like San Francisco, as WWI does.
Given these dierences between Walk Score and WWI, and especially given the importance of
crime in WWI and its absence from Walk Score, it is no surprise that the two indices show a statisti-
cally signicant negative correlation of moderate strength (Spearman’s rho values of -0.525 to -0.585,
depending on WWI model). e Tenderloin neighborhood, for example, has some of the lowest WWI
block scores in the city, however its Walk Score is 99 (out of 100) making it in the top 5% of neighbor-
hoods in the city in terms of walkability.
is study has limitations. First and foremost, the index weights are based on the subjective prefer-
ences of focus group participants and repeating this study with other participants in San Francisco, or
with participants in other locations, might yield dierent weights. e participant recruitment method
may have introduced a self-selection bias, with women who are particularly interested in walking more
likely to volunteer. Since none of the participants lived in the urban core areas of the city, where crime
and homelessness are most pronounced, it could be argued that participants were more sensitive to these
issues than women who do live in the urban core. However, many of the participants work downtown
and travel there daily. Focus group discussions around crime and homelessness may have inuenced
participants’ perception of these variables’ importance. Some of the datasets used in the analysis, particu-
larly homelessness and street and sidewalk cleanliness, are based on data reported to the city by residents
(i.e., SF311 reports). Such reports may be more likely in certain neighborhoods, or by certain residents,
than others, and it is impossible to know whether absence of reporting indicates absence of occurrences,
or whether occurrences are simply under-reported in certain places. e exclusion of nighttime crimes
from analysis may have created a bias, since it may very well be that areas known to have many night-
time crimes become less walkable for women during the daytime too, even if crimes are not committed
during the day.
Finally, the Women’s Walkability Index models walkability in an additive way, assuming a decision-
making process that considers all walkability-inuencing factors simultaneously. However, anecdotal
evidence from focus group discussions suggests that women’s decision-making process may be com-
prised of several binary decisions, where some minimum factors—like an acceptably low level of crime
or homelessness—must be met, before other factors of lower importance can be weighed additively.
Future research should perhaps explore the idea of a “minimal threshold” for women’s walking.
is study was a rst attempt at creating a women-specic walkability index. Future studies should
repeat this study either with a larger group of women or in other cities to identify issues of broader sig-
nicance. Additional variables may also need to be included to fully account for women’s unique walk-

523Gendered walkability: Building a daytime walkability index for women
ing preferences. To better understand women’s walkability, future research could also explore women’s
daily routes, especially as they relate to familial and caregiving duties, which in many families are still
being carried out primarily by women.
5 Conclusions
e Women’s Walkability Index is heavily inuenced by crime, homelessness and sidewalk cleanliness. It
seems that the Index under-values the presence of businesses on the street, which is traditionally believed
to increase walkability, leading to low walkability scores in areas usually considered highly walkable and
highly appealing for pedestrians. Still, there is no reason to suspect that the index signicantly over-rep-
resents the importance of fear of crime, homelessness and cleanliness, and anecdotal evidence suggests
that this is, indeed, a fair representation of the most important considerations for women pedestrians.
Most of the variables comprising the Women’s Walkability Index have a negative association with
walking, thus WWI could be interpreted as an index that captures deterrents to walking. Walk Score, on
the other hand, is heavily inuenced by the proximity of walking destinations, i.e., attractants. Perhaps
WWI should be used to inform Walk Score, or vice versa. A walkability model integrating both Walk
Score and WWI could oer the most accurate representation of women’s walkability to date.
is is a rst attempt at creating a women-specic walkability index. Future iterations of this index
may provide more insight into women’s walking. is study’s main contributions to the walkability
discussion are its focus on women’s walkability, and its mixed-method approach, combining the added
value of hearing from women rst-hand what matters to them, with the analytical capabilities of GIS
software, which provides the opportunity for a large-scale, replicable analysis. Another major contribu-
tion of this study is the understanding that women’s daytime walking is not much dierent than night-
time walking in that it, too, is governed by fear of crime and the search for a sense of personal safety.
is understanding should guide policy makers, urban planners, landscape architects and walkability
advocates in designing urban environments that are safe and welcoming for women. Policy makers hop-
ing to improve walkability for women should focus on designing neighborhoods and streets that feel
safe. Streets and building designs should enable human interaction and maximize the number of “eyes
on the street” while eliminating or minimizing the number of spaces that feel deserted or unsafe. Trac
calming measures on major pedestrian streets, including a better separation between cyclists and pedes-
trians and certain limitations on freight movement; prioritizing underground parking garages over large
o-street parking lots which reduce the street’s vibrancy and may be seen as a hazard by women; and
adding greenery and shrubbery to urban streets and sidewalks to make them feel more inviting could
also enhance walkability for women.
Future research should expand this study’s scope to look at other cities and geographies, additional
variables not included here, and nighttime walking. e notion of women’s decision-making process as
a set of binary decisions resulting in a “minimal threshold” necessary for women’s walking should also
be further investigated. Finally, the idea that women perceive walking as the most vulnerable means of
transport, expressed by some focus group participants, is an interesting concept that warrants further
investigation.
Understanding the mechanisms and built-environment features that inuence women’s walking,
especially in a walkable city like San Francisco, can have important implications for cities trying to im-
prove women’s pedestrian experience, increase overall walkability, and create better pedestrian environ-
ments that encourage walking over driving for both genders. As many cities around the world become
increasingly more interested in minimizing vehicular usage, improving walkability for women may go a
long way in achieving such goals.

524 JOURNAL OF TRANSPORT AND LAND USE 12.1
References
Brookeld, K., & Tilley, S. (2016). Using virtual street audits to understand the walkability of older
adults’ route choices by gender and age. International Journal of Environmental Research and Public
Health, 13(11), 1061. doi.org/10.3390/ijerph13111061
Calise, T. V., Heeren, T., DeJong, W., Dumith, S. C., & Kohl, H. W. (2013). Do neighborhoods make
people active, or do people make active neighborhoods? Evidence from a planned community in
Austin, Texas. Preventing Chronic Disease, 10, E102. oi.org/10.5888/pcd10.120119
Carr, L., Dunsiger, S., & Marcus, B. (2010). Walk ScoreTM as a global estimate of neighborhood walkabil-
ity. American Journal of Preventive Medicine, 39(5), 460–3. doi.org/10.1016/j.amepre.2010.07.007
Cervero, R., & Duncan, M. (2003). Walking, bicycling, and urban landscapes: Evidence from the
San Francisco Bay Area. American Journal of Public Health, 93(9), 1478–1483. doi.org/10.2105/
AJPH.93.9.1478
Clifton, K. J., & Dill, J. (2005). Women’s travel behavior and land use: Will new styles of neighborhoods
lead to more women walking? Transportation Research Board Conference Proceedings, 35, 89-99.
Clifton, K. J., & Livi, A. D. (2005). Gender dierences in walking behavior, attitudes about walking,
and perceptions of the environment in three Maryland communities. Research on Women’s Issues in
Transportation, 2, 79–88.
Cortright, J. (2009). Walking the walk: How walkability raises home values in U.S. cities. CEO’s for Cit-
ies. Retrieved from http://www.icr.org/article/3110/293/
Craig, C. L., Brownson, R. C., Cragg, S. E., & Dunn, A. L. (2002). Exploring the eect of the environ-
ment on physical activity. American Journal of Preventive Medicine, 23(2), 36–43. doi.org/10.1016/
S0749-3797(02)00472-5
Davis, J. D. (2014). LineSlope.py [Esri ArcPy geoprocessing script]. San Francisco, CA: Institute for
Geographic Information Science, San Francisco State University.
Diao, M., & Ferreira, J. (2014). Vehicle miles traveled and the built environment: Evidence from ve-
hicle safety inspection data. Environment and Planning A, 46(12), 2991–3009. doi.org/10.1068/
a140039p
Duncan, D. T., Aldstadt, J., Whalen, J., & Melly, S. J. (2012). Validation of Walk Scores and transit
scores for estimating neighborhood walkability and transit availability: A small-area analysis. Geo-
Journal, 78(2), 407–416. doi.org/10.1007/s10708-011-9444-4
Durand, C., Andalib, M., Dunton, G., Wolch, J., & Ma, P. (2011). A systematic review of built envi-
ronment factors related to physical activity and obesity risk: Implications for smart growth urban
planning. Obesity Reviews, 12(5), e173–e182. doi.org/10.1111/j.1467-789X.2010.00826.x
Foster, S., & Giles-Corti, B. (2008). e built environment, neighborhood crime and constrained
physical activity: An exploration of inconsistent ndings. Preventive Medicine, 47(3), 241–251. doi.
org/10.1016/j.ypmed.2008.03.017
Frank, L. D., Sallis, J. F., Conway, T. L., Chapman, J. E., Saelens, B. E., & Bachman, W. (2006).
Many pathways from land use to health: Associations between neighborhood walkability and active
transportation, body mass index, and air quality. Journal of the American Planning Association, 72(1),
75–87. doi.org/10.1080/01944360608976725
Getis, A., & Ord, J. K. (1992). e analysis of spatial association by use of distance statistics. Geographi-
cal Analysis, 24(3), 189–206. doi.org/10.1111/j.1538-4632.1992.tb00261.x
Handy, S. L. (2007). Community design and travel behavior: Exploring the implications for women.
UC Davis: Institute of Transportation Studies (UCD). Retrieved from https://escholarship.org/uc/
item/8jx6x28p

525Gendered walkability: Building a daytime walkability index for women
Jacobs, J. (1961). e death and life of great American cities. New York: Vintage Books.
Kerr, J., Norman, G., Millstein, R., Adams, M. A., Morgan, C., Langer, R. D., & Allison, M. (2014).
Neighborhood environment and physical activity among older women: Findings from the San Di-
ego cohort of the Women’s Health Initiative. Journal of Physical Activity and Health, 11(6), 1070–
1077. doi.org/10.1123/jpah.2012-0159
Koschinsky, J., & Talen, E. (2015). Aordable housing and walkable neighborhoods: A national urban
analysis. Cityscape: A Journal of Policy Development and Research, 17(2), 13–56.
Koskela, H., & Pain, R. (2000). Revisiting fear and place: Women’s fear of attack and the built environ-
ment. Geoforum, 31(2), 269–280. doi.org/10.1016/S0016-7185(99)00033-0
Leinberger, C., & Alfonzo, M. (2012). Walk this way: e economic promise of walkable places in metro-
politan Washington, DC. Washington, DC: Brookings Institution.
Leslie, E., Coee, N., Frank, L., Owen, N., Bauman, A., & Hugo, G. (2007). Walkability of local
communities: Using geographic information systems to objectively assess relevant environmental
attributes. Health & Place, 13(1), 111–22. doi.org/10.1016/j.healthplace.2005.11.001
Loukaitou-Sideris, A. (2006). Is it safe to walk? Neighborhood safety and security consider-
ations and their eects on walking. Journal of Planning Literature, 20(3), 219–232. doi.
org/10.1177/0885412205282770
Manaugh, K., & El-Geneidy, A. (2011). Validating walkability indices: How do dierent households
respond to the walkability of their neighborhood? Transportation Research Part D: Transport and En-
vironment, 16(4), 309–315. doi.org/10.1016/j.trd.2011.01.009
Owen, N., Humpel, N., Leslie, E., Bauman, A., & Sallis, J. F. (2004). Understanding environmental
inuences on walking; Review and research agenda. American Journal of Preventive Medicine, 27(1),
67–76. doi.org/10.1016/j.amepre.2004.03.006
Pain, R. (2001). Gender, race, age and fear in the city. Urban Studies, 38(5–6), 899–913. doi.
org/10.1080/00420980120046590
Park, A., & Calvert, T. (2008). A social agent pedestrian model. Computer Animation And Virtual Worlds,
19(August), 331–340. doi.org/10.1002/cav
Park, A., Clare, J., Spicer, V., Brantingham, P. L., Calvert, T., & Jenion, G. (2012). Examining con-
text-specic perceptions of risk: Exploring the utility of “human-in-the-loop” simulation models
for criminology. Journal of Experimental Criminology, 8(1), 29–47. doi.org/10.1007/s11292-011-
9132-x
Pelclová, J., Frömel, K., & Cuberek, R. (2013). Gender-specic associations between perceived neigh-
borhood walkability and meeting walking recommendations when walking for transport and recre-
ation for Czech inhabitants over 50 years of age. International Journal of Environmental Research and
Public Health, 11(1), 527–536. doi.org/10.3390/ijerph110100527
Saaty, T. L. (2008). Relative measurement and its generalization in decision making why pairwise com-
parisons are central in mathematics for the measurement of intangible factors the analytic hierarchy/
network process. Revista de La Real Academia de Ciencias Exactas, Fisicas y Naturales. Serie A.
Matematicas, 102(2), 251–318. doi.org/10.1007/BF03191825
Salzman, Z. (2017). Walkscore-client-js [Computer software]. https://github.com/zivzs/walkscore-cli-
ent-js
San Francisco Fire Department. (n.d.). 5.01 Street widths for emergency access. Fire Department. Re-
trieved from http://sf-re.org/501-street-widths-emergency-access
San Francisco Planning Department. (2010). Ch. 4, Approach: Designing great streetscapes. San Francisco
better streets. Policies and guidelines for the pedestrian realm. San Francisco. Retrieved from http://
www.sf-planning.org/ftp/BetterStreets/docs/Draft_BSP_for_Adoption_4_Designing_Great_
Streetscapes.pdf

526 JOURNAL OF TRANSPORT AND LAND USE 12.1
San Francisco Planning Department. (2011). Walk rst. Where people walk: Methodology for determining
Pedestrian activity Factors. San Francisco: SF Planning Department.
SFMTA. (2014). San Francisco Municipal Transportation Agency (SFMTA) travel decisions survey 2014.
San Francisco: SFMTA.
Smith, J. (2016, November 2). Explosion of complaints of feces and syringes across San Francisco
streets. Daily Mail Online. Retrieved from http://www.dailymail.co.uk/news/article-3897480/Ex-
plosion-complaints-feces-syringes-San-Francisco-streets-city-s-homeless-epidemic-grows.html
Sreetheran, M., & Van Den Bosch, C. C. K. (2014). A socio-ecological exploration of fear of crime in
urban green spaces—A systematic review. Urban Forestry and Urban Greening. doi.org/10.1016/j.
ufug.2013.11.006
Sugiyama, T., Salmon, J., Dunstan, D. W., Bauman, A. E., & Owen, N. (2007). Neighborhood walk-
ability and TV viewing time among Australian adults. American Journal of Preventive Medicine,
33(6), 444–449. doi.org/10.1016/j.amepre.2007.07.035
Talen, E., & Koschinsky, J. (2013). e walkable neighborhood: A literature review. International Jour-
nal of Sustainable Land Use And Urban Planning, 1(1), 42–63.
Talen, E., & Koschinsky, J. (2014). Compact, walkable, diverse neighborhoods: Assessing eects on
residents. Housing Policy Debate, 24(4), 717–750. http://doi.org/10.1080/10511482.2014.900102
U.S. Department of Transportation. (2011). 2009 National Household Travel Survey. Retrieved from
http://nhts.ornl.gov/2009/pub/stt.pdf
Vale, D. S., Saraiva, M., & Pereira, M. (2015). Active accessibility: A review of operational measures of
walking and cycling accessibility. Journal of Transport and Land Use, 9(1), 209–235. doi.org/10.5198/
jtlu.2015.593
Valentine, G. (1989). e geography of women’ s fear. Area, 21(4), 385–390.
Van Dyck, D., Cerin, E., Conway, T. L., De Bourdeaudhuij, I., Owen, N., Kerr, J., … Sallis, J. F. (2013).
Perceived neighborhood environmental attributes associated with adults’ leisure-time physical activ-
ity: Findings from Belgium, Australia and the USA. Health & Place, 19, 59–68. doi.org/10.1016/j.
healthplace.2012.09.017
Van Dyck, D., Cerin, E., De Bourdeaudhuij, I., Salvo, D., Christiansen, L. B., Macfarlane, D., … Sal-
lis, J. F. (2015). Moderating eects of age, gender and education on the associations of perceived
neighborhood environment attributes with accelerometer-based physical activity: e IPEN adult
study. Health and Place, 36, 65–73. doi.org/10.1016/j.healthplace.2015.09.007
Walk Score. (n.d.). Walk Score methodology. Retrieved from http://www.walkscore.com/methodology.
shtml
Walk Score. (2016). Most walkable cities in the United States, Canada, and Australia on Walk Score.
Retrieved from https://www.walkscore.com/cities-and-neighborhoods/
Weinstein Agrawal, A., Schlossberg, M., & Irvin, K. (2008). How far, by which route and
why? A spatial analysis of pedestrian preference. Journal of Urban Design, 13(1), 81–98. doi.
org/10.1080/13574800701804074
Zahran, S., Brody, S. D., Maghelal, P., Prelog, A., & Lacy, M. (2008). Cycling and walking: Explain-
ing the spatial distribution of healthy modes of transportation in the United States. Transportation
Research Part D: Transport and Environment, 13(7), 462–470. doi.org/10.1016/j.trd.2008.08.001