Transport Reviews

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/ttrv20

Crowdsourced data for bicycling research and

practice

Trisalyn Nelson , Colin Ferster , Karen Laberee , Daniel Fuller & Meghan
Winters

To cite this article: Trisalyn Nelson , Colin Ferster , Karen Laberee , Daniel Fuller & Meghan
Winters (2020): Crowdsourced data for bicycling research and practice, Transport Reviews, DOI:
10.1080/01441647.2020.1806943

To link to this article: https://doi.org/10.1080/01441647.2020.1806943

© 2020 The Author(s). Published by Informa

UK Limited, trading as Taylor & Francis
Group

Published online: 15 Aug 2020.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=ttrv20
TRANSPORT REVIEWS
https://doi.org/10.1080/01441647.2020.1806943

Crowdsourced data for bicycling research and practice

Trisalyn Nelsona, Colin Fersterb, Karen Labereeb, Daniel Fullerc and Meghan Wintersd
a
School of Geographical Sciences and Urban Planning, Arizona State University, Tempe, AZ, USA;
b
Department of Geography, University of Victoria, Victoria, BC, Canada; cSchool of Human Kinetics and
Recreation, Memorial University, St. John’s, NL, Canada; dFaculty of Health Sciences, Simon Fraser University,
Burnaby, BC, Canada

ABSTRACT ARTICLE HISTORY

Cities are promoting bicycling for transportation as an antidote to Received 11 December 2019
increased traffic congestion, obesity and related health issues, and Accepted 31 July 2020
air pollution. However, both research and practice have been
KEYWORDS
stalled by lack of data on bicycling volumes, safety, infrastructure, Crowdsourced; bicycling;
and public attitudes. New technologies such as GPS-enabled exposure; safety;
smartphones, crowdsourcing tools, and social media are changing infrastructure; attitudes
the potential sources for bicycling data. However, many of the
developments are coming from data science and it can be
difficult evaluate the strengths and limitations of crowdsourced
data. In this narrative review we provide an overview and critique
of crowdsourced data that are being used to fill gaps and
advance bicycling behaviour and safety knowledge. We assess
crowdsourced data used to map ridership (fitness, bike share, and
GPS/accelerometer data), assess safety (web-map tools), map
infrastructure (OpenStreetMap), and track attitudes (social media).
For each category of data, we discuss the challenges and
opportunities they offer for researchers and practitioners. Fitness
app data can be used to model spatial variation in bicycling
ridership volumes, and GPS/accelerometer data offer new
potential to characterise route choice and origin-destination of
bicycling trips; however, working with these data requires a high
level of training in data science. New sources of safety and near
miss data can be used to address underreporting and increase
predictive capacity but require grassroots promotion and are
often best used when combined with official reports.
Crowdsourced bicycling infrastructure data can be timely and
facilitate comparisons across multiple cities; however, such data
must be assessed for consistency in route type labels. Using social
media, it is possible to track reactions to bicycle policy and
infrastructure changes, yet linking attitudes expressed on social
media platforms with broader populations is a challenge. New
data present opportunities for improving our understanding of
bicycling and supporting decision making towards transportation
options that are healthy and safe for all. However, there are
challenges, such as who has data access and how data
crowdsourced tools are funded, protection of individual privacy,
representativeness of data and impact of biased data on equity in

CONTACT Trisalyn Nelson trisalyn.nelson@asu.edu

© 2020 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group
This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivatives License
(http://creativecommons.org/licenses/by-nc-nd/4.0/), which permits non-commercial re-use, distribution, and reproduction in any
medium, provided the original work is properly cited, and is not altered, transformed, or built upon in any way.
2 T. NELSON ET AL.

decision making, and stakeholder capacity to use data given the

requirement for advanced data science skills. If cities are to
benefit from these new data, methodological developments and
tools and training for end-users will need to track with the
momentum of crowdsourced data.

Introduction
Around the world, cities are increasingly prioritising bicycling. Bicycling for transportation
is being hailed as a solution to transportation congestion and air pollution, and active
travel provides health benefits related to physical activity (Fishman, Schepers, & Kamphuis,
2015). As a result, cities are setting goals to increase mode share of bicycling and are
investing in policies and infrastructure to increase bicycling ridership. Decisions about
which policies to implement and where to prioritise investment have created an unprece-
dented need for data and information on bicycling in cities.
The lack of data on many aspects of bicycling have limited our ability to research and
implement pro-bicycling policy (DiGioia, Watkins, Xu, Rodgers, & Guensler, 2017). Although
cities invest heavily in tracking vehicular volumes, safety, and infrastructure, the budgets
for collection of bicycling data are limited. For example, bicycling volume data are usually
only available for a handful of locations and limited time periods (Roy, Nelson, Fothering-
ham, & Winters, 2019). Bicycling safety incidents are underreported, on the order of >85%
underreported in some cities (Winters & Branion-Calles, 2017). Given that changes to bicy-
cling infrastructure are often incremental, they can be difficult to track.
Gaps left by traditional sources of bicycling data are being filled with crowdsourcing.
Crowdsourcing is a general term that means users generate data (Xu & Nyerges, 2017);
within this the term “citizen science” is often used to encompass activities that broaden
participation in research and practice (Eitzel et al., 2017). In the case of bicycling, this
means that all road users can contribute data on bicycling levels, safety, and conditions.
Crowdsourcing and citizen science are being used, with growing popularity, to generate
data on everything from bike safety incidents to infrastructure. As well, people are
sharing their perceptions of bicycling, infrastructure projects, and response to policies
through social media, which creates an archive of data of interest for bicycling research
and practice.
Our goal is to overview and critique crowdsourced data that are being used to fill gaps
and advance bicycling research around bicycling volumes, safety, infrastructure, and
public attitudes. As a collaboration between data scientists and transportation researchers,
we aim to explore the strengths and weaknesses of emerging crowdsourced data as they
apply to bicycling research. We build on a previous review of bicycling and big data that
focused on ridership data (Romanillos, Austwick, Ettema, & De Kruijf, 2016). Nearly five
years on in a rapidly changing environment, we provide updates on ridership data and
further introduce new domains of crowdsourced data for bicycling (Figure 1). We focus
on data issues; however, we also touch on methods, given that issues of data and
methods are inevitably intertwined. This is not an attempt to provide a systematic
review of all literature; rather we aim to highlight dominant trends. We set the stage for
discussion of each data type by reviewing the current state of more traditional data
 TRANSPORT REVIEWS 3

Figure 1. An overview of crowdsourced data for bicycling topics covered in this review and their jux-
taposition with traditional datasets.

collection. Finally, we conclude with recommendations for advancing the use of novel data
in bicycling research and planning practice.

Ridership
Bicycling volume data are needed for planning bicycling infrastructure, monitoring
changes in ridership, characterising network connectivity, and as exposure data for
safety and health studies (Vanparijs, Panis, Meeusen, & de Geus, 2015). Yet collecting bicy-
cling volumes at the street level has been limited by the availability of tools for data col-
lection. Typically, data are collected using temporary counts at numerous locations (i.e.
pneumatic tube counters or video), permanent counts at a few locations (i.e. automated
in-ground counters), or periodically through count programmes facilitated by volunteers.
Count programmes provide critical and complete data on bicycling volumes at a particular
location and time; however, they lack the spatial and temporal detail that is often desirable
to meet bicycling information needs (Roy et al., 2019).
GPS-enabled smartphones and wearables are transforming how data on movement,
including bicycling, are collected and in Table 1 we provide details of some key papers
that use crowdsourcing tools to collect data and analyse or map ridership. Fitness apps
like Strava, MapMyRide, and Garmin Connect are used by many bicyclists to track and
monitor their bicycling activity. As a result, massive global databases on bicycling ridership
now exist. Strava data, for example, provide the number of bicyclists using Strava on every
road segment with 15-minute resolution; thus, Strava data are continuous across space
and through time (Sun & Mobasheri, 2017). Of course, Strava bicyclists are a sample of
all bicyclists, and their proportion relative to all bicyclists varies across infrastructure,
road type, and neighbourhood characteristics (Griffin & Jiao, 2015). However, in dense
urban cores, the patterns of Strava ridership are a good proxy for overall bicycling
(Jestico, Nelson, & Winters, 2016), suggesting that despite demographic bias (Strava
riders are disproportionately middle-aged and men) the spatial patterns may be
 4
Table 1. Literature on crowdsourced data for bicycling ridership.
Type of
crowdsourced Select challenges that arose with

T. NELSON ET AL.
data Reference Research paper goal Opportunities of using crowdsourced data crowdsourced data
Strava Roy et al. (2019) Develop an approach to statistically correcting bias in Strava was used to predict bicycling ridership within Method was sufficient to predict categorical
Strava data. ±100 average annual bicyclists for 86% of road cycling levels, not exact volumes.
segments.
Strava Hochmair, Identify built environment and sociodemographic High resolution of crowdsourced data, combined with Using patterns in big data it is not possible to
Bardin, and factors associated with bicycling ridership. a technique for bias made it possible to characterise determine causality and further
Ahmouda relationships between ridership and other factors experimental research and data are
(2019) across an entire network. needed.
Strava (+OSM) Orellano and Explore the effect spatial configuration of street Strava data enabled evaluation of a bicycling network. Sampling bias in Strava data was not
Guerrero networks on movement patterns using Strava Metro accounted for.
(2019) and OpenStreetMap.
Strava Boss et al. (2018) Detect changes in bicycling volumes after Installation of bicycling infrastructure changes Difficult to bias correct data at a very high
(+Infrastructure) infrastructure changes. volumes of bicyclists at multiple locations in a city. temporal resolution, due to limited data
for training that has the same temporal
resolution.
Smartphone app Garber et al. Quantify race and gender bias in users of smartphone Presents an approach to using app data in a more Demonstrates bias in standard app
(2019) apps to record bicycling ridership. representative way. generated data.
Smartphone app Pritchard et al. Study route and mode choice of study participants Using an app it is possible to collect complete routes Small samples associated with recruitment
(2019) before and after new cycling infrastructure using a of individual bicyclists with high accuracy. limited representativeness of sample
passive smartphone app (Moves).
Smartphone app Blanc and Crowdsourced data collected by an app (developed by Using an app it is possible to collect observed route Possible overrepresentation by year-round
Figliozzi (2017) state DOT) to model cyclists’ comfort. information and quantify impacts on comfort levels bicycle commuters. Transferability to other
of trip purpose and sources of stress along a route. areas and cities is unknown.
GPS/ Brondeel et al. Construct an algorithm to reliably predict Using machine learning, GPS/accelerometry made it Mode detection remains challenging for
Accelerometry (2015) transportation mode at trip level with data collected possible to correctly predict transportation mode in bicycling and method inappropriate for
from GPS/accelerometers. 90% of trips. multi-modal trips.
Bike share Winters, Hosford, Characterise the super-users of a bikeshare to System data with userID enabled the identification of Demographic data not included in bike share
and Javahera understand who is capitalising to understand the “super-users” – 10% of users who made 50% of bike ridership data; analyses relies on linkage to
(2019) equity implications of a public programme. share trips. From user survey with demographic user survey data.
data, determined super-users were lower income
with fewer transportation options than other
members.
Bike share Scott and Ciuro Effect of weather conditions, hub attributes, temporal Because data are collected continuously, there is a full High resolution data was buffered which
(2019) variables on ridership. year of data across space and time to compare to creates issues with modifiable areal units
weather and temporal variables. and scale.
Bike share (+GPS) Wergin and Examine the routes taken by bike share users to By using the on-bike GPS and the bike share system Accuracy of the GPS pings may be limited in
Buehler (2017) identify geographic areas and individual road data, researchers were able to identify differences in urban areas; spatial granularity lacking to
segments used; correlate routes with infrastructure trips by membership type discern sidewalk from road riding.
and highlight gaps; and describe stops.
 TRANSPORT REVIEWS 5

representative. A unique benefit of Strava data is that because it is continuously collected

and archived, it can be drawn on for retrospective pre–post studies. For example, if an
unexpected controversy arises over new infrastructure, Strava provides a unique ability
to evaluate usage before and after (Boss, Nelson, Winters, & Ferster, 2018).
Beyond fitness apps, researchers use GPS and/or accelerometry data to track bicycling
trips, often to supplement or replace more travel diary tools (Kerr, Duncan, & Schipperijn,
2011). These studies invite participants to use research- and practice-oriented apps (e.g.
Ride Report, Cycle Atlanta) or provide participants with GPS devices for a study period.
The resulting datasets comprise millions of second-by-second data on individual travel
behaviours. One of the advantages of using researcher apps is that the individual tracks
of bicyclists can be mapped. Strava, for example, provides the number of bicyclists per
road segment, rather than tracks, to protect the privacy of individual bicyclists who
could easily be identified by frequent origin and destination locations. Yet with accelero-
metry activity, it remains difficult to identify bicycling activities (Broach, Dill, & McNeil,
2019). As well, use of GPS and accelerometry data require recruitment of participants to
download and use an app of study-specific technology. Even recruiting a few hundred par-
ticipants can require huge effort and recruiting a representative sample of participants can
be extremely difficult.
Bike share is another source of crowdsourced data on ridership. An excellent source for
bike share data is Bike Share Research (https://bikeshare-research.org/), a collaborative,
open data platform linked to system data from hundreds of bike share programmes glob-
ally and allowing API accessibility. There are many papers on bike share, in part because it
was one of the first detailed datasets on ridership. Recent reviews provide insights from
bike share data (e.g. Hirsch et al., 2019), and there is potential for GPS movement data
from bike share bicycles (Romanillos, Moya-Gómez, Zaltz-Austwick, & Lamíquiz-Daudén,
2018). However, most systems provide only origin-destination data (not routes), and the
geographic coverage of systems is often limited to downtown cores. Select systems
provide gender and age categories for members, demographic information that is
rarely available in ridership data.
Fitness app and GPS and/or accelerometry datasets can be massive. For example, Strava
data from 2016 for the Phoenix region has ∼44 million records. While Strava has outstand-
ing spatial and temporal coverage and resolution, it is difficult to wrangle the raw data into
information products, as data are not stored in GIS-ready formats, and they include mul-
tiple measures of bicycling by direction travelled and differently for street segments and
intersections. Cleaning data directionality and connecting aspatial data to a road network
can be time consuming and require skills with handling large volumes of data because
there are many time periods of data included. Similarly, with GPS and/or
accelerometry data there are not yet reliable and transferable tools to detect trips/
dwells and identify the mode of travel. Bicycle trips may be the most challenging to ident-
ify because bicyclists can go as fast as slow cars or as slow as fast pedestrians (Broach et al.,
2019; Brondeel, Pannier, & Chaix, 2015). If these data sources are to be tapped widely for
research and practice applications, substantial effort is needed to develop and standardise
tools for data processing. At present, strong data science skills are needed to put the data
to their best use. For example, there are legitimate concerns about bias of the Strava
sample, which is disproportionally used by middle-aged, white men (Garber, Watkins, &
Kramer, 2019), that have led to development of many studies of how best to use Strava
6 T. NELSON ET AL.

(Mcnair & Arnold, 2016). As with many big data analyses, it is critical that when using Strava
data for research and practice applications that expert opinion, local knowledge, and
appropriate goals and metrics are also considered (Griffin, Mulhall, Simek, & Riggs,
2020). Additionally, methods to correct the bias in Strava data have been developed
(Roy et al., 2019) to map all ridership. Using official counts to train and test a model, as
well as geographic covariates, average annual daily bicycling can be predicted within
100 bicyclists for 86% of street segments. Limitations of the modelling include lower accu-
racies in low ridership areas and the precision of data makes the results best presented as
categorical ridership (low/medium/high). However, implementation of bias correction
requires advanced programming and statistical expertise. The uses of these data are
just beginning to be understood and patterns in data have potential to help stratify
count programmes (Brum-Bastos, Ferster, Nelson, & Winters, 2019) and monitor change
(Boss et al., 2018; Hong, McArthur, & Livingston, 2019). At present, cell phone GPS is not
accurate enough to pick up sidewalk riding (Wergin & Buehler, 2017) and using GPS/accel-
erometry data to differentiate multimodal trips is problematic (Brondeel et al., 2015). For
the full potential of fitness app and GPS and/or accelerometer data to be used for main-
stream planning new methods and tools are needed to easily ingest and work with data.

Safety
Bicycling incidents are underreported (Medury, Grembek, Loukaitou-Sideris, & Shafizadeh,
2019) and detailed reports of safety incidents are difficult for researchers to obtain. Based
on hospital reports, it has been shown that only 15% (Winters & Branion-Calles, 2017) to
30% (Teschke et al., 2014) of bicycling incidents, even those that require hospital attention,
are reported to official sources such as insurance or police. Of the bicycling incidents that
do get reported, most include vehicles, whereas single bicycle crashes or those that
involve infrastructure or other bikes are rarely reported (Branion-Calles, Nelson, &
Winters, 2017). Fatality data are collected more completely, but often lack georeferencing
in detail making it difficult to link fatality data with other spatial datasets. Finally, near miss
data are rarely collected, although they have been shown to cause some bicyclists to have
serious psychological barriers to continued bicycling (Sanders, 2015). Near miss data can
be used to address underreporting and increase predictive power of modelling (Poulos
et al., 2017). Even when data has been collected, accessing data can be difficult for
researchers due to limited access and licensing. A typical scenario is that the data are
stripped of incident descriptions to protect the privacy of the people involved. Alterna-
tively, the data may be spatially aggregated to intersections, to mid-block, or summarised
by larger areas. As a result, safety data used for research and planning are often incom-
plete and lack details needed to advance planning and knowledge.
Crowdsourcing has become a tool for filling gaps in safety data and key papers are out-
lined in Table 2. Using web-maps, bicyclists can self-report crashes and near misses, and
the data are stored as GIS or other geolocated sources. Examples are BikeMaps.org, a
global tool for mapping crashes and near misses (Nelson, Denouden, Jestico, Laberee, &
Winters, 2015) and Collideoscope (collideoscope.org.uk), a tool used in the UK. Related
tools include www.badintersections.com and social media campaigns such as #Near-
DeathTO to report near misses in Toronto, Canada. Using a combination of structured
and open-ended questions, these websites allow anonymous reports of bicycling
 TRANSPORT REVIEWS 7

incidents and reports include details of crash and near miss outcomes, such as the pres-
ence and severity of injuries (Nelson et al., 2015). From our team’s experience with Bike-
Maps.org data, we have observed that crowdsourced data on bicycling crashes
highlight locations where there are a lot of bicycles, whereas data that comes from
police or insurance tends to highlight areas with high amounts of vehicle traffic. Risks
associated with features like multi-use paths, regional trails, and micro barriers (e.g. rail-
road tracks and loose surfaces) are more likely to emerge in analyses of crowdsourced
data (Jestico, Nelson, Potter, & Winters, 2017). Other apps, similar to BikeMaps.org in func-
tionality, have been developed but tend to be more regionally specific, like ORcycle (Blanc
& Figliozzi, 2017). New technology that could be of interest for crowdsourcing safety data
include apps that use GPS data to track unsafe riding behaviour (Gu et al., 2017).
Crowdsourcing favours data collection from people with access to technology (Ferster,
Nelson, Winters, & Laberee, 2017). Therefore, these tools do not capture the experience
from some of the most vulnerable road users, such as older adults, children, and people
with lower incomes. The amount of data submitted is also tied to promotion of the tool
and requires consistent and ongoing promotion. As such, the data are not ideal for track-
ing change through time, but rather can represent spatial variation in bicycling safety and
can be used to understand safety barriers to bicycling. The longevity of crowdsourced
tools is also problematic, and many tools come and go due to lack of funding or energy
for ongoing promotion and maintenance. While many jurisdictions want access to
better safety data, investment in, and promotion of the data collection tool can be
difficult to fund. BikeMaps.org is the longest running project that we are aware of and
has been operating for more than five years. While the initial tools were developed with
industry support, operations have been federally supported and are funded by research
projects that use the data. Tools that are centrally supported and long-standing can fill
critical gaps in safety data and ideally be used in combination with other existing datasets.

Infrastructure
Cities small and large have increasing capacity to maintain spatial data files of their bicy-
cling infrastructure. With the Open Data movement, many communities are making their
bicycling infrastructure data publicly available. However, there are no standard naming
conventions for bicycle infrastructure. As an example, an investigation of open datasets
in 45 Canadian municipalities revealed ∼100 different terms used for bicycle infrastructure
(Winters, Zanotto, & Butler, 2020). There was also great variability in the timeliness of the
data, with some open datasets already up to five years old when accessed for the study.
These realities mean that compiling bicycle infrastructure data directly from cities is labor-
ious, requiring the assembly of data from disparate sources and reconciling the resolution
of differences in naming conventions, timeliness, and accuracy. As such, multi-city studies
that relate bicycle infrastructure to mode share and health outcomes are limited and have
rarely been repeated even with substantial investments across North America in recent
years (Pucher & Buehler, 2012).
OpenStreetMap (OSM – openstreetmap.org) is an emerging data source with the
potential to serve as a single source of infrastructure data with global coverage. OSM is
a crowdsourced map of the world that provides free spatial data for the natural and
built environment, including active transportation infrastructure. With data quality
8 T. NELSON ET AL.

Table 2. Literature on crowdsourced data for bicycling safety.

Type of Select challenges that
crowdsourced Opportunities of using arose with crowdsourced
data Reference Research paper goal crowdsourced data data
Safety – Nelson et al. Development and New technology enables Data collection requires
BikeMaps.org (2015) deployment of a tool for collection of self- continued promotion of
crowdsourced bicycling reported bicycling crash crowdsourced tools.
safety data. and near miss data.
Safety – Strauss, Estimate bicycle volumes Continuous spatial Exposure maps were not
Exposure & Miranda- (AADB) by combining coverage for ridership bias corrected.
Risk Mireno, and GPS (smartphone app) enabled exposure and
Morency with manual counts to risk to be mapped
(2015) map injury risk through across the entire
a network. network to be mapped.
Safety Saad, Abdel- A demonstration of the Continuous spatial Bias corrections benefit
(intersections) Aty, Lee, issues of how bicycle coverage for ridership from additional SES and
– Exposure and Cai crash modelling suffers enabled exposure and behavioural variables.
(2019) from a lack of exposure risk to be mapped
data and that exposure across the entire
data limitations can be network. Study data
overcome using statistically adjusted to
crowdsourced data. address demographic
bias.
Safety Aldred and UK Near Miss Project, a Using crowdsourcing it is Crowdsourcing impacts
Goodman crowdsource project of possible to collect representativeness of
(2018) self-reported near miss diverse individual data. More experienced
bicycling incidents, used narratives. bicycling and men aged
to characterise 30–49.
individual predictors
that influence scariness
of an incident.

enforced through community standards, OSM data are contributed by a wide range of
people and interests, including but not limited to hobbyists, professional mappers, and
hired mappers at companies that use OSM data for delivery services and navigation
apps (Anderson, Sarkar, & Palen, 2019). OSM has a rapidly growing user community and
completeness is quickly improving: OSM launched in 2004 and by 2019 ∼5.7 million con-
tributors have created a database with ∼600 million features. Recent investigations have
found that globally more than 80% of roads are complete on OSM; upwards of 95% in
North American cities (Barrington-Leigh & Millard-Ball, 2017).
Applications of OSM data in research and practice are extensive and diverse. As OSM
data are provided for free use, they underlie many other applications; for example,
Strava Metro uses OSM data to aggregate activities across the street and trail network.
In research applications, OSM provides a single global data source spanning across city,
state, and national boundaries. At national scales, OSM may be used to measure bicycling
infrastructure in cities across Canada (Ferster, Fisher, Manaugh, Nelson, & Winters,
2019) (Table 3). At global scales, OSM can provide insight into urbanisation around the
world (Barrington-Leigh & Millard-Ball, 2017). Researchers are already tapping into OSM
as a core input for developing walkability metrics at the national scale in Canada
(Hermann et al., 2019). In terms of bicycle infrastructure, a study of six Canadian cities
found the length of bicycle infrastructure mapped in OSM was similar to cities’ open
data sources in some cases (<±2%), in others differed more (± 30%) (Ferster et al.,
2019). An investigation into the differences indicated that cities’ data often were not
up-to-date, and that more informal paths may also be mapped in OSM. Further, the
 TRANSPORT REVIEWS 9

Table 3. Literature on crowdsourced data for characterising bicycling infrastructure.

Type of
crowdsourced Opportunities of using Select challenges that arose
data Reference Research paper goal crowdsourced data with crowdsourced data
Infrastructure – Ferster et al. Compare the OSM provides a single data Inconsistent labelling and
OSM (2019) representation of bike source for all cities. definitions in both OSM
facilities on OSM with Agreement with city data and open data.
open data provided by was high.
six cities in Canada.
Infrastructure – Hochmair Completeness of bicycle OSM was more complete To obtain the most
OSM et al. trails and lanes in two for bicycle trails. complete representation
(2019) American cities on OSM of bicycle facilities OSM
and Google Maps. was ideally supplemented
with other data sources
Infrastructure – Collins and Develop a predictive By using a data network it OSM lacked information on
open data Graham model for cycling is possible to show bus parking and loading
(2019) collisions using road lanes, 30 mph speed facilities. Open data also a
infrastructure variables. limits, junction density snapshot at one time
and multi-lane roads point whereas collisions
affected the collision occurred over many years.
counts.

OSM tags (open-ended and flexible labels assigned to features by citizen-mappers) were
queried to differentiate types of bicycle infrastructure (cycle tracks, on-street bicycle lanes,
paths (bicycle-only or multi-use), and local street bikeways). Across cities, paths (bicycle-
only or multi-use) and painted bike lanes were the most common types of facility, and
importantly, these were also the most well-mapped and easiest to query in OSM. Fre-
quently the OSM data were more up-to-date than cities’ data, as new infrastructure is
often mapped on OSM before it is released on open data from cities. Further, OSM pro-
vides an edit history for every feature, which allows measurement of change in the built
environment (Zhang & Pfoser, 2019), though few researchers have used this feature.
In crowdsourced data such as OSM, consistency can be a challenge. Similar to open
data provided by cities, citizen-mappers may employ diverse labelling practices, with vari-
ation within and across cities. While it is possible to develop queries on the data, this
requires cross-checking in Google Street View, and repeated training of queries for
them to be valid across cities. It will be important to standardise tagging methods for
different infrastructure types, especially for types of bicycling infrastructure that may be
less common. Multiple tools are available to extract and process OSM data within standard
GIS software. Methods of accessing OSM data include downloading OSM world files, acces-
sing the data using application programming interfaces (APIs), or using third party data
products. Downloading world files (e.g. https://planet.osm.org/) provides the most
timely and direct route of obtaining data, but requires a relatively large global download
and importing the data into traditional GIS applications may be a barrier to some research-
ers. APIs (e.g. http://overpass-api.de/) provide access to spatial and thematic subsets of the
data, but these APIs rely on servers that can be overwhelmed by large or frequent
requests, there may be technical barriers for using the APIs with traditional GIS software
is plugins. Third parties offer data products that deliver spatial subsets of the data in
formats that are easily imported into traditional GIS applications (e.g. http://download.
geofabrik.de/); however, third party processing steps influence the data format (e.g. the
data structure of tags). Further, this is a dynamic space, with companies and products
always shifting. With respect to data extraction and processing, publishing detailed
10 T. NELSON ET AL.

extraction and processing steps (including queries) would increase clarity and repeatabil-
ity. On a promising note, it is expected that OSM data will become increasingly complete
(Barrington-Leigh & Millard-Ball, 2017), given its wide acceptance and use in a wide range
of applications (including commercial, research, and more) and new tools and pro-
grammes to increase the ease of contribution.

Attitudes
Understanding public attitudes about bicycling is critical for understanding how and when
to implement bicycling policies and investments. Even when there is evidence that infra-
structure is a good decision for mobility and health reasons, projects cannot be success-
fully implemented without support from public and elected officials. Often survey data
are the predominant source of data on attitudes toward bicycling (Handy, van Wee, &
Kroesen, 2014). Population-wide samples may provide the most representative data
including those who bicycle and those who do not; however, these are expensive and
increasingly challenging as landlines become increasingly obsolete. Convenience
samples, recruited through word-of-mouth or a listserv, are ultimately biased towards
those most connected to the issues. Field intercept surveys will capture the users (and
the most regular of users) of the infrastructure but brings few insights on those who
are not current users. Generally, cities and researchers must invest on a project-by-
project basis or city-specific efforts, as there are few questions related to attitudes
towards national or regional travel surveys in the North American context. Further, such
survey efforts happen only every few years. Including survey questions in household
travel surveys or other national surveys on perceptions and attitudes is one way to
ensure that representative samples can be collected with sufficient frequency to under-
stand changes in attitudes, although endeavours such as the National Highway Traffic
Safety Administration Attitudes and Behavior Survey are few and far between (Federal
Highway Administration, 2017). Over the past decade, there have been increasing chal-
lenges to reach representative samples via traditional telephone phone surveys, with
researchers and local governments turning to online panels, either those compiled by
market research firms, or panels maintained by the organisations themselves (Dillman,
Smyth, & Christian, 2014).
Social media research methods can provide timely data about attitudes towards active
transportation programmes (Table 4). Social media functions as an extension of public,
urban spaces (Brighenti, 2012), and social media research tools can measure communi-
cation in these digital-urban spaces. Common social media research methods include
natural language processing (NLP; which uses word frequencies and co-occurrence of
words in phrases to quantify meaning), sentiment analysis (where dictionaries are used
to assign positive or negative sentiments to words and phrases), topic modelling
(where topics are classified based on the occurrence of words), and social network
graphs (where connections on social networks are analysed). Social media has been
used to evaluate transportation attitudes. For example, Schweitzer (2014) found that the
sentiments expressed towards public transportation agencies and patrons on Twitter
were more negative than sentiments for other types of public services. However, public
transit agencies that interacted with Twitter users (by answering questions and respond-
ing to statements rather than disseminating information from the top-down) had more
 TRANSPORT REVIEWS 11

Table 4. Literature on crowdsourced data for characterising attitudes towards bicycling.

Type of Select challenges that
crowdsourced Opportunities of using the arose with crowdsourced
data Reference Research paper goal crowdsourced data. data.
Media coverage in Ralph, Examine how local Using social media it was Media typically gathers
crash reporting Iacobucci, news outlets report possible to show that information about
Thigpen, car crashes local news coverage crashes from police – not
and involving subtly but consistently planners, engineers or
Goddard pedestrians and blames VRUs for crashes. other experts.
(2019) bicyclists.
Social media data Hollander and Using a case study By comparing a random Twitter data only collected
& urban Shen (2017) evaluates how sample of tweets to for 5 weeks, and difficult
attitudes to bicycling facilities geolocated bike-focused to obtain retrospectively.
bicycling and city plans tweets, it was possible to
influence people’s relate attitudes to
attitudes. infrastructure.
Twitter attitudes Ferster et al. Evaluate attitudes on Attitudes on Twitter Twitter is the most
to new bike (2020) Twitter toward new changed from bike accessible platform to
infrastructure bicycling facilities advocacy before the lanes researchers, but it is only
in two Canadian opened, to controversy part of the discussions
cities. after the lanes opened, and messaging on social
towards broader media.
acceptance over time.

positive sentiments that might impact how voters and stakeholders think about future
investments. To understand user experiences for a bike share programme in Washington
D.C., Das, Sun, and Dutta (2015) used Twitter data to measure sentiments, which extend
beyond usual measures of service (e.g. number of trips) and can be obtained from a
large audience more rapidly than traditional surveys. Ferster et al. (2020) evaluated discus-
sion about new protected bike lanes in Edmonton and Victoria, Canada, and found that
opposition to the projects was framed around weather in Edmonton and impacts to
small business in Victoria. The discussion on Twitter changed over time, where Twitter
was first used by bicycling advocates to promote new bike infrastructure, engaged
larger audiences in sharing news articles and general discussion once the lanes were
opened, and finally indicated greater general acceptance of the new infrastructure.
Social media research methods tools can rapidly measure communication from large
audiences on social networks, but face challenges related to data access and sample repre-
sentativeness. Key considerations for social media research are data access and data repre-
sentiveness. In part due to access limitations, most social media research papers in active
transportation have used Twitter because the Twitter Search application programming
interface (API) and Twitter Streaming API provide data in a format that is accessible to
researchers for free. The data provided by these tools are a subset and include biases
due to internal filters, such as overrepresenting central accounts (accounts that are
widely retweeted or mentioned) compared to less central accounts that represent a
much larger mass of people (González-Bailóna, Wang, Riveroc, Borge-Holthoefer, &
Morenoc, 2014). Continued access to the data depends on platform policies, changes to
filters, and monetisation of data access. Active transportation research using other social
media platforms have potential for increasing understanding of these socio-technical
systems, yet research access to other platforms can be challenging. Another critical issue
is representativeness. Compared to the general population of the United States, Twitter
users are younger, more educated, and more often located in urban centres (Duggan,
12 T. NELSON ET AL.

2015). However, this younger demographic is often missing from traditional feedback
mechanisms, such as townhall meetings (Einstein, Palmer, & Glick, 2019), so social media
research may help provide voices in planning processes for some underrepresented
groups. As well, some neighbourhoods, topics, and bicycling projects are underrepre-
sented, as social media does not reflect any of these aspects of urban life equally. Research
is needed to understand the spatial and thematic completeness of social media discussions
at a neighbourhood scale. Failure to reflect representative voices, through use of social
media or other crowdsourced data, can lead to decision making that is biased towards
those who use technology, which are typically already privileged groups.

Discussion
Transportation and planning research and practice need to keep pace with the backdrop
of smart cities, big data, and burgeoning technology. Trends in bicycling data are an
example of what is happening with city data in general; sources, variety, and volume of
data are exploding, offering huge potential to help us better understand how city charac-
teristics vary over space and through time. While emerging sources of bicycling data
create opportunities, they also bring challenges. Many of the challenges of using crowd-
sourced data are common regardless of application (e.g. see Mooney & Pejaver, 2018), but
here we frame five broad challenge areas with specific regard to bicycling.

Challenge 1 – access and funding

Emerging bicycling data, like many new datasets on cities, bring challenges of access,
ownership, and funding. Companies collecting data as part of smartphone apps have
business models that include packaging and selling data. In the case of Strava, data
costs are based on the size of the Strava data sample (number of users and recorded
rides). The costs of data purchase can seem high but may be efficient as compared to
manual count programmes if the information needs benefit from data that are continuous
in space and time.
Alternately, projects that use crowdsourcing to improve safety data are open and
anyone can access and benefit from data. However, these projects lack a business
model, and as a result many have come and gone. While cities clearly want more complete
and diverse safety data, it is unclear who will fund the long-term maintenance of crowd-
sourced data collection tools and cities must invest in promotion for ongoing data collec-
tion. In contrast, projects like OSM have taken off globally. The sheer breadth of
applications of OSM may guarantee the longevity and ongoing popularity of such projects.
One potential solution to funding crowdsourced projects may be for cities to join together
and form consortiums around projects that are working well. Rather than build one-off
tools for crowdsourced data collection, tools that have the buy in of many cities will be
beneficial and can have a shared model.

Challenge 2 – privacy
Most of the technology associated with crowdsourced bicycling data includes smart-
phones or mapping. When location information is captured, privacy of individuals can
 TRANSPORT REVIEWS 13

be compromised. Businesses selling data, like Strava, protect privacy by providing

only aggregated data (Roy et al., 2019). While aggregated data are useful for
volume estimates, the need to protect privacy makes it difficult to use data for ana-
lyses of individual paths or trajectories. The only way to access individual level data is
to use apps that have a research purpose, although such studies may struggle with
the number of participants due to recruitment issues. The ethics requirements
facing university-based research projects serve to protect privacy (i.e. BikeMaps.org),
by ensuring anonymity in data collection (Nelson et al., 2015). In the current era,
we have seen a huge shift toward sharing of information on the internet. Twitter
and other social media are good examples of the willingness of people to share per-
sonal details and even their location in digital public space. While Twitter users can
turn off their geolocations, hashtags and place names can still be used to link
people to places. We urge researchers and practitioners to use good ethical judge-
ment in how they use and protect data (Breslin, Shareck, & Fuller, 2019). Issues of
privacy will grow more complex as smart phone technology adds face and voice rec-
ognition. Privacy is a critical area of research that requires legal and ethical research.
Research into and guidelines on the characteristics of technology that have appropri-
ate protection for privacy are urgently needed and will be a useful guide to the indus-
try as we determine which technologies to adopt.

Challenge 3 – representativeness and equity

Data collection efforts that opportunistically use technology result in biased depictions
of the population. When transportation decisions are made using data biased towards
those who use technology the most, vulnerable road users – children, older adults, and
people with low incomes, may be excluded (Garber et al., 2019). Using biased data may
exacerbate urban equity issues; it is insufficient to invest in infrastructure only where
those with access report issues. Research that uses crowdsourced ridership data con-
tinue to identify bias as problematic (e.g. Orellano & Guerrero, 2019) and identify
tradeoffs between sampling bias and sample size (Pritchard, Bucher, & Frøyen, 2019).
However, with effort it can be possible to reduce bias in data using statistical
approaches. There is promising research into bias correction, however bias correction
is only possible when there are good data for training correction models and results
should be used as categorical (low, medium, high ridership) rather than considered
precise to the number of bicycles (Garber et al., 2019; Roy et al., 2019). To improve
representativeness during data collection, in our experience, also requires someone
to go into the community and work one-on-one with people that do not have
access to technology or who do not know how to contribute data. Depending on
the nature of the phenomena represented by data, it can also be possible to correct
the bias from sampling by analytic methods. Bias correction is only possible when
“truth” can be sampled – this may be the case of ridership (e.g. counts; Roy et al.
(2019)), but not for other data needs such as safety. Finally, when possible, more
sources of data are often better and using crowdsourced data in complement with tra-
ditional data sets will allow a more comprehensive understanding of the challenges and
can be less biased than a single source.
14 T. NELSON ET AL.

Challenge 4 – analytics and data uncertainty

The focus of our paper is crowdsourced data, but it is a natural next step to consider ana-
lytics, especially given that crowdsourced datasets are usually large and can be complex.
There are several emerging issues with analysis of crowdsourced bicycling data. First, most
of the tools used do not sufficiently manage data uncertainty (Woodcock, Givoni, &
Morgan, 2013). Lack of robust sampling design increase the uncertainties in data and
the outcome can be analytical results that are more uncertain than they initially appear
(Rojas-Rueda, de Nazelle, Tainio, & Nieuwenhuijsen, 2011). Bayesian machine learning
methods are one group of approaches that can be used to better address uncertainty.
Across fields that use crowdsourced data there is a need to develop tools that handle
data uncertainty in more robust ways.

Challenge 5 – open methods

In this rapidly emerging space, researchers should be packaging tools to be usable as
widely as possible. Built-in data visualisation tools can often help and more automated
tools such as Tableau and Esri dashboards are helping fill this gap. Sharing annotated
code (e.g. markdown documents) on Github and other platforms and developing standar-
dised tools (i.e. packages or libraries for open source software) will help facilitate repeata-
ble research and advance the capacity of planners and researchers to use data in future.
These open tools could include machine learning and data visualisation methods specific
to the various emerging data sources we discuss. While open data and methods are
helping advance bicycling research and practice, collaboration between data scientists
are needed to ensure that the tools developed are generating meaningful solutions and
can be deployed in support of decision making.

Challenge 6 – stakeholder capacity

Crowdsourced data in bicycling is beneficial when it supports better decision making or
enables monitoring of policy outcomes, but skilled practitioners are needed to translate
data into information. Technology for data collection has outpaced analytics and meth-
odological developments are needed to convert data into information products that
can be utilised by practitioners. Most cities that need better bicycling data for planning
do not have the in-house data science capacity to use emerging data to their full potential.
Capacity is needed, in the form of both tools and training. Even with packaged tools, urban
planning programmes need to be training future practitioners with data science skills. As
well, partnerships between universities and regions can help fill the gap in knowledge
needed to make good decisions from data.

Conclusion
In sum, crowdsourced data are a huge opportunity for advancing bicycle research and
practice. There are emerging datasets relevant to bicycling ridership, safety, infrastructure,
and attitudes. Tapping into these requires an understanding of their strengths and weak-
nesses. As well, we need to be proactive in championing solutions to challenges of access
 TRANSPORT REVIEWS 15

and funding, privacy, representativeness and equity, analytics, open methods, and stake-
holder capacity. More specifically we recommend the following be considered to advance
the effective use of crowdsourced data for supporting research on and for pro-bicycling
policy development:

(1) Crowdsourced data collection tools should be developed with many stakeholders to
ensure adequate funding and longevity.
(2) Research into and guidelines on the characteristics of technology that have appropri-
ate protection for privacy are urgently needed.
(3) Representativeness of data can be improved through engagement of underrepre-
sented populations and/or modelling. However, we must avoid the temptation to
treat crowdsourced data in the same way we would a statistically robust sample. In
other words, the data are not precise, but are more suitably represented as categorical
data.
(4) Across fields that use crowdsourced data, including bicycling research, there is a need
to develop tools that handle data uncertainty in more robust ways.
(5) Open data and methods are required to advance our understanding and use of bicy-
cling data.
(6) Use of crowdsourced data often relies on data scientists, but we must translate the
tools to enable use by practitioners.

Of course, data are just one of the components needed to shape the cities of our future
to be sustainable, equitable, and healthy places. Supports are needed for tools and training
such that data can be packaged in a timely manner into stakeholder-relevant products.
Through such efforts, crowdsourcing projects that engage people to share their infor-
mation, stories, and experiences, can translate into compelling evidence for decision
making to support bicycling.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Funding
This work was supported by a Public Health Agency of Canada [grant number #1516-HQ-000064];
Canada Research Chairs program [grant number # 950-230773]; Michael Smith Foundation for
Health Research Scholar Award and the Arizona State University Foundation.

References
Aldred, R., & Goodman, A. (2018). Predictors of the frequency and subjective experience of cycling
near misses: Findings from the first two years of the UK Near Miss Project. Accident Analysis &
Prevention, 110, 161–170.
Anderson, J., Sarkar, D., & Palen, L. (2019). Corporate editors in the evolving landscape of
OpenStreetMap. ISPRS International Journal of Geo-Information, 8(5), 232.
Barrington-Leigh, C., & Millard-Ball, A. (2017). The world’s user-generated road map is more than 80%
complete. PLoS One, 12, e0180698.
16 T. NELSON ET AL.

Blanc, B., & Figliozzi, M. (2017). Safety perceptions, roadway characteristics and cyclists’ demo-
graphics: A study of crowdsourced smartphone bicycle safety data. Transportation Research
Board 96th Annual Meeting, 17-03262.
Boss, D., Nelson, T., Winters, M., & Ferster, C. J. (2018). Using crowdsourced data to monitor change in
spatial patterns of bicycle ridership. Journal of Transport & Health, 9, 226–233.
Branion-Calles, M., Nelson, T., & Winters, M. (2017). Comparing crowdsourced near-miss and collision
cycling data and official bike safety reporting. Transportation Research Record, 2662(1), 1–11.
Breslin, S., Shareck, M., & Fuller, D. (2019). Research ethics for mobile sensing device use by vulnerable
populations. Social Science & Medicine, 232, 50–57.
Brighenti, A. M. (2012). New media and urban motilities: A territoriologic point of view. Urban Studies,
49(2), 399–414.
Broach, J., Dill, J., & McNeil, N. W. (2019). Travel mode imputation using GPS and accelerometry data
from a multi-day travel survey. Journal of Transport Geography, 78, 194–204.
Brondeel, R., Pannier, B., & Chaix, B. (2015). Using GPS, GIS, and accelerometer data to predict trans-
portation modes. Medicine & Science in Sports & Exercise, 47(12), 2669–2675.
Brum-Bastos, V., Ferster, C., Nelson, T., & Winters, M. (2019). Where to put bike counters? Stratifying
bicycling patterns in the city using crowdsourced data. Transport Findings, November, 10828.
Collins, D. J., & Graham, D. J. (2019). Use of open data to assess cyclist safety in London. Transportation
Research Record, 2673(4), 27–35.
Das, S., Sun, X., & Dutta, A. (2015). Investigating user ridership sentiments for bike sharing programs.
Journal of Transportation Technologies, 5(2), 69–75.
DiGioia, J., Watkins, K. E., Xu, Y., Rodgers, M., & Guensler, R. (2017). Safety impacts of bicycle infrastruc-
ture: A critical review. Journal of Safety Research, 61, 105–119.
Dillman, D. A., Smyth, J. D., & Christian, L. M. (2014). Internet, phone, mail, and mixed-mode surveys. The
tailored design method. Hoboken, NJ: John Wiley & Sons.
Duggan, M. (2015). Mobile messaging and social media – 2015. Pew Research Center. Retrieved from:
https://www.pewinternet.org/2015/08/19/mobile-messaging-and-social-media-2015/
Einstein, K. L., Palmer, M., & Glick, D. M. (2019). Who participates in local government? Evidence from
meeting minutes. Perspectives on Politics, 17(1), 28–46.
Eitzel, M. V., Cappadonna, J. L., Santos-Lang, C., Duerr, R. E., Virapongse, A., West, S. E., & Jiang, Q.
(2017). Citizen science terminology matters: Exploring key terms. Citizen Science Theory and
Practice, 2, 1–20.
Federal Highway Administration. (2017). Exposure analysis on specific transportation facilities.
Synthesis of Methods for Estimating Pedestrian and Bicyclist Exposure to Risk at Areawide Levels
and on Specific Transportation Facilities. Ch. 3. Technical Report. Retrieved from: https://safety.
fhwa.dot.gov/ped_bike/tools_solve/fhwasa17041/ch3.cfm#ss2
Ferster, C. J., Fisher, J., Manaugh, K., Nelson, T., & Winters, M. (2019). Using OpenStreetMap to inven-
tory bicycle infrastructure: A comparison with open data from cities. International Journal of
Sustainable Transportation, 14(1), 64–73.
Ferster, C. J., Laberee, K., Nelson, T., Thigpen, C., Simeone, M., & Winters, M. (2020). From advocacy to
acceptance: social media discussions of protected bike lane installations. Manuscript submitted
for publication.
Ferster, C. J., Nelson, T., Winters, M., & Laberee, K. (2017). Geographic age and gender representation
in volunteered cycling safety data: A case study of BikeMaps.org. Applied Geography, 88, 144–150.
Fishman, E., Schepers, P., & Kamphuis, C. B. M. (2015). Dutch cycling: Quantifying the health and
related economic benefits. American Journal of Public Health, 105(8), e13–e15.
Garber, M. D., Watkins, K. E., & Kramer, M. R. (2019). Comparing bicyclists who use smartphone apps to
record rides with those who do not: Implications for representativeness and selection bias. Journal
of Transport & Health, 15, 100661.
González-Bailóna, S., Wang, N. B., Riveroc, A., Borge-Holthoefer, J., & Morenoc, Y. (2014). Assessing the
bias in samples of large online networks. Social Networks, 38, 16–27.
Griffin, G. P., & Jiao, J. (2015). Where does bicycling for health happen? Analysing volunteered geo-
graphic information through place and plexus. Journal of Transport and Health, 2(2), 238–247.
 TRANSPORT REVIEWS 17

Griffin, G. P., Mulhall, M., Simek, C., & Riggs, W. W. (2020). Mitigating bias in Big Data for transpor-
tation. Journal of Big Data Analytics in Transportation, 2, 49–59.
Gu, W., Liu, Y., Zhou, Y., Zhou, Z., Spanos, C. J., & Zhang, L. (2017). BikeSafe: bicycle behavior moni-
toring via smartphones. In Proceedings of the 2017 ACM International Joint Conference on
Pervasive and Ubiquitous Computing and Proceedings of the 2017 ACM International Symposium
on Wearable Computers (pp. 45–48).
Handy, S., van Wee, B., & Kroesen, M. (2014). Cycling for transport. Research needs and challenges.
Transport Reviews, 34(1), 4–24.
Hermann, T., Gleckner, W., Wasfi, R. A., Thierry, B., Kestens, Y., & Ross, N. A. (2019). A pan-Canadian
measure of active living environments using open data. Health Reports, 30(5), 16–25.
Hirsch, J. A., Stratton-Rayner, J., Winters, M., Stehlin, J., Hosford, K., & Mooney, S. J. (2019). Roadmap for
free-floating bikeshare research and practice in North America. Transport Reviews, 39(6), 706–732.
Hochmair, H. H., Bardin, E., & Ahmouda, A. (2019). Estimating bicycle trip volume for Miami-Dade
county from Strava tracking data. Journal of Transport Geography, 75, 58–69.
Hollander, J. B., & Shen, Y. (2017). Using social media data to infer urban attitudes about bicycling: An
exploratory case study of Washington DC. In A. Karakitsiou, S. Rassia, & P. Pardalos (Eds.), City net-
works. Springer optimization and ITS applications (Vol. 128). Cham: Springer.
Hong, J., McArthur, D. P., & Livingston, M. (2019). The evaluation of large cycling infrastructure
investments in Glasgow using crowdsourced cycle data. Transportation, doi:10.1007/s11116-
019-09988-4
Jestico, B., Nelson, T. A., Potter, J., & Winters, M. (2017). Multiuse trail intersection safety analysis: A
crowdsourced data perspective. Accident Analysis & Prevention, 103, 65–71.
Jestico, B., Nelson, T., & Winters, M. (2016). Mapping ridership using crowdsourced cycling data.
Journal of Transport Geography, 52, 90–97.
Kerr, J., Duncan, S., & Schipperijn, J. (2011). Using global positioning systems in health research. A
practical approach to data collection and processing. American Journal of Preventive Medicine,
41(5), 532–540.
Mcnair, H., & Arnold, L. (2016). Crowd-sorting: reducing bias in decision making through consensus
generated crowdsourced spatial information. In International Conference on GIScience Short Paper
Proceedings (Vol. 1, No. 1).
Medury, A., Grembek, O., Loukaitou-Sideris, A., & Shafizadeh, K. (2019). Investigating the underreport-
ing of pedestrian and bicycle crashes in and around university campuses− a crowdsourcing
approach. Accident Analysis & Prevention, 130, 99–107.
Mooney, S. J., & Pejaver, V. (2018). Big data in public health: Terminology, machine learning, and
privacy. Annual Review of Public Health, 39, 95–112.
Nelson, T. A., Denouden, T., Jestico, B., Laberee, K., & Winters, M. (2015). Bikemaps.org: A global tool
for collision and near miss mapping. Frontiers in Public Health, 3, 53.
Orellano, D., & Guerrero, M. L. (2019). Exploring the influence of road network structure on the spatial
behaviour of cyclists using crowdsourced data. Environment and Planning B: Urban Analytics and
City Science, 46(7), 1314–1330.
Poulos, R. G., Hatfield, J., Rissel, C., Flack, L. K., Shaw, L., Grzebieta, R., & McIntosh, A. S. (2017). Near miss
experiences of transport and recreational cyclists in New South Wales, Australia. Findings from a
prospective cohort study. Accident Analysis and Prevention, 101, 143–153.
Pritchard, R., Bucher, D., & Frøyen, Y. (2019). Does new bicycle infrastructure result in new or rerouted
bicyclists? A longitudinal GPS study in Oslo. Journal of Transport Geography, 77, 113–125.
Pucher, J., & Buehler, R. (Eds.). (2012). City cycling. Cambridge, MA: MIT Press.
Ralph, K., Iacobucci, E., Thigpen, C. G., & Goddard, T. (2019). Editorial patterns in bicyclist and ped-
estrian crash reporting. Transportation Research Record, 2673(2), 663–671.
Rojas-Rueda, D., de Nazelle, A., Tainio, M., & Nieuwenhuijsen, M. J. (2011). The health risks and
benefits of cycling in urban environments compared with car use: Health impact assessment
study. British Medical Journal, 343, d4521.
Romanillos, G., Austwick, M. Z., Ettema, D., & De Kruijf, J. (2016). Big data and cycling. Transport
Reviews, 36(1), 114–133.
18 T. NELSON ET AL.

Romanillos, G., Moya-Gómez, B., Zaltz-Austwick, M., & Lamíquiz-Daudén, P. J. (2018). The pulse of the
cycling city: Visualising Madrid bike share system GPS routes and cycling flow. Journal of Maps, 14
(1), 34–43.
Roy, A., Nelson, T. A., Fotheringham, A. S., & Winters, M. (2019). Correcting bias in crowdsourced data
to map bicycle ridership of all bicyclists. Urban Science, 3(2), 62.
Saad, M., Abdel-Aty, M., Lee, J., & Cai, Q. (2019). Bicycle safety analysis at intersections from crowd-
sourced data. Transportation Research Record, 2673(4), 1–14.
Sanders, R. L. (2015). Perceived traffic risk for cyclists: The impact of near miss and collision experi-
ences. Accident Analysis & Prevention, 75, 26–34.
Schweitzer, L. (2014). Planning and social media: A case study of public transit and stigma on twitter.
Journal of the American Planning Association, 80(3), 218–238.
Scott, D. M., & Ciuro, C. (2019). What factors influence bike share ridership? An investigation of
Hamilton, Ontario’s bike share hubs. Travel Behaviour and Society, 16, 50–58.
Strauss, J., Miranda-Moreno, L. F., & Morency, P. (2015). Mapping cyclist activity and injury risk in a
network combining smartphone GPS data and bicycle counts. Accident Analysis & Prevention,
83, 132–142.
Sun, Y., & Mobasheri, A. (2017). Utilizing crowdsourced data for studies of cycling and air pollution
exposure: A case study using Strava data. International Journal of Environmental Research and
Public Health, 14(3), 274.
Teschke, K., Frendo, T., Shen, H., Harris, M. A., Reynolds, C. C., Cripton, P. A., & Winters, M. (2014).
Bicycling crash circumstances vary by route type: A cross-sectional analysis. BMC Public Health,
14(1), 1205.
Vanparijs, J., Panis, L. I., Meeusen, R., & de Geus, B. (2015). Exposure measurement in bicycle safety
analysis: A review of the literature. Accident Analysis & Prevention, 84, 9–19.
Wergin, J., & Buehler, R. (2017). Where do Bikeshare bikes actually go? Analysis of Capital Bikeshare
trips with GPS data. Transportation Research Record, 2662(1), 12–21.
Winters, M., & Branion-Calles, M. (2017). Cycling safety: Quantifying the under reporting of cycling
incidents in Vancouver, British Columbia. Journal of Transport and Health, 7, 48–53.
Winters, M., Hosford, K., & Javahera, S. (2019). Who are the ‘super-users’ of public bike share? An
analysis of public bike share members in Vancouver, BC. Preventive Medicine Reports, 15, 100946.
Winters, M., Zanotto, M., & Butler, G. (2020). The Canadian Bikeway comfort and safety (Can-BICS)
classification system: A proposal for developing common naming conventions for cycling infra-
structure. Health Promotion and Chronic Disease Prevention in Canada: Research, Policy and
Practice (the HPCDP Journal), 40(9).
Woodcock, J., Givoni, M., & Morgan, A. S. (2013). Health impact modelling of active travel visions for
England and Wales using an Integrated Transport and Health Impact Modelling Tool (ITHIM). PloS
One, 8(1), e51462.
Xu, J., & Nyerges, T. L. (2017). A framework for user-generated geographic content acquisition in an
age of crowdsourcing. Cartography and Geographic Information Science, 44(2), 98–112.
Zhang, L., & Pfoser, D. (2019). Using openstreetmap point-of-interest data to model urban change – A
feasibility study. PLoS One, doi:10.1371/journal.pone.0212606