Estimation and validation of hybrid choice models to identify the role of perception in the choice to cycle

Abstract

Cycling is one of the most sustainable and ecofriendly modes of travel and a good form of exercise. Many government and public health authorities recommend cycling to stay fit as well as to reduce air and noise pollution, CO2 emissions, traffic congestion, and other negative consequences of car use. In light of these benefits, a major challenge for researchers today is how to promote cycling. However, in countries where cycling is not common, apart from the need for proper cycling facilities, one major issue concerns people’s perception of cycling for sport or recreational activities rather than as a mode of transport. The aim of this paper is to explore the role of perception in the likelihood of the bike being used for utilitarian purposes. We focus on the perception of: the bicycle as a means of transport; bikeability (in terms of usefulness and safety) and of bike infrastructure. Hybrid Choice Models (HCMs) have been used to estimate the effect of people’s perception on the propensity to bike. The HCM also accounts for the serial correlation between error terms in the discrete and latent perceptions, to allow for agent-common unknown factors. Furthermore, we also validate the model results using a hold-out sample and discuss some policy measures aimed at changing travel behavior. The results suggest that, besides individual characteristics, latent aspects related to the perception of the context and of the bicycle as a means of transport strongly affect the propensity to cycle.

Full text

Full Terms & Conditions of access and use can be found at
http://www.tandfonline.com/action/journalInformation?journalCode=ujst20
International Journal of Sustainable Transportation
ISSN: 1556-8318 (Print) 1556-8334 (Online) Journal homepage: http://www.tandfonline.com/loi/ujst20
Estimation and validation of hybrid choice models
to identify the role of perception in the choice to
cycle
Eleonora Sottile, Benedetta Sanjust di Teulada, Italo Meloni & Elisabetta
Cherchi
To cite this article: Eleonora Sottile, Benedetta Sanjust di Teulada, Italo Meloni & Elisabetta
Cherchi (2018): Estimation and validation of hybrid choice models to identify the role of
perception in the choice to cycle, International Journal of Sustainable Transportation, DOI:
10.1080/15568318.2018.1490465
To link to this article: https://doi.org/10.1080/15568318.2018.1490465
Published online: 05 Sep 2018.
Submit your article to this journal
View Crossmark data

ORIGINAL ARTICLE
Estimation and validation of hybrid choice models to identify the role of
perception in the choice to cycle
Eleonora Sottile
a
, Benedetta Sanjust di Teulada
a
, Italo Meloni
b
, and Elisabetta Cherchi
c
a
CRiMM –Centre of Research of Mobility Models, University of Cagliari, Cagliari, Italy;
b
Dipartimento di Ingegneria Civile, Ambientale e
Architettura, University of Cagliari, Cagliari, Italy;
c
Dipartimento di Ingegneria Civile, Ambientale e Architettura, Newcastle University,
Newcastle, UK
ABSTRACT
Cycling is one of the most sustainable and ecofriendly modes of travel and a good form of exer-
cise. Many government and public health authorities recommend cycling to stay fit as well as to
reduce air and noise pollution, CO
2
emissions, traffic congestion, and other negative consequences
of car use. In light of these benefits, a major challenge for researchers today is how to promote
cycling. However, in countries where cycling is not common, apart from the need for proper
cycling facilities, one major issue concerns people’s perception of cycling for sport or recreational
activities rather than as a mode of transport. The aim of this paper is to explore the role of per-
ception in the likelihood of the bike being used for utilitarian purposes. We focus on the percep-
tion of: the bicycle as a means of transport; bikeability (in terms of usefulness and safety) and of
bike infrastructure. Hybrid Choice Models (HCMs) have been used to estimate the effect of peo-
ple’s perception on the propensity to bike. The HCM also accounts for the serial correlation
between error terms in the discrete and latent perceptions, to allow for agent-common unknown
factors. Furthermore, we also validate the model results using a hold-out sample and discuss
some policy measures aimed at changing travel behavior. The results suggest that, besides individ-
ual characteristics, latent aspects related to the perception of the context and of the bicycle as a
means of transport strongly affect the propensity to cycle.
ARTICLE HISTORY
Received 3 July 2017
Revised 14 June 2018
Accepted 14 June 2018
KEYWORDS
Choice of cycling; hybrid
choice model; serial
correlation; simula-
tion; validation
1. Introduction
The promotion of cycling can help to abate environmental
and traffic-related problems caused by motorized forms of
transport. Considering the benefits of cycling, a major chal-
lenge for researchers today is how to encourage this form of
transportation. Denmark and The Netherlands are famous
for their cycling culture (Carstensen & Ebert, 2012).
However, in the majority of other European countries,
cycling mode share is very low: only 2% of trips are made
by bicycle in Great Britain, 3% in Ireland and the Czech
Republic and 5% in France (Kurt, 2008), figures confirmed
also at the Capitals level (ECF
1
). In Italy, though 50% of
daily trips do not cover more than 5 km, cycling is the least-
used mode of transportation, accounting for just 3.8% of
daily trips
2
. In countries with low cycling mode share,
bicycles are used above all for sport and recreational activ-
ities. The lack of proper infrastructure has precluded the
development of a cycling culture, strengthening the percep-
tion of the bicycle as a form of recreation (in urban areas
not accessible to motor vehicles) rather than as a mode of
transport. Thus, in order to increase the propensity to cycle,
user perception needs to be changed. In addition to
providing proper cycling facilities, it is important to evaluate
individuals’perception of bikes as an alternative mode
of travel.
Promoting bicycle use requires understanding those fac-
tors underpinning the propensity to cycle and also the struc-
tural and psychosocial barriers that may contribute to
hindering use of the bike (Fern
andez-Heredia, Monz
on, &
Jara-D
ıaz, 2014; Heinen, Van Wee, & Maat, 2010, Pucher,
Dill, & Handy, 2010). There exists a broad literature on the
objective factors (context characteristics, facilities, socio-eco-
nomic and demographic factors) affecting the propensity to
cycle (see for example, Broach, Dill, & Gliebe, 2012; Calvey,
Shackleton, Taylor, & Llewellyn, 2015; Dickinson, Kingham,
Copsey, & Hougie, 2003; Dill and Voros, 2007; Heinen
et al., 2010; Hunt and Abraham, 2007; Kingham, Dickinson,
& Copsey, 2001; Krizek, 2012; Parkin, Wardman, & Page,
2007; Pucher & Buehler, 2006; Vandenbulcke et al., 2011)as
well as on the psycho-social factors. For example, positive
attitudes toward cycling (Li, Wang, Yang, & Ragland, 2013;
Willis, 2015) convenience (flexible, efficient) and exogenous
restrictions (danger, vandalism, facilities), (Fern
andez-
Heredia et al., 2014) are indicated as having a positive effect
on the choice to cycle. Social norms explain why cycling in
CONTACT Eleonora Sottile elesottile@gmail.com CRiMM –Centre of Research of Mobility Models, University of Cagliari, Via San Giorgio 12, Cagliari
09124, Italy
1
https://ecf.com/resources/cycling-facts-and-figures
2
http://www.isfort.it/sito/statistiche/Congiunturali/Annuali/RA_2014.pdf, in Italian.
ß2018 Taylor & Francis Group, LLC
INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION
https://doi.org/10.1080/15568318.2018.1490465

some areas in Northern Europe is particularly common
(Pucher, Komanoff, & Schimek, 1999; Wardman, Tight, &
Page, 2007). The importance that individuals attach to the
health benefits of cycling, for example, also has an effect on
cycling; the same goes for environmental beliefs (Emond &
Handy, 2012; Gatersleben & Appleton, 2007; Heinen, Maat,
& Van Wee, 2011; Hunecke, Bl€
obaum, Matthies, & H€
oger,
2001; Pooley et al., 2012). Dill and Voros (2007) found that
social norms affect cycling behavior: people living in house-
holds with other adults who cycle regularly, had co-workers
who cycled to work, or who frequently saw adults cycling
on their street were more likely to be regular cyclists them-
selves. According to Stinson and Bhat (2005), cycling more
in leisure time could increase the frequency of bicycle use
for commuting.
Among the psycho-social factors, individuals’perception
of the system plays an important role in biking. Gatersleben
and Appleton (2007) observed that non-cyclists perceive
more barriers to cycling than utility cyclists and vice versa
(Bamberg & Schmidt, 1994). Majumdar, Mitra and Pareekh
(2015) studied several perceived benefits (physical fitness,
environmental awareness, travel reliability, travel flexibility,
psychological safety, affordability, desire for pollution free
roads) and found safety hazards, social barriers and road
conditions to influence the choice to travel by bike the
most. A comparative study of cyclists in Brisbane and
Copenhagen on the perceptions of safety (Chataway, Kaplan,
Nielsen, & Prato, 2014) showed that cyclists in Brisbane per-
ceived mixed traffic environments as less safe, and felt more
apprehensive of traffic than cyclists in Copenhagen. Ma Dill,
& Mohr (2014) explored the relationships between the
objectively measured environment, perceptions of the envir-
onment, and cycling behavior. The results of their study
showed that the perception of the environment had a direct
and significant effect on cycling behavior, while the direct
effect of the objective environment on cycling behavior
became insignificant when controlling for perception. They
concluded that a good cycling environment was necessary
but not sufficient for using the bike. Akar and Clifton
(2009) explored the perceptions of the campus community
regarding possible cycling infrastructure improvements, pol-
icy, and programme innovations. Kaplan, Manca, Nielsen,
and Prato (2015) investigated the behavioral factors underly-
ing tourist intentions to use urban bike-sharing for recre-
ational cycling while on holiday. Sigurdardottir, Kaplan,
Møller, and Teasdale (2013) focused on the intentions of
adolescents to commute by car or bicycle as adults. Mu~
noz,
Monzon, and L
opez (2016) proposed a methodology for
including cycling-related indicators in mobility surveys
based on the theory of planned behavior.
None of these works studies the perception of the bike as
a mode of transport. They also use factor analyses alone or
in conjunction with structural equation models to study the
relationship between psycho-social factors and biking.
Only a few applications of Hybrid Choice Models
(HCMs) for estimating the effect of psycho-social factors on
the choice to cycle are reported in the literature.
Kamargianni and Polydoropoulou (2013) estimated the
influence of the latent variable willingness to walk or cycle
on mode choice. Motoaki and Daziano (2015) investigated
the effects of weather (temperature, rain, and snow), cycling
time, slope, cycle facilities (bike lanes), and traffic on cycling
decisions. Maldonado-Hinarejos, Sivakumar, & Polak (2014)
incorporated attitudes towards cycling, perceptions of the
image associated with cycling and the stress arising from
safety concerns in a choice model for cycling. La Paix,
Cherchi, and Geurs (2015) estimated the impact of the per-
ception of the quality of bicycle interchanges and attitudes
towards cycling on the mode choice (including the bike) to
access/egress rail stations. Habib, Mann, Mahmoud, and
Weiss (2014) studied the effect of comfort, safety conscious-
ness and perceptions of bikeability (in terms of quality of
cycling facilities), on the choice of biking for utilitarian or
recreational purposes.
This paper aims specifically to explore the role of percep-
tion in the likelihood of utility cycling. We focus on the per-
ception of: the bicycle as a means of transport; bikeability
(in terms of usefulness and safety) and of bike infrastruc-
ture. HCMs have been used to estimate the effect of people’s
perception on the propensity to bike. Unlike all earlier
works, our HCM also accounts for the serial correlation
between error terms in the discrete and latent perceptions,
to allow for agent-common unknown factors that equally
affect both the discrete choice to use the bike and the latent
perceptions of it. We also use revealed preference instead of
stated preference data, to represent the actual context. In
this way, we can capture individuals’current perceptions so
as to define the correct measures to be adopted for increas-
ing the propensity to cycle. Furthermore, we also validate
the model results, using a hold-out sample. The validation
phase is crucial and highly recommended for assessing the
quality of the models estimated, but it has been widely
neglected in transportation research. Few exceptions are
Cherchi and Cirillo (2010), Mabit, Cherchi, Jensen, &
Jordal-Jørgensen (2015) and Klapper, Ebling, and
Temme (2005).
The remainder of the paper is structured as follows:
Section 2 describes the methodology adopted. Section 3
presents the application and provides a descriptive analysis
of the dataset including both the objective characteristics
and perceptions. Section 4 discusses the modeling frame-
work and Section 5 presents the model results. Section 6
concludes the paper.
2. Methodological framework
The methodology followed envisaged a preliminary phase
for survey design. The survey aimed, on the one hand, to
gather information on current travel habits focusing on
bicycle mode, and on the other hand, to identify and meas-
ure (using a Likert scale) those factors underpinning the
choice to use/not to use the bike. We focused in particular
on analyzing how people perceive the bicycle and related
aspects (safety, infrastructure, etc.). We then performed a
confirmatory factor analysis on the survey data so as to pin-
point which of the items contained in the questionnaire
2 E. SOTTILE ET AL.

better represented the a priori assumption of perceptions.
Lastly, and most importantly we analyzed the effect of per-
ceptions on the propensity to use/not to use the bicycle. We
achieved this objective by estimating and validating a hybrid
discrete choice model that allowed us to identify those
socio-economic characteristics influencing to the greatest
extent the propensity to cycle/not to cycle and at the same
time the role played by perceptions on that propensity.
3. Application
The data used in the analysis were drawn from a web-sur-
vey, called "Bicimipiaci" ("BikeIlikeyou"), conducted, between
2014 and 2016, by the University of Cagliari (Italy) in col-
laboration with public authorities, among a sample of
Sardinian Regional government, University and munici-
pal employees.
E-mails were sent to around 9600 individuals inviting
them to participate in a web-survey. A number of prizes
3
were offered as an incentive to fill in the questionnaire. A
total of 4691 individuals completed the survey, a very high
response rate (48.9%). Questionnaires were carefully ana-
lyzed and 2752 observations were used for modeling pur-
poses (corresponding to 28.6% of the individuals contacted
and to 58.6% of respondents). Clearly, the sample was not
intended to be representative of the general population, but
sample size was large enough to permit interesting analyses.
The questionnaire was divided into four sections. The
first section separated out cyclists from non-cyclists and
aimed to identify why and how the former chose to cycle
for utilitarian purposes. The second section was designed to
measure perceptions using the 5-point Likert scale, specific-
ally (1) positive and negative perceptions about cycling in
general, (2) the "perception of context characteristics",
intended as the importance assigned to policies for increas-
ing bicycle use, and (3) the "perception of bikeability and
safety" of bike lanes and paths and to detect the main bar-
riers to cycling (non-cyclists). The third section consisted of
the description of the daily commute. The last part was
dedicated to gathering information about individuals’socio-
economic characteristics. (More details in Meloni, Sanjust, &
Sottile, 2015).
The study context concerned small and medium-sized
towns which did not have a complete urban cycle system
when the survey was conducted. The majority of existing
cycle lanes had been created on the roadway between the
parking lane and the pavement. The lack of a capillary net-
work meant that it was not possible to design complete
routes between a given origin and destination. A bike shar-
ing service was introduced in the capital Cagliari in 2010,
initially with four stations which were increased to 10 in
2013 for a total of 100 bikes, of which 30 pedal assisted.
However, by February 2015 this service was no lon-
ger accessible.
3.1. Data analysis
The final sample used for the estimation comprises the same
proportion of non-cyclists (50.1%), and frequent cyclists
(49.9%). Despite making up half the sample, only 20.7% of
cyclists use the bike several times a week or every day. Four
Table 1. Socioeconomic characteristics.
Socio-economic characteristics Total sample (2752 obs.) Estimation sample (2202 obs.) Validation sample (550 obs.)
Variables AVG No. %AVG No. %AVG No. %
Bicycle use 1373 49.9 1095 49.73 278 50.55
Age 48.00 47.93 48.30
Age_18–30 104 3.78 87 3.95 17 3.09
Age_31–40 458 16.64 370 16.80 88 16.00
Age_41–60 2004 72.82 1603 72.80 401 72.91
Age >60 186 6.76 142 6.45 44 8.00
Gender: male 1399 50.84 1129 51.27 270 49.09
Level of Education
Middle school or lower 113 4.11 96 4.36 17 3.09
High school 1042 37.86 813 36.92 229 41.64
Specialization 63 2.29 49 2.23 14 2.55
Undergraduate and master's degree 915 33.25 733 33.29 182 33.09
Post lauream (phd, etc.) 619 22.49 511 23.21 108 19.64
Marital status: married 2007 72.93 1601 72.71 406 73.82
With children 1517 55.12 1214 55.13 303 55.09
# of members in the household 2.88 2.88 2.89
Driving licence 2688 97.67 2158 98.00 530 96.36
Personal car available 2491 90.52 1992 90.46 499 90.73
# of cars in the household 1.71 1.71 1.70
# of bikes in the household 1.53 1.53 1.53
Not informed about bike sharing 1009 36.66 786 35.69 223 40.55
Informed about bike sharing, non-subscriber 1349 49.02 1091 49.55 258 46.91
Informed about bike sharing, subscriber 39 1.42 28 1.27 11 2.00
Informed about bike sharing 355 12.90 297 13.49 58 10.55
No Income 49 1.78 38 1.73 11 2.00
Low level (<1000 e) 124 4.51 105 4.77 19 3.45
Medium level (1000 e–2000 e) 1823 66.24 1445 65.62 378 68.73
Medium–High level (2000 e–3000 e) 375 13.63 302 13.71 73 13.27
High level (>3000 e) 381 13.84 312 14.17 69 12.55
3
The lottery comprised: one bicycle, 15 one-year bike-sharing cards, 1 one-
year car-sharing card, and various cycling gadgets.
INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 3

different purposes for cycling are investigated: commuting,
shopping, leisure and traveling to a public transport stop.
Frequency of biking was measured with a 5-level scale: (1)
never, (2) 1–10 times a year, (3) 1–5 times per month, (4)
several times a week, and (5) every day. The frequency ana-
lysis confirms that the bicycle is used mostly for recreational
purpose (98.1%), whereas 66.7% indicated "never" for cycling
to work, and 88.3% for traveling to a public transport stop.
This is not surprising since in the context under study the
culture of utility cycling is inexistent; the bike is used
for leisure.
Turning to individual and household characteristics,
Table 1 shows that the sample is practically equally divided
between males and females. Average age is 48 years, though
74.6% of the sample are aged between 41 and 60. As
expected, the sample is relatively educated, and the majority
are wage earners. The largest proportion have a monthly
income of between e1000 2000, are married with children
and the average number of household members is around
three. 97.6% of the respondents have a driving licence and
90.5% are car owners. The average number of cars is higher
than the average number of bicycles per household, though
interestingly surprisingly high, considering that cycling is
not common in Italy. The sample was divided into two sub-
samples, one for model estimation ("Estimation sample")
and the other for validation ("Validation sample"). The val-
idation sample includes 550 observations and was obtained
by random sampling 20% of observations from the total.
The remaining 80% (corresponding to 2202 observations)
was used for model estimation. Table 1 shows the socio-
economic characteristics for both the estimation and valid-
ation subsamples which, as can be seen, do not differ
significantly.
3.2. Perceptions
The latent variables have been defined on the basis of the
items reported in Table 2. Each respondent was asked to
express her/his level of agreement or disagreement with
each item on a Likert scale from 1 to 5 (1 ¼Totally disagree
to 5 ¼Totally agree). In particular, as mentioned, we focused
on the three latent variables relating to: (1) perception of
the bicycle as a means of transport, defined at the personal
and societal level by twelve items; (2) perception of context
characteristics, described by four items; and (3) perception
of bikeability, defined by nine items.
A first descriptive statistics analysis revealed that the sam-
ple perceived positively the bicycle as a means of transport,
though cycling in traffic was considered dangerous and bikes
unsuitable for carrying heavy items. As was to be expected,
the perception of bikeability confirmed the dangers users
associate with cycling in traffic. This could create a barrier
to bike use, given the lack of adequate cycling infrastructure
in the reference context. Added to this is the fact that exist-
ing bike lanes are not perceived as useful for traveling in
urban areas and that motorists encroach on cyclists’
road spaces.
As for the perception of context characteristics in the
propensity to cycle, this was judged to be important, espe-
cially the presence of a dedicated bicycle network within
Table 2. Factor analysis results (weights for the first factor). Underline values refer to items that define the latent variables.
Factor analysis results (weights for the first factor)
Total sample Estimation sample Validation sample
Items (Indicators of latent variables) Name (2752 obs.) (2202 obs.) (550 obs.)
Latent variable: Perception of bicycle as a means of transport
It is a rapid means of transport (avoids queues and traffic) Perc_1 0.417 0.451 0.343
Cycling in traffic is not dangerous Perc_2 0.011 0.022 0.118
It is likely to be stolen and parking areas are inadequate Perc_3r 0.038 0.006 0.030
It is not expensive Perc_4 0.574 0.537 0.717
It involves exposure to bad weather and air pollution Perc_5r 0.030 0.001 0.073
It avoids wasting time looking for parking Perc_6 0.580 0.557 0.645
It is healthy Perc_7 0.776 0.767 0.754
It is difficult to carry heavy items Perc_8r 0.109 0.125 0.070
It allows one to appreciate historic centers and increases accessibility to city services Perc_9 0.691 0.689 0.699
Need for cycling gear Perc_10r 0.087 0.038 0.130
It contributes to reducing polluting emissions Perc_11 0.740 0.746 0.684
It limits daily activity patterns Perc_12r 0.041 0.104 0.067
Latent variable: Perception of bikeability (in terms of usefulness and safety)
Existing bike lanes are not useful for traveling Bikeab_1 0.792 0.788 0.82
Existing bike lanes and crossings are safe, comfortable and well-marked Bikeab_2r 0.701 0.682 0.728
It is better to ride in traffic than use the existing bike paths Bikeab_3 0.582 0.609 0.548
Motorists often encroach on dedicated bike lanes Bikeab_4 0.23 0.233 0.084
Latent variable: Perception of context characteristics in propensity to cycle
A dedicated cycle network in urban areas Context_1 0.938 0.997 0.932
The presence of bike racks and secure parking Context_2 0.953 0.962 0.939
Extending RTZ or pedestrian zones Context_3 0.679 0.659 0.728
A bike-sharing station close to home or at public transport stops Context_4 0.356 0.371 0.453
If other people use it Context_5 0.211 0.203 0.138
Dedicated facilities at work / study (parking, showers, lockers for equipment, etc.) Context_6 0.351 0.413 0.255
An integrated ticket for bike-sharing and public transport services Context_7 0.162 0.255 0.046
Combination with public transport services Context_8 0.209 0.318 0.018
Increase of parking fees Context_9 0.241 0.258 0.158
The rmeans it was analyzed in reverse.
4 E. SOTTILE ET AL.

urban areas, of bike racks and secure parking and extending
restricted traffic zones or pedestrian zones. These are a
necessary condition for the bicycle to be regarded as an
alternative transport mode. Without proper infrastructure
and facilities people are unlikely to consider the bicycle as
an available travel mode option (see Table 2). However, the
presence of appropriate cycling facilities is not the only con-
dition for choosing to cycle. Indeed, individuals who do not
recognize the attributes associated with bicycle use are less
likely to choose to cycle.
We performed two types of factor analysis to identify the
latent dimensions underpinning our set of items: Principal
Component Analysis (PCA) and Factor Analysis (FA). To
determine the factorability of the data and the number of
factors to be extracted, we used the Kaiser-Meyer-Olkin
(KMO) test (>0.5), the scree or Cattell test (Cattell, 1966)
and parallel analysis. Tables 2 and 3show the factor analysis
results, in particular the first one shows the weight of each
item for the first factor identified by the FA for the total
sample, for the estimation sample (2202 individuals) and for
the validation sample (550 individuals), and the second one
the factorability and the reliability of the three latent varia-
bles. However only two of the three latent variables were
found to be significant: perception of the bicycle as a means
of transport and perception of the context (defined both by
the items underlined in the table) whereas perception of
bikeability (in terms of usefulness and safety) was below the
reliability threshold as shown in Table 3. As will be seen
later, the model estimation confirmed that this latent vari-
able was indeed problematic.
Figure 1 shows the distribution of responses for the esti-
mation sample (2202 individuals) for those items identifying
the perception of the bicycle as a means of transport and for
perception of the context.
It clearly emerges that the majority of this subsample
(60–90%) has a positive perception of the bicycle as a form of
transport, recognizing the associated benefits. The same goes
for perception of the context characteristics, though to a lesser
extent, that are indeed encourage the propensity to cycle.
4. Modeling framework
The hybrid choice model used here is a binary logit that mod-
els the choice to cycle vs. the choice not to cycle as a function
of a set of socioeconomic characteristics and three latent vari-
ables that measure individuals’perception of different aspects
of biking. In addition, we also account for serial-correlation
between the error terms. Since both the discrete and the latent
part apply to the same individual making the choice and pro-
vided an evaluation of the indicators, the error terms of these
sub-models may be correlated, as they potentially share unob-
served variables specific to each individual. Following
Bierlaire (2016) we deal with serial correlation by incorporat-
ing an agent effect in the model specification. This is an error
component appearing in all the sub-models involved, distrib-
uted across the individuals. Let us denote with DU
q
the differ-
ence between the utilities that the individual qassociates to
the alternatives of cycling and not cycling:
DUq¼ASC þbSEqþXnknþhnSEq

CnþanSE0
qþxqn þgq

þeqþgq
(1)
Table 3. Factorability and reliability.
Latent variables Sample KMO No. Indicators factor 1 Alpha factor 1 % of variance explained
Perception of bicycle as a means of transport Total sample 0.776 5 0.680 22.891
Estimation sample 0.767 0.724 22.660
Validation sample 0.775 0.724 24.133
Perception of bikeability (in terms of usefulness and safety) Total sample 0.559 3 0.489 37.774
Estimation sample 0.566 0.486 37.761
Validation sample 0.528 0.486 38.011
Perception of context characteristics in propensity to cycle Total sample 0.814 3 0.773 45.293
Estimation sample 0.823 0.776 46.042
Validation sample 0.775 0.776 42.363
0.00
10.00
20.00
30.00
40.00
50.00
60.00
70.00
80.00
90.00
100.00
Perc_4 Perc_6 Perc_7 Perc_9 Perc_11
Percentage
Items
Perception of bicycle as a means of transport
1
2
3
4
5
0.00
10.00
20.00
30.00
40.00
50.00
60.00
70.00
80.00
90.00
100.00
Context_1 Context_2 Context_3
Percentage
Items
Perception of context characteristics in propensity to cycle
1
2
3
4
5
Figure 1. Distribution of responses of the estimation sample.
INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 5

where SE
q
is a vector of socioeconomic characteristics, bthe
respective vector of coefficients, ASC is the constant specific
for the cycling alternative, and LVnq ¼CnþanSE0
qþ
xqn þgqis the n-th latent variable (with n¼1,2,3) that
depends on a vector of socioeconomic characteristics (SE
q
0),
which can be different from that included in the discrete
choice, with a
n
the associated coefficients. C
n
is the intersect,
x
qn
is a normally distributed error term with zero mean
and standard deviation r
x
, and k
n
is the coefficient associ-
ated with each LV. Lastly, h
n
is a vector of coefficients asso-
ciated with the interaction between LV and SE, while e
q
is
the logistically distributed error term and g
q
is a normally
distributed error component with zero mean and standard
deviation r
g
,which is common between the LVs and the
utility of the discrete alternatives.
The items reported in Table 2 are used as indicators of
the latent variables and are related thereto by means of the
following measurement equation:
Iqnk ¼ckþfkLVqn þtqk k¼1; :::; K
(2)
where I
qnk
is the k-th indicator for the n-th latent variable,
c
k
is the intersect, f
k
is the coefficient associated with the
latent variable (cand fare normalized to zero and 1 for the
first indicator for identification purposes), and 
qk
is the
normally distributed error term with zero mean and stand-
ard deviation r
t
.
The distributions of the latent variable and the indicators
are respectively:
fLV LVqnjSE;a;rx

¼1
rx
/
LVqnCnþanSE0
qþgq

rx
0
@1
A
fIIqnkjLVqn ;c;f;rt

¼1
rtnk
/IqnkckfkLVqn
ðÞ
rtk
!
The probability that individual q will choose alternative j
is given by:
Pqj ¼ðxðg
Pqj LVqn xq;gq
ðÞ

YnfLVqn gq
ðÞ
YkfIqnk
LVqn xq;gq
ðÞ

fx
ðÞ
fg
ðÞ
dxdg
Models are estimated using PythonBiogeme (Bierlaire and
Fetiarison, 2009).
Table 4. Model results.
Name
HCM serial correlation HCM HCM
(Estimation
subsample)
(Estimation
subsample)
(Validation
subsample)
Value
Robust
Value
Robust
Value
Robust
t-test t-test t-test
Discrete part
Constant 2.480 7.050 2.380 6.670 1.880 2.520
Age 0.014 2.230 0.014 2.330 0.024 1.840
Male (Female is the base) 0.893 7.960 0.866 7.800 0.803 3.510
Children in household (No children is the base) 0.713 4.530 0.675 4.210 0.982 3.560
# of bikes in household 1.580 20.880 1.560 21.130 1.320 10.090
Informed about bike sharing
non-subscriber (Not informed is the base) 0.528 4.780 0.516 4.730 0.456 2.060
subscriber (Not informed is the base) 2.940 3.850 2.830 3.600 1.530 2.280
Perception
of bicycle as a means of transport 1.090 18.860 0.980 6.310 1.250 6.910
of context characteristics in propensity to cycle 1.340 17.530 1.530 7.940 0.946 4.310
of bicycle as a means of transport# of members in the household 0.064 50.074 4.690 ––
of context characteristics in propensity to cycle# of cars in the household 0.058 2.360 0.068 2.440 ––
Latent variable: Perception of bicycle as a means of transport
Level of education (from 1 to 5) 1.030 21.460 0.868 25.640 1.020 13.390
# of bikes in the household 0.965 13.840 0.778 15.450 0.600 6.450
Standard Deviation 0.646 9.520 0.684 14.870 0.709 6.590
Latent variable: Perception of context characteristics
Level of education (from 1 to 5) 0.480 13.900 0.519 14.160 0.509 6.240
Children in household (No children is the base) 0.165 1.630 0.213 2.270 0.075 0.400
Available car 0.573 4.380 0.193 1.550 0.489 1.880
# of bikes in household 0.567 10.040 0.595 11.120 0.490 4.700
Income per month 0–1000 e1.010 4.310 0.781 3.710 1.300 3.100
Income per month 1000–2000 e0.591 6 0.328 3.460 0.443 2.300
Standard Deviation 9.480 31.630 0.509 8.440 0.554 4.130
Actual biker, formerly car as driver (Past behavior) 0.417 1.340 0.566 2.110 1.550 2.270
Serial Correlation (Error Component) 1.910 24.600 ––––
Number of estimated parameters: 43 42 40
Sample size: 2202 2202 550
Number of draws 500 ––
Init log-likelihood: 21,282.880 23,496.830 5865.876
Final log-likelihood: 11,891.360 12,090.430 3001.590
Likelihood ratio test for the init. model: 18,783.040 22,812.800 5728.573
Rho for the init. model: 0.441 0.485 0.488
Rho bar for the init. model: 0.439 0.484 0.481
6 E. SOTTILE ET AL.

5. Estimation results
Tables 4 and 5present the results of the final hybrid model
4
,
estimated separately for the estimation and validation sam-
ples. To test whether disregarding the serial-correlation
effect influenced the estimation results, both the models esti-
mated without serial-correlation (first 2 models in Tables 4
and 5) and including serial-correlation (last column in both
tables) are shown. Table 4 gives the results for the binary
logit model, Table 5 those for the indicators of
latent variables.
First of all, we estimated the discrete and latent models
(one for each latent variable) separately. Having identified
the best discrete model specification and the three best
latent models, the hybrid choice model was then estimated
simultaneously. The model presented in Tables 4 and 5
includes only two latent variables "perception of bicycle as a
means of transport" and "perception of context characteris-
tics", as the third, "perception of bikeability", was significant
only when estimated alone; it was not identified when esti-
mated together with the other two. This result is not sur-
prising because the factor analysis also revealed this factor
to be below the reliability threshold.
As far as the discrete part of the model is concerned, sev-
eral socioeconomic characteristics have a direct influence on
the propensity to cycle. In particular, men and younger peo-
ple are more likely to cycle than women and older persons
respectively (the age variable is included in the specification
as continuous, so the propensity to cycle decreases as age
increases). Most studies conclude that men cycle more than
women (Heinen et al., 2010), but this tendency seems to be
related to cycling culture: Garrard, Rose, and Lo (2008)
found that in countries with low rates of cycling males are
more likely to cycle than females; by contrast, in countries
with high cycling rates, such as the Netherlands and
Denmark, cycling is more evenly spread across the two gen-
ders (Heinen, Maat, & Van Wee, 2013). A likely explan-
ation, in this case study, is that women usually have more
family responsibilities, such as shopping and picking up
children, activities that are less likely to be undertaken by
bike. This is confirmed by the negative sign of the presence
of children in the household (dummy variable that takes
value 1 if children are present, 0 otherwise). Indeed, families
with children are less likely to cycle. By contrast, in the
existing literature the relation between cycling and age is
not so clear. Some researchers indicate that cycling levels
decline with age (Dill and Voros 2007; Pucher et al. 1999)
while others (De Geus, 2007; Wardman et al., 2007) have
found that age is not a significant factor. In the context of
this study, the age-based result probably depends on the fact
that the majority of young adults have low incomes and
cannot afford more expensive travel mode options. Indeed,
individuals with greater purchasing power appear to be less
likely to cycle. On the other hand, the relationship between
cycling and income reported in the literature is ambiguous.
While some workers (Dill & Voros, 2007; Stinson & Bhat,
2005) report a positive relationship between income and
cycle use, Parkin et al. (2007) found that in England there is
a link between lower incomes and lower bicycle share.
Furthermore, people who are informed about bike shar-
ing are more likely to cycle (and of course yearly subscribers
more than non-subscribers), than the uninformed, confirm-
ing that the motivation to cycle positively influences the
propensity to cycle.
The two latent variables have positive sign, indicating
that increased perception of the context (i.e. the availability
of cycling infrastructure and facilities), increases the propen-
sity to cycle (positive sign of the latent variable). The pres-
ence of cycling facilities is a significant factor in the choice
to cycle. Furthermore, almost all earlier studies have
Table 5. Model results: Indicators of latent variables.
HCM serial correlation HCM HCM validation
Value Robust t-test Value Robust t-test Value Robust t-test
Perception of bicycle as a means of transport
delta1 1.270 16.930 1.080 17.740 1.030 7.880
delta2 1.440 25.090 1.270 26.100 1.420 13.610
lambdaPerc_6 0.776 18.490 0.836 20.380 0.829 10.590
delta1Perc_6 0.753 10.120 0.682 10.070 0.624 4.740
delta2Perc_6 1.050 15.210 0.962 15.340 1.030 7.740
lambdaPerc_7 1.400 14.040 1.390 15.780 1.300 8.310
delta1Perc_7 0.594 4.550 0.480 4.600 0.563 2.600
delta2Perc_7 1.750 11.250 1.490 11.360 1.350 5.920
lambdaPerc_9 1.020 17.020 1.140 18.390 0.956 10.530
delta1Perc_9 1.050 9.630 1.030 9.620 0.950 5.510
delta2Perc_9 1.510 15.310 1.440 15.260 1.150 7.740
lambdaPerc_11 1.360 12.820 1.450 13.680 1.300 9.420
delta1Perc_11 0.611 5.030 0.571 5.090 0.655 2.890
delta2Perc_11 0.642 5.910 0.599 5.940 1.240 5.450
Perception of context characteristics for propensity to cycle
lambdaContext_2 2.290 10.400 1.530 21.580 1.530 9.630
delta1Context_2 2.050 7.820 1.120 10.660 1.140 4.890
delta2Context_2 2.420 9.980 1.420 14.230 1.650 6.640
lambdaContext_3 0.858 20.680 0.926 23.510 0.898 11.740
delta1Context_3 0.996 14.550 0.982 14.460 0.892 7.040
delta2Context_3 1.400 21.740 1.350 21.190 1.500 10.920
4
The variables described in Section 4 were all tested in the model
specification, retaining only the significant ones. All continuous variables were
also tested as dummy variables for different ranges to test for non-
linear effects.
INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 7

revealed that the existence of a bike network could increase
the propensity to cycle for utilitarian purposes (Dill &
Voros, 2007; Hunt & Abraham, 2007).
The same can be said of the perception of the bicycle as
a means of transport (the latent variable has a positive sign).
Regarding the magnitude of the estimates of the two latent
variables, the need for cycling infrastructure and facilities
affects the propensity to cycle to a greater extent than the
perception of the bicycle as a means of transport (1.53 vs.
0.98). In other words, under the same perception of the con-
text, a person who perceives the bicycle as a means of trans-
port is more likely to cycle.
The latent variables are also included in the interaction
with socioeconomic variables. Particularly, the interaction
between the perception of bicycle as a means of transport
and the number of household members is negative and
highly significant. This indicates that the effect of the per-
ception of the bicycle as a means of transport is positive but
less so for larger households. This is not surprising, as it
could be related to the responsibilities of households with
children where many activities (pickup and drop off, shop-
ping, etc.) are difficult to do by bike.
Another significant interaction is between the latent vari-
able perception of the context and the number of cars per
household. This interaction can be interpreted similarly to
the above (negative sign of the parameter). As the number
of available cars increases, so the effect of the perception of
the context on the propensity to cycle decreases. This may
be because in multi-car households members are less con-
cerned about the context characteristics for cycling. This
result is in agreement with other studies in which car avail-
ability is widely reported as negatively related to bicycle
choice (Mu~
noz et al., 2016).
As far as the latent part of the model is concerned, the
socioeconomic variables that contribute to specifying the
latent variable “perception of context”are high level of edu-
cation, low-medium income, car availability, a certain num-
ber of bikes in the household and no children. This is easily
understood as those who are more likely to cycle (number
of bikes in the household, absence of family impediments,
low-medium income) assign importance to the availability
of cycling facilities. In this context, 90.5% of the sample
owns a car, so this variable does not single out a certain tar-
get. For "perception of bicycle as a means of transport", the
socioeconomic variables defining the latent variable indicate
that a high level of education and bike ownership corres-
pond to high perception of the benefits of cycling.
We estimated the validation model (550 obs.) using the
same specification as above, apart from the interactions of
latent variables with socioeconomic variables, as the sample
size was too small to allow identification. The validation
results indicate that the model performs well with the hold-
out sample. All the coefficients estimated are significant at
95%, with only two exceptions. More importantly, the coeffi-
cients do not differ significantly from those estimated with
the estimation sample. Furthermore, in order to verify
model performance, we simulated the choice probabilities
using the two models estimated for the estimation and valid-
ation samples. As illustrated in Table 6, the two models per-
form similarly, in particular the probability to cycle is
practically identical (49% vs. 49.7%) for both models.
Finally note that in the model that takes into account ser-
ial correlation, the coefficients do not differ significantly
from those estimated in the model that does not. However,
standard deviation that accounts for serial correlation is
highly significant, suggesting it cannot be disregarded.
6. Conclusions
Experience gained in utility cycling in major European cities
has shown a correlation between travel mode choice in
urban environments and measures that encourage or hinder
the use of a given mode. This is particularly true of the
bicycle, a marginal and not generally recognized form of
transport for utilitarian purposes, precisely because much of
the existing infrastructure has been designed for use prefer-
entially by other means of transport. Indeed, in many con-
texts, such as the case at hand, where cycling culture does
not exist, the bicycle may not be perceived as an alternative
means of travel, in the same way as the private car and pub-
lic transport. Thus the decision to cycle entails a more rad-
ical change compared to the switch between traditional
travel modes.
For this reason it is essential, when implementing meas-
ures for increasing bike commuting, to identify and under-
stand those factors influencing travel choice, in order to
explain why certain people choose to bike commute whereas
others prefer to travel by traditional means of transport.
This will enable planners to develop comprehensive mobility
systems, radically different from existing ones. In particular,
exploring travel behavior and mobility styles of potential
and existing cyclists serve as the basis for formulating pro-
grammes that aim to change commuter preferences in urban
mobility choices.
In this work, we studied the role of perception in the
propensity to commute by bike. In particular, we focused on
three aspects of perception: bike as a means of transport,
bikeability (in terms of usefulness and safety), and bike
infrastructure. The context of application concerns utility
cycling by public employees in Cagliari (Italy). To measure
the impact of individuals’perception on the probability to
Table 6. Simulation results.
Simulation results
AVG HCM (Estimation subsample) HCM (Validation subsample)
Elasticity to # of bikes in household 0.761 0.636
Conditional probability 0.486 0.490
Unconditional probability 0.337 0.344
Integrated unconditional probability 0.490 0.497
8 E. SOTTILE ET AL.

cycle, we estimated HCM that account for serial correlation
between error terms in the discrete and latent perceptions.
We also validated the model results, using a hold-out sam-
ple. The results suggest that, beside individual characteristics
(young adults, males, with no children in the household are
more willing to cycle), the perception of the context charac-
teristics (i.e. the availability of cycling infrastructure and
facilities) and of the bicycle as a means of transport (benefits
of cycling) affect the propensity to cycle. Particularly the
results show, on the one hand, that the lack of proper infra-
structure and facilities create a strong barrier to bicycle use,
and on the other that the greater propensity to cycle is dir-
ectly correlated with the perception of the bike as a means
of transport. Indeed, for the same context characteristics,
individuals who perceive the bike as a mode of travel are
more likely to cycle. Thus, also in view of the large sample
size that made it possible to obtain a validation sample and
hence robust results, the model results provide scientific evi-
dence, often disregarded, of the importance not only to
eliminate barriers through the creation of dedicated infra-
structure and facilities, but also to raise the awareness of the
bicycle as an alternative commute mode. This is especially
true of Italy where the bicycle is perceived more as a form
of exercise, or for leisure than as a means of transport. This
result provides a useful tool for establishing preliminary pol-
icy directions for increasing the propensity to commute by
bike. These need to account for individuals’intrinsic charac-
teristics through the provision of information aimed at pro-
moting bicycle use.
Thus, arguably, policies for promoting the bicycle as an
alternative mode of transport to motorized vehicles need to
combine hard and soft measures.
Funding
This research was sponsored by the Sardinian Regional Government.
Disclosure statement
No potential conflict of interest was reported by the authors.
References
Akar, G., & Clifton, K. J. (2009). The influence of individual percep-
tions and bicycle infrastructure on the decision to bike.
Transportation Research Record,2140(1), 165–172.
Bamberg, S., & Schmidt, P. (1994). Car or bicycle-an empirical-test of a
utility-theory-approach for predicting the choice between means of
transportation. Kolner Zeitschrift fur Soziologie und Sozialpsychologie,
46(1), 80–102.
Bierlaire, M. (2016). Estimating choice models with latent variables with
PythonBiogeme. (No. EPFL-REPORT-221361). Transport and Mobility
Laboratory School of Architecture, Civil and Environmental
Engineering. Ecole Polytechnique Federale de Lausanne.
Bierlaire, M., & Fetiarison, M. (2009). Estimation of discrete choice
models: Extending BIOGEME. 9th Swiss transport research confer-
ence. Switzerland: Monte Verit
a.
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: Policy and Practice,46(10),
1730–1740.
Calvey, J. C., Shackleton, J. P., Taylor, M. D., & Llewellyn, R. (2015).
Engineering condition assessment of cycling infrastructure: Cyclists’
perceptions of satisfaction and comfort. Transportation Research
Part A: Policy and Practice,78, 134–143.
Carstensen, T. A., & Ebert, A. K. (2012). Chapter 2 Cycling Cultures in
Northern Europe: From ‘Golden Age’to ‘Renaissance’,inJohn
Parkin (ed.) Cycling and Sustainability (Transport and Sustainability),
Emerald Group Publishing Limited, 1,23–58.
Cattell, R. B. (1966). The scree test for the number of factors.
Multivariate Behavioral Research,1(2), 245–276.
Chataway, E. S., Kaplan, S., Nielsen, T. A. S., & Prato, C. G. (2014).
Safety perceptions and reported behavior related to cycling in mixed
traffic: A comparison between Brisbane and Copenhagen.
Transportation Research Part F: Traffic Psychology and Behaviour,
23,32–43.
Cherchi, E., & Cirillo, C. (2010). Validation and forecasts in models
estimated from multi-days travel survey. Transportation Research
Record: Journal of the Transportation Research Board,2175(1),
57–64.
De Geus, B. (2007). Cycling to work: psychosocial and environmental fac-
tors associated with cycling and the effect of cycling on fitness and
health indexes in an untrained working population. Vrije Universiteit,
Department of Human Physiology and Sports Medicine, Vrije
Universiteit Brussel, Pleinlaan 2, 1050 Brussels, Belgium.
Dickinson, J. E., Kingham, S., Copsey, S., & Hougie, D. J. P. (2003).
Employer travel plans, cycling and gender: will travel plan measures
improve the outlook for cycling to work in the UK? Transportation
Research Part D: Transport and Environment,8(1), 53–67.
Dill, J., & Voros, K. (2007). Factors affecting bicycling demand: initial
survey findings from the Portland, Oregon, region. Transportation
Research Record: Journal of the Transportation Research Board,
2031(1), 9–17.
Emond, C. R., & Handy, S. L. (2012). Factors associated with bicycling
to high school: Insights from Davis, CA. Journal of Transport
Geography,20(1), 71–79.
Fern
andez-Heredia, 
A., Monz
on, A., & Jara-D
ıaz, S. (2014).
Understanding cyclists’perceptions, keys for a successful bicycle pro-
motion. Transportation Research Part A: Policy and Practice,63,1–11.
Garrard, J., Rose, G., & Lo, S. K. (2008). Promoting transportation
cycling for women: the role of bicycle infrastructure. Preventive
Medicine,46(1), 55–59.
Gatersleben, B., & Appleton, K. M. (2007). Contemplating cycling to
work: Attitudes and perceptions in different stages of change.
Transportation Research Part A: Policy and Practice,41(4), 302–312.
Habib, K. N., Mann, J., Mahmoud, M., & Weiss, A. (2014). Synopsis of
bicycle demand in the City of Toronto: Investigating the effects of
perception, consciousness and comfortability on the purpose of bik-
ing and bike ownership. Transportation Research Part A: Policy and
Practice,70,67–80.
Heinen, E., Maat, K., & Van Wee, B. (2011). The role of attitudes
toward characteristics of bicycle commuting on the choice to cycle
to work over various distances. Transportation Research Part D:
Transport and Environment,16(2), 102–109.
Heinen, E., Maat, K., & van Wee, B. (2013). The effect of work-related
factors on the bicycle commute mode choice in the Netherlands.
Transportation,40(1), 23–43.
Heinen, E., Van Wee, B., & Maat, K. (2010). Commuting by bicycle:
An overview of the literature. Transport Reviews,30(1), 59–96.
Hunecke, M., Bl€
obaum, A., Matthies, E., & H€
oger, R. (2001).
Responsibility and environment ecological norm orientation and
external factors in the domain of travel mode choice behaviour.
Environment and Behavior,33(6), 830–852.
Hunt, J., & Abraham, J. (2007). Influences on bicycle use.
Transportation,34(4), 453–470.
Kamargianni, M., & Polydoropoulou, A. (2013). Hybrid choice model
to investigate effects of teenagers’attitudes toward walking and
cycling on mode choice behavior. Transportation Research Record:
Journal of the Transportation Research Board,2382(1), 151–161.
INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 9

Kaplan, S., Manca, F., Nielsen, T. A. S., & Prato, C. G. (2015).
Intentions to use bike-sharing for holiday cycling: An application of
the Theory of Planned Behavior. Tourism Management,47,34–46.
Kingham, S., Dickinson, J., & Copsey, S. (2001). Travelling to work:
Will people move out of their cars. Transport Policy,8(2), 151–160.
Klapper, D., Ebling, C., & Temme, J. (2005). Another look at loss aversion
in brand choice data: can we characterize the loss averse consumer?.
International Journal of Research in Marketing,22(3), 239–254.
Krizek, K. J. (2012). Cycling, urban form and cities: What do we know
and how should we respond?: Cycling and Sustainability, 1, 111–130
Kurt, V. H. (2008). Literature search bicycle use and influencing factors
in Europe. Universiteit Hasselt –Instituut voor mobiliteit, Belgium.
La Paix, L., Cherchi, E., & Geurs, K. (2015). Integration of unobserved
effects in access/egress mode choice models. In NECTAR
Conference, Chicago, USA.
Li, Z., Wang, W., Yang, C., & Ragland, D. R. (2013). Bicycle commut-
ing market analysis using attitudinal market segmentation approach.
Transportation Research Part A: Policy and Practice,47,56–68.
Ma,L.,Dill,J.,&Mohr,C.(2014).Theobjectiveversustheperceived
environment: What matters for bicycling? Transportation,41(6),
1135–1152.
Mabit, S., Cherchi, E., Jensen, A., & Jordal-Jørgensen, J. (2015). The
effect of attitudes on loss aversion: estimation and validation for the
case of alternative-fuel vehicles. Transportation Research Part A:
Policy and Practice,82,17–28.
Majumdar, B. B., Mitra, S., & Pareekh, P. (2015). Methodological
framework to obtain key factors influencing choice of bicycle as a
mode. Transportation Research Record: Journal of the Transportation
Research Board,2512, 110–124.
Maldonado-Hinarejos, R., Sivakumar, A., & Polak, J. W. (2014).
Exploring the role of individual attitudes and perceptions in predict-
ing the demand for cycling: a hybrid choice modelling approach.
Transportation,41(6), 1287–1304.
Meloni, I., Sanjust, B., & Sottile, E. (2015). Modeling the propensity to
cycle. An experimental analysis. In World conference on transport
research –WCTR 2016 Shanghai. 10-15 July 2016.
Motoaki, Y., & Daziano, R. A. (2015). A hybrid-choice latent-class
model for the analysis of the effects of weather on cycling demand.
Transportation Research Part A: Policy and Practice,75, 217–230.
Mu~
noz, B., Monzon, A., & L
opez, E. (2016). Transition to a cyclable
city: Latent variables affecting bicycle commuting. Transportation
Research Part A: Policy and Practice,84,4–17.
Parkin, J., Wardman, M., & Page, M. (2007). Estimation of the deter-
minants of bicycle mode share for the journey to work using census
data. Transportation,35(1), 93–109.
Pooley, C., Horton, D., Scheldeman, G., Tight, M., Harwatt, H.,
Jopson, A., & Mullen, C. (2012). The role of walking and cycling in
reducing the impacts of climate change. Transport and Climate
Change,2, 175.
Pucher, J., & Buehler, R. (2006). Why Canadians cycle more than
Americans: A comparative analysis of bicycling trends and policies.
Transport Policy,13(3), 265–279.
Pucher, J., Dill, J., & Handy, S. (2010). Infrastructure, programs, and
policies to increase bicycling: An international review. Preventive
Medicine,50, S106–S125.
Pucher, J., Komanoff, C., & Schimek, P. (1999). Bicycling renaissance
in North America?: Recent trends and alternative policies to pro-
mote bicycling. Transportation Research Part A: Policy and Practice,
33(7-8), 625–654.
Sigurdardottir, S. B., Kaplan, S., Møller, M., & Teasdale, T. W. (2013).
Understanding adolescents’intentions to commute by car or bicycle
as adults. Transportation Research Part D: Transport and
Environment,24,1–9.
Stinson, M. A., & Bhat, C. R. (2005). A comparison of the route prefer-
ences of experienced and inexperienced bicycle commuters. In TRB
84th annual meeting compendium of papers (No. 05-1434),
Washington, DC: Transportation Research.
Vandenbulcke, G., Dujardin, C., Thomas, I., de Geus, B., Degraeuwe,
B., Meeusen, R., & Panis, L. I. (2011). Cycle commuting in Belgium:
Spatial determinants and ‘re-cycling’strategies. Transportation
Research Part A: Policy and Practice,45(2), 118–137.
Wardman, M., Tight, M., & Page, M. (2007). Factors influencing the
propensity to cycle to work. Transportation Research Part A: Policy
and Practice,41(4), 339–350.
Willis, J. R. (2015). Effect of recycled materials on pavement life-cycle
assessment: A case study. In Transportation research board 94th
annual meeting (No. 15-4109). Washington, DC: Transportation
Research Board.
10 E. SOTTILE ET AL.