Content uploaded by Ray Pritchard
Author content
All content in this area was uploaded by Ray Pritchard on May 15, 2019
Content may be subject to copyright.
Contents lists available at ScienceDirect
Journal of Transport Geography
journal homepage: www.elsevier.com/locate/jtrangeo
Does new bicycle infrastructure result in new or rerouted bicyclists? A
longitudinal GPS study in Oslo
Ray Pritchard
a,⁎
, Dominik Bucher
b
, Yngve Frøyen
a
a
Department of Architecture and Planning, Faculty of Architecture and Design, NTNU—Norwegian University of Science and Technology, 7491 Trondheim, Norway
b
Institute of Cartography and Geoinformation, Department of Civil, Environmental and Geomatic Engineering, ETH Zürich, 8093 Zürich, Switzerland
ARTICLE INFO
Keywords:
Bicycle infrastructure
Transport planning
Urban planning
Route choice
Mode choice
GPS tracking
ABSTRACT
Well-connected bicycle infrastructure networks are widely accepted to be an important factor for increasing the
level of bicycling in urban environments where motorised and active transport modes must co-exist. However,
little is known about the extent to which new bicycle infrastructure results in changes of route amongst existing
bicyclists as opposed to changes in the mode of transport. This article addresses the route-mode research gap
through a panel study in which participant travel behaviour (n= 113) is recorded with a smartphone Global
Positioning System (GPS) application. The study observes short-term changes to route and mode choice of
participants before and after the establishment of a contraflow bicycle lane in Oslo, Norway. Video and radar-
based traffic counting are used as supplementary methods to affirm bicycle volume changes in the broader
population.
The bicycle lane intervention resulted in a shift in the preferred route in the neighbourhood. The intervention
street saw increased numbers of bicycle trips taken whilst the two nearest parallel routes in the same neigh-
bourhood witnessed a decrease. For bicycle trips taken on the intervention street, the mean deviation from the
shortest path increased (from 171 to 221 m, p< .05). Bicycle counts based on video observations also support
the route shift finding. Bicycle modal share did not significantly increase when comparing the panel sub-group
exposed to the intervention (n= 39) with a quasi-control group (n= 47) who were not exposed but had made at
least one trip in the near vicinity of the intervention in both time periods.
This natural experiment study provides evidence to suggest that route substitution from nearby streets and
paths can explain more of the change in bicycling levels than modal shifts to bicycling in the short term following
the opening of the bike lane.
1. Introduction
High quality and separate bicycle infrastructure has been frequently
established as a precondition for achieving high levels of utility bicycle
use (Dill, 2009;Hull and O'Holleran, 2014;Wahlgren and Schantz,
2014). Many studies of environmental correlates have established a link
between cycling rates and infrastructure (Mertens et al., 2017;Nielsen
et al., 2013;Saelens et al., 2003;Schneider and Stefanich, 2015),
however, the empirical data is somewhat limited with respect to single
project infrastructural impacts within bicycle networks (Handy et al.,
2014;Yang et al., 2010).
This panel study analyses the route and mode choice effects of a
contraflow bicycle lane built in August 2017 in Oslo, Norway. GPS-
based tracking is used to identify changes before and after the inter-
vention for a group of participants who were recruited specifically for
this study. Video observations and radar traffic counts provide volume
changes as a supplementary data source to the GPS panel.
This paper is structured as follows: the background introduces ex-
isting research connected to bicycle interventions, the methods section
describes the data collection approach, including a description of the
intervention area. The timeline of the data collection and intervention
is also described here. This is followed by the results section, which
reports the changes observed within the GPS panel and comparisons
with bicycle count data. Finally, the discussion and conclusion of this
paper summarise the main findings, limitations of the study together
with recommendations for future research.
2. Background
This paper's study design makes use of GPS for data collection, a
https://doi.org/10.1016/j.jtrangeo.2019.05.005
Received 20 September 2018; Received in revised form 3 May 2019; Accepted 7 May 2019
⁎
Corresponding author.
E-mail address: Ray.Pritchard@cantab.net (R. Pritchard).
Journal of Transport Geography 77 (2019) 113–125
0966-6923/ © 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/BY/4.0/).
T
longitudinal natural experiment for bicycling and a focus on both route
and mode choice behaviour. To the authors' knowledge, the combina-
tion of all these three methodological elements in a single study has not
been previously published. Existing research has, however, addressed
these component elements individually and this is summarised below.
Firstly, a number of review studies connected to GPS and bicycle re-
search are reported on, followed by summaries of the relevant results
from three systematic reviews on bicycle infrastructure interventions.
Subsequently five descriptive studies are introduced with focus on route
changes resulting from bicycle infrastructure interventions, whilst the
final section of the background summarises two studies that use the
same type of contraflow bicycle lane as this case study.
The use of GPS in bicycle research is becoming increasingly
common and is now utilised in approximately two-thirds of all studies
connected to bicycle route choice (Pritchard, 2018). The use of GPS
within active transportation and bicycling has been the subject of two
comprehensive reviews (Patricia J. Krenn et al., 2011;Loveday et al.,
2015), whilst GPS in combination with other methods have been re-
viewed by several other researchers, covering more recent combina-
tions of GPS in studies using crowdsourcing, ‘big app’data aggregators,
instrumented bicycle setups and bike sharing operator data (Buehler
and Dill, 2015;Pritchard, 2018;Romanillos et al., 2016).
In a 2017 systematic review of built environment effects on physical
activity and active transport, 11 of 28 reviewed articles had levels of
cycling as a specific outcome (Smith et al., 2017). The reviewed articles
used natural experiments or prospective, retrospective, experimental or
longitudinal study designs and all but one demonstrated either a posi-
tive or non-significant relationship between infrastructure provision
and levels of cycling. Infrastructure types found to have a positive effect
on cycling include: combined pedestrian and bicycle access bridges and
boardwalks (Goodman et al., 2014), urban trails (Fitzhugh et al., 2010),
traffic calming (Morrison, 2004) and bicycle lanes (Lott et al., 1978;
Parker et al., 2013). In Portland, USA, the effect of bicycle boulevards
was evaluated, however, the length and frequency of bicycle trips
performed decreased following the intervention (Dill et al., 2014).
A second systematic review concerning the physical activity impact
of built environment infrastructural changes reviewed eight articles
that reported on changes in levels of bicycling (Stappers et al., 2018).
Positive effects were found for separate bicycle paths which are
sometimes also referred to as bikeways (Heesch et al., 2016;Rissel
et al., 2015).
Three cross-sectional bicycle infrastructure intervention studies
from the grey literature are discussed in a systematic review of 25 cy-
cling interventions studies, with all three found to result in increased
cycling frequency (Yang et al., 2010). Evidence regarding net effects on
cycling modal share was also reported in two of the three studies. The
first, based in Delft in The Netherlands revealed a 3% increase in bi-
cycle modal share in the intervention area compared to a 1% increase
elsewhere in the city (Wilmink & Hartman, 1987). The second study
from Odense, Denmark revealed a 3.4% increase in cycling modal share
from a combination of initiatives including infrastructure improve-
ments but did not have a control group (Troelsen et al., 2004).
Early evaluations of Dutch bicycle planning policies in Tilburg and
The Hague in the 1970s and 1980s contributed in part to the wide-
spread development of bicycle infrastructure across much of the
Netherlands (van Goeverden et al., 2015). Both cities experienced
greatly increased cycling volumes along the routes which received bi-
cycle infrastructure (140% in Tilburg and 76% in The Hague) whilst
only a 10–20% increase was observed in the corridor bicycle volumes
for both cities. Comparable although less significant changes were ob-
served from a before-after study in Davis, California, where a bicycle
volume increase of 87% was observed on the intervention bicycle lane
versus 57% for the corridor (Lott et al., 1978). Furthermore, up to 45%
of the interviewed bicyclists that took alternative routes prior to the
intervention modified their route post-completion to use the new lane.
A traffic count study performed in New Orleans demonstrated increase
bicycle volumes on a new bicycle lane and a simultaneous reduction in
bicycle volumes in the streets parallel to the intervention (Parker et al.,
2013). With a large increase in corridor bicycle volumes, this study's
findings suggest that a significant mode and route change occurred as a
result of the bicycle lane.
Concerning route change effects, a cross-sectional Global
Positioning System (GPS) study from San Francisco found evidence of
route substitution through significantly increased bicycle volumes on
four intervention streets whilst a decline was observed in neighbouring
streets (Fitch et al., 2016). A separate bicycle route choice model using
GPS data from the same city quantified the preference for bicycle in-
frastructure using the Marginal Rate of Substitution (MRS) (Hood et al.,
2011). The model estimated an MRS of 0.49, meaning that the average
cyclist would rather cycle on 100 m along bicycle lanes to avoid cycling
on 49 m of ordinary roads. In addition, the model estimated an MRS of
4.02 for cycling the wrong way down a one-way street, meaning that
cyclists will only ride against the trafficflow if it saves them more than
four times the distance of a conventional street. This is assumed to
apply to streets for which contraflow cycling is not permitted.
Two studies specifically on the effects of contraflow bicycle lanes
were uncovered, the first of which demonstrated significant increases in
the use of contraflow bicycle lanes and simultaneous reduction in
footpath cycling in Oslo, Norway (Bjørnskau et al., 2012). The second
study involved an intercept survey of bicyclists in Washington, D.C.
which revealed that participants' weekly usage of the bidirectional
contraflow bicycle lane street increased from 15% pre-intervention to
80% post-intervention (Goodno et al., 2013).
This paper contributes both to the knowledge regarding this specific
type of initiative and more importantly, to the empirical knowledge
regarding intervention studies and bicycle route choice. The literature
reveals that whilst there are several studies that demonstrate a gen-
erally positive association between bicycle infrastructure provision and
bicycle modal share, the state of knowledge regarding changes in route
choice is less mature. This applies particularly for longitudinal inter-
vention studies, since most of the research presented up to this point
uses forms for cross-sectional study design such as traffic counting.
Several reviews of research on bicycle travel behaviour have noted the
rarity of longitudinal studies using control groups (Handy et al., 2014;
Smith et al., 2017;Yang et al., 2010). This paper has made an effort to
capture the intervention effects separate from population changes
through the use of a quasi-control respondent group.
3. Methods
3.1. Study area
A contraflow bicycle lane (i.e. in the opposite direction to one-way
vehicular traffic) in Markveien in Oslo, Norway, was opened for cyclists
at the end of August 2017. Markveien extends north-south through the
district of Grünerløkka and is one of several parallel streets connecting
the suburb of Torshov with Oslo city centre. The contraflow bicycle lane
is a part of the City of Oslo's City Route 1 bicycle infrastructure project
which commenced in 2016. City Route 1 is one of eight City Route
bicycle infrastructure projects in Oslo covering 55 km of streets within
Oslo's outermost ring road: Ring 3. The planned completion of the City
Routes is 2020 and is seen by the City of Oslo as its most important
bicycle promotion initiative. The changes are pictured in Fig. 1 whilst
the map in Fig. 2 illustrates the bicycle lane together with the existing
bicycle infrastructure in Grünerløkka and Torshov.
The ‘intervention’(or natural experiment) is a 400 m long section of
Markveien, between Grüners gate and Øvrefoss (59°55′32.2″N,
10°45′25.6″E), in which a 2.4 m wide red asphalt bicycle lane sub-
stituted parallel car parking on the eastern side of the street. Parallel car
parking on the western side of the street remained unchanged.
Bicyclists have been permitted to ride contraflow in this street since
2015. There are no bicycle lanes in the same direction as traffic,
R. Pritchard, et al. Journal of Transport Geography 77 (2019) 113–125
114
meaning cyclists must ride on the road lane. The intervention extends
the total length of contraflow bicycle lanes on Markveien from 447 m to
847 m, as shown in Fig. 2. Following the intervention, only 100 m of the
City Route 1 section of Markveien lacks contraflow bicycle lanes.
Two other streets in the same neighbourhood received bicycle in-
frastructure modifications during the analysis period (thus making the
isolation of the intervention effects harder since they also affect bicycle
behaviour). The first was a 245 m segment of Sandakerveien, a one-way
street 1 km to the north of Markveien, which received the same treat-
ment as the intervention site in late September 2017 (contraflow bi-
cycle lane in lieu of parallel car parking). Sandakerveien is also part of
Oslo's City Route 1 project. The second infrastructure upgrade involved
the recolouring (from black to red) and widening of 745 m of bicycle
lanes along both sides of Toftes gate in June 2017, a parallel street two
blocks to the east of Markveien. Both Toftes gate and Sandakerveien are
illustrated together with Markveien in Fig. 2.
3.2. Participants
This study tracked the mobility behaviour of a panel of residents
from the northern suburbs of Oslo who would be most exposed to a new
bicycle lane constructed in Markveien, Grünerløkka. Participants were
recruited to the study using multiple approaches. 3000 personalised
invitational letters were mailed to addresses < 400 m from the northern
section of City Route 1. The mailing area was entirely north of the in-
tersection between Markveien and Grüners gate, where the intervention
begins. This was done since it was assumed that the dominant desti-
nation for cyclists in the neighbourhood would be central Oslo, south of
the intervention.
The study was also distributed through a local newspaper adver-
tisement, flyers, posters and social media connected with the area of
interest. Except for social media targeting specific interest groups, the
recruitment process was randomised. In total 113 Oslo residents par-
ticipated in both data collection rounds, 51 of whom were recruited via
the letters and unknown numbers recruited via other means.
Fig. 1. Before and after changes in Markveien (top and bottom images respectively), completed in August 2017 (view to the north from the intersection with
Seilduksgata). Source: the City of Oslo Agency for Urban Environment.
R. Pritchard, et al. Journal of Transport Geography 77 (2019) 113–125
115
The bicycle lane intervention was constructed between the 14th and
31st August 2017. The bicycle lane and the study's focus on bicycle
travel behaviour were deliberately not referenced in the invitational
material in the interest of reducing response bias (Envall, 2007, p. 164).
The study purpose was instead described as being related to long-
itudinal travel behaviour changes in the local environment. Participant
travel behaviour was recorded in two four-week periods pre-interven-
tion between 13th May and 9th June and post-intervention from 12th
September to 9th October 2017.
3.3. Instrumentation: GPS-enabled smartphone application (app)
Participants' own smartphones with integrated Global Positioning
System (GPS) were used for gathering panel mobility data from the
participant panel. 91% of the Norwegian population had access to a
smartphone in 2017 and thus selection bias through this choice of
method was considered minimal (Vaage, 2018).
Whilst a number of travel survey-specific commercial apps exist
(Berger and Platzer, 2015;Flügel et al., 2017), a more affordable so-
lution was found that built upon a passive physical activity monitoring
app called Moves®(shut down in July 2018). A second app, GoEco!
Tracker,
1
was required to extract information from Moves®and re-
classify the mode of transport used for motorised journeys, which are
classified in Moves®as ‘transport’. GPS data is recorded first in Moves®,
and via an application programming interface (API), is automatically
collated to a secure server by the GoEco! Tracker app (Bucher et al.,
2016). This required participants to download both apps and authorise
the transfer of data from Moves®to GoEco! Tracker. More detailed in-
formation on the data collection protocol (approved for this study by
the Norwegian Centre for Research Data) can be found in the metho-
dological paper from the GoEco! project team (Bucher et al., 2016).
Fig. 2. The intervention street Markveien in Oslo together with existing bicycle infrastructure in Oslo's inner northern suburbs of Grünerløkka and Torshov. Arrows
indicate the one-way direction for cars since bicycles are permitted in both directions on all streets.
1
www.goeco-project.ch
R. Pritchard, et al. Journal of Transport Geography 77 (2019) 113–125
116
3.4. Pre-processing GPS data
Map-matching is a necessary procedure in the preparation of GPS
data for subsequent analysis, to determine the distance travelled and to
be able to count the number of trips along a specific street or path.
Hidden Markov Model-based map-matching was performed on the raw
data (after mode validation in GoEco! Tracker) using the Open Source
Routing Machine (OSRM) matching profiles for car and walking trips
(Project OSRM, 2018). Additional matching profiles were created for
trains, trams and buses, and the profiles for bicycle journeys were
adapted by the GoEco! Tracker developers to allow matching to both
bicycle-specific and generic routes within OpenStreetMap.
To handle the variable raw data quality (due to different tracking
resolution from dissimilar recording devices), several map-matching
strategies were used to pre-process the GPS trajectories, as illustrated in
Fig. 3 below. By default, OSRM applies a matching algorithm similar to
Fig. 3. Examples of the different matching approaches used to handle the varying route data quality. Red lines indicate raw data and blue are matched to the street
network. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
R. Pritchard, et al. Journal of Transport Geography 77 (2019) 113–125
117
the one described by Newson and Krumm (2009) (Panel A in Fig. 3).
The matching process first locates network nodes in the proximity of
each raw GPS point by searching within a radius determined by the
tracking device's reported accuracy (for example 20 m). The map-mat-
ched route must pass through at least one of the nearby nodes from
each GPS point. OSRM maps all possible combinations of nodes be-
tween consecutive GPS points and repeats this procedure for the full
GPS trajectory. From the many combinations created, an optimum
route is determined based primarily on the time difference between
consecutive GPS points and the typical speeds of the transport mode.
In case there are too few GPS points along a recorded route, OSRM
simply “routes”journeys on the shortest path between the start and end
points (see panels B and C in Fig. 3 with two different scales) (Huber
and Rust, 2016). In some cases, the opposite situation occurs in which
there are very large numbers of GPS points (due to high resolution and/
or long journey distance). This results in problems for the computation
of a map-matched route in OSRM –potentially due to limitations in the
memory for storing all node combinations for routes with large num-
bers of GPS points. For these journeys, as illustrated in panel D, the
number of GPS points is repeatedly simplified using a Douglas-Peucker
algorithm (Douglas and Peucker, 1973). This procedure removes the
least critical GPS points (based on proximity to consecutive points),
yielding a smaller number of possible node-to-node combinations until
the matching algorithm delivers a route. Despite the reduced data re-
solution, this procedure was found to provide satisfactory results. Fi-
nally, for recorded journeys in which there are large gaps between
consecutive GPS points (more than three kilometres, E), the gaps are
individually routed while the parts without large gaps are matched (i.e.
a combination of A and B is applied).
The panel produced 36,153 trips (across all modes) during the two
months of data collection, 2.7% of which were taken outside of Norway
and were not considered for matching. The approach described above
allowed the direct matching (A) of 87.3% of all trips, routing (B and C)
of 6.0%, simplification and matching (D) of 1.6%, ‘large gap’routing of
0.3% (E) whilst the remaining 2.1% had missing mode information or
failed. The data collection approach was found to correctly identify the
travel mode in approximately 80% of cases in a test of Swiss GoEco!
Tracker travel data in which participants were requested to confirm
travel mode (Bucher et al., 2016).
3.5. Supplementary data collection: video observations and automated
traffic counting
In addition to GPS data collection, two further before and after
methods were used: bicycle counts extracted from video observations
and automated traffic counting of bicycles and motorised vehicles with
Doppler radar traffic counters.
In the interest of capturing route choice changes, an elevated
Miovision Scout camera (720 × 480 pixels, 30 fps) was temporarily
installed above a forked intersection near to the intervention street (see
Fig. 2). The forked intersection was chosen as it forms a natural decision
point where bicycle users can select one of two alternative routes when
cycling towards the city centre (one of which is the intervention street).
Similarly, bicycle movements along the same two alternative routes
coming from the city centre merge at this point when continuing further
north. Cyclist movements in the video recordings were extracted by
Miovision through their automated traffic data processing tool. With
the configuration shown in Fig. 4,Miovision guarantees ≥85% inter-
section count accuracy (an accurate count correctly registers a cyclist's
movement between any two of the three coloured zones). Video data
was uploaded to the Miovision Traffic Data Online server and bicycle
counts were received in 15-minute intervals going into and out of the
two streets of interest.
Radar-based traffic counting was also deployed in three locations
including the intervention street Markveien and two nearest parallel
alternative streets Thorvald Meyers gate and Toftes gate (see Fig. 2).
The ViaCountII mobile traffic counters use integrated Doppler radar
devices (24,165 GHz/100 mW EIRP) to determine the speed, length,
vehicle class (including bicycle) and direction of travel (Via Traffic
Controlling GMBH, 2016). The accuracy of the counters is not stated in
the technical product specifications, but are regularly used by the City
of Oslo for traffic counting.
3.6. Analytical approach
Data from the three sources were recorded before and after the in-
tervention completion during the time periods illustrated in Table 1.
Pre-processed GPS data (after conversion to .shp format) were pro-
cessed using a combination of software including a Geographic In-
formation System (GIS) program, statistical software and spreadsheets.
The automated traffic counts from the video footage (recorded from
6 am to 9 pm excluding start and end days) and radar traffic data (24 h
per day) were analysed in spreadsheets.
In order to observe changes in route choice, all bicycle trips (as
classified by the GoEco! Tracker app) taken by the panel participants
were accumulated for each link in the transport network in the before
and after time periods. For any given link, this resulted in two counts
for the number of bicycle trips that passed the link during the before
and after periods respectively. Thereafter the number of link bicycle
trips (num) in each period was normalised by dividing by the sum of all
link volumes from the corresponding period for the map extent in-
dicated in Fig. 2. The change in bicycle volumes is calculated in GIS
using the expression below for each link in the transport network where
the before period is 1 and the after period is 2. This mitigates for po-
tential confounding factors such as weather variability or other sea-
sonal variation between the two data collection periods.
∆=−
∙
Adjusted bicycle volume num num num num
num
(/Σ /Σ )
Σ
link x xN xN
N
22 11
1(1)
The scale of the intervention and limited time to adjust behaviour is
such that short term modal changes cannot be expected for all journeys
taken by the panel. To account for potential modal changes, it was,
therefore, necessary to remove journeys that are not in the immediate
vicinity of the intervention (defined as being the area bounded by the
four nearest parallel streets, two on each side of Markveien). This was
done by creating a modal analysis zone (a polygon) in ArcMAP covering
this immediate vicinity and selecting only those GPS journeys which
intersect with this zone. This zone is shown in Fig. 7 with the red
shaded polygon. In this manner, only the subset of journeys that are
taken in proximity to the intervention is considered. This is an im-
portant consideration given the dataset covers trips taken by the par-
ticipants across the whole of Oslo and beyond.
Checking for mode substitution was performed by firstly selecting
panel participants who had taken at least one journey in the modal
analysis zone in both periods (n= 86). From this group, a subset of
respondents (n= 39) was exposed to the intervention, whilst the re-
mainder are considered as a quasi-control group (n = 47). Exposure
was defined as having used at least one segment of the 400 m inter-
vention section of Markveien in the after period with any mode (ex-
cluding trips that cross Markveien since the bicycle lane does not ex-
tend through intersections). In other words, the criterion for exposure
requires intervention link utilisation (to travel on or alongside the
contraflow bicycle lane). This approach was adopted since it is not
guaranteed that users crossing Markveien will register changes in side-
street appearance if they are more occupied with traffic hazards (and
given the dark red bicycle lane has low conspicuity in wet weather and
at night).
Existing approaches for exposure typically rely upon area or
proximity based measures, often categorised using distance from the
intervention (Stappers et al., 2018). Alternative approaches attempt to
demonstrate the diminishing influence of the intervention with
R. Pritchard, et al. Journal of Transport Geography 77 (2019) 113–125
118
proximity through the use of the negative square root of distance
(Heinen et al., 2017). The strict link-utilisation definition used in this
paper was chosen in favour of the broader definitions above due to a)
the short time frame of post-intervention travel behaviour measure-
ment, b) smaller scale of the intervention compared to the examples
reviewed by Stappers et al. (2018) above and c) the ability to be able to
select participants based on their actual use of a road (due to the GPS
data).
The journeys that intersected the modal analysis zone were sum-
marised into a modal share for each user in this sub-group for the before
and after periods. Paired samples t-tests were then used to compare the
change in bicycle modal share for the exposure group and the non-
exposure group (a quasi-control group) between the before and after
periods.
3.7. Difference in differences
Since both the quasi-control and exposure group experience in-
creases in bicycle modal share, the difference in differences approach is
used to quantify the changes. This involves considering the difference
between the trends (such that when the two groups of interest increase,
it is the differences in the increase that are measured).
Since the intention of this paper is to measure the significance of the
changes, the classic regression approach is used to calculate the dif-
ference in differences for the dependent variable bicycle modal share
given by y
it
in equation 1 below (Donald and Lang, 2007).
=+∙ +∙ +∙ ∙ +y α β Exposure β Post β Exposure Post()ϵ
it itit
123
(2)
Exposure
i
and Post
t
are dummy variables introduced to distinguish
group membership in which Exposure
i
equals one for the participants in
the exposure group (n= 39) and is zero for the quasi-control group,
and Post
t
equals one for the post-intervention time period and is zero for
the pre-intervention period. Running this as linear regression in SPSS
provides an estimate for the difference in differences given by the
parameter β
3
together with the necessary outputs to report statistical
significance.
4. Results
4.1. Characteristics of study participants
The numbers of men and women participating in the panel
(n= 113) were approximately equal, although men were generally
older as can be seen in Fig. 5. The education level of the sample was
considerably higher than that of the local population. Eigty five percent
of the panel had some form of higher education, compared with census
records for Sagene and Grünerløkka that show 60% of the intervention
area population had higher education (Holseter, 2018). Before the in-
tervention, 86 members of the GPS panel had conducted 4 or more trips
by bicycle during the first month of data collection (or an average of
Fig. 4. Video camera perspective with bicycle counting zones for automated counting of bicycle movements between the three zones.
Table 1
Data sources recording periods.
Source Before After Data processing
GPS 28 days 13th May –9th Jun 28 days 12th Sep –9th Oct ESRI ArcMap 10.6, Microsoft Excel, IBM SPSS Statistics 25
Video observation 39 h 12th, 13th, 15th, 16th May 86 h 21st Sep –26th Sep Microsoft Excel
Radar traffic counting 7 days 8th May –14th May 7 days 18th –24th Sep Microsoft Excel
0
2
4
6
8
10
12
14
16
18
Female Male
Fig. 5. Age distribution of GPS panel.
R. Pritchard, et al. Journal of Transport Geography 77 (2019) 113–125
119
one or more trips per week). There were 83 panel members who took 4
or more bicycle journeys following the intervention (also over a period
of one month). As a proportion (76% and 73% respectively) this is
significantly higher than the weekly cycling levels for the Grünerløkka
(52%) and Sagene (49%) city districts where most participants live
(Bayer, 2018).
Seasonal variation in Scandinavia as with many other countries with
snowy winters results in variability in the levels of bicycling. The GPS
panel modal share data for each month was compared with travel
survey data from Ruter, the public transportation authority in Oslo.
Ruter's market information system, a type of continuous travel survey
has a sample size of approximately 3400 Oslo residents spread
throughout the year. The comparison of the GPS data with the popu-
lation sample from Ruter is shown in Fig. 6 below. Minimal seasonal
variation is observed during the before and after data collection per-
iods, however cyclists and pedestrians are greatly overrepresented
whilst car drivers and public transport users are underrepresented.
In addition, Fig. 6 displays the modal split for the recruitment
neighbourhood (defined as the zone in which invitation letters were
distributed). This data is taken from the 2013/2014 Norwegian Na-
tional Travel Survey (NNTS) (Hjorthol et al., 2014). This reveals that
the (average annual) neighbourhood modal shares of public transport
(31%) and cycling (5%) are approximately equal to that of Ruter's
sample in Oslo. However, walking is more common in the neighbour-
hood (38%) than the Ruter sample (27%), whilst car journeys are less
common (26% versus 32%).
4.2. Route substitution
Positive values for changes in normalised bicycle volume, depicted
in light turquoise in Fig. 7, indicate the approximate increase in bicycle
trips made by the panel after the intervention compared to before.
Negative values, drawn in dark orange, show the corresponding re-
ductions in panel bicycle volumes. The intervention street Markveien
has clearly increased in popularity amongst the panel, whilst neigh-
bouring streets Thorvald Meyers gate and the riverside shared path
experienced a reduction. Although infrastructural changes were made
in Sandakerveien and Toftes gate (as depicted in Fig. 2) during ap-
proximately the same time interval as Markveien, mixed results are
observed in these streets with a smaller change in travel behaviour.
Monthly volumes are used preferentially to daily volumes since the data
comes from two one-month-long periods, first in May/June 2017 and
afterwards in September/October 2017.
4.3. Deviation rate
A form of quantification for the change in bicycle route choice can
be made by considering the deviation distance from the shortest path
(calculated in ArcMAP) (Krenn et al., 2014). An independent samples t-
test was performed using all the bicycle trips taken on Markveien before
and after the intervention. The deviation from the shortest path (in
metres) after the intervention was built in Markveien was greater
(mean = 221, SE = 18), than before (mean = 171, SE = 15), and the
difference, −50, 95% CI [−96, −4] was significant t (289) = −2.16,
p= .032. In other words, the upgraded Markveien was able to induce a
221 m deviation from the shortest path (compared to 171 m before).
This demonstrates that the average bicycle user of Markveien had a
significantly increased detour from the shortest path in order to use the
contraflow bicycle lane configuration than the same street pre-inter-
vention. Existing users presumably continued to use Markveien, so the
increase in the mean suggests that the new cyclists who began to use
Markveien took greater detours than 221m to use the intervention in-
frastructure.
4.4. Video comparison
More than 100 h of video footage was processed by Miovision to
count the number of bicycles taking Øvrefoss, which leads directly to
the intervention street Markveien, and the alternative street Thorvald
Meyers gate. Since only bicycles were counted in the footage, the video
data cannot be used to determine any changes in modal share –but
allows observation of any changes to bicycle route choice. In Table 2
below the percentages of cyclists choosing each of these two streets is
shown and compared with the GPS panel counts on the same two
streets. It should be noted that not all traffic through the intervention
goes through this intersection, and therefore it is only indicative of
changes that occur in the intervention. Immediately apparent in Table 2
however is that the scale of the change for the video observations is
much less than the GPS panel.
4.5. Directional changes
The contraflow bicycle lane undoubtedly improved the bicycling
conditions for northbound cyclists using the intervention, since the
replacement of a parking lane with a bicycle lane provided much
greater separation from the flow of one-way southbound traffic. The
directional flows are displayed in Table 3 below for those routes passing
through the directional analysis zone indicated in yellow in Fig. 7. The
0%
10%
20%
30%
40%
50%
60%
70%
Walking Bicycle Public
transport
Car
Oslo (Ruter) before
Oslo (Ruter) aer
Recruitment neighbourhood (NNTS)
0%
10%
20%
30%
40%
50%
60%
70%
Walking Bicycle Public
transport
Car
Modal share of all trips
GPS panel before
GPS panel aer
Fig. 6. Transport modal share for the GPS panel (left) relative to the general Oslo population (right) for the before and after data collection intervals. NNTS data is
additionally shown to the right for the recruitment neighbourhood in Oslo.
R. Pritchard, et al. Journal of Transport Geography 77 (2019) 113–125
120
directional analysis zone is a single cross-section of streets surrounding
Markveien and all trips that intersect it were counted and sorted by
street and direction. This included two parallel streets to the west of
Markveien: Fossveien and Steenstrups gate and three to the east:
Thorvald Meyers gate, Bjerkelundgata and Toftes gate.
Markveien is found to become a more popular choice amongst the
six streets in both the northbound and southbound direction, with a
near-doubling in the percentage of trips taken on this street. No evi-
dence is found in the GPS data to suggest that northbound cycling in-
creased any more than southbound cycling. Video data also supports
this finding in which the proportion of northbound cyclists entering the
Fig. 7. Change in the number of monthly recorded bicycle trips taken before and after intervention adjusted for seasonal variation. The intervention stretchof
Markveien is shown by the dashed violet line. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this
article.)
Table 2
Average daily number of observed trips taken by bicycle (in both directions) at
the intersection of Øvrefoss and Thorvald Meyers gate (see video camera lo-
cation in Fig. 2).
GPS panel (n= 113) Video observation
(population)
Time period Intervention
‘tributary’
(Øvrefoss)
Thorvald
Meyers
gate
Intervention
‘tributary’
(Øvrefoss)
Thorvald
Meyers
gate
Pre-intervention 4.19 (43%) 5.60 (57%) 374 (46%) 439 (57%)
Post-intervention 5.69 (70%) 2.41 (30%) 563 (50%) 566 (50%)
Table 3
Percentage of bicycle journeys on Markveien relative to the total number of
trips that cross the directional analysis zone.
Time period Northbound Southbound
Pre-intervention 16.0% of 318 16.4% of 372
Post-intervention 29.2% of 226 31.1% of 302
R. Pritchard, et al. Journal of Transport Geography 77 (2019) 113–125
121
intersection Øvrefoss increases from 48% to 52% (compared only with
Thorvald Meyers gate rather than the 5 other streets). Southbound
cycling from the intersection into Øvrefoss also increases from 45% to
48% following the intervention. The difference in proportions between
the GPS data and the video data is a limitation of the method in which
the video observations can only record directional preferences against
one other street. More importantly, however, is the similar increase in
cycling independent of direction within each method.
The small difference in north and southbound cycling in both data
sources is contrary to expectations, given that the conditions for cycling
southbound were largely unchanged. However, the lack of change in
directional utilisation of Markveien can potentially be explained by the
change in contraflow bicycle direction on the sections of City Route 1
both north and south of the intervention (see Fig. 2). The contraflow
bicycle lane alternates between the west and east sides of the road
(given the shift in one-way direction for cars). This means that cyclists
who are unwilling to share a street with cars are unlikely to utilise City
Route 1 since there is no bicycle infrastructure in the car travel direc-
tion. The low degree of directional difference may also be the result of
the improved perceived safety and comfort of Markveien also when
travelling southbound with cars due to the removal of parked cars on
the east side of the street (see Fig. 1).
4.6. Mode substitution to bicycle
For the exposure group (n = 39), the modal share is calculated
based on the 2032 trips they made in the modal analysis zone in both
periods. The exposure group had a higher bicycle modal share in the
after period (mean = 0.499, SE = 0.056), than the before period
(mean = 0.422, SE = 0.053), however the difference, −0.077, 95% CI
[−0.166, 0.012] was only weakly significant t (38) = −1.743,
p= .089. The modal shares above are presented as decimal values but
indicate the percentage of all trips taken by bicycle: 42.2% before and
49.9% after for the exposure group.
The quasi-control group (n =4) for the modal analysis is the subset
of the panel that was not exposed to the intervention but still performed
at least one trip in the modal analysis zone in the before and after
period. For the quasi-control, the modal share is calculated based on the
1193 trips they made in the modal analysis zone in both periods. They
had a higher bicycle modal share in the after period (mean = 0.342,
SE = 0.058) than the before period (mean = 0.312, SE = 0.053). The
difference, −0.030, 95% CI [−0.11, 0.05] was not significant t
(46) = −0.728, p= .471.
Since both the exposure and quasi-control groups experience an
increase in bicycle modal share, the difference in differences approach
can be used to reveal the effect of the intervention. Linear regression in
SPSS provided the difference in differences coefficient β
3
= 0.047, or a
treatment effect of 4.7% change in bicycle modal share. The difference
(95% CI [−0.065, 0.159]) was not significant t (171) = 0.419,
p= .676. The 4.7% difference in differences was confirmed through the
manual calculation of the means from each combination of the dummy
variables (Lechner, 2010).
To make a comparison with volume data in which modal informa-
tion is available the radar data can be used. The radar counting device
registered traffic volumes in each of the three parallel streets for one
week at a time in two time intervals as detailed in Table 1. The data is
not directly comparable as only three north-south streets are compared
instead of all trips in the modal analysis zone, but it approximates the
same conditions. The corridor bicycle volumes across the three streets
increase from 13.7% to 16.8%. Markveien meanwhile observed a de-
crease in the bicycle modal share amongst the three streets from 31.5%
to 27.8%. The finding does not corroborate either the video evidence or
GPS data. Inconsistencies in the radar data are further discussed in the
following section.
5. Discussion
5.1. New infrastructure: rerouted or new bicyclists?
Study designs for longitudinal bicycle infrastructure evaluation
studies such as this vary widely; however, few studies register changes
of bicycle route choice as well as mode choice. This paper provides
evidence for route substitution both through the GIS-plotted changes in
Fig. 7, bicycle counts from the video observations and through a sig-
nificant increase in deviation from the shortest path (by 50 m) on the
intervention street. However, the increase in the rates of cycling fol-
lowing the intervention was not found to be significant for the group
exposed to the intervention using the difference in differences approach
(4.7% increase in modal share, p= .676). This is despite reducing the
number of trips under consideration to those in the immediate vicinity
of the intervention and taking into consideration only the panel sub-
group directly exposed to the change.
The lack of significant modal increase may be a result of the small
sample size in the exposure group (n= 39). Alternatively, it may simply
be a function of the relatively minor scale of the intervention –400 m of
bicycle lanes on one side of a street, or the short period of time (one
month) residents had to adapt to the intervention changes in the after
period. It could be that alternative study designs (including a longer
follow-up period) would be able to demonstrate a significant modal
shift.
Although route substitution of bicyclists has not been thoroughly
researched, existing literature suggests that it can vary greatly de-
pending on the type of intervention and context (Fitch et al., 2016;Lott
et al., 1978;Parker et al., 2013;van Goeverden et al., 2015). The
aforementioned studies principally use volume or cross-sectional
methods to assess changes across two time periods rather than long-
itudinal study designs, making a generalised assessment of route sub-
stitution difficult. The phenomenon is of key importance for regional
and national transport models, which until now have rarely considered
other effects than minimisation of travel time when routing cyclists
(van Wee and Börjesson, 2015). For this study, the intervention did not
provide a new network connection but improved the quality of an ex-
isting route. Travel time benefits are therefore marginal, however,
benefits in terms of traffic safety and thereby attractiveness to existing
cyclists are worth considering in future research seeking to model the
route substitution effect.
From a theoretical perspective, the observation of changes in route
but no (significant) changes in bicycle modal share can be partly ex-
plained by the concept of utility maximisation (or optimisation). Utility
maximisation is a central concept in microeconomic theory in which
actors always make optimal decisions. The assumption is that people
make rational decisions which offer a level of utility (or satisfaction)
that is greater than or equal to any other option open to them. The
theory therefore implies that new bicycle infrastructure will only result
in changes to route or mode if it provides a more attractive transport
option compared to existing alternatives. Thus should bicycle infra-
structure be developed near to competing routes, the marginal utility
can be expected to be reduced according to this approach (Broach,
2016). Although information about the intervention was unlikely to be
known by all study participants, it was able to provide a degree of
utility sufficient to cause route change. Since cyclists have many similar
options available to them in this gridded street suburb of Oslo, small
changes on the intervention street can make this a superior alternative.
The similarity between modes meanwhile is less pronounced for most
travellers - thereby requiring a greater change in utility to result in
significant change. That route change was clearly witnessed whilst
mode change did not significantly change is in line with utility max-
imisation theory and the relative differences within route and mode
choice sets.
A similar study to this paper in the Norwegian context required
users to draw their typical routes rather than have their travel
R. Pritchard, et al. Journal of Transport Geography 77 (2019) 113–125
122
behaviour tracked by GPS. It demonstrated significant changes to both
route and mode choice, however the initiative was for bi-directional
cycling and was longer (1.8 km versus 400 m), objectively safer (phy-
sically separated bicycle path versus contraflow bicycle lane) and in-
cluded greater restrictions to car usage (two of four road lanes replaced
and no-through driving restriction versus substitution of parking lane)
(Vasilev et al., 2018). Considering these substantial contextual differ-
ences, a much larger change in utility can be expected compared to this
paper's intervention –thereby possibly accounting for significant
(p = .0014) changes also in travel mode. The drawn routes study does
have weaknesses in terms of sample representativity, a post-interven-
tion only evaluation (with routes recalled from pre-intervention phase)
and lack of complete travel mode information (such as a travel diary).
Combining the approaches from this paper and Vasilev et al. (2018)
over multiple post-intervention follow-ups would make for a more
rigorous bicycle infrastructure intervention study design that can state
travel behaviour effects with greater certainty.
The remainder of the discussion section highlights the considera-
tions made in selecting this study design, limitations and makes re-
commendations for future studies.
5.2. Strengths and weaknesses of selected methods
A passive smartphone app was selected for this study as it runs in
the phone background, reducing participant burden relative to active
start-stop apps and more easily enabling the capture of all travel be-
haviour (Pritchard, 2018). Such apps have the advantage of counting all
traffic movements rather than only bicycle journeys, thus providing an
indication of modal effects in addition to route changes. The dis-
advantage with Moves®and many other passive apps is high battery use
and a low GPS sampling rate, with GPS points recorded on average once
every 76 s for bicycle journeys. The frequency was higher for journeys
associated with physical activity (walking GPS points every 45 s) –than
motorised travel (105 s between consecutive car GPS points). This is
perhaps unsurprising given the measurement of physical activity is the
principal aim of Moves®. Since cycling journeys have an average origin-
destination speed of 13.1 kph, the mean spacing between consecutive
GPS points is 277 m. Given typical distance between parallel streets in
the gridded study area are around 100 m, nearly three city blocks can
be traversed in the time between GPS points.
A literature review of bicycle route choice data collection methods
(Pritchard, 2018) revealed three papers which use passive smartphone
GPS, however only one of these stated the GPS sampling rate: one point
per second (Sandsjö et al., 2015). For this study, Moves®did not state
the GPS point frequency but early trials revealed that the GPS sampling
rate to be considerably lower than 1 Hz. The trials suggested that bi-
cycle route choice would remain clear despite the lower sampling rate,
however the 76-second period between GPS points was greater than
expected (corresponding with an average frequency of 0.013 Hz), po-
tentially due to wide variability between smartphone models.
Although the point frequency from the GPS method used in this
paper is low, the process for mode and route matching is automated,
thus providing a consistent means of analysing the data across the two
time periods. The point frequency did not appear to be highly proble-
matic for mode identification, however walking trips were found to be
correctly matched at a higher rate than other trips (most likely due to
the combination of characteristic accelerometer movements and low
speeds) (Bucher et al., 2016). For map-matching, slightly > 6% of GPS
routes required the routing engine in OSRM as described in methods
Section 3.4. This uses a shortest path search on the OpenStreetMap
network, thereby providing a consistent approach for routing (Huber
and Rust, 2016). Comparison of GPS data collected before and after the
bicycle lane intervention in Fig. 7 should therefore effectively cancel
the impact of potential routing errors that result from low GPS point
frequency.
Despite the challenges this created for map-matching and route
quality at higher speeds, the adopted method had many benefits (Moves
was shut down in July 2018): compatibility with both Android and
iPhone smartphones, automatic trip segmentation, partial mode clas-
sification, a freely available API and no need for technical support. The
smartphone GPS methodology is, however, challenging in terms of re-
cruitment as data privacy concerns made response rates very low (152
responses from 3000 mailed invitation letters –51 of whom provided
sufficient data for inclusion in the panel).
Portable GPS units have also been used in bicycle route choice re-
search. A review of 21 bicycle route choice studies employing such
units found the median rate of geo-location to be one point per second,
however concurrent data collection would require the acquisition of
many GPS devices, thereby being very costly for a study with similar
numbers of participants (Pritchard, 2018).
The average number of daily trips recorded for each panel partici-
pant was 6.00 pre-intervention and 5.46 post-intervention. By com-
parison, 3.40 daily trips were made per person amongst inner Oslo
residents in the Norwegian National Travel Survey (NNTS) from 2013
to 2014 (Ellis et al., 2015). The discrepancy is likely the result of two
factors: over-segmentation of trips from the app and under-reporting of
(especially short) trips in telephone-based travel surveys like the NNTS.
The video recordings provided a means with which the route choice
changes of the GPS panel could be compared with population route
choice in an intersection. The volumes of bicycles counted on Øvrefoss
increased but not to the same degree as the GPS panel, as shown in
Table 2. This is likely a result of a combination of factors, including the
small sample size, different time periods for recording and a lower trip
rate in the GPS panel after the intervention was completed. The video
data is reliable, however, only one location is available for any re-
cording, limiting the comparison opportunities with GPS data.
The radar traffic counts on the other hand were problematic from a
data consistency perspective. The post-intervention data collection in
Markveien revealed an 83% decrease in volumes of northbound cyclists
despite the contraflow bicycle lane specifically providing for this group.
Directional data, whilst not obviously inconsistent in the two parallel
streets could not be used as a result. When considering overall volumes,
the intervention street Markveien experienced a reduction as discussed
in the results section whilst neighbouring streets experienced an in-
crease in cycling levels. Such a finding conflicts with the GPS and video
data and is likely a result of improper radar installation. The manu-
facturers of the ViaCountII device do not recommend the use of their
product where parked cars or other objects may cause reflection of the
radar beam from the opposite side of the road. In this highly urban area,
video, manual or pneumatic tube counts may have been more appro-
priate options to understand volume changes in parallel streets.
5.3. Potential other causes of variability
Before and after travel behaviour studies must be considerate of
several other confounding factors. The intervention was selected as a
natural experiment due to the absence of nearby planned bicycle in-
frastructure projects in early 2017. However as previously mentioned,
two other streets received bicycle infrastructure modifications as illu-
strated in Fig. 2. Sandakerveien was completed in late September and
was thus still under construction during the second phase of GPS data
collection, which may have led to the modest increases in bicycle vo-
lumes here (see Fig. 7). The existing bicycle lane in Toftes gate was
widened and marked red, however, this did not lead to travel behaviour
changes as substantial as the primary intervention.
Variable weather can strongly impact the modal share of bicycles
with cycling rates typically three to four times lower in the winter
months compared to the summer in Norway (Hjorthol et al., 2014). For
this study, it was a specific aim to avoid data collection during the
winter months. The public transport operator Ruter's Market Informa-
tion Survey shows that the bicycle modal share was not greatly different
between the before (8.4%) and after (7.4%) periods in Oslo as
R. Pritchard, et al. Journal of Transport Geography 77 (2019) 113–125
123
illustrated in Fig. 6. The slight difference can, however, partly explain
the reduction in corridor volumes of bicyclists observed in Table 2.
Long term effects are typically larger than short term ones, as col-
lective improvements begin to improve connectivity in the neighbour-
hood and the level of exposure to infrastructure changes increases.
Cross-sectional travel behaviour surveys commissioned by the City of
Oslo in 2013 and 2017 show that the two city districts of Sagene and
Grünerløkka had statistically significant increases in the numbers of
residents who cycled at least once per week. For Sagene, north of the
intervention area, this corresponded to an increase from 39 to 49%,
whilst for Grünerløkka, the city district containing the intervention, the
proportion of residents who used a bicycle once or more per week in-
creased from 40 to 52% (Bayer, 2018). Approximately 0.5% of the adult
population of these city districts were sampled (in 2017 this corre-
sponded to 240 of 48,158 residents in Grünerløkka and 168 of 35,377
residents in Sagene). Although a significant change in the number of
residents who regularly cycle is observed over the four-year time in-
terval –it is not possible to determine which factors had the most in-
fluence on the change using this approach.
Within the infrastructure intervention literature, follow-up periods
of up to two years are not uncommon (Smith et al., 2017). A paper
which reviewed 17 natural experiments and their impact on physical
activity revealed that studies with positive results generally had follow-
up times of > 6 months (Mayne et al., 2015). Only one of the 17 studies
reviewed had a comparable timeframe to this paper. It evaluated a 23-
mile-long multi-use trail (converted from an unused railway) in North
Carolina two months after opening and found no statistically significant
changes in the levels of physical activity or walking for transportation
amongst residents located within 2 miles of the intervention. In addi-
tion, 11% of the survey sample was not aware of the trail's presence
whilst 23% had made use of it (Evenson et al., 2005). Although the
study did not assess travel behaviour in the same manner as this paper
(using mode or route choice), it highlights that even relatively large
infrastructural changes are not noticed by the entire population. This is
supported by feedback provided at the conclusion of the study (in Oc-
tober 2017) from a small selection of the participants (n= 14) in which
8 participants reported that they had noticed the contraflow bicycle
lane installation in Markveien when prompted: ‘Did you observe any
changes in your neighbourhood between the two data collection per-
iods? If so, please describe.’
The importance of differences in context, intervention types and
follow-up timings makes it difficult to precisely determine the im-
portance of post-intervention follow-up time (Smith et al., 2017). One
study which performed two follow-ups of travel behaviour is the UK
iConnect study. The iConnect project found that residents located
within one kilometre of three selected bicycle infrastructure interven-
tion sites had increased their average weekly physical activity by
45 min after two years, a finding which was not reflected in the one-
year post-completion survey (Goodman et al., 2014). Future research
should consider adopting this approach with multiple follow-ups in
order to provide insights into short-term versus long-term effects of
bicycle infrastructure.
6. Conclusion
The aim of the study was to observe bicycle route and mode choices
in a panel of residents. A natural experiment study design was used in
which residents were recruited specifically in connection with the
construction of a contraflow bicycle lane in Oslo. The study's principal
finding is the demonstration of the route substitution effect. The study
additionally shows that the observed increase in the modal share of
bicycles was not statistically significant. Route substitution of existing
bicyclists is critically important when estimating the network impacts
of new bicycle infrastructure (change of route has a very different
meaning for the transport network than change of mode). Failing to
account for route substitution can lead to an overestimation of the
benefits of bicycle infrastructure development (since more cyclists are
estimated than are present).
The paper outlines a smartphone GPS approach to collecting in-
depth travel behaviour data from a respondent panel, however
achieving satisfactory numbers of responses was troublesome, detri-
mentally impacting the ability to assess the significance of the inter-
vention. With a panel participation rate of only 2% from the mailed
invitations, alternative means of recruitment may be necessary when
using similar approaches going forward. Natural experiments are re-
ceiving increased attention in the literature, furthering our knowledge
about the effects of specific types of bicycle infrastructure provision.
Future research efforts should attempt to compare such initiatives and
control for contextual differences where possible.
To date, existing research on the impact of bicycle infrastructure has
been mostly focussed on either mode or route change. This study con-
tributes to a small but growing body of research that maintains a hol-
istic perspective considering both mode and route factors in the eva-
luation of bicycle infrastructure over time. Future studies of this nature
will assist in bettering our understanding of how bicycle infrastructure
is utilised, assisting planners, policymakers and engineers in their ef-
forts to create safe and attractive people-focussed (rather than car-
centric) urban areas.
Acknowledgements
The authors would like to thank Siv Linette Solheim Grann at the
City of Oslo Agency for Urban Environment for her assistance with case
selection and radar counts in addition to Aliaksei Laureshyn, Torkel
Bjørnskau and Carl Johnsson from the Institute of Transport Economics
in Oslo for their assistance with video data collection.
Declaration of interest
Financial support for this study was provided by the Norwegian
Public Roads Administration (reference number 17/122038-2), the
Nordic Road Association and the City of Oslo Agency for Urban
Environment. All authors declare that they had: (1) No financial sup-
port for the submitted work from anyone other than their university
and the other funding sources listed above; (2) No financial relation-
ships with commercial entities that might have an interest in the sub-
mitted work. (3) Apart from assistance in case selection received by the
City of Oslo, none of the funding organisations listed above has been
involved in the study design, analysis or review of this study. The au-
thors declare no other conflicts of interest.
References
Bayer, S.B., 2018. IRIS Report 2018/252. Reisevaneundersøkelse for Oslo 2017 {Travel
Behaviour Survey for Oslo 2017}. International Research Institute of Stavanger,
Stavanger (Retrieved from). https://www.oslo.kommune.no/getfile.php/13314342/
Innhold/Gate%2C transport og parkering/Sykkel/Sykkelstrategier og dokumenter/
Undersøkelser og rapporter/Reisevaneundersøkelse høsten 2017.pdf.
Berger, M., Platzer, M., 2015. Field evaluation of the smartphone-based travel behaviour
data collection app “smartMo”. Transp. Res. Procedia 11, 263–279. https://doi.org/
10.1016/j.trpro.2015.12.023.
Bjørnskau, T., Fyhri, A., & Sørensen, M. W. J. (2012). TØI report 1237/2012. Sykling mot
enveiskjøring. Effekter av å tillate toveis sykling i enveisregulerte gater i Oslo.
{Contraflow Cycling. Effects of Allowing Two-Way Cycling in One-Way Streets in
Oslo}. (Retrieved September 1, 2018, from https://www.toi.no/getfile.php/
1325062/Publikasjoner/TØI rapporter/2012/1237-2012/1237-2012-elektronisk.pdf
Broach, J., 2016. Travel Mode Choice Framework Incorporating Realistic Bike and Walk
Routes (Doctoral thesis). Portland State Universityhttps://doi.org/10.15760/etd.
2698.
Bucher, D., Cellina, F., Mangili, F., Raubal, M., Rudel, R., Rizzoli, A.E., Elabed, O., 2016.
Exploiting fitness apps for sustainable mobility –challenges deploying the GoEco!
app. In: 4th International Conference on ICT for Sustainability (ICT4S 2016). Atlantis
Press (Retrieved from). https://pdfs.semanticscholar.org/16c7/
ba4702ec81529d2410ac30468ecce61cfbbe.pdf.
Buehler, R., Dill, J., 2015. Bikeway networks: a review of effects on cycling. Transp. Rev
(August 2015), 1–19. https://doi.org/10.1080/01441647.2015.1069908.
Dill, J., 2009. Bicycling for transportation and health: the role of infrastructure. J. Public
R. Pritchard, et al. Journal of Transport Geography 77 (2019) 113–125
124
Health Policy 30 (Suppl. 1), S95–S110. https://doi.org/10.1057/jphp.2008.56.
Dill, J., McNeil, N., Broach, J., Ma, L., 2014. Bicycle boulevards and changes in physical
activity and active transportation: findings from a natural experiment. Prev. Med. 69
(S), S74–S78. https://doi.org/10.1016/j.ypmed.2014.10.006.
Donald, S.G., Lang, K., 2007. Inference with difference-in-differences and other panel
data. Rev. Econ. Stat. 89 (2), 221–233. https://doi.org/10.1162/rest.89.2.221.
Douglas, D.H., Peucker, T.K., 1973. Algorithms for the reduction of the number of points
required to represent a digitized line or its caricature. Cartographica Int. J. Geogr. Inf.
Geovisualization 10 (2), 112–122. https://doi.org/10.3138/FM57-6770-U75U-7727.
Ellis, I.O., Søgnen Haugsbø, M., Johansson, M., Berglund, G., Haug, T.W., 2015. PROSAM
report 218. Reisevaner i Osloområdet. En analyse av den nasjonale
reisevaneundersøkelsen 2013/14. {Travel Behaviour in the Oslo Region. An Analysis
of the National Travel Survey 2013/2014} (Retrieved September 1, 2018, from).
http://www.prosam.org/index.php?page=report&nr=218#.
Envall, P., 2007. Accessibility Planning: A Chimera? (Doctoral Thesis). University of
Leeds (Retrieved from). http://etheses.whiterose.ac.uk/id/eprint/11279.
Evenson, K.R., Herring, A.H., Huston, S.L., 2005. Evaluating change in physical activity
with the building of a multi-use trail. Am. J. Prev. Med. 28 (2), 177–185. https://doi.
org/10.1016/j.amepre.2004.10.020.
Fitch, D., Thigpen, C., Cruz, A., Handy, S.L., 2016. Bicyclist Behavior in San Francisco: A
Before-and-After Study of the Impact of Infrastructure Investments (Retrieved
September 1, 2018, from). http://ncst.ucdavis.edu/project/ucd-ct-to-012.
Fitzhugh, E.C., Bassett, D.R., Evans, M.F., 2010. Urban trails and physical activity: a
natural experiment. Am. J. Prev. Med. 39 (3), 259–262. https://doi.org/10.1016/j.
amepre.2010.05.010.
Flügel, S., Hulleberg, N., Fyhri, A., Weber, C., Ævarsson, G., 2017. Empirical speed
models for cycling in the Oslo road network. Transportation (1), 1–25. https://doi.
org/10.1007/s11116-017-9841-8.
Goodman, A., Sahlqvist, S., Ogilvie, D., 2014. New walking and cycling routes and in-
creased physical activity: one- and 2-year findings from the UK iConnect study. Am.
J. Public Health 104 (9), e38–e46. https://doi.org/10.2105/AJPH.2014.302059.
Goodno, M., McNeil, N., Parks, J., Dock, S., 2013. Evaluation of innovative bicycle fa-
cilities in Washington, D.C. Transp. Res. Rec. J. Transp. Res. Board 2387, 139–148.
https://doi.org/10.3141/2387-16.
Handy, S., van Wee, B., Kroesen, M., 2014. Promoting cycling for transport: research
needs and challenges. Transp. Rev. 34 (1), 4–24. https://doi.org/10.1080/01441647.
2013.860204.
Heesch, K.C., James, B., Washington, T.L., Zuniga, K., Burke, M., 2016. Evaluation of the
Veloway 1: a natural experiment of new bicycle infrastructure in Brisbane, Australia.
J. Transp. Health 1–11. https://doi.org/10.1016/j.jth.2016.06.006.
Heinen, E., Harshfield, A., Panter, J., Mackett, R., Ogilvie, D., 2017. Does exposure to new
transport infrastructure result in modal shifts? Patterns of change in commute mode
choices in a four-year quasi-experimental cohort study. J. Transp. Health 6 (July),
396–410. https://doi.org/10.1016/j.jth.2017.07.009.
Hjorthol, R., Engebretsen, Ø., Uteng, T.P., 2014. TØI report 1383/2014. Den nasjonale
reisevaneundersøkelsen 2013/2014 –nøkkelrapport {2013/14 National Travel
Survey –Key Results}. Institute of Transport Economics, Oslo (Retrieved from).
https://www.toi.no/getfile.php?mmfileid=39511.
Holseter, A.M.R., 2018. Educational Attainment of the Population (Retrieved September
1, 2018, from). https://www.ssb.no/en/statbank/table/09434.
Hood, J., Sall, E., Charlton, B., 2011. A GPS-based bicycle route choice model for San
Francisco, California. Transp. Lett 3 (1), 63–75. https://doi.org/10.3328/TL.2011.
03.01.63-75.
Huber, S., Rust, C., 2016. Calculate travel time and distance with openstreetmap data
using the open source routing machine (OSRM). Stata J. 16 (2), 416–423. https://doi.
org/10.1177/1536867X1601600209.
Hull, A., O'Holleran, C., 2014. Bicycle infrastructure: can good design encourage cycling?
Urban Plan. Transp. Res 2 (1), 369–406. https://doi.org/10.1080/21650020.2014.
955210.
Krenn, P.J., Titze, S., Oja, P., Jones, A., Ogilvie, D., 2011. Use of global positioning
systems to study physical activity and the environment: a systematic review. Am. J.
Prev. Med. 41 (5), 508–515. https://doi.org/10.1016/j.amepre.2011.06.046.
Krenn, P.J., Oja, P., Titze, S., 2014. Route choices of transport bicyclists: a comparison of
actually used and shortest routes. Int. J. Behav. Nutr. Phys. Act. 11 (1), 7. https://doi.
org/10.1186/1479-5868-11-31.
Lechner, M., 2010. The estimation of causal effects by difference-in-difference methods.
Found. Trends Econ 4 (3), 165–224. https://doi.org/10.1561/0800000014.
Lott, D.F., Tardiff, T., Lott, D.Y., 1978. Evaluation by experienced riders of a new bicycle
lane in an established bikeway system. Transp. Res. Rec. J. Transp. Res. Board 683,
40–46. (Retrieved from). http://www.john-s-allen.com/research/davis_studies/Lott,
Tardiff, and Lott.pdf.
Loveday, A., Sherar, L.B., Sanders, J.P., Sanderson, P.W., Esliger, D.W., 2015.
Technologies that assess the location of physical activity and sedentary behavior: a
systematic review. J. Med. Internet Res. 17 (8). https://doi.org/10.2196/jmir.4761.
Mayne, S.L., Auchincloss, A.H., Michael, Y.L., 2015. Impact of policy and built environ-
ment changes on obesity-related outcomes: a systematic review of naturally occurring
experiments. Obes. Rev. 16 (5), 362–375. https://doi.org/10.1111/obr.12269.
Mertens, L., Compernolle, S., Deforche, B., Mackenbach, J.D., Lakerveld, J., Brug, J., ...
Van Dyck, D., 2017. Built environmental correlates of cycling for transport across
Europe. Health Place 44, 35–42. https://doi.org/10.1016/j.healthplace.2017.01.007.
Morrison, D.S., 2004. Evaluation of the health effects of a neighbourhood traffic calming
scheme. J. Epidemiol. Community Health 58 (10), 837–840. https://doi.org/10.
1136/jech.2003.017509.
Newson, P., Krumm, J., 2009. Hidden Markov map matching through noise and sparse-
ness. In: Proceedings of the 17th ACM SIGSPATIAL International Conference on
Advances in Geographic Information Systems –GIS ’09, pp. 336–343. https://doi.
org/10.1145/1653771.1653818.
Nielsen, T.A.S., Olafsson, A.S., Carstensen, T.A., Skov-Petersen, H., 2013. Environmental
correlates of cycling: evaluating urban form and location effects based on Danish
micro-data. Transp. Res. Part D: Transp. Environ. 22, 4044. https://doi.org/10.1016/
j.trd.2013.02.017.
Parker, K.M., Rice, J., Gustat, J., Ruley, J., Spriggs, A., Johnson, C., 2013. Effect of bike
lane infrastructure improvements on ridership in one New Orleans neighborhood.
Ann. Behav. Med. 45 (S1), 101–107. https://doi.org/10.1007/s12160-012-9440-z.
Pritchard, R., 2018. Revealed preference methods for studying bicycle route choice—A
systematic review. Int. J. Environ. Res. Public Health 15 (3), 1–30. https://doi.org/
10.3390/ijerph15030470.
Project OSRM, 2018. Open Source Routing Machine Application Programming Interface
Documentation v5.15.2 (Retrieved September 1, 2018, from). http://project-osrm.
org/docs/v5.15.2/api/#match-service.
Rissel, C., Greaves, S., Wen, L.M., Crane, M., Standen, C., 2015. Use of and short-term
impacts of new cycling infrastructure in inner-Sydney, Australia: a quasi-experi-
mental design. Int. J. Behav. Nutr. Phys. Act. 12 (1), 129. https://doi.org/10.1186/
s12966-015-0294-1.
Romanillos, G., Zaltz Austwick, M., Ettema, D., De Kruijf, J., 2016. Big data and cycling.
Transp. Rev. 36 (1), 114–133. https://doi.org/10.1080/01441647.2015.1084067.
Saelens, B.E., Sallis, J.F., Frank, L.D., 2003. Environmental correlates of walking and
cycling: findings from the transportation, urban design, and planning literatures.
Ann. Behav. Med. 25 (2), 80–91. https://doi.org/10.1207/S15324796ABM2502_03.
Sandsjö, L., Sjöqvist, B.A., Candefjord, S., 2015. A concept for naturalistic data collection
for vulnerable road users using a smartphone-based platform. In: International
Technical Conference on the Enhanced Safety of Vehicles (ESV). Gothenburg,
Sweden, pp. 6. (Retrieved from). http://www-esv.nhtsa.dot.gov/Proceedings/24/
isv7/main.htm%0A.
Schneider, R.J., Stefanich, J., 2015. Neighborhood characteristics that support bicycle
commuting. Transp. Res. Rec. J. Transp. Res. Board 2520 (2520), 41–51. https://doi.
org/10.3141/2520-06.
Smith, M., Hosking, J., Woodward, A., Witten, K., MacMillan, A., Field, A., ... Mackie, H.,
2017. Systematic literature review of built environment effects on physical activity
and active transport –an update and new findings on health equity. Int. J. Behav.
Nutr. Phys. Act. 14 (1), 158. https://doi.org/10.1186/s12966-017-0613-9.
Stappers, N.E.H., Van Kann, D.H.H., Ettema, D., De Vries, N.K., Kremers, S.P.J., 2018. The
effect of infrastructural changes in the built environment on physical activity, active
transportation and sedentary behavior –a systematic review. Health Place 53 (July),
135–149. https://doi.org/10.1016/j.healthplace.2018.08.002.
Troelsen, J., Jensen, S.U., Andersen, T., 2004. Evaluering af Odense –Danmarks
Nationale Cykelby {Evaluation of Odense –Denmark's National Cycle City}. Odense
Kommune (Retrieved from). http://arkiv.cykelviden.dk/filer/cykel_inet.pdf.
Vaage, O.F., 2018. Norwegian Media Barometer 2017 (Retrieved September 1, 2018,
from). https://www.ssb.no/kultur-og-fritid/artikler-og-publikasjoner/_attachment/
346186?_ts=162d7feae58.
van Goeverden, K., Nielsen, T.S., Harder, H., van Nes, R., 2015. Interventions in bicycle
infrastructure, lessons from Dutch and Danish cases. Transp. Res. Procedia 10 (July),
403–412. https://doi.org/10.1016/j.trpro.2015.09.090.
van Wee, B., Börjesson, M., 2015. How to make CBA more suitable for evaluating cycling
policies. Transp. Policy 44, 117–124. https://doi.org/10.1016/j.tranpol.2015.07.
005.
Vasilev, M., Pritchard, R., Jonsson, T., 2018. Trialing a road lane to bicycle path
redesign—Changes in travel behavior with a focus on users' route and mode choice.
Sustainability 10 (12), 1–18. https://doi.org/10.3390/su10124768.
Via Traffic Controlling GMBH, 2016. ViaCount II Specifications (Retrieved September 1,
2018, from). https://www.viatraffic.de/fileadmin/viatraffic-content/downloads/
katalog2016/en/viatraffic_2016_GB_viacountII.pdf.
Wahlgren, L., Schantz, P., 2014. Exploring bikeability in a suburban metropolitan area
using the active commuting route environment scale (ACRES). Int. J. Environ. Res.
Public Health 11 (12), 8276–8300. https://doi.org/10.3390/ijerph110808276.
Wilmink, A., Hartman, J.B., 1987. Evaluation of the Delft bicycle network plan. Final
summary report. In: Ministry of Transport and Public Works. Transportation and
Traffic Engineering Division, The Hague, Netherlands.
Yang, L., Sahlqvist, S., McMinn, A., Griffin, S.J., Ogilvie, D., 2010. Interventions to pro-
mote cycling: systematic review. BMJ 341 (oct18 2), 1–10. https://doi.org/10.1136/
bmj.c5293.
R. Pritchard, et al. Journal of Transport Geography 77 (2019) 113–125
125
