Identifying cycling-inducing neighborhoods: A latent class approach

Abstract

Understanding how spatial attributes of cities and neighborhoods induce cycling is relevant for urban planning and policy making. In this work, ordered logit and latent class models are specified and estimated to analyze how the built environment affects bicycle-commuting frequency. Data come from a survey to 1,487 people in the city of Santiago, Chile, including sociodemographic information, travel behavior patterns and place of residence and work. Using geographic information systems tools, the built environment was characterized with variables calculated for a 500-m-radius buffer around the residential and work locations of each individual. Two models are estimated, first an ordered logit model confirms that built environment variables effect on cycling is similar to what has been reported in the literature, with some new findings such as an increase in cycling when public transport accessibility is low and the role of built environment attributes at the destination. Second, a latent class ordered logit is used to identify two classes of neighborhood in term of their cycling patterns, as a function of their density, presence of cycling infrastructure and distance to the main activity center of the city. This result allows to map the class membership probabilities, potentially helping to identify neighborhoods that encourage cycling and providing relevant information for policy making and infrastructure decisions.

Full text

Identifying cycling-inducing neighborhoods: A latent class approach
Ignacio Oliva
a
, Patricia Galilea
a
, and Ricardo Hurtubia
a
,
b
a
Departamento de Ingenier

ıa de Transporte y Log

ıstica, Vicu~
na Mackenna, Macul, Santiago, Chile;
b
School of Architecture, Pontificia Universidad
Cat
olica de Chile, Santiago, Chile
ARTICLE HISTORY
Received 6 June 2017
Revised 19 January 2018
Accepted 20 January 2018
ABSTRACT
Understanding how spatial attributes of cities and neighborhoods induce cycling is relevant for urban
planning and policy making. In this work, ordered logit and latent class models are specified and
estimated to analyze how the built environment affects bicycle-commuting frequency. Data come from a
survey to 1,487 people in the city of Santiago, Chile, including sociodemographic information, travel
behavior patterns and place of residence and work. Using geographic information systems tools, the built
environment was characterized with variables calculated for a 500-m-radius buffer around the residential
and work locations of each individual. Two models are estimated, first an ordered logit model confirms
that built environment variables effect on cycling is similar to what has been reported in the literature,
with some new findings such as an increase in cycling when public transport accessibility is low and the
role of built environment attributes at the destination. Second, a latent class ordered logit is used to
identify two classes of neighborhood in term of their cycling patterns, as a function of their density,
presence of cycling infrastructure and distance to the main activity center of the city. This result allows to
map the class membership probabilities, potentially helping to identify neighborhoods that encourage
cycling and providing relevant information for policy making and infrastructure decisions.
KEYWORDS
Bicycle commuting; built
environment; GIS; latent class
model; ordered logit
1. Introduction
Due to its characteristics, cycling has been identified as a trans-
portation mode likely to solve several problems cities are facing
and will face in the future. It does not require any kind of fuel,
being a “carbon zero”transportation mode (Chapman, 2007);
it improves population’s health (Pucher & Buehler, 2010) and
can solve traffic congestion problems (Pucher & Dijkstra,
2003). Based on these facts, urban planning and the design of
public spaces should encourage cycling.
The influence of the built environment in travel behavior
has been widely researched. In particular, mode choice and
the intensity of the use of nonmotorized modes have been
studied, confirming that the built environment has a strong
influence in travel patterns (Cao, Mokhtarian, & Handy,
2009). For example, Cervero (2002) analyzes the influence
of built environment on mode choice, while Handy, Boar-
net, Ewing, and Killingsworth (2002) found that certain
aspects of built environment promote more cycling and
walking among the population.
Notwithstanding the extensive literature linking built envi-
ronment and bicycle usage, we identify some aspects that have
not been deeply analyzed and that could provide a better
understanding of this phenomenon. First of all, most literature
focuses on bicycle usage for any purpose, mixing commuting,
utilitarian, and strolling trips. Only few studies focus exclu-
sively in commuting trips, which have distinctive characteris-
tics, such as occurring during specific hours and being longer
than utilitarian and strolling trips. In addition, the focus on
bike usage has been placed mostly on the probability of choos-
ing this mode over others instead of, for example, the frequency
of its usage.
Since strolling and utilitarian trips tend to be made within
the user’s neighborhood, the built environment characteristics
at the place where people work or study are not usually ana-
lyzed. Analyzing built environment characteristics at the desti-
nation might also be relevant when analyzing commuting trips
between different neighborhoods, as found by Winters, Brauer,
Setton, and Teschke (2010).
Additionally, most of the studies found in the literature refer
to cities located in Europe, the United States, and Canada,
whereas investigation applied to cities in developing countries
falls short. Developing countries exhibit different trip patterns,
with a lower modal split for motorized modes and a great
potential to increase bicycle usage. Therefore, it is interesting to
analyze how built environment affects bicycle usage in cities in
emerging countries.
Taking these aspects into consideration, we hereby propose
to model the weekly bicycle commuting frequency as a function
of socioeconomic characteristics and built environment varia-
bles, for both the origin and the destination points. In addition,
we estimate a latent class model (LCM) where the class mem-
bership function depends on attributes of the user’s residential
location. Hence, classes will describe neighborhoods instead of
type of users.
CONTACT Ignacio Oliva iroliva@uc.cl Departamento de Ingenier

ıa de Transporte y Log

ıstica, Vicu~
na Mackenna 4860, Macul, Santiago 7820436, Chile.
Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/ujst.
© 2018 Taylor & Francis Group, LLC
INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION
2018, VOL. 0, NO. 0, 1–13
https://doi.org/10.1080/15568318.2018.1431822

In this study, we identify relevant variables influencing bicy-
cle usage in the city of Santiago. Most of these variables confirm
trends reported in literature, but some reveal new factors that
are correlated with bicycle usage, such as the accessibility to
public transport at the origin and office density at the destina-
tion. After performing a latent class analysis, we find types of
neighborhoods which seem to induce different cycling behav-
iors among their population. The results of this analysis are
consistent with patterns identified in previous studies, yet they
propose a new way to define classes in this kind of analysis, as a
function of built environment attributes.
This article is organized as follows: Section 2reviews existent
literature on built environment and bicycle usage. Section 3
describes the methodology developed for this study’s purpose.
Section 4describes the case study, the data collection process
and analyzes some relevant statistics for the sample. Section 5
shows model estimation results. Finally, Section 6summarizes
the conclusions and suggests future research.
2. Built environment and cycling
Several studies have analyzed the relationship between built envi-
ronment and travel behavior. Many of them, conducted in the
United States (Badoe & Miller, 2000;Cervero,1988,1996;Ewing&
Cervero, 2001;Frank&Pivo,1994; Frank, Stone, & Bachman,
2000; Kitamura, Mokhtarian, & Laidet, 1997; McNally & Kulkarni,
1997; Rajamani, Bhat, Handy, Knaap, & Song, 2003), suggest the
built environment has an important influence on transport mode
choice. Specifically, higher residential densities and more diverse
land uses in the territory are related with a reduced use of car and a
higher use of transit, walking, and cycling. These variables are also
relatedwithlesskilometerstraveledbycarandafewerrateofsingle
occupant vehicles. Studies conductedinEuropehavefoundsimilar
results (van Wee & Handy, 2016).
Further studies have analyzed whether there is self-selection
when analyzing the built environment and travel behavior. Peo-
ple with particular travel habits may choose a certain kind of
neighborhood that suits their mobility preferences. In this con-
text, Krizek (2003) found that households that relocate change
their travel patterns, in part due to the changes in their nearby
built environment. In addition, cross-sectional studies have
been conducted in neighborhoods in different cities from the
United States (Cao, Handy, & Mokhtarian, 2006; Handy, Cao,
& Mokhtarian, 2005)finding that, after controlling for self-
selection, the effect of built environment on travel behavior can
be proven for different modes and travel purposes.
2.1. Influence of built environment on bicycle usage
The relationship between built environment and bicycling has
also been widely investigated. First of all, it has been shown
that the presence of bike lanes is related with higher rates of
cycling (Buehler & Dill, 2015; Heinen, van Wee, & Maat, 2010),
especially in the presence of well interconnected networks
(Buehler, 2012; Titze, Stronegger, Janschitz, & Oja, 2008). In
addition, larger travel distances imply longer travel times and a
higher effort to be made by the user, making bicycle less attrac-
tive. Therefore, longer travel distances, more likely found in
low density or spread cities, have a negative effect on bicycling
(Cervero & Duncan, 2003; Cui, Mishra, & Welch, 2014; Handy
& Xing, 2011).
A positive relation between residential density and bicycle
usage has been established (Pucher & Buehler, 2006; Sallis et al.,
2013), with cycling becoming more attractive due to the traffic
congestion associated to higher density (Forsyth, Oakes,
Schmitz, & Hearst, 2007). Land use diversity is also associated
with more cycling, since this implies there are activities and serv-
ices provided at short distances, making the use of bicycle more
attractive (Ewing & Cervero, 2010; Moudon et al., 2005). Never-
theless, land use mix has been associated with less bicycle usage
in one case: the city of Curitiba, Brazil (Hino, Reis, Sarmiento,
Parra, & Brownson, 2014). In this case, neighborhoods with a
higher diversity in land uses were also located in high-income
areas of the city, where people were more likely to travel by car.
Other interesting findings are that zones of the city with bet-
ter overall accessibility show a positive correlation with bicycle
usage (Cui et al., 2014; Kockelman, 1997; Rajamani et al.,
2003). Also, neighborhoods with more intersections seem to
induce more cycling (Sallis et al., 2013; Winters et al., 2010).
This can be explained because a more permeable urban form,
or denser road network, allows cyclist to find shorter routes,
making this transport mode more attractive.
Nevertheless, as mentioned before, sociodemographic varia-
bles play a key role when explaining bicycle usage. For example,
men tend to cycle more than women (Heinen, Maat, & van
Wee, 2011; Rodr
ıguez & Joo, 2004). When it comes to age, the
studies reviewed present contradictory results (Heinen et al.,
2010; Pucher & Buehler, 2010) and, in some cases, it is a non-
significant variable (Kitamura et al., 1997; Plaut, 2005). When
it comes to income, contradictory results are also found (Cer-
vero & Kockelman, 1997; Cui et al., 2014; Fern
andez-Heredia,
Jara-D
ıaz, & Monz
on, 2016; Kitamura et al., 1997). The same
situation is detected when controlling for education level (Cer-
vero & Gorham, 2009; Handy, Xing, & Buehler, 2010; Kockel-
man, 1997; Piatkowski & Marshall, 2015; Plaut, 2005). These
contradictory findings can be explained by historical and cul-
tural differences between the places where the studies were con-
ducted. Last, but not least, car ownership or access is related
with a lower use of the bicycle, while bicycle possession has the
opposite effect (Buehler, 2012; Cervero & Duncan, 2003).
2.2. Built environment analysis
Over the last few decades, several indicators have been pro-
posed to characterize the built environment. For example, Cer-
vero and Kockelman (1997) proposed to categorize these
indicators into three categories: “density,”“diversity,”and
“design.”Twelve years later, two additional categories of indi-
cators were proposed: “destination’s accessibility”and “distance
to public transport”(Cervero, Sarmiento, Jacoby, Gomez, &
Neiman, 2009). As reviewed by Ewing and Cervero (2010),
these make the “five D’s”categories for analyzing the built envi-
ronment and have been used in several studies (Larra~
naga,
Rizzi, Arellana, Strambi, & Cybis, 2016; Sehatzadeh, Noland, &
Weiner, 2011; Stewart & Moudon, 2014; Winters et al., 2010).
Under this analysis, density corresponds to the “amount of
one activity in a determined area”(Handy et al., 2002); diversity
is the “number of different land uses in a determined area”
2 I. OLIVA ET AL.

(Ewing & Cervero, 2010); design is the shape conformed by
blocks, streets, and sidewalks which compose a specific neigh-
borhood or area; destination accessibility indicates how easy is
to reach attractive places in the destination area; and distance
to public transport, as its name states, corresponds to the dis-
tance a person has to travel in order to access to public trans-
port services.
3. Methodology
We propose to use ordered logit to models to measure the rela-
tion of weekly cycling frequency with built environment attrib-
utes and socioeconomic characteristics. These models are
coherent with the ordinal nature of the dependent variable and,
besides providing a benchmark, are used to verify if the behav-
ior of the individuals in our case study is consistent with what
is reported in the literature. We also propose to use LCMs to
account for heterogeneity in user behavior, under the assump-
tion that, instead of user’s characteristics, what explains hetero-
geneity are neighborhood attributes.
3.1 Model for cycling frequency
An ordered logit model (McKelvey & Zavoina, 1975) is used to
model cycling frequency and following the notation used by
Sawkins, Seaman, and Williams (1997), the latent preference
y

of an individual can be modeled as:
yDb0xCe(1)
where bis a vector of parameters to be estimated, xis the vector
of characteristics of the individual and built environment
attributes at the origin and destination of her trip. The term, e
is an error term accounting for measuring errors and unob-
served factors, If ycorresponds to an ordered discrete variable,
it can be assumed that its value will vary according to certain
thresholds of perceived achieved utility mk:
yD0ifym0
yD1ifm0<ym1
yD2ifm1<ym2
yDKif mk¡1<y
(2)
In our case of study, ygoes from 0 to 5, which corresponds
to the amount of work days a person can commute to work,
not considering weekends.
Assuming a logistic standard distribution for the error term,
the probability of observing yDk, where kis the amount of
days commuted by bike, can be written as follows:
Py
nDkðÞD1
1Cexp ¡mkCb0xn

¡1
1Cexp ¡mk¡1Cb0xn

(3)
From Eq. (3) a maximum log-likelihood function can be
derived and maximized in order to estimate the parameters b
and thresholds mk. A detailed explanation of this can be found
on Greene (2003).
3.2. Latent class models
LCMs introduce taste heterogeneity by probabilistically seg-
menting the decision makers into groups of homogeneous
behavior. Thus, a specific set of parameters can be estimated
for each group or class (Kamakura & Russell, 1989). LCMs are
an interesting and powerful modeling instrument, because they
allow the capture of unobserved heterogeneity (Walker & Ben-
Akiva, 1999) in a way that is easier to interpret than other
approaches such as mixed logit or latent variable models (Hess,
Shires, & Jopson, 2013; Hurtubia, Nguyen, Glerum, & Bierlaire,
2014).
Due to their characteristics, LCMs have been widely used for
marketing studies. Nevertheless, during recent decades,
researchers have been using them to segment populations and
improve travel behavior analysis. For instance, Ben-Akiva et al.
(1999) highlight the opportunity this methodology presents in
hybrid choice models. In addition, LCM segmentation has been
used successfully to classify population by their characteristics,
lifestyle, and beliefs in studies of residential choice (Walker &
Li, 2007), car ownership (Bhat & Guo, 2007), route selection
(Greene & Hensher, 2003), and bicycle demand (Motoaki &
Daziano, 2015).
We have found few studies linking built environment and
travel behavior using class membership models. For example,
HOSHINO (2010), Olaru, Smith, and Taplin (2011), and Meng,
Taylor, and Scrafton (2016) indentify behavioral classes associ-
ated to stated and revealed residential location preferences, while
Smith and Olaru (2013) identify classes related to different “life-
stages”which have different sensitivities to built environment
attributes. To the extent of our literature review, no author has
yet proposed a LCM where segmentation responds explicitly to
built environment variables. Instead of segmenting population,
we propose to segment residential neighborhoods according to
their attributes. As there are aspects of individuals which we are
not able to see, segmentating by urban attributes while keeping
socioeconomic characteristics as explanatory variables may help
to understand how different urban configurations are related to
certain travel behavior patterns.
To introduce latent clases, we define an ordinal preference
latent variable similar to that of Eq. (1), but conditional on
belonging to a class s, hence having a class-specific vector of
parameters bs. This means that the ordinal probability of
Eq. (3.3) will also be conditional on belonging to a class:
Py
nDkjsðÞD1
1Cexp ¡ms
kCbs0xn

¡1
1Cexp ¡ms
k¡1Cbs0xn

(4)
Since latent classes cannot be deterministically assigned to a
specific individual or land buffer in the case of this study,
the proposed methodology assumes that the class membership
is probabilistic, depending on characteristics of the built
INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 3

environment in each individual’s residential location BEn. This
relation can be expressed as a class membership equation fas
follows:
Fns Df BEn;gs

Cens (5)
where Fns is the continuous latent variable which relates to the
probability of belonging to class sand gsis a set of parameters
to be estimated. Assuming that ens distributes i.i.d EV (0,1), the
probability of individual nbelonging to a determined class sis:
PnsðÞDexp f BEn;gs

Xr2Sexp f BEn;gr
 (6)
Finally, from Eqs. (4) and (6), the probability of an individ-
ual nchoosing to perform kcommuting trips by bicycle in a
week can be written as:
Py
nDkðÞDX
s2S
Py
nDkjsðÞPnsðÞ (7)
4. Case study and data collection
Santiago is a city with more than seven million inhabitants and is
the capital of Chile. It has experienced a dramatic increase in the
share of trips performed by bicycle, from 2% in 2001 to 4% in 2012
(SECTRA, 2015), adding up to approximately 750,000 daily trips
and still growing. Since 2010, a lot of cycling infrastructure has
been built in Santiago, which may explain this trend, together with
asignificant increase in congestion in central areas of the city.
However, the subdivision of the city in 37 independent administra-
tive zones (comunas) has produced significant spatial heterogeneity
in the quantity and quality of the infrastructure and other cycling
oriented policies, as well as urban and transport planning in gen-
eral. This turns Santiago into an interesting case study, because the
location of the infrastructure does not necessarily respond to typi-
cal planning variables such as predicted demand (due to high den-
sity), road hierarchy or connectivity to activity centers. This
translates into a significant heterogeneity of combinations of attri-
bute levels in the area of study, especially for relevant variables such
as density, zonal income, job accessibility and presence of cycling
infrastructure.
Additionally, Santiago is still growing at a fast pace, with an
important densification of its central areas and urban sprawl in
its periphery. Therefore, it is relevant to know how to orient
this urban growth to induce more sustainable mobility patterns
in the near future. The models proposed in this research may
be used as tools for this objective.
Travel behavior and socioeconomic data for the estimation
of the proposed models were obtained from a survey specially
developed for this purpose. Land use and transport network
data describing locations and their built environment were col-
lected from secondary sources. The following subsections
describe these efforts.
4.1. Survey
The instrument developed consists of a 30 questions survey,
which was divided in four parts to be completed by the
interviewees in approximately 10 min. The first section con-
sisted of socioeconomic information related to the inter-
viewees and their households. The second section gathered
information about their travel behavior, the place of resi-
dence and studies or work location. The last two sections
were part of a different research project (Rossetti, 2017),
consisting of a revealed preferences questionnaire about
bike lane design and questions about attitudinal and percep-
tual beliefs of the interviewee.
The survey was conducted between March 21 and April 26,
2016. This period corresponds to the transition between sum-
mer and autumn in the Southern Hemisphere, guaranteeing a
good cycling environment characterized by warm tempera-
tures, sunny days, and a relative absence of rain.
In order to get cyclist and noncyclist commuters’information, a
three-step strategy was designed. First, cyclists were intercepted in
11 different bike lanes in the city (Figure 1). Trained interviewers,
equipped with tablets, approached cyclists while they were passing
through the interception point and invited them to answer the sur-
vey. This field work was conducted during three different time peri-
ods: between 07:30 and 09:30, between 13:00 and 15:00, and
between 18:00 and 20:00, corresponding to peak hours of the day.
If the cyclist did not have time to answer the survey, they were
offered to answer it later, through an e-mail sent to their personal
address. During this step, 1,050 answers were obtained on the field
and 355 by e-mail. Interception points were placed around the his-
torical city center and the central business district. This allowed to
capture commuters coming in and out of the main activity centers
of the city.
The second and third steps were designed to capture infor-
mation about people who did not use the bicycle. In this con-
text, flyers were distributed in households of selected
neighborhoods, among car drivers in intersections and by plac-
ing them in strategic places, such as parking lots. An additional
295 observations were obtained this way. Finally, the third step
consisted in an online survey distribution, reaching 905 people.
In total, 2,605 observations were collected. However, after data
cleaning, only 1,487 surveys were identified as correctly com-
pleted and included in the sample for estimation.
4.2. Sample statistics
Collected respondent data used as explanatory variables in the
models are gender, age, education, occupation, income, cars,
and bicycles per household and household size. The dependent
variable is the weekly number of commuting trips made by
bicycle, which goes from zero to five. For this study, we did not
consider commuting trips during weekends. Descriptive varia-
bles of sociodemographic variables can be found in Table 1.
A modal share analysis shows that 59.94% of respondents
are bicycle commuters, as expected due to the way the survey
was applied (Table 2). A total of 7.29% corresponds to car com-
muters and 21.98 to transit commuters.
4.3. Built environment indicators
By using geographic information systems (GIS), households
and workplaces were georeferenced, as it can be seen in
Figure 2. With data from the National Census (INE, 2011),
4 I. OLIVA ET AL.

OpenStreetMaps (OpenStreetMaps contributors, 2017), and the
National Tax Agency (SII, 2014), built environment indicators
were estimated within a 500-m-radius buffer. While this scale
has been used and validated in previous studies for active trans-
port (Cervero et al., 2009; Hino et al., 2014; Larra~
naga et al.,
2016; Winters et al., 2010; Zegras, 2010), we also explored other
possible buffer sizes (between 250 and 750 m). Preliminary esti-
mation results showed that variables calculated at the 500-m
scale allowed for much higher significance of the parameters,
hence confirming that this was an adequate scale to describe
the built environment around each location.
As mentioned in Section 2.2, the variables are classified into
five categories: density, diversity, design, accessibility and dis-
tance to public transport. We included a sixth category—
measures of travel time and distance, which are relevant when
analyzing bicycle commuting (Handy & Xing, 2011; Yang &
Zacharias, 2016) and for short-term transport modeling (Ben-
Akiva & Bierlaire, 1999). This information is presented in
Table 3.
It is important to notice that Altitude difference corre-
sponds to the subtraction between the altitude at the desti-
nation minus the altitude at the origin, hence a positive
value indicates uphill cycling from home to work. This is
relevant since Santiago’s altitude goes from 400 to 800 m
above sea level, from west to east in an approximately
30 km span.
The Entropy Index was built based on Zegras’(2010) work
for the city of Santiago, which is similar to others found in
Figure 1. Survey interception points in Santiago.
Table 1. Descriptive statistics for the dependent and socioeconomic variables.
Mean
*
SD N
Dependent variable
Weekly commuting bicycle trips 3.11 2.21 —
Sociodemographic
Female (1,0) 0.37 0.48 548
Age (years) 32.10 10.54 —
Household size 3.32 1.73 —
Sons or daughters (1,0) 0.35 0.48 521
Occupation student (1,0) 0.25 0.43 372
Occupation formal or informal work (1,0) 0.70 0.46 1,048
Non occupation (1,0) 0.05 0.21 67
Low and middle educational level (1: elementary, high school comp/incomp, 0: other) 0.10 0.30 150
High educational level (1: college or higher education comp/incomp, 0: other) 0.73 0.44 1,085
Graduate educational level (1: postgraduate studies comp/incomp, 0: other) 0.17 0.38 252
Low income (1: <CLP 450.000, 0: other) 0.09 0.28 132
Middle income (1: CLP 450.000 to 1.000.000, 0: other) 0.31 0.46 461
Middle-high income (1: CLP 1.000.000 to 3.000.000, 0: other) 0.44 0.50 658
High income (1: CLP >3.000.000, 0: other) 0.16 0.37 236
One car at home (1, 0) 0.44 0.50 650
Two or more cars at home (1,0) 0.20 0.40 302
Bicycles at home 2.17 1.30 —

Corresponds to the percentage of the observations with a value of 1 in the categorical variable.
INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 5

literature (Bhat & Gossen, 2004; Bhat & Guo, 2007; Rajamani
et al., 2003) and is presented in Eq. (8). In this index, scorre-
sponds to the entropy index, vto square meters of dwellings, c
to commerce, tto offices, ito industry, eto education, and oto
others. Tcorresponds to the sum of these variables.
5. Results
In the following section, results are analyzed for two mod-
els. First, we estimate an ordered logit model (see Sec-
tion 3.1) to understand commuting bicycle frequency, as
explained by socioeconomic and built environment varia-
bles. Second, we estimate a LCM (see Section 3.2)to
understand the role of neighborhood types in cycling
behavior.
5.1. Cycling frequency
First, a base-line model is proposed, explaining cycle frequency
as a function of individual and travel characteristics. After this
exercise, the effect on the built environment is included in the
analysis.
As it can be seen in Table 4 (first column), the availability of
different transport modes has a statistically significant effect.
People who have one car at home is less likely to commute by
bicycle over the week. The effect of a second or more cars is
even bigger, since car availability per person increases. On the
other hand, households with more bicycles are more likely to
use them, probably for analogue reasons as the car. The size of
the household has a negative impact on bicycle commuting,
suggesting that larger homes have more complex trip patterns,
hence preferring other transport modes.
When analyzing the individuals’occupation, students and
workers, are significantly more likely to use the bicycle for com-
muting. Finally, the effect of income and gender is confirmed
for Santiago’s case, with high-income individuals and females
being less likely to frequently commute by bicycle. While the
reason behind the income effect is unclear, there are two possi-
ble explanations for the gender effect. First, women may be
more aware of the risks bicycling implies (Baker, 2009) and,
secondly, because of family dynamics (specially in developing
Table 2. Sample’s modal share.
Mode NPercent
Car 129 7.29%
Transit 229 21.98%
Bicycle 1,061 59.94%
Walking 139 7.85%
Taxi 32 1.81%
Other 20 1.13%
sD1¡jv
T¡1
6jCjc
T¡1
6jCj t
T¡1
6jCj i
T¡1
6jCje
T¡1
6jCjo
T¡1
6j
5
3

(8)
Figure 2. Origins and destinations of the sample (red dots) and origins and destinations of people who commutes by bike at least once a week (blue dots).
6 I. OLIVA ET AL.

countries), women tend to deal with more household-related
responsibilities than men, such as shopping or taking care of
children, which are not compatible with bicycle commuting if
proper infrastructure is not provided.
The negative value of the parameter for altitude difference
means that a person is more likely to use the bicycle when the
trip to work is downhill, since a smaller effort is required and
the likelihood of sweating is reduced. As expected, longer trips
are less likely to be made by bicycle. However, an interesting
counter-effect takes place when commuting time by public
transport is high.
5.1.1. Cycling frequency explained by built environment
variables
Having estimated the base-line model, built environment varia-
bles at the origin and destination are included (third column of
Table 4). It is important to notice that an incremental approach
was followed, first including only origin attributes (second col-
umn of Table 4) in order to be coherent with most of the litera-
ture on built environment, which focuses on the surroundings
of dwellings.
When estimating the model with built environment varia-
bles, no significant differences with the base-line model are
observed, with the exception of two parameters: household
size, which reduces its significance, and the middle-high
income dummy variable, which becomes significant.
There are three statistically significant attributes at the ori-
gin: number of dwellings within the buffer (residential density),
distance to the nearest bus stop, and the average length of the
bike lanes intersecting the buffer. The positive effect of density
was expected, as denser zones tend to be more congested, mak-
ing more attractive to use nonmotorized modes (Forsyth et al.,
2007). Additionally, low-density zones tend to be related with
suburban parts of the city, which are generally far from the
main activity centers and with low access to public transport,
hence encouraging the use of car (Cervero, 1988).
Neighborhoods with more and longer bike lanes tend to
facilitate cycling to their inhabitants, especially if there is a well-
connected network (Buehler, 2012). On the other hand, with
longer distances to the nearest bus stop, the likelihood of cycling
decreases. This result should be analyzed together with the
(opposite) effect of commuting time by transit, and may be sug-
gesting that physical access to public transport is correlated with
overall accessibility, hence inducing cycling, while poor access to
opportunities through public transport has the opposite effect.
The inclusion of built environment variables describing the
destination zone strengthens the model in terms of fit and
explanatory power. Four destination-specific and statistically
significant variables are included. In this context, the only fac-
tor that promotes bicycle usage is office-density. For similar
reasons to residential density, places with a higher concentra-
tion of offices may turn cycling into a good alternative to avoid
Table 3. Descriptives for built environment variables at the origin and destination of the trip.
Origin Destination
Mean SD Mean SD
Density variables
Number of dwellings 5,389.76 3,209.80 5,118.64 2,977.94
Number of offices 96.87 102.81 121.75 112.81
Number of commerce stores 270.98 299.96 357.09 381.74
Diversity variables
Entropy index 0.41 0.15 0.51 0.13
Parks area (m
2
) 31,786.72 43,253.08 42,131.34 53,339.36
Design variables
Bike lane average length (m) 1,220.47 1,172.47 1,329.42 1,211.64
Meters of bike lanes within the buffer 737.20 689.72 729.31 639.75
Number of bike lanes 1.49 1.55 1.58 1.41
Distance to closest bike lane (m) 483.72 597.65 406.38 437.31
Street average length (m) 57.10 17.80 50.29 15.45
Meters of street within the buffer 16,926.88 3,912.09 17,442.23 3,566.10
Number of street intersections 39.66 25.86 46.93 31.43
SD of angles formed in every street intersection 101.55 10.20 101.30 8.97
SD street length 53.85 21.68 51.63 16.84
Average block size (m
2
) 41,863.90 245,035.30 23,106.03 54,702.97
Accessibility variables
Distance to Alameda–Providencia Av. 5,497.77 38,543.44 3,645.34 17,392.51
Distance to central square 6,596.34 4,612.99 5,738.84 4,375.12
Distance to public transport
Metro stations within the buffer 0.49 0.66 0.82 0.74
Distance to closest metro station (m) 1,006.03 1,048.37 629.32 852.66
Number of bus stops within the buffer 16.94 6.99 17.56 7.43
Number of bus services within the buffer 60.49 42.67 73.23 51.82
Distance to closest bus stop (m) 101.17 122.87 93.88 90.99
Travel variables
Altitude difference (m) ¡3.62 79.72
Travel time by car (min) 16.78 7.13
Travel time by transit (min) 34.86 16.00
Travel time by walking (min) 86.25 52.98
Distance by car (km) 8.80 6.36
Distance by transit (km) 8.23 5.31
Distance by walking (km) 6.86 4.27
INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 7

traffic congestion. As negative factors, the number of subway
(metro) stations within the buffer reveals to what extent indi-
viduals rely on this service in Santiago and how, for many,
cycling is not a competitive alternative. Distance to the closest
bike lane has, as expected, a negative effect meaning that (as in
the origin) the presence of a dense network of bike lanes
encourages cycling. This responds to the fact that people do
not want to ride without proper infrastructure (Buehler & Dill,
2015), especially in zones that tend to be congested and to have
narrow streets. Finally, entropy (land use mix) appears as a fac-
tor that discourages cycling, which contradicts the existing lit-
erature. One plausible explanation for this is that people prefer
not to use their bikes when they have the possibility (or need)
of conducting several different activities (Heinen, Maat, & van
Wee, 2013). However, this could also be because high-entropy
places are mostly located in the historical city center, which is
highly accessible through public transport.
5.2. Latent class model
A LCM was estimated to categorize neighborhoods according
to their cycling behavior, as a function of built environment
characteristics of the residential location of users. Considering
the available variables, several different model specifications
were explored until finding the one with the best goodness of
fit and coherent segmentation. Models with more than two
classes were specified and estimated, but none rendered signifi-
cant or meaningful parameters. As a result, a two latent classes
specification was selected as the final one, where the class mem-
bership model is a function of the number of dwellings (resi-
dential density) within the buffer, the distance to the main
activity center of the city,
1
and the logarithm of the total length
of bike lanes that pass through the buffer. Estimation results
are shown in Table 5.
Interpretation of the parameters allows to label each class.
Locations belonging to latent class 1 (LC1) are characterized for
being further away from the main activity centers, having a
higher dwelling density and lower presence of bike lanes than
those of latent class 2 (LC2). Dwellers of places belonging to LC1,
despite being less sensible to distance and slope, are less likely to
commute by bike under several socioeconomic circumstances
(being female, of high income, having a car or a large household).
This can be interpreted as bicycle users that are less diverse. Nev-
ertheless, they seem to be willing to bike longer distances (which
is expected since these neighborhoods tend to be far from activity
Table 4. Bicycle frequency to work explained by built environment variables.
Base-line model Origin variables Origin and destination variables
Variable Value t-test Value t-test Value t-test
Socioeconomic variables
1 Car at home ¡0.303 ¡2.390 ¡0.276 ¡2.170 ¡0.293 ¡2.290
2 or more cars at home ¡0.677 ¡3.880 ¡0.632 ¡3.610 ¡0.718 ¡4.210
Number of bicycles at home 0.531 10.580 0.532 10.560 0.503 10.690
Student 1.160 3.890 1.140 3.820 1.370 4.540
High income ¡0.724 ¡4.040 ¡0.746 ¡4.140 ¡0.732 ¡4.020
Middle-high income ¡0.230 ¡1.88
*
¡0.260 ¡2.090 ¡0.270 ¡2.150
Female ¡0.500 ¡4.490 ¡0.520 ¡4.650 ¡0.513 ¡4.520
Household size ¡0.084 ¡2.210 ¡0.065 ¡1.68
*
--
Employed 1.870 6.500 1.860 6.380 1.880 6.490
Destination built environment variables
Number of offices within the buffer/1,000 ————1.480 2.610
Distance to closest bike lane (km) ————¡0.795 ¡6.170
Entropy ————¡1.590 ¡3.280
Number of metro stations within the buffer ————¡0.174 ¡1.990
Origin built environment variables
Number of dwellings within buffer 1,000 ——0.037 2.020 0.044 2.400
Distance to closest bus stop (km) ——¡0.001 ¡2.120 ¡0.001 ¡2.000
Average length of bike lanes that go through the buffer (km) ——0.109 2.310 0.119 2.500
Travel variables
Altitude difference (km) ¡3.990 ¡5.440 ¡3.330 ¡4.360 ¡3.280 ¡4.230
Distance (km) ¡0.141 ¡7.010 ¡0.145 ¡7.090 ¡0.128 ¡6.070
Commuting time by transit (min) 0.021 3.940 0.026 4.750 0.028 4.740
Thresholds
a
m
0
(one trip) 0.587 1.76
*
0.371 2.740 0.330 0.77
**
d
1
(two trips) 0.107 5.150 0.021 5.150 0.113 5.150
d
2
(three trips) 0.168 6.660 0.026 6.660 0.178 6.660
d
3
(four trips) 0.370 10.440 0.036 10.440 0.391 10.450
d
4
(five trips) 0.441 11.930 0.037 11.930 0.461 11.940
Final log likelihood ¡1,697.257 ¡1,688.579 ¡1,658.152
r
2
0.179 0.183 0.198
Adjusted r
2
0.171 0.173 0.186

Not significant at 95%.

Not significant at 90%.
a
mkDm
k-1
Cd
k
.
1
Conformed by Alameda, Providencia, and Apoquindo avenues, which concen-
trate most work places in the city (Niehaus, 2016), including the historical city
center and several business districts.
8 I. OLIVA ET AL.

centers) and are less sensitive to slope. Therefore, we assign the
label “Homogeneous Riders Neighborhoods”to LC1.
On the other hand, LC2 corresponds to lower density
neighborhoods with a good presence of bicycle infrastruc-
ture and closer to the main activity center. Dwellers of LC2
places are not affected in a negative way by any socioeco-
nomic characteristic, suggesting more diverse cycling com-
muters are spawn from these areas. However, they are more
sensitive to distance and slope, probably because they are
likely to dwell in places that are already close to activity
centers. We assign the label “Diverse Riders Neighbor-
hoods”to LC2.
When analyzing the value of thresholds, an unexpected neg-
ative value for d2in LC1 is found, which is not coherent with
an ordered logit model. Nevertheless, it is not statistically sig-
nificant, which implies its real value is cero. Therefore, for LC1,
commuting by bike 2 or 3 days a week is considered as the
same in this model. A similar situation is detected for d1in
LC2, implying that commuting by bike once or twice a week is
statistically the same for this class.
We conclude that LC2 denotes a type of neighborhood that
is friendlier for cycling, inducing it in a more diverse group of
users. All but one of the spatial variables explaining member-
ship to this class are consistent in this regard to what is found
in the literature, since residential density is systematically
reported as an inducer of cycling and walking, which contra-
dicts our results. However, Santiago’s high-rise residential
buildings are not characterized by being very friendly with
cycling, often lacking safe street-level parking and forcing its
dwellers to park their bikes in locked-down underground facili-
ties or to carry their bikes up and down through elevators or
stairs. Therefore, living in a high-rise residential building in
Santiago often discourages cycling. Moreover, Santiago is quite
polarized in this regard, with high density explained mostly by
high rise. We believe that this particular characteristic of the
residential supply of Santiago explains the negative parameter
for density in the membership to the class that encourages
more cycling (LC2).
The estimation of a LCM based on built environment attrib-
utes may be controlling for residential self-selection since it
mimics, to some extent, the location choice process of individu-
als. While choosing to locate in a LC2 neighborhood may be
due to preexisting preferences for cycling commuting, we know
that the remaining (socioeconomic and travel) variables are
likely to be free of self-selection bias. This, however, requires
further validation.
This approach should allow to explore “differences
within differences,”for example, by analyzing the role of
density within each latent class (already defined by density).
This was attempted but results were not satisfactory due to
correlation issues that triggered numerical estimation prob-
lems. Further research will explore methods to achieve this.
5.3. LC1 and LC2 distribution in the city of Santiago
The latent class segmentation analysis was applied for the
whole city of Santiago. The city was divided in a squared grid
where each edge is 500 m long. In the centroid of each cell, var-
iables within a 500-m-radius buffer were calculated. Once vari-
ables where calculated for each cell, the class membership
probability was estimated using Eq. (3.6) and the parameters
from Table 5.
Results of this analysis are shown in Figure 3.Adarkercolor
represents a higher probability of belonging to LC2. It is interest-
ing to notice the influence of bike lanes, which can be clearly seen
in dark orange in the map. This confirms that investment in
Table 5. Latent class model estimation.
LC1 LC2
Variable Value t-test Value t-test
Socioeconomic characteristics
1 or more car at home ¡0.805 ¡3.1 ——
Number of bicycles at home 0.95 5.89 0.266 2.88
Female ¡1.27 ¡4.66 ——
Household size ¡0.277 ¡3.41 ——
High income ¡1.05 ¡3.58 ——
Travel variables
Altitude difference (km) ¡2.96 ¡1.9
*
¡8.01 ¡4.95
Distance (km/10) ¡0.571 ¡2.01 ¡1.49 ¡4.73
Class membership variables
ASC_2 ——0.542 1.12
**
Number of dwellings within the buffer/1,000 ——¡0.15 ¡2.37
Distance to main activity centers (m/100) ——¡0.00699 ¡3.18
LN of sum of length of bike lanes that pass through the buffer ——0.083 2.59
Thresholds
a
m0(one trip) ¡1.35 ¡3.58 ¡1.77 ¡3.88
d1(two trips) 0.11 2.07 0.113 1.79
*
d2(three trips) ¡0.367 ¡1.67
*
0.713 3.17
d3(four trips) 0.275 2.36 0.475 4.24
d4(five trips) 0.439 3.4 0.483 4.08
Final log likelihood ¡1,727.159
r20.164
Adjusted r20.153

Not significant at 95%.

Not significant at 90%.
a
m
kD
m
k-1
Cd
k
.
INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 9

dedicated bicycle infrastructure is crucial for making cycling more
attractive for a wider group of users. The spatial distribution of the
variables considered in the latent class are shown in Figure 4.
Finally, it is interesting to see how the LCM relates with
what is actually seen in the city of Santiago in terms of cycling
behavior. Figure 5 shows the bicycle daily trip-generation rates
per person for each comuna of the city. As it can be seen, there
are larger trip-generation rates in central and eastern comunas.
Especially in those which are close to the main city axis. If com-
pared with the LCM results (Figure 3), there is some similarity
between zones with a higher probability to belong to LC2 and
communes with higher bike trip-generation rates.
6. Conclusions
After controlling for sociodemographic variables and trip dis-
tance, the effect of the built environment on bike commuting
frequency was analyzed. The influence will vary depending on
the type of analyzed location: residential place (trip origin) or
work place (trip destination).
From the ordered logit model, it was found that, at the ori-
gin, residential density and bike lanes length have a positive
effect on cycling commuting, confirming previous findings
from the literature. A novel contribution of this work is the
exploration of the effect of built environment attributes on the
destination (work locations), finding that presence of bike lanes
and high density of offices are likely to induce frequent com-
mute by bicycle.
Another contribution is the specification and estimation of a
LCM based on residential neighborhoods characteristics. This
methodological innovation explores a new way to segment and
identify neighborhoods, according to the cycling behavior
observed in them, as a function of built environment attributes.
Two latent classes were identified. The “Homogeneous
Riders Neighborhoods”are likely to have more male, single
and lower income commuters who are willing to cycle for lon-
ger distances. The Diverse Riders Neighborhoods are likely to
have more miscellaneous commuters and to be friendlier to
cycling in general, as they are closer to activity centers and have
a higher presence of cycling infrastructure. An interesting result
is the fact that this type of neighborhood also tends to be of
lower density, which may be explained by the way in which
high-density materializes in Santiago.
Results confirm the importance of cycling infrastructure as
an element that induces cycling, although residential self-selec-
tion could be playing a role in this. However, the use of latent
classes applied to locations may be controlling for this effect.
Figure 4. Graphic representation of the class membership variables in the city of Santiago.
Figure 5. Bicycle trips per person generated in communes of Santiago.
Figure 3. Probability of belonging to LC2 for Santiago de Chile.
10 I. OLIVA ET AL.

Further research will explore and validate this, together with
the inclusion of latent variables related to individual indicators
of attitudes toward cycling (already collected in the survey). As
studied by Heinen and Handy (2012), these variables have a
significant effect in the decision of commuting by bicycle and
might help to account for self-selection (Mokhtarian & Cao,
2008).
Future research should attempt to include more variables
describing urban design, such as qualitative attributes of urban
spaces or type of cycling infrastructure, which may also have a
relevant influence on cycling behavior (Rossetti, 2017; Rossetti,
Saud, & Hurtubia, 2017)
Finally, as identified by Aldred, Woodcock, and Goodman
(2015) for the city of London, more cycling among the popula-
tion does not necessarily imply further diversity of cyclists.
Taking this into consideration, future research should analyze
how the built environment affects specific groups, finding ways
to encourage cycling in those that are currently less likely to do
it, such as women and children.
Acknowledgments
The authors want to thank to Tom
as Cox for his valuable help with the cal-
culation of built environment variables. This research was partially funded
by FONDECYT (Project number 1180605), the Complex Engineering Sys-
tems Institute (ICM: P-05-004-F, CONICYT:FBO16) the Center for Sus-
tainable Urban Development (CEDEUS, CONICYT/FONDAP 15110020)
and supported by the BRTCCentre of Excellence funded by the Volvo
Research and Educational Foundations (VREF).
Funding
Fondo de Fomento al Desarrollo Cient
ıfico y Tecnol
ogico (1180605).
References
Aldred, R., Woodcock, J., & Goodman, A. (2015). Does more cycling mean
more diversity in cycling? Transport Reviews,1647(July), 1–17.
doi:10.1080/01441647.2015.1014451
Badoe, D. A., & Miller, E. J. (2000). Transportation–land use interaction:
Empirical findings in North America and their implications for model-
ing. Transportation Research –D,5, 235–263. doi:10.1016/S1361-9209
(99)00036-X
Baker, L. (2009). How to get more bicyclists on the road: To boost urban
cycling, figure out what women want. Scientific American,301(4), 28–
29.
Ben-Akiva, M., & Bierlaire, M. (1999). Discrete Choice Methods and their
Applications to Short Term Travel Decisions. In: Hall R.W. (eds)
Handbook of Transportation Science. International Series in Opera-
tions Research & Management Science, vol 23. Boston, MA: Springer.
Ben-Akiva, M., Walker, J., Bernardino, A. T., Gopinath, D. A., Morikawa,
T., & Polydoropoulou, A. (1999). Integration of Choice and Latent Var-
iable Models, forthcoming, International Association of Traveler Behav-
ior Research (IATBR) book from the 1997 Conference in Austin, Texas.
Bhat, C. R., & Gossen, R. (2004). A mixed multinomial logit model analysis
of weekend recreational episode type choice. Transportation Research
Part B: Methodological,38(9), 767–787. doi:10.1016/j.trb.2003.10.003
Bhat, C. R., & Guo, J. Y. (2007). A comprehensive analysis of built environ-
ment characteristics on household residential choice and auto owner-
ship levels. Transportation Research Part B: Methodological,41(5),
506–526. doi:10.1016/j.trb.2005.12.005
Buehler, R. (2012). Determinants of bicycle commuting in the Washing-
ton, DC region: The role of bicycle parking, cyclist showers, and free
car parking at work. Transportation Research Part D: Transport and
Environment,17(7), 525–531. doi:10.1016/j.trd.2012.06.003
Buehler, R., & Dill, J. (2015). Bikeway networks: A review of effects on
cycling. Transport Reviews,1647(July 2015), 1–19. doi:10.1080/
01441647.2015.1069908
Cao, X., Handy, S., & Mokhtarian, P. (2006). The influences of the built
environment and residential self-selection on pedestrian behavior: Evi-
dence from Austin, TX. Transportation,33(1), 1–20. doi:10.1007/
s11116-005-7027-2
Cao, X., Mokhtarian, P., & Handy, S. (2009). Examining the impacts of res-
idential self selection on travel behaviour: A focus on empirical find-
ings. Transport Reviews, 29. doi:10.1080/01441640802539195
Cervero, R. (1988). Land-use mixing and suburban mobility. Transporta-
tion Quarterly,42(3), 429–446. doi:10.1068/a201285
Cervero, R. (1996). Mixed land-uses and commuting: Evidence from the
American housing survey. Transportation Research Part A: Policy and
Practice,30(5 PART A), 361–377. doi:10.1016/0965-8564(95)00033-X
Cervero, R. (2002). Built environments and mode choice: Toward a nor-
mative framework. Transportation Research Part D: Transport and
Environment,7(4), 265–284. doi:10.1016/S1361-9209(01)00024-4
Cervero, R., & Duncan, M. (2003). Walking, bicycling, and urban land-
scapes: Evidence from the San Francisco bay area. American Journal of
Public Health,93(9), 1478–1483. doi:10.2105/AJPH.93.9.1478
Cervero, R., & Gorham, R. (2009). Commuting in transit versus automo-
bile neighborhoods. Journal of the American Planning Association,61
(2), 210–225. doi:10.1080/01944369508975634
Cervero, R., & Kockelman, K. (1997). Travel demand and the 3Ds: Density,
diversity, and design. Transportation Research Part D: Transport and
Environment,2(3), 199–219. doi:10.1016/S1361-9209(97)00009-6
Cervero, R., Sarmiento, O. L., Jacoby, E., Gomez, L. F., & Neiman, A.
(2009). Influences of built environments on walking and cycling: Les-
sons from Bogot
a. International Journal of Sustainable Transportation,
3(4), 203–226. doi:10.1080/15568310802178314
Chapman, L. (2007). Transport and climate change: A review. Journal of
Transport Geography,15(5), 354–367. doi:10.1016/j.jtrangeo.2006.11.008
Cui, Y., Mishra, S., & Welch, T. F. (2014). Land use effects on bicycle rider-
ship: A framework for state planning agencies. Journal of Transport
Geography,41, 220–228. doi:10.1016/j.jtrangeo.2014.10.004
Ewing, R., & Cervero, R. (2001). Travel and the built environment: A syn-
thesis. Transportation Research Record,1780(Paper No. 01-3515), 87–
114. doi:10.3141/1780-10
Ewing, R., & Cervero, R. (2010). Travel and the built environment. Journal
of the American Planning Association,76(3), 265–294. doi:10.3141/
1780-10
Fern
andez-Heredia, 
A., Jara-D
ıaz, S., & Monz
on, A. (2016). Modelling
bicycle use intention: The role of perceptions. Transportation,43(1), 1–
23. doi:10.1007/s11116-014-9559-9
Forsyth, A., Oakes, J. M., Schmitz, K. H., & Hearst, M. (2007). Does resi-
dential density increase walking and other physical activity? Urban
Studies,44(4), 679–697. doi:10.1080/00420980601184729
Frank, L. D., & Pivo, G. (1994). Impacts of Mixed Use and Density on Uti-
lization of Three Modes of Travel: Single-Occupant Vehicle, Transit,
and Walking.
Frank, L. D., Stone, B., & Bachman, W. (2000). Linking land use with
household vehicle emissions in the central puget sound: Methodologi-
cal framework and findings. Transportation Research Part D: Transport
and Environment,5(3), 173–196. doi:10.1016/S1361-9209(99)00032-2
Greene, W. H. (2003). Econometric analysis. Upper Saddle River, N.J.:
Prentice Hall.
Greene, W. H., & Hensher, D. A. (2003). A latent class model for discrete
choice analysis: Contrasts with mixed logit. Transportation Research Part
B: Methodological,37(8), 681–698. doi:10.1016/S0191-2615(02)00046-2
Handy, S., Boarnet, M., Ewing, R., & Killingsworth, R. (2002). How the
built environment affects physical activity: Views from urban planning.
American Journal of Preventive Medicine,23(2 SUPPL. 1), 64–73.
doi:10.1016/S0749-3797(02)00475-0
Handy, S., Cao, X., & Mokhtarian, P. (2005). Correlation or causality
between the built environment and travel behavior? Evidence from
Northern California. Transportation Research Part D: Transport and
Environment,10(6), 427–444. doi:10.1016/j.trd.2005.05.002
INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 11

Handy, S., & Xing, Y. (2011). Factors correlated with bicycle commuting: A
study in six small U.S. cities. International Journal of Sustainable
Transportation,5(2), 91–110. doi:10.1080/15568310903514789
Handy, S., Xing, Y., & Buehler, T. (2010). Factors associated with bicycle
ownership and use: A study of six small U.S. cities. Transportation,37
(6), 967–985. doi:10.1007/s11116-010-9269-x
Heinen, E., & Handy, S. (2012). Similarities in attitudes and norms and the
effect on bicycle commuting: Evidence from the bicycle cities davis and
delft. International Journal of Sustainable Transportation,6(5), 257–
281. doi:10.1080/15568318.2011.593695
Heinen, E., Maat, K., & van Wee, B. (2011). Day-to-day choice to commute
or not by bicycle. Transportation Research Record: Journal of the Trans-
portation Research Board,2230,9–18. doi:10.3141/2230-02
Heinen, E., Maat, K., & van Wee, B. (2013). The effect of work-related fac-
tors on the bicycle commute mode choice in the Netherlands. Trans-
portation,40(1), 23–43. doi:10.1007/s11116-012-9399-4
Heinen, E., van Wee, B., & Maat, K. (2010). Commuting by bicycle: An
overview of the literature. Transport Reviews,30(1), 59–96.
doi:10.1080/01441640903187001
Hess, S., Shires, J., & Jopson, A. (2013). Accommodating underlying pro-
environmental attitudes in a rail travel context: Application of a latent
variable latent class specification. Transportation Research Part D:
Transport and Environment,25,42–48. doi:10.1016/j.trd.2013.07.003
Hino, A. A. F., Reis, R. S., Sarmiento, O. L., Parra, D. C., & Brownson, R. C.
(2014). Built environment and physical activity for transportation in
adults from Curitiba, Brazil. Journal of Urban Health,91(3), 446–462.
doi:10.1007/s11524-013-9831-x
Hoshino, T. (2010). Formulation and Application of A Restricted Multi-
Level Latent Segment Model in Modeling Residential Preferences, 1–
20. Retrieved from http://www.soc.titech.ac.jp/info/english/docs/
dp2009-03.pdf
Hurtubia, R., Nguyen, M. H., Glerum, A., & Bierlaire, M. (2014). Integrat-
ing psychometric indicators in latent class choice models. Transporta-
tion Research Part A: Policy and Practice,64, 135–146. doi:10.1016/j.
tra.2014.03.010
INE. (2011). National pre-census. Gobierno de Chile: Instituto Nacional de
Estad
ısticas.
Kamakura, W., & Russell, G. (1989). A probabilistic choice model for mar-
ket segmentation and elasticity structure. Journal of Marketing
Research,26(4), 379–390. doi:10.2307/3172759
Kitamura, R., Mokhtarian, P. L., & Laidet, L. (1997). A micro-analysis of
land use and travel in five neighborhoods in San Francisco bay area.
Transportation,24(November), 125–158. doi:10.1023/A
Kockelman, K. (1997). Travel behavior as function of accessibility, land use
mixing, and land use balance: Evidence from San Francisco bay area.
Transportation Research Record: Journal of the Transportation Research
Board,1607(970048), 116–125. doi:10.3141/1607-16
Krizek, K. J. (2003). Residential relocation and changes in urban travel:
Does neighborhood-scale urban form matter? Journal of the American
Planning Association,69(3), 265–281. doi:10.1080/01944360308978019
Larranaga, A. M., Rizzi, L. I., Arellana, J., Strambi, O., & Cybis, H. B. B.
(2016). The influence of built environment and travel attitudes
on walking: A case study of Porto Alegre, Brazil. International Journal
of Sustainable Transportation, 10(4), 332–342. doi:10.1080/155
68318.2014.933986
McKelvey, R. D., & Zavoina, W. (1975). A statistical model for the analysis
of ordinal level dependent variables. The Journal of Mathematical Soci-
ology,4(1), 103–120. doi:10.1080/0022250X.1975.9989847
McNally, M. G., & Kulkarni, A. (1997). Assessment of influence of land
use-transportation system on travel behavior. Transportation Research
Record: Journal of the Transportation Research Board,1607(1), 105–
115. doi:10.3141/1607-15
Meng, L., Taylor, M. A. P., & Scrafton, D. (2016). Combining latent class
models and GIS models for integrated transport and land use plan-
ning—A case study application. Urban Policy and Research, (February),
1–25. doi:10.1080/08111146.2015.1118372
Mokhtarian, P., & Cao, X. (2008). Examining the impacts of residential
self-selection on travel behavior: A focus on methodologies. Transpor-
tation Research Part B: Methodological,42(3), 204–228. doi:10.1016/j.
trb.2007.07.006
Motoaki, Y., & Daziano, R. A. (2015). A hybrid-choice latent-class model
for the analysis of the effects of weather on cycling demand. Transpor-
tation Research Part A: Policy and Practice,75(2015), 217–230.
doi:10.1016/j.tra.2015.03.017
Moudon, A. V., Lee, C., Cheadle, A. D., Collier, C. W., Johnson, D.,
Schmid, T. L., & Weather, R. D. (2005). Cycling and the built environ-
ment, a US perspective. Transportation Research Part D: Transport and
Environment,10(3), 245–261. doi:10.1016/j.trd.2005.04.001
Niehaus, M. (2016). Accesibilidad y equidad: Herramientas para ampliar la
evaluaci
on social de proyectos de transporte. Pontificia Universidad
Cat
olica de Chile.
Olaru, D., Smith, B., & Taplin, J. H. E. (2011). Residential location and tran-
sit-oriented development in a new rail corridor. Transportation Research
Part A: Policy and Practice,45(3), 219–237. doi:10.1016/j.tra.2010.12.007
OpenStreetMaps contributors. (2017). Planet dump [Data file from March
2017]. Retrieved from https://planet.openstreetmap.org
Piatkowski, D. P., & Marshall, W. (2015). Not all prospective bicyclists are
created equal: The role of attitudes, socio-demographics, and the built
environment in bicycle commuting. Travel Behaviour and Society,2(3),
166–173. doi:10.1016/j.tbs.2015.02.001
Plaut, P. O. (2005). Non-motorized commuting in the US. Transportation
Research Part D: Transport and Environment,10(5), 347–356.
doi:10.1016/j.trd.2005.04.002
Pucher, J., & Buehler, R. (2006). Why Canadians cycle more than Ameri-
cans: A comparative analysis of bicycling trends and policies. Transport
Policy,13(3), 265–279. doi:10.1016/j.tranpol.2005.11.001
Pucher, J., & Buehler, R. (2010). Walking and cycling for healthy cities.
Built Environment,36(4), 391–414.
Pucher, J., & Dijkstra, L. (2003). Promoting safe walking and cycling to
improve public health walking and cycling: The MOST sustainable
transport modes. American Journal of Public Health,93(9), 1509–1516.
doi:10.1016/j.ypmed.2009.07.028
Rajamani, J., Bhat, C. R., Handy, S. L., Knaap, G., & Song, Y. (2003).
Assessing impact of urban form measures on nonwork trip mode
choice after controlling for demographic and level-of-service effects.
Transportation Research Record,1831(1), 158–165. doi:10.3141/1831-18
Rodr
ıguez, D., & Joo, J. (2004). The relationship between non-motorized
mode choice and the local physical environment. Transportation
Research Part D: Transport and Environment,9(2), 151–173.
doi:10.1016/j.trd.2003.11.001
Rossetti, T. (2017). Modelaci
on de preferencias por dise~
no de infraestruc-
tura ciclista utilizando variables latentes perceptuales. (tesis de Mag
ı-
ster). Santiago, Chile: Pontificia Universidad Cat
olica de Chile.
Rossetti, T., Saud, V., & Hurtubia, R. (2017). I want to ride it where I like:
Measuring design preferences in cycling infrastructure. Transportation,
1–22. doi:10.1007/s11116-017-9830-y
Sallis, J. F., Conway, T. L., Dillon, L. I., Frank, L. D., Adams, M. A., Cain, K.
L., & Saelens, B. E. (2013). Environmental and demographic correlates
of bicycling. Preventive Medicine,57(5), 456–460. doi:10.1016/j.
ypmed.2013.06.014
Sawkins, J. W., Seaman, P. T., & Williams, H. C. S. (1997). Church atten-
dance in Great Britain: An ordered logit approach. Applied Economics,
29(2), 125–134. doi:10.1080/000368497327209
SECTRA. (2015). Encuesta origen y destino de viajes 2012.
Sehatzadeh, B., Noland, R., & Weiner, M. (2011). Walking frequency, cars,
dogs, and the built environment. Transportation Research Part A: Pol-
icy and Practice,45(8), 741–754. doi:10.1016/j.tra.2011.06.001
SII. (2014). Catastro de Bienes Ra
ıces del Servicio de Impuestos Internos.
Retrieved May 15, 2017, from https://zeus.sii.cl/cvc_cgi/dfmun/dfmun_
repGobierno.cgi
Smith, B., & Olaru, D. (2013). Lifecycle stages and residential location
choice in the presence of latent preference heterogeneity. Environment
and Planning A,45(10), 2495–2514. doi:10.1068/a45490
Stewart, O. T., & Moudon, A. V. (2014). Using the built environment to over-
sample walk, transit, and bicycle travel. Transportation Research Part D:
Transport and Environment,32,15–23. doi:10.1016/j.trd.2014.06.012
Titze, S., Stronegger, W. J., Janschitz, S., & Oja, P. (2008). Association of
built-environment, social-environment and personal factors with bicy-
cling as a mode of transportation among Austrian city dwellers. Preven-
tive Medicine,47(3), 252–259. doi:10.1016/j.ypmed.2008.02.019
12 I. OLIVA ET AL.

van Wee, B., & Handy, S. (2016). Key research themes on urban
space, scale and sustainable urban mobility. International Journal
of Sustainable Transportation,10(1), 18–24. doi:10.1080/
15568318.2013.821008
Walker, J., & Ben-Akiva, M. (2002). Generalized random utility model.
Mathematical Social Sciences,43(3), 303–343. doi:10.1016/S0165-4896
(02)00023-9
Walker, J., & Li, J. (2007). Latent lifestyle preferences and household loca-
tion decisions. Journal of Geographical Systems,9(1), 77–101.
doi:10.1007/s10109-006-0030-0
Winters, M., Brauer, M., Setton, E. M., & Teschke, K. (2010). Built envi-
ronment influences on healthy transportation choices: Bicycling versus
driving. Journal of Urban Health,87(6), 969–993. doi:10.1007/s11524-
010-9509-6
Yang, M., & Zacharias, J. (2016). Potential for revival of the bicycle in
Beijing. International Journal of Sustainable Transportation, 10(6),
517–527. doi:10.1080/15568318.2015.1012281
Zegras, C. (2010). The built environment and motor vehicle ownership and
use: Evidence from Santiago de Chile. Urban Studies,47(July), 1793–
1817. doi:10.1177/0042098009356125
INTERNATIONAL JOURNAL OF SUSTAINABLE TRANSPORTATION 13