In what is one of the first visual preference surveys using Google Street View through a free tool StreetSeen (http://streetseen.osu.edu), adult students viewed a series of paired slides of image of city streets. Participants were asked to choose which image from the pair they preferred based on which street they would prefer to ride a bicycle. Subsequent analyses showed that differences in continent of the respondent impact preferences. This research demonstrates the extent to which certain segment-level factors such as presence of trees along the street, width of the street, presence of sidewalks, and other features are preferred using discrete choice models. The models reveal that increasing vehicle traffic, number of lanes, streetscapes with dense trees, and presence of parking lots decrease the probability of being chosen. Having sidewalks, presence of pedestrians, trees set back from the street, and traffic calming devices are positively associated with respondents’ preferences. The results related to trees may relate to perceptions of safety. For example, dense trees close to a street may limit visibility along a roadway. The models also reveal significant differences in preferences based on respondents’ locations. We conclude that this method is effective in capturing information about bicycling preferences. The survey methodology and analysis techniques introduced in this study can help city planners design streets that are preferred by bicyclists.
StreetSeen: Factors Influencing the Desirability of a Street for Bicycling
1. StreetSeen:
Factors Influencing the Desirability of a Street for Bicycling
Jennifer Evans-Cowley & Gulsah Akar, City and Regional Planning, The Ohio State University
TRB 93rd Annual Meeting
January 12-16, 2014
Washington, D.C.
2. Introduction
AIM: understand the street characteristics
that are most important to support cycling
Bicyclists face various choices of links to
travel from their origins to destinations.
Street characteristics contribute to
individuals’ bicycling choices
Understanding street characteristics can
lead to street design that is preferred by
bicyclists.
3. Methods
Used Free Tool:
http://streetseen.osu.edu
Anyone can use to create, collect data, and analyze a
pairwise visual survey using geo-tagged images from
Google Street View
Images from Columbus, Ohio,
metropolitan area.
Images were categorized based on
specific segment-level attributes.
5. Variables of Interest
Traffic on street
(including parked and
moving vehicles)
Parking
Roadway surface
condition
Roadway surface
material
Roadway grade
Presence of pedestrians
Presence of bicyclists
Land use
Streetscape
Number of lanes
Presence of bicycle lane
Sidewalk
Presence of traffic
calming devices
6. Respondents
Students enrolled and active in TechniCity (a
massive open online course) were invited to
participate in the StreetSeen survey.
7. Image Preferences
Images scored based on the fraction of times
that they were selected over other images,
correcting by the “win” and “loss” ratios of all
images with which they were compared.
10. Choice Models
Choice models are estimated to analyze
the effect of each street feature on
individuals bicycling choice.
As each observation is the choice
between two images, binary logit models
are estimated taking into account the
characteristics of both chosen and not
chosen images.
14. Model with Region Specific
Interactions (*)
*Selected results. As compared to N. America as the base case.
15. Conclusions
The models reveal that increasing vehicle traffic,
number of lanes, streetscapes with dense trees,
and presence of parking lots decrease the
probability of being chosen.
Having sidewalks, presence of pedestrians, trees
set back from the street, and traffic calming
devices are positively associated with
respondents’ preferences.
The models also reveal significant differences in
preferences based on respondents’ locations.
16. Contributions
This work provides a mechanism to understand the tradeoffs
among various attributes in a clean, quantitative framework.
The survey methodology and analysis techniques introduced in
this study can help city planners design streets that are preferred
by bicyclists.
17. Future Work
Including other segment-level factors.
Including questions regarding respondent
specific factors which are known to affect cycling
decisions (for instance being a beginner,
intermediate or expert cyclist, frequency of
biking, etc.)
Aiming larger samples from different locations
to provide a more robust study.
Testing preferences for walking along a street.
20. Variables of Interest
Traffic on street (including
parked and moving
vehicles)
parking lot
Roadway surface condition
excellent
none
good
1-2 vehicles visible
poor
3-5 vehicles visible
Roadway surface material
6-9 vehicles visible
asphalt
10+ vehicles visible
concrete
Parking
no on-street parking
parallel parking on one side
parallel parking on both sides
pull-in parking
brick
21. Variables of Interest
Roadway grade
flat and straight
hilly and straight (where a grade
change is clearly visible)
flat and curved (where a curve in
the roadway is clearly visible)
Presence of pedestrians
Presence of bicyclists
Land use
vacant/not visible (no structures
are visible from the street view)
widely spaced apart)
suburban residential (homes have
a 25 foot are larger setback)
suburban commercial (strip
commercial)
in-town residential (single family
homes that are close together)
medium density residential
(apartments and townhomes)
medium density
commercial/industrial (businesses
are located close together)
manufactured home park
mixed use (a mix of uses are
visible)
rural residential (homes are
widely spaced apart)
high density (high rise buildings)
rural commercial (businesses are
22. Variables of Interest
Streetscape
no trees
one way
Presence of bicycle lane
street trees
Sidewalk
trees set back from roadway
none
dense trees
one side
Number of lanes
Special Road type
alley
narrow two way (a narrow
roadway, without markings,
typically in a residential area
that is intended for two way
traffic)
both sides
Presence of traffic calming
devices
School crosswalk, textured
crosswalk, traffic circle, speed
humps
Increasingly cities are promoting bicycling for both recreation and daily transport. Cities have pre-existing street networks that may or may not be able to accommodate additional bicycling infrastructure. Cities are heterogeneous and vary in the suitability of roadways for the purposes of bicycling. Bicyclists face various choices of links to travel from their origins to destinations. Cities may offer different combinations of bicycle infrastructure, such as dedicated multiuse paths, bicycle boulevards, roads with sharrows, and bicycle lanes combined with routes where there is no bicycle infrastructure. For instance, the shortest path could require principally traveling on a high-traffic and high-speed-limit road that has on-street parking. The longest might be a multiuse trail. One other choice may include a blend of primarily residential streets and a multiuse path. It is important to understand the effects of street characteristics that contribute to individuals’ bicycling choices in order to make informed investment decisions and design streets that are preferred by bicyclists.
There are a variety of methods for measuring attributes in the built environment, such as visual surveys. Visual surveys ask people to rate images on a scale or choose an image over some other paired images (29-42). These visual surveys are intended to capture uniqueness in the built environment.In the past building visual surveys was time consuming and difficult. Over the last few years Google, Nokia, and other companies have undertaken extensive efforts to collect panoramic imagery of streets. This is obtained through multiple directional cameras at a consistent height of approximately 8.2 feet. GPS units capture the positioning (1). second bullet -- Images were immediately eliminated from the study if the conditions were unfavorable, such as a view in the rain or a fuzzy image. third bullet -- For example, a no-traffic condition in a rural, suburban and urban context.
59 images in the pairwise survey
A total of 260 people whose latitude and longitude could be detected participated, contributing 15,759 votes. Each participant contributed an average of 60 votes. After all votes were collected each latitude and longitude was coded to determine the country and continent of each vote using latlong.net. Table 1 shows the distribution of votes across regions. -- Talk about technicity class.
Top images respondents chose are residential streets with trees along them and a few parked cars.
The images respondents were less likely to choose were the ones with five or more lanes and significant traffic visible
The calculated Q scores give information on the most and least desirable streets on which to bicycle, however, they do not reveal information on the effects of each single street feature. For instance, is it the number of lanes that make one street desirable, or the presence of traffic calming devices? Although the answer may be a combination of both, which features have the highest impact and how do these impacts vary across different survey regions? To be able to answer such questions, discrete choice models are estimated.
Having pedestrians generally increases the probability of choosing an image The effect is less for respondents from Asia and South America. Having sidewalks on one side of the road will increase the choice probability more in Asia. An image with dense trees is less likely to be chosen, and this affects the respondents from Europe even more. the effect is the opposite for respondents from Africa. Respondents from Africa have slight preference for having dense trees. Being close to a parking lot decreases the probability of being chosen, and this effect is significantly higher for respondents from Africa, Asia, and Europe as compared to North America. Having a street with a curve increases the choice probability for North American respondents, and even more for respondents from Australia, while a street with a curve is less likely to be chosen by respondents from Africa.
This research demonstrates that pairwise surveys can be effective for understanding preferences for bicycling.
It provides a virtually limitless number of images based on a limitless set of attribute data that can be collected from snapshot streetview images. This provides an innovative contribution that simplifies the process for researchers.
Including other segment-level factors.Testing preferences for walking along a street. Including questions regarding respondent specific factors which are known to affect cycling decisions (for instance being a beginner, intermediate or expert cyclist, frequency of biking, etc.) may enhance future studies by revealing how effects of certain street features vary across individuals with different characteristics. Studies could be undertaken with homogeneous samples with equal familiarity with the kinds of situations represented to understand how people who are familiar respond. Additionally, using multiple evaluators of each image to rate each attribute prior to deploying the survey would address the reliability and validity of the categorization of each image’s attributes. Future studies could include larger samples of people from different locations to provide a more robust study.Future studies could also integrate demographic questions and individual perceptions to better understand the respondents including such questions may also help identify the underlying reasons for differences across continents and possibly countries; for instance the preferences of respondents from a country where the bicycling mode share is high and the culture is well-established versus a country where bicycling is perceived as dangerous and not very common. To be able to test the differences across countries, more data will be required to achieve substantial samples from each location.