Content uploaded by Hebenstreit Cornelia
Author content
All content in this area was uploaded by Hebenstreit Cornelia on Jun 27, 2018
Content may be subject to copyright.
Hebenstreit & Fellendorf / TRA2018, Vienna, Austria, April 16 -19, 2018
1
Proceedings of 7th Transport Research Arena TRA 2018, April 16-19, 2018, Vienna, Austria
Multi- and intermodal Trip Chain Simulation for individual daily
Routines, using Bicycles
Cornelia Hebenstreita, Martin Fellendorfa
a Graz University of Technology, Institute of Highway Engineering and Transport Planning, Rechbauerstraße 12, 8010 Graz, Austria
Abstract
In recent years, the funding of bicycle traffic especially in European countries has increased substantially.
Nevertheless, bicycles are often not depicted in detail within traffic models. Especially for bicycle traffic, research
just started to contribute better understandings of spatio-temporal distribution of cycle trips. Without realistic
bicycle models, neither the initial state can be consistently evaluated, nor can an assessment of the effectiveness
measures be carried out. Within this paper, we focus on a framework for routing and utility calculation of bicycle
trips within agent-based simulation environment. Therefore, we consider influencing factors, coming from
environment or infrastructure side. As bicyclists do not necessarily take the shortest path, we developed an edge
weight calculation to feed the Dijkstra Algorithm. With it a unique route for each person is found. Every person
perceives a utility for the chosen route, dependent on the conceived friendliness of cycling. In further consequence
this paper represents the base work for the simulation of e-bike sharing and intermodal trips involving bicycles.
Keywords: Bicycle Traffic; Bicycle Routing; Utility-Function; agent-based modelling
Hebenstreit & Fellendorf / TRA2018, Vienna, Austria, April 16 -19, 2018
2
1. Introduction
Plenty of traffic, little of space, especially in cities: To maintain a high standard of mobility, bicycle traffic, as well
as intermodal trips get more and more important in urban areas all over the world. However, especially for bicycle
traffic, research just started to contribute better understandings of spatio-temporal distribution of cycle trips. To
know when, where, how many and which cyclists are on the road is important for various applications. Therefore
this paper focusses on the development of a simulation tool which includes bicycle flows and bike sharing. The
work described here forms the foundation for the future implementation of intermodal trips. This work is an
ongoing development of the open source, agent based simulation tool MATSim.
The objective is to route and score bicycles realistically and in further consequence to include intermodal trips.
Especially intermodal trips involving bicycles are rarely known in detail and get often neglected. Further, they can
strengthen the environmental alliance and bring positive effects for the city’s traffic. Depicting bicycle traffic
within a model, can bring new information about distribution, bicycle traffic volumes as well as weaknesses and
strengths within the bicycle network, which is usually just imprecisely or selectively known. Bicycle traffic gets
promoted and an increase is aspired not only in Austria, due to the fact that bicycles are ecofriendly, quiet, health
enhancing and space-saving, just to name a few benefits. Moreover with e-bikes, speed and scope can be increased
and new users are addressed. Because the potential user, scope and speed widen, bicycle traffic should be included
within multimodal or intermodal mobility analysis. E.g. in Austria, bicycle traffic of cities holds a share of about
60% of the total bicycle trips. This means more attention should be given to bicycle traffic especially within cities.
To reproduce multimodal transport systems, bicycle traffic needs to be modelled precisely. In most simulation
tools bicycle traffic is not realistically implemented. Within this paper, we focus on the integration of bicycle
traffic within agent-based transport models. Beforehand the essential needs for bicycle simulation are gathered by
reviewing numerous research literature and conducting a comparative survey in Graz, the second largest city of
Austria.
2. Literature Review and Survey
In Austria, Sweden, Canada and the USA researchers established an index, called “bikeability”, where the aptitude
of the surrounding for cycling is used to describe bicycle friendly or unfriendly zones. In further consequence this
index is used as basis e.g. for new cycle way planning. To calculate the zones, different categories of environmental
features like safety, attractiveness of the surrounding, land use and infrastructure are analyzed. (Hoedl et al. 2010,
Winters et al. 2010) Although the bikeability-index shows areas where people would prefer to bicycle, the index
does not provide real bicycle traffic behavior data and cannot measure or predict real usage. The categories, and
category members, included in the bikeability index were investigated by other researchers for bicycle research as
well.
Safety matters a lot for cyclists. Four out of five persons are anxious about traffic accidents while riding their bike.
Street lighting is seen as essential to increase safety (Sener et al. 2009, Menghini et al. 2010), nevertheless within
our own survey evaluation, street-lighting was not strongly decisive for bicycle route choice in Graz. This might
be due to the fact that in Graz street-lighting is existing area-wide. Other research claimed additional factors which
influence the route choice of bicyclists such as land use, characteristics of motorized traffic (Segadilha & Sanches
2014, Handy & Clifton 2001, Ortúzar et al. 2000, Parkin et al. 2008), population density, housing or employment,
street connectivity (Saelens et al. 2003, Crane 2000), cycle infrastructure (Boufous et al. 2017, Petritsch et al.
2006, Shankwiler 2006, Broach et al. 2012, Snizek et al. 2013), pavement conditions (Noland & Kunreuther 1995,
Landis et al. 1997, Stinson et al 2004) or a combination of safety, comfort and speed (Broach et al. 2012, Boufous
et al, 2017). In Atlanta, USA, it was found that female and senior cyclists more likely take longer routes and the
deviation from the shortest route depends on presence of cycle infrastructure, bicycle friendly streets, speed limits,
confidence and comfort (Misra & Watkins 2017). For Graz, Austria, within two independent surveys we found,
that driving speed, age and fitness are most important for the importance level of the different route choice
attributes.
In the first survey, 48 persons participated and recorded their daily bike-trips for two weeks with the cell-phone
app “Bike Citizens Fahrrad Navi GPS”. Further they filled out a bicycle trip diary and a questionnaire. Within this
survey, route choice, health and fitness attributes were queried. Fitness was retrieved from the body-mass-index
as well as the people’s self-evaluation. Speed was analyzed through the recorded GPS data. Unfortunately mainly
people aged between 20 and 35 participated within this study. Therefore we surveyed further 103 persons once-
only, with ages fitting better to Graz’ populations age distribution. No matter which study or person-type we
analyzed, the most important attributes for route choice was a fluid cycling route (and rapid progression), followed
by detour-poorness. Generally, low traffic volumes of motorized traffic are more desirable, than low speeds.
Nevertheless for slow or unfit bicyclists low speed of motorized traffic is seen almost as important as low traffic
volumes, while bicycling in mixed traffic. Using the self-evaluation of fitness brought better results than using the
Hebenstreit & Fellendorf / TRA2018, Vienna, Austria, April 16 -19, 2018
3
body-mass-index. Moreover bike paths, safety and storage sidings were generally also quite important. Other
criteria we requested (space for overtaking, low gradients, low effort, illumination, avoidance of streetcar rails,
width of cycle infrastructure or road, low conflict potential to pedestrians or other road users) were mentioned as
important, but not as strongly. The most significant differences were found between gender, slow and fast
bicyclists, fit and unfit ones or persons aged over 50/being retired and bicyclists of other ages. E.g. males tend
more to avoid streetcar rails, but illumination is less important than for females. Other route-choice-factors were
almost equally rated by each gender. Bicyclists of age 50 or older, or retired ones rated e.g. illumination and green
areas equally important as a good maintenance of infrastructure and fluid cycling (see Figure 1, b and d). Males
and females ranked attributes in nearly the same sequence, while females quote a higher importance level of all
attributes (Figure 1, a). Both studies clarify, that it would be inappropriate to assume that every bicycling person
behaves equally. Also literature showed that practiced and less practiced individuals behave differently (Miosra &
Watkins 2017). Fast cyclists avoid e.g. detours and denote a smooth surface as well as overtaking possibilities as
most important factors for their route choice, subsequently to rapid progression and fluid cycling.
a) Total and Gender,
total= 120, female = 54, male = 67
b) Professional Life
Retiree=7, Job-seeking=14, Student=44,
Employed=53
c) Education
A-Levels=63, University=28, MainSchool=29
d) Age Pattern
<19=19, 20-29=48, 30-39=25, 40-49=12, 50+=10
Figure 1: Onetime survey – Gra z, n = 120 (Average value over all persons of the group, Question: How important are the following attributes
for your route choice? 1 unimportant, 2 rather unimportant, 3 rather important, 4 important, this means average scores over 2 .5 can be seen as
desired route choice attributes
3. Methodology
Since the results of both surveys and literature showed that there are certain route choice differences within user
groups, we implemented an approach which allows different and adjustable routing preferences. The objective was
to choose bicycle routes differently, so we decided to use user groups. Consequently we developed a person
specific edge weight calculation. Therefore we considered bikeability (friendliness of bicycling) and speed values,
as the survey stated different preferences for persons bicycling at different speeds. This means, with our person
Hebenstreit & Fellendorf / TRA2018, Vienna, Austria, April 16 -19, 2018
4
specific edge weight calculation, the Dijkstra Algorithm will reveal different routes, not only for every user group,
but also for different speeds. With this educated guess, the best possible route for each agent is found individually.
As the network must be routable, a route is always found, regardless of whether there are good or poor cycle
conditions. An equally long route in terms of time, must not be equally attractive. Routes with bicycle friendly
environment and good infrastructure need to be regarded more positive, than routes with heavy traffic and no
bicycle infrastructure. Nevertheless, each and every agent is considered individually with its own settings, as route
choice preferences vary.
As mentioned before, there are different individual needs and attitudes, which should also be considered for the
evaluation of the chosen route. Therefore every route gets its own score (utility). Scores are calculated particularly
for every individual, taking chosen routes, user groups and specific speed into account. Fast and experienced
cyclists may take routes where no bike infrastructure exists, as long as they are fast. Slow bicyclists tend to choose
safe routes and feel uncertain within mixed traffic segments. This means such roads should be scored increasingly
negative (high disutility) for slow bicyclists, while fast ones will perceive such routes as adequate. After we
finished speed drawing, user group definition, routing and scoring, which we introduce within this paper, we are
now working on the integration of e-bike sharing as well as on intermodal trips, involving public transport, bike
sharing and bicycle traffic.
3.1. Gathering data for cruise speed verification
As speed values are important for routing and scoring, we explain the speed assignment in this section. An Austrian
and German survey analysed free flow and journey speed, respectively (Jellinek et al. 2013, Schleinitz et al. 2015).
Within the Austrian survey the bicycle speed was measured during uninfluenced movement for the bike-types
conventional bike, e-bike and racing bike. The average speed of conventional bikes was 17.7 km/h, (e-bikes: 20.0
km/h, racing bikes: 24.0 km/h) (Jellinek et al. 2013).
Figure 2: Probability density functions for three bicycle types based on empirical data of Jellinek et al. 2013 (=Survey). Speed as input for the
model will be reduced by 1.5 k m/h due to results for average trip speed of the GPS-tracks-analysis of Graz (=Model).
The results of the German study (Schleinitz et al. 2015) reveal an average journey speed of 15.3 km/h on
conventional bicycles and 17.4 km/h on Pedelecs respectively. Both study results were verified with data of the
Graz survey. As they fit quite well, we used the results of the Austrian free flow measurements and superimposed
them with a normal distribution function (see Figure 2).
Stops and traffic signals were not considered, therefore the mean travel speed (free flow) was reduced by 1.5 km/h
to fit the overall journey speed. After this reduction the speed matches the results of the GPS-track analysis of the
Graz study where we tracked 48 persons, riding conventional bikes. The calculated trip speeds of all recorded trips
reached a median of 16.2 km/h (see Figure 3). Taking only the slowest, median or fastest recorded track of each
person, a median speed of 9.73 | 15.9 | 21.1 km/h (successively) was reached.
The tracks included waiting times at traffic signals or interaction with other road users. With the normal
distribution and assigned bike type, we are able to draw the desired speed for every person (vp).
Hebenstreit & Fellendorf / TRA2018, Vienna, Austria, April 16 -19, 2018
5
Figure 3: Boxplot of the minimum, median, maximum and total speed (cruise speed) of the GPS-track analysis (n = 48, recording during 2
weeks, Graz, Austria). The average trip speed according to this, lies between 15.9 (median of median per person) and 16.2 km/h (median of
all recorded trip).
3.2. Routing and edge-weight calculation
MATSim, the agent based transport simulation, applies the well-known Dijkstra algorithm to solve the shortest
path problem, and so we do. The goal is to route bicyclists according to literature and survey results. Therefore,
we established a generalized cost function and introduced an edge-weight ,, specific for bicyclists (since they
do not only decide on trip distance ∑ or travel time). This edge-weight is used by the Dijkstra Algorithm, to
solve the shortest path problem. The edge weight ,, of each link (a link is an edge), comprises four link
attributes (see Formula 1):
,
,
=
,
∗
,
(
+
)
(1)
where , is a bikeability, a link () or person () specific speed and is the length of the link.
Prospectively we want to include , a link disutility considering noise and pollution, which uses the travel
volumes of motorized traffic to calculate a noise and pollution disutility. Thus, the applied generalized cost
function considers travel time (/), link quality (,) and environmental influences (). A regular
navigation network is supplemented by these link attributes, if bicyclists are allowed to travel on a link. Motorways
e.g. need to be excluded, but we admit bicyclists to use footpaths with very low speed (walking speed), as bicyclists
may walk a bike. The bikeability
,
=
∗
,
=
∗
,
+
∗
,
+
∗
,
+
∗
,
+
∗
,
(2)
of a link and user group is determined by various attribute categories (gradient (), travel safety (),
comfort (), environment (), additional ()) (see Formula 2). ∈ [0,10] defines the friendliness of cycling for
each category. 0 indicates the most attractive cycle conditions while 10 shows the poorest conditions. The set of
all bicyclists is classified in a limited, arbitrary number of groups with similar behaviour and perception of link
attractiveness, called user group (). Our objective at this time is the model-reproduction and not the model-
development. The values of each category are multiplied with the share of each attribute, which is specific
to every user group. defines the importance of every , ∑, has to be 1 (or 0, with it only speed and
length are taken into account).
The bikeability , captures the aptitude of the surrounding and infrastructure for cycling for every link and user
group. Once the edge weight ,, was calculated, the well-known Dijkstra-Algorithm is fed with it and solves
the shortest path problem, resulting in different routes for different persons due to specific speed and user groups
(specific bikeability).
The link travel time is subject to speed (travel time = distance / speed). If the infrastructure allows only slow
cycling the route choice is influenced, especially for the fast bicyclists. This may happen e.g. in pedestrian zones
and on narrow bike lanes shared by pedestrians. The link speed is , = min(,). is the personal desired
speed and is the speed assumed to a link ().
Hebenstreit & Fellendorf / TRA2018, Vienna, Austria, April 16 -19, 2018
6
3.3. (Dis)Utility calculation for travelling by bike
Before the utility calculation is explained in detail, some program specific information will be given here. For
simulation, we use the open source program MATSim, which runs multiple iterations until a utility-equilibrium is
reached (Kickhöfer 2009). In general, MATSim uses positive utilities for performing activities and negative
utilities for travelling. This means travelling is seen as a penalty. As shown in Formula 3, the terms are multiplied
and agents try to reach the highest possible utility.
,
=
+
+
+
(3)
Scores of a full agent plan consist of positive utilities for performing activities , negative utilities for
being late or for waiting and of negative utilities for travelling . For bicycle scoring, the part
of is important. Formula 4 shows the utility of travelling in general:
,
=
∗
,
(4
)
where is the marginal, normally negative utility of travelling and , is the travel time in hours (Horni et
al. 2016, Kickhöfer 2009).
As the utility of travelling is generally negative (penalty), it is also called disutility. Kickhöfer (2009) defined
utility functions for car, public transport and walk in detail. For bicycle traffic the monetary costs ( * ) are
zero. For bike sharing, they must reflect the costs of the bike sharing system. For bicycle traffic, it would be
inappropriate to use the standard function (formula 4) or functions of Kickhöfer for a realistic behavior. By using
the real travel time, bad cycling conditions would not yield a smaller value than good ones. This means for two
persons, having an equally long cycle trip in terms of time, the score would be the same even though varying cycle
conditions are present. Therefore we introduced the felt travel time ,, which takes the bike attributes of all
travelled links of a route into account. If a link has good cycle conditions, the felt travel time will be smaller than
the real travel time, resulting in a lower disutility (higher utility). Vice versa, bad cycle conditions result in a higher
disutility. The felt travel time is calculated by
,
=
∗
ℎ
∗
,
(5)
where ℎ depends on the characteristics of every traveled link () (see Formula 5). This means a route
gets scored with the help of the felt travel time, keeping original departure or arrival times, for activity scoring. If
a route has just short parts of worse cycle conditions, good scores can be achieved, nevertheless slow bicyclists
will in general reach poorer scores than fast ones.
4. First Results
The bicycle edge-weight calculation and routing was proved with several case examples. Therefore we established
different small networks to test the route choice of the agents within the network. One test case is given here in
detail. We defined four user groups, and three bike types, using the survey results. For every bike type
(conventional, e-bike and racing bike) we threw the speed using the normal distribution (Figure 2). This speed was
used for user group allocation. With the different speeds, user groups and network attributes (see Figure) agents
trustworthy choose their best routes.
For all test scenarios, at this time simplified attribute estimations were conducted. E.g. we valued each attribute
category for each link () and user group () assuming that every link, having the same bicycle infrastructure
type, achieves equal attribute rating. This means we did not use width, surface, or state of maintenance, which
would result in various bikeability values for the same bicycle infrastructure type (which would be more realistic).
Hebenstreit & Fellendorf / TRA2018, Vienna, Austria, April 16 -19, 2018
7
Figure 4: Simple test network with the simulated network attributes. All persons (= 100 agents) use their bicycle to get from Home (H) to
Work (W). They had four different options where they could choose from. For the link attributes 0 indicates the best possible cycle
conditions, 10 the poorest. This means a value of 5 shows average cycle conditions, which would not bring and increase or decrease of travel
time for the felt travel time calculation. For average cycle conditions, the real travel time gets used.
Share of User Groups
and Bike Types in %
Share of User Groups in %
Share of Bike Types in %
Figure 5: Agent distribution for the use-case of figure 4. There were 4 different user groups defined. For user group definition the bike-type
as well as the speed values of desired trip speed were taken into account. Also the shares for each attribute group is stated . (p = share, a =
additional, g = gradient, e = environment, s = safety, c = comfort)
Depending on the user group type and the previously drawn speed, different routes will be considered as least cost
path. Therefore attributes describing the bicycle conditions for every link need to be calculated carefully. For the
given example, using the input shown in Figure 4 and 5 the following routes were chosen by the agents (see Figure
6). As shown, two thirds of the agents took the bike path, while the other third divides almost equally to pedestrian
zones, shared-use route and bus lane. Further it can be seen, that mainly agents of user group 1 (faster bicyclists,
who, in our case, do not care too much about safety or comfort) took the bus lane or the shared use path, while
mainly agents of user group 3 (slow bicyclists, who want to bicycle safe and within a nice environment) chose the
pedestrian zone and bike path.
This means the edge weight calculation and Dijkstra works stable and infrastructure is taken into account,
respectively to a person (user group and speed dependent). E.g. taking just the shortest path would exclude using
the bike path, because it is 100 meter longer, which in our case was most frequently seen as the best path.
Hebenstreit & Fellendorf / TRA2018, Vienna, Austria, April 16 -19, 2018
8
User
group 1
User
group 3
All u
ser
groups
Figure 6: Chosen Routes of the User groups 1|3 and all users in total in percent. This figure shows the finally chosen routes of the in Figure 4
and 5 represented test scenario. This means the bike path is chosen most often, because it has the best cycle conditions, although its length is
slightly longer than all other choice options. Only agents of group 1 took the shorter route, as they mainly want to be fast. Bicyclists having a
lower speed (vp) than the rideable speed of the shared use path, chose this one. Faster cyclists of user group 1 chose the bus lane. In this
example, in total about 2/3 chose the bike path, while the other third divides onto the other three types.
After a route has been chosen, every person gets its score, again dependent onto attributes, speed and user group.
This means, as already explained, not only travel time is taken into account. A route getting good scores for one
agent, can also be scored poorly for another one, resulting in mode changes.
5. Further Work and Outlook
The work presented in this paper allows to simulate bicycles. Distribution within the bicycle network was
generated through the use of bicycle user group, individual speed and ambient attributes. For evaluation of a found
route, bicycle user groups speed and ambient attributes were used to derive a felt travel time. With it routes were
scored, to declare if this route is cycle friendly or not. If a route has just short parts of worse cycle conditions, good
scores can be achieved, nevertheless slow bicyclists will in general reach poorer scores than fast ones. These steps
were necessary to simulate daily trip chains within intermodal context. Therewith we will analyze the effects of
good or poor bicycle infrastructure onto bicycle volumes and mode choice.
However, in future, we want to analyze influences on/of bike sharing usage, e.g. if right or wrong siting has effects
not only on bike usage, but also on the use of public transport. At the moment we work on the implementation of
(e-)bike sharing. Therefore we use the foundations (routing and scoring) presented within this paper. Modelling e-
bike sharing is a bit more complex than conventional bike sharing, because the bikes must not only be rented and
returned, they also need to be charged and discharged.
Our objective is to develop a universally operational framework, which can also handle different bicycle models
(battery types). Therefore we use an adaptable and customizable input file for the bike sharing-bicycles and
stations. The input file contains the parameters which are needed for the charging/ discharging and rental process.
The single methods (renting, returning, charging, and discharging) are already existing, but not fully integrated
into the simulation. At the moment only conventional bike sharing can be fully operated. Additionally, after
finalization of bike- and e-bike sharing we will involve intermodal trips, focusing on the combination of public
transport and bicycle traffic. In a first step we focus mainly on methodology and a properly working simulation.
In a second step we need to proof our implementations also with larger applications, where we will look at different
scenarios for the city of Vienna.
6. Acknowledgement
The work presented within this paper are partly developed within the projects (R)adOmnes, MatchSim and FamoS,
supported by the Austrian Ministry for Transport, Innovation and Technology (BMVIT) within the national
funding program “Mobilität der Zukunft”.
Hebenstreit & Fellendorf / TRA2018, Vienna, Austria, April 16 -19, 2018
9
7. Literature
Boufous S; Hatfield J; Grzebieta R, 2017: The impact of environmental factors on cycling speed on shared paths.
Broach J.; Dill J.; Gliebe J., 2012: Where do cyclists ride? A route choice model developed with revealed
preference GPS data. Transportation Research Part A, n46, pp. 1730–1740.
Crane R.; 2000: The influence of urban form on travel: an interpretive review. J. Plan. Literature 15(1). 3-23.
Handy S. L.; Clifton K. J.; 2001: Local shopping as a strategy for reducing automobile travel. Transportation 28
(4). 317-346.
Hoedl S.; Titze S.; Oja P., 2010: The Bikeability and Walkability Evaluation Table: Reliability and Application,
American Journal of Preventive Medicine, 39(5):457-459.
Horni, A.; Nagel, K. and Axhausen, K.W. (eds.) 2016 The Multi-Agent Transport Simulation MATSim. London:
Ubiquity Press. DOI: http://dx.doi.org/10.5334/baw. License: CC-BY 4.0
Jellinek R.; Hildebrandt B.; Pfaffenbichler P.; Lemmerer H., 2013, Auswirkungen der Entwicklung des Marktes
für E-Fahrräder auf Risiken, Konflikte und Unfälle auf Radinfrastrukturen, Forschungsarbeiten des
österreichischen Verkehrssicherheitsfonds, Band 019.
Kickhöfer B., 2009: Die Methodik der ökonomischen Bewertung von Verkehrsmaßnahmen in
Multiagentensimulationen, Master Thesis, Technical University Berlin, Berlin.
Landis B. W.; Vattikuti V. R.; Brannick M. T., 1997: Real-time human perceptions toward a bicycle level of
service. Transportation Research Record n1578, pp. 119-126.
Menghini G.; Carrasco N.; Schüssler N.; Axhausen K.W., 2010: Route choice of cyclists in Zurich. Transportation
Research Part A, vol 44, pp. 754-765.
Misra A.; Watkins K., 2017: Modeling Cyclists‘ Willingness to Deviate from Shortest Path Using Revealed
Preference Data; TRB 2017
Noland, R. B.; Kunreuther H., 1995: Short-run and long-run policies for increasing bicycle transportation for daily
commuter trips. Transport Policy, pp. 67–79.
Ortúzar J.; Iacobelli A.; Valeze C., 2000: Estimating demand for cycle-way network. Transportation Research Part
A, 34(5). pp. 353-373.
Parkin J.; Wardman M.; Page M., 2008: Estimation of the determinants of bicycle mode share for the journey to
work using census data. Transportation 35 (1). 93-109.
Petritsch T. A.; Landis B.; Huang H.; Challa S., 2006: Sidepath Safety Model: Bicycle Sidepath Design Factors
Affecting Crash Rates. Transportation Research Record, n1982-25.
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. Annals of Behavioral Medicine 25 (2). 80-92.
Schleinitz, K.; Petzoldt, T.; FrankBartholdt, L.; Krems, J.F.; Gehlert, T, 2015: The German Naturalistic Cycling
Study - Comparing cycling speed of riders of different e-bikes and conventional bicycles. Safety Science.
dx.doi.org/10.1016/j.ssci.2015.07.02
Segadilha A. B. P.; Sanches S. d. P., 2014: Identification of factors that influence Cyclists’ route choice, Social
and Behavioral Sciences vol 160, pp.372- 380.
Sener I. N.; Eluru N.; Bhat C. R., 2009: Who are Bicyclists? Why and how much are they bicycling? Transportation
Research Record 2134, pp. 63-72.
Shankwiler, K. D., 2006: Developing a framework for behavior assessment of bicycling commuters: a cyclist-
centric approach. School of Industrial Design, Atlanta: Georgia Institute of Technology. pp. 87.
Snizek B.; Nielsen T. A. S.; Skov-Petersen H., 2013: Mapping bicyclist’s experiences in Copenhagen, Journal of
Transport Geography, Volume 30, June 2013, p227-233.
Stinson M.A.; Bhat C. R., 2004: Frequency of Bicycle Commuting: Internet-Based Survey Analysis.
Transportation Research Record: Journal of the Transportation Research Board, nº1878, pp.122-130.
Winters, M.; Brauer, M.; Setton E.M. et al., 2010: Built Environment Influences on Healthy Transportation
Choices: Bicycling versus Driving, J Urban Health 87: 969. https://doi.org/10.1007/s11524-010-9509-6