You are on page 1of 14

Transportation Research Part A 130 (2019) 452–465

Contents lists available at ScienceDirect

Transportation Research Part A


journal homepage: www.elsevier.com/locate/tra

In-vehicle crowding: Integrating tangible attributes, attitudes, and


T
perceptions in a choice context between BRT and metro☆

Luis Márquez , Julieth V. Alfonso A, Juan C. Poveda
Universidad Pedagógica y Tecnológica de Colombia, Avenida Central del Norte 39-115, Tunja, Colombia

A R T IC LE I N F O ABS TRA CT

Keywords: Understanding in-vehicle crowding is crucial to improving public transportation service levels.
In-vehicle crowding Although in-vehicle crowding has usually been studied as a tangible attribute, previous research
Perceived discomfort has shown that the standing passenger density cannot fully explain the experience of crowding. In
Perceived insecurity the present work, we used the hybrid discrete choice modeling approach in order to provide a
Standing passenger density
richer explanation of in-vehicle crowding in a choice context between Bus Rapid Transit (BRT)
Hybrid discrete choice model
BRT
and Metro, using Bogotá as a case of study. Modeling results showed that perceived discomfort,
perceived insecurity, waiting time, fare, travel time, and its interaction with the positive attitude
towards crowding, are the main variables in explaining the choice process. We found that the
positive attitude towards crowding affects the perception of the standing passengers’ density
level. We demonstrated that this attitude depends on non-transportation information since
people who live in larger spaces, as well as people who live in apartments, have a greater positive
attitude towards crowding. We obtained crowding multipliers in regarding the users’ attitude for
different levels of overcrowding and analyze several policies related to latent variables.

1. Introduction

In a general sense, crowding occurs when the number of people concentrated in a space limits the freedom of movement. From an
objective point of view, crowding emerges as a physical density, that is to say a numerical measurement of the concentration of
individuals within a given geographical unit. In this regard, it is possible to speak of regional density, residential density, and people
density, among many different connotations (Cheng, 2010). Although it is a tangible attribute, an individual may have some per-
ception of physical density, depending on spatial characteristics and interaction between the individual and the environment, as well
as individual attitudes that also could change the way people perceive such interaction (Alexander, 1993).
In the context of public transportation systems, crowding is becoming a matter of concern by transportation service providers and
users because public transportation needs a minimum utilization rate in order to be profitable and efficient and some level of
crowding is unavoidable. This is the reason why public transportation systems are designed with a specific standing passenger density
(Batarce et al., 2016; Pel et al., 2014), though sometimes it can be exceeded in periods of high demand, affecting the service quality of
public transportation and travel behavior (Tirachini et al., 2016). Crowding can have a potential effect both on the efficiency of the
public transportation industry and passenger perceptions since the efficiency of the system is affected by delays while passengers may
perceive the loss of service quality (Cox et al., 2006).


On behalf of all authors, the corresponding author states that there is no conflict of interest.

Corresponding author.
E-mail addresses: luis.marquez@uptc.edu.co (L. Márquez), julieth.alfonso@uptc.edu.co (J.V. Alfonso A),
juan.poveda@uptc.edu.co (J.C. Poveda).

https://doi.org/10.1016/j.tra.2019.09.061
Received 28 August 2018; Received in revised form 28 August 2019; Accepted 30 September 2019
0965-8564/ © 2019 Elsevier Ltd. All rights reserved.
L. Márquez, et al. Transportation Research Part A 130 (2019) 452–465

In a typical public transportation trip, users face different types of crowding when they walk from their origin to the boarding
station, wait for the vehicle, ride in the vehicle, transfer (if necessary), and walk from the alighting station to the destination. As each
of these parts of the same trip occurs in a different physical space, i.e. in-vehicle, waiting platforms, transfer stations, and access
facilities, physical density could be perceived differently by users in each part of the trip. In this regard, it is possible to speak of in-
vehicle crowding (Li et al., 2017), platform crowding (Qin, 2014; Duduta and Subedi, 2015) and access-way crowding (Li and
Hensher, 2011). Additionally, users may experience different levels of crowding when boarding and alighting vehicles (Sun et al.,
2014).
Traveler satisfaction levels typically decrease with increasing in-vehicle crowding which, moreover, is a significant attribute
influencing traveler’s mode choices (Li et al., 2017). In this context, in-vehicle crowding has typically been specified as a tangible
attribute (passenger density, load factor, occupancy rate) through the parameters of utility function as can be seen in the works of
Basu and Hunt (2012), Tirachini et al. (2013), Batarce et al. (2015), Batarce et al. (2016), Tirachini et al. (2016), Björklund and
Swärdh (2017), Hörcher et al. (2017), and Li et al. (2017), among others. Often, though not in all cases, discrete choice models have
been used to study in-vehicle crowding. Recently, Tirachini et al. (2017) used a latent class model to study heterogeneity in in-vehicle
crowding perception highlighting substantial heterogeneity in sensitivities to in-vehicle crowding levels across individuals.
Previous research has mainly focused on studying the willingness to pay (WTP) for crowding reduction on urban public trans-
portation systems. It is common in the literature, for the WTP measure, to be calculated as multipliers of the value of travel time
savings for a reference trip condition (Wardman and Whelan, 2011; Haywood and Koning, 2015; Björklund and Swärdh, 2017).
However, the present study is more in line with authors that have promoted further research directions aimed at obtaining a more
comprehensive understanding of in-vehicle crowding in public transportation services. For example, Tirachini et al. (2013) suggest
integrating psychological aspects that influence in-vehicle crowding perception. In this vein, Li and Hensher (2013) conclude that in-
vehicle crowding as a tangible attribute (e.g. passenger density) cannot completely represent the experience of crowding, given that
the perception of in-vehicle crowding is subjective. They recommend that, in addition to tangible attributes, additional studies should
conduct perception surveys to obtain further user information.
This paper, unlike previous works, contributes to the in-vehicle crowding modeling literature by being the first that integrates
tangible attributes (Bahamonde-Birke et al., 2010), attitudes, and perceptions, in a choice context between BRT and Metro, using
Bogotá as a case of study. Our behavioral hypothesis is that, in addition to traditional tangible attributes, such as fare, waiting time
and travel time, the modal choice depends on in-vehicle crowding, which could emerge in various dimensions. We hypothesize that
standing passenger density not only explains the alternatives’ systematic utility but also affects the perceived discomfort and the
perceived insecurity. We also hypothesize that the positive attitude towards crowding affects the perception of the density level
considered and thus, this attitude could enter the systematic utility of alternatives in interaction with standing passenger density.
Finally, we hypothesize that the positive attitude towards crowding could be influenced by some housing attributes, such as a larger
living space area and living in an apartment.
The remainder of this paper is organized as follows: in Section 2, we briefly review existing studies on in-vehicle crowding from a
passengers’ perspective. Section 3 explains the context, survey, population, sample, and the modeling framework of this research. In
Section 4, we present the modeling results and discussion, and finally, Section 5 summarizes the main findings of the study, discusses
policy implications, and provides the outlook for future work.

2. In-vehicle crowding from a passenger’s perspective

During traveling, experienced in-vehicle crowding leads to increased dissatisfaction (Li and Hensher, 2013). Typically, in-vehicle
crowding is manifested when passengers on the vehicle impede the flow of other people boarding and alighting (Fletcher and El-
Geneidy, 2013). In addition to the discomfort due to proximity to others (Haywood et al., 2017), from a passenger’s perspective, in-
vehicle crowding has consequences in terms of exacerbation in the risk perception (Cox et al., 2006), anxiety (Cheng, 2010), stress
(Mahudin et al., 2011), increase in feeling of exhaustion (Mahudin et al., 2012), dwell time increments (Fletcher and El-Geneidy,
2013), restrictions on user privacy (Feife, 2014), sexual harassment concern (Kim et al., 2015), and feeling of wasted time (Mahudin
et al., 2011; Haywood et al., 2017), among other phenomena.
Previous research that has studied crowding valuation from the user’s perspective has shown that it has effects on in-vehicle time,
waiting time, travel time reliability, wellbeing, route choice, optimal public transportation fare, and valuation of travel time savings
(Tirachini et al., 2013). It has also been shown that differences observed in the crowding valuation literature can be explained by
local circumstances among factors that can range from socio-demographic attributes to performance activities when traveling,
personal attitudes, cultural values and context (Tirachini et al., 2016). In this way, previous research has shown that socioeconomic
characteristics, such as gender, income, and age, are significant attributes in explaining heterogeneity in crowding valuation
(Tirachini et al., 2017).
Around the world, standing passenger density is a tangible standard measurement for in-vehicle crowding used by BRT service
operators. However, no clear consensus exists on the unacceptable standing passenger density threshold. Fletcher and El-Geneidy
(2013) suggest that in-vehicle crowding appears after the vehicle passes 60% capacity, but some authors have suggested that the
perception of service quality varies across different countries, in general, and between developed and developing countries (Das and
Pandit, 2013), particularly. For instance, in Australia, four passengers per square meter is the benchmark (Li and Hensher, 2013)
while in Bogotá, the density of BRT is designed to a standard of seven passengers per square meter. However, in practice, higher
values are observed—even reaching cases, such as in China, where standing passenger density reaches eight users per square meter
(Varghese and Adhvaryu, 2016). Depending on the users’ expectations, some of these passenger densities would be unacceptable in

453
L. Márquez, et al. Transportation Research Part A 130 (2019) 452–465

the context of certain BRT systems.


Although standing passenger density is commonly used as a tangible measure of in-vehicle crowding by many transportation
authorities, it has long been recognized that this tangible attribute is not completely able to represent the in-vehicle crowding
phenomenon (Li and Hensher, 2013), particularly because it is not capable of reflecting individual perceptions (Cox et al., 2006). In a
subjective approach, crowding is perceived when there is excessive social interaction (Evans et al., 2000) and can affect the user’s
satisfaction and the perceived quality of public transportation services as shown in the works of Eboli and Mazzulla (2007), Dell’Olio
et al. (2011), and Imre and Celebi (2017). Furthermore, as passenger perceptions are subjective, they can be influenced by individual
characteristics and other factors such as previous experience.

3. Data and methods

To test our hypotheses, we conducted a face-to-face survey during May 2018, in which individuals faced a stated preference
experiment in a choice context between the Bogotá BRT system and Metro.

3.1. Context

A key part of the public transportation system in Bogotá is the BRT system that runs all over the city. Despite the prestige gained
by the BRT system during its first years of operation, which has been widely recognized in the transportation literature (see, for
instance, Hidalgo et al., 2013a,b), the truth is that, at present, its level of service has been declining, especially for the low-income
groups that locate mainly on the south and west sides of the city. In general, the public transportation system in Bogotá exhibits
significant problems of overcrowding, unreliable service, and insufficient supply, especially during peak hours (Guzman and
Bocarejo, 2017).
Unlike other contexts, where individual perceptions about whether a seat is guaranteed are important (Bouscasse and de
Lapparent, 2018), in our study context, given the high in-vehicle crowding levels, these specific perceptions seem to be not sig-
nificant. In Bogotá, the system’s buses usually have such high occupancy levels that for the users it is not relevant whether they travel
seated or not. According to our observations, the standing passenger density is much higher than 6 passengers per square meter at
peak times in Bogotá. Although a density of 6 or more passengers per square meter is an engineering standard, normally it can be
exceeded at peak times in some corridors, which can result in problems such as crashes loads, possible injuries, and forced movement
(Vuchic, 2005). In our context, at the beginning of the 2000s, previous studies had assumed a density of 6.5 for the BRT system
(Hidalgo et al., 2013a), which is evidently higher than the optimum level proposed by Guerra and Bocarejo (2013), who found 5.7
passengers per square meter. Recently, numerous press releases have reported a density of 8 passengers per square meter, which is
clearly extreme but a reality in Bogotá.
Recently, the Bogotá City Council has approved a proposed agreement to fund the construction of the Colombian capital’s first
metro line. This elevated metro line was chosen to improve city center access for those living in the capital’s outlying northern and
southern districts. The first metro line will have a maximum capacity of 72,000 passengers per hour and will stretch from the depot in
Bosa to 72nd Street. It is expected that the metro will be integrated with the Bogotá transportation system and will substantially
reduce commute times for users as well as decrease air pollution in the city. It is in this context that we designed a stated preference
experiment that faced the traditional BRT system and the future elevated metro line.

3.2. Survey

Our survey consisted of three parts that were relevant to test the hypotheses made. At the beginning of the survey, individuals
faced a stated preference experiment in a choice context between the Bogotá BRT system and the Metro, then rated a set of indicators
of latent variables and, at the end of the survey, individuals reported their main socioeconomic characteristics and housing attributes.
Based upon a pre-questionnaire among some focus groups, in which we asked individuals to state the most influencing factors for the
alternatives, we selected the following attributes to be part of the stated preference experiment: travel time, waiting time, fare, and,
density, i.e. standing passengers per square meter. Table 1 shows these attributes and their levels.
According to the expectations that users declared in the focus groups and after the proper changes we made to the pilot test, we
decided to use the levels of 4, 6, and 8 passengers per square meter. We considered it reasonable suppose that a density of 6
passengers per square meter was the middle level. After having ascertained the likely locations of the boundary values based on the
way people valued the density in focal groups, we proposed a threshold of 2 passengers per square meter, that is 33% with respect to
the value of the middle level. Finally, as usual, we used equal increments between attribute levels for a reasonable presentation.
Although a density of 8 passengers per square meter is far superior to the world standard, it was a reasonable value that allowed
achieving realistic choice situations in our study context. Finally, we decided to use a complementary set of neutral images to
describe each level in the same manner as the show-cards used by Haywood and Koning (2015) and Haywood et al. (2017). Fig. 1
shows crowding representation diagram used to present passenger densities to surveyed individuals.
We obtained a fractional factorial design, comprised of 27 treatments, by using the Ngene (ChoiceMetrics, 2012) software. As the
number of treatments was too large to give all these choice situations to a single respondent, we also assigned three blocks using a
minsum search that minimizes the total correlation values between the blocking column and all of the attributes. Therefore, the stated
preference experiment consisted of nine choice situations for each individual. Per repetition, we displayed the two experimental
alternatives with the relevant attributes previously mentioned. The experimental design maintained attribute level balance to ensure

454
L. Márquez, et al. Transportation Research Part A 130 (2019) 452–465

Table 1
Attributes and levels of the experimental design.
Attribute BRT Metro

Travel time (min) 20 10


30 20
40 30

Waiting time (min) 4 4


8 8
12 12

Fare (COP) $2000 $2300


$2300 $2800
$2600 $3300

Density (standing passengers/square meter) 4 4


6 6
8 8

4 passengers/m2 6 passengers/m2 8 passengers/m2


Fig. 1. Crowding representation.

that the parameters could be estimated well on the whole range of levels, instead of just having data points at only one or a few of the
attribute levels.
In the second part of the survey, individuals rated a set of effect indicators (Bollen, 2002), which are hypothesized to be man-
ifestations of the underlying latent behavioral constructs. According to focus groups and previous interviews, the main interests of the
BRT users were discomfort and insecurity. We defined these perceptions as a priori latent variables and we added the positive attitude
towards crowding that we found important to include in the latent variables set to better explain individual behavior in our choice
context. Following the work of Bahamonde-Birke et al. (2017), according to the nature of the proposed latent variables, the positive
attitude towards crowding is a “non-alternative related individual-specific latent attribute”, while discomfort perceived and in-
security perceived are “alternative-specific latent attributes”. In our case study, these perceptions are related to the BRT alternative.
In order to allow for the identification of latent the variables, we introduced the effect indicators shown in Table 2. For each
indicator, the table shows the expected signs according to our belief regarding the direction in which each latent variable will
manifest in the measurement model. We used at least three indicators per latent variable and the format of a typical Likert item to
measure the respondents’ level of opinion on the most common 5-point agree-disagree scale, in line with the most frequent practice in

Table 2
Indicators of latent variables.
Latent variable Indicator Stem statement Expected sign

Positive attitude towards crowding A1 It does not bother me to take an elevator with a lot of people +
A2 I enjoy going to public places where there are a lot of people +
A3 I like shopping when the stores are not crowded –

Perceived discomfort C1 I am exposed to touching and fondling inside the vehicle +


C2 It is uncomfortable for me to move inside the vehicle +
C3 I feel I can maintain a comfortable distance from other passengers –
C4 It is easy to bring packages or belongings inside the vehicle –

Perceived insecurity I1 I feel I can be robbed inside the vehicle +


I2 I feel safe carrying my belongings or packages inside the vehicle –
I3 I avoid using my cell phone for fear of it being stolen inside the vehicle +

455
L. Márquez, et al. Transportation Research Part A 130 (2019) 452–465

the field of transportation modeling (Márquez et al., 2018a).


The three attitude indicators were rated by respondents without considering any specific density level. Instead, according to our
hypotheses, in which the standing passenger density could affect the perceived discomfort and the perceived insecurity, we asked
individuals to rate the perceptual indicators for a certain density level, i.e. 4, 6, or 8 standing passengers per square meter. In this
way, we were able to simultaneously investigate how standing passenger density affects the latent variables as well as how these
constructs affect choice behavior. We had been careful not to ask on the metro’s perceptions of discomfort and insecurity because we
fully understand that individuals do not have an opinion formed about them. Therefore, individuals rated all the perceptual in-
dicators for the BRT service only.
At the end of the survey, we asked individuals for the main socioeconomic characteristics of individuals. In addition, the survey
also collected information on the type of housing, that is if people live in a house, an apartment or other living space. We also
gathered data on living space area (in square meters) and the locality where people live, in order to relate this information to
residential density, giving us the individual’s housing attributes which can help explain individual attitudes regarding crowding.

3.3. Population and sample

The population of interest for our study consisted of BRT users across all localities in Bogotá. In order to test the hypotheses of the
study, using a convenience sampling, we surveyed a representative sample of these users, who are also potential Bogotá Metro users.
Due to the survey complexity, it was conducted not in stations but in the places where respondents were carrying out their activities,
that is, in their workplaces, study sites, and shopping centers, among others. To reach our objective population, a filter question was
included at the beginning of the survey regarding project metro knowledge to obtain a convenience sample of potential users. We did
not consider car users because it was necessary that the respondents were familiar with the BRT service to proper rate the indicators
of the perceptual latent variables.
The survey was applied by hired staff, who had experience in the application of surveys in Bogotá, under the supervision of the
research team. The information was gathered during May 2018, in business hours, using a face-to-face questionnaire comprised by a
stated preference experiment, a set of effect indicators, and a section of socioeconomic characteristics, as previously explained. In the
case of workers, the surveys were authorized by their bosses, in order to have the necessary time for their application in each
workplace. In contrast, surveys applied to students and non-workers were conducted with the consent of each individual in the places
of study and shopping centers, intercepting them in rest areas. Sampling was controlled to correctly represent population distribu-
tions by age, gender, occupation, and level of education, in order to complete a target sample size of 300 individuals. Furthermore,
the surveys were applied in all the localities of Bogotá in order to obtain a sufficient variation of the individuals’ housing densities.
After data cleaning, the survey yielded 2781 choice responses corresponding to 309 BRT users. Table 3 gives an overview of both

Table 3
Profile of the population and sample.
Variable Category Sample (%) All BRT users (%)

Age < 18 1.95 7.37


18–27 44.01 41.24
28–37 23.30 22.24
38–47 15.21 16.23
48–59 10.68 9.94
> 59 4.85 2.98

Gender Male 48.54 46.20


Female 51.46 53.80

Occupation Work 56.95 53.81


Study 20.39 20.90
Work-study 12.30 6.30
Home 5.18 5.07
Unemployed 2.91 3.05
Retired 1.62 2.06
Other 0.65 8.81

Education level No studies 0.33 1.74


Primary 3.88 6.18
High school 40.45 48.25
Technical 14.56 13.67
University 30.42 28.03
Post-graduate 10.36 2.13

Socioeconomic strata Strata 1 4.53 5.13


Strata 2 33.98 35.75
Strata 3 42.72 46.25
Strata 4 15.86 10.44
Strata 5 2.59 1.82
Strata 6 0.32 0.61

456
L. Márquez, et al. Transportation Research Part A 130 (2019) 452–465

Table 4
Variables used in the model.
Group Variable Type Description

Socioeconomic characteristics Female gender Dummy 1: Female


0: Other
High education level Dummy 1: University or postgraduate
0: Other
People aged over 28 Dummy 1: Older than 28
0: Other

Housing attributes High-density locality Dummy 1: Larger than 161 inhabitants/ha


0: Other
Living in apartment Dummy 1: Apartment
0: Other living space
Larger living space area Dummy 1: Larger than 62 square meters
0: Other

Latent variables Perceived discomfort Continuous Not observable


Perceived insecurity Continuous Not observable
Positive attitude towards crowding Continuous Not observable

Design attributes Travel time Continuous Minutes


Waiting time Continuous Minutes
Fare Continuous Fare (COP) * 0.001
Standing passenger density 6 Dummy 1: 6 passengers/square meter
0: Other
Standing passenger density 8 Dummy 1: 8 passengers/square meter
0: Other

our sample and the population frequencies regarding the main socioeconomic characteristics of users. Table 3 also presents dis-
tributions by age, gender, occupation, education level, and socio-economic strata, which describe the relative social position of
individuals based upon their income. We used the same categories displayed in mobility studies by transportation authorities to draw
comparisons. As Table 3 shows, the main variables distribute in a similar way for both our sample and the population. In regard to
residential density, the sample was comprised of people from all localities, with a median of 161 inhabitants per hectare. Regarding
living space area, a median of 62 square meters per living space was found.
It is important to recognize that it was not possible to interview a sufficient number of under-age passengers (under 18 for
Colombia) and for this reason, both age distribution and distribution by education level show small differences between population
and sample. This is a common issue when sampling public transportation users as you can see in the work of Tirachini et al. (2017)
who stated that “The fact that our survey was applied only to adults partially explains the under-representation of young users in our
sample”. The proportion of people who study and work simultaneously also exhibits a small difference between population and
sample, which is explained in some answering issues experienced by the people who make up this category because some of them
reported only what they perceived to be the main activities (study or work), despite doing both. It is for this reason that in the
sampling we obtained a higher percentage of people who work, although we consider that this small variation does not reduce the
validity of the sample. In fact, we are very satisfied with the sampling because we have obtained data that adequately represent the
target population of our study.

3.4. Variables used in modeling

Table 4 shows the variables used in modeling arranged into four groups: socioeconomic characteristics, housing attributes, latent
variables, and design attributes. We transform some continuous variables into dummy with the purpose of identifying certain ca-
tegories with a significant effect on the explanation of latent variables. All socioeconomic characteristics entered the model as dummy
variables, even the variables related to the housing attributes such as high-density locality and larger living space area that we
grouped into two categories taking the median as a cutoff. All other variables, i.e. the latent variables and the design attributes, were
entered into the model as continuous variables. We changed the scale of the fare by dividing its original values by one thousand
Colombian pesos. At this point, we consider it convenient to clarify that, according to the exchange rate at the time of the study, an
American dollar was equivalent to 2900 Colombian pesos.

3.5. Modeling framework

The hybrid discrete choice (HDC) modeling framework postulates that users choose the alternative that maximizes their perceived
utility, in the same way as random utility theory framework does (Ben-Akiva et al., 2002). The advantage of this approach is that, in
addition to specification of observable individual characteristics, alternatives’ attributes and other tangible variables, this modeling
framework extends the traditional models allowing us to form latent variables in order to include them as part of the systematic
utility of alternatives, also providing a more productive explanation of the choice process.

457
L. Márquez, et al. Transportation Research Part A 130 (2019) 452–465

Fig. 2. Hybrid discrete choice modeling approach.

Fig. 2 shows our modeling approach. Model representation follows the convention that latent variables are shown in ovals,
observable variables in rectangles, causal relationships by solid arrows, measurement relationships by dashed arrows, and error terms
are shown by dotted arrows. The symbol ⊗ represents interactions between variables. In the Fig. 2, two sub-models are distinguished:
a discrete choice sub-model and a latent variable sub-model in which the latent variables were incorporated through structural
equations (Ashok et al., 2002). We included two types of latent variables (Bahamonde-Birke et al., 2017): a non-alternative related
individual-specific attribute, i.e. the positive attitude towards crowding, and two alternative-specific latent attributes, i.e. perceived
discomfort and perceived insecurity of the BRT. The positive attitude towards crowding was specified as a function of observable
including housing attributes that vary among individuals. This variable entered the model in interaction with travel time and each
level of standing passenger density.
Here below we present the equations of the models. In the equations, q relates to an individual, l to a perceptual latent variable, A
to the positive attitude towards crowding, and r to an explanatory variable. In general, the notation used for the latent variables is ηl ,
except for the positive attitude towards crowding whose notation is ηA . As seen in (1), the perceptual latent variables were specified as
a function of standing passenger density (Dq ) and observable individual characteristics (Sq ), where, α is a set of parameters to be
estimated. The structural equation for the positive attitude towards crowding was specified as a function of housing attributes (Hq ).
Error terms υiq were assumed to follow a Normal distribution with mean zero. For identification issues, the variance of the structural
equation was set at one.

ηlq = αDl Dq + ∑ αrl Srq + υlq


r (1)

ηAq = ∑ αrA Hrq + υAq


r (2)

The measurement equations were specified as ordered logit models to capture the nature of the data (Daly et al., 2012). We
hypothesized that each discrete response k observed within each indicator p is obtained from the latent variables plus an error term
through a censoring mechanism that defines different categories of response, according to (3) and (4), where each categorical
response in the indicator Cpq was defined by a set of thresholds (τ ) to be estimated.

⎧ 1 if(−∞) < Cpq ≤ τp1
⎪ ∗
Cpq = 2 ifτp1 < Cpq ≤ τp2
⎨⋯
⎪ K ifτp (K − 1) < Cpq

≤∞
⎩ (3)

458
L. Márquez, et al. Transportation Research Part A 130 (2019) 452–465


Cpq = ∑ γlp·ηlq + ζpq
l (4)
The error terms (ζ ) were assumed to follow a logistic distribution and are independent from the set of parameters γ . If F is the
cumulative distribution function, the probability of observing Cpq within a discrete indicator or category k, can be written as (5) and
(6), where τp0 = (−∞) and τpK = ∞.


P {Cpq∈k|ηq} = F ⎜τpk − ∑ γlp·ηlq⎞⎟ − F ⎛⎜τp (k −1) − ∑ γlp·ηlq⎞⎟
⎝ l ⎠ ⎝ l ⎠ (5)

1 1
P {Cpq∈k|ηq} = −
1 + e−(τpk −∑l γlp·ηlq ) 1 + e−(τp (k − 1)−∑l γlp·ηlq ) (6)
We entered the latent variables in the utility function, as shown in (7) and (8), where θ and β are sets of parameters to be
estimated and they are associated with the modal attributes Xq , including the set of latent variables ηq . According to our hypotheses,
TT
the positive attitude towards crowding (ηA ) entered the discrete choice model in interaction with travel time ( Xiqt ) and each level of
6 8
standing passenger density (Diqt and Diqt ).
TT TT 6 TT 8
UBRTqt = ASCBRT + ∑ θk XkBRTqt + βA ηAq XBRTqt + βA6 ηAq XBRTqt DBRTqt + βA8 ηAq XBRTqt DBRTqt + ∑ βl ηlq + εBRTq
k l (7)
TT TT 6 TT 8
UMetroqt = ASCMetro + ∑ θk XkMetroqt + βA ηAq XMetroqt + βA6 ηAq XMetroqt DMetroqt + βA8 ηAq XMetroqt DMetroqt + εMetroq
k (8)
We observed a sequence of nine choices made by each individual and we supposed unobserved factors that affect decision makers
to be correlated over the repeated choices, as shown in (7) and (8). This means that factors that are not observed by us persist over
choice situations and are the same for each individual. In each alternative's utility function, the correlation over the repeated choices
was captured by an error term. Hence, our specification allowed us to model the repeated choices by treating the alternative’s specific
constants as varying over individuals but being constant over the nine choice situations for each person. We did not consider panel
disturbances for the beta coefficients because we were interested in specifying the correlation over the repeated choices in a par-
simonious fashion so as to model sufficiently realistic individual behavior by integrating latent variables.
The individual choices, given the choice set Aq , were expressed as a function of the utilities according to (9).

1 if Uiqt ≥ Ujqt , ∀ j ∈ Aq
yiqt = ⎧

⎩ 0 otherwise (9)

4. Modeling results and discussion

We performed the maximum simulated likelihood to estimate the unknown parameters of the HDC model. The model was
estimated using all the information simultaneously, which prevented the overestimation of the weights of the latent variables
(Bahamonde-Birke and Ortúzar, 2014). The joint probability of observing choice and latent variable indicators was built as seen in
(10), where P (·) is the choice probability of selecting alternative i , f (·) is the density function of the indicators, and g (·) is the density
function of the latent variables.

P¯ (yiqt , Cq |Xqt , Dqt , S q, Hq, θ , β , τ , γ , α, ϕ, Σε , Σζ , Σ υ) = ∫ P (yiqt |Xqt , ηq , θ, β, Σε ) f (Cq |ηq , γ, τ , Σζ ) g (ηq |Dqt , S q, Hq, α, ϕ, Συ) d η
q
η

(10)
Eq. (10) assumes independence between yiqt and Cq . The key assumption is not so much the shape of the distribution as that the
errors are independent of each other. This independence means that the unobserved portion of utility for one alternative is unrelated
to the error term for each categorical response in the latent variables’ indicators. This assumption, while restrictive, provides a very
convenient form for the joint probability. However, we recognized that the assumption of independence can be inappropriate in some
situations.
To simulate the likelihood, we used 500 draws for each individual obtained from a modified Latin Hypercube sampling. In order
to deal empirically with the identifiability issues of our model, we verified that the estimated parameters did not lie outside the range
of reasonable values. Also, the model was estimated multiple times, employing different starting values for the parameters for each
estimation run, verifying that in all cases we reached the same solution, thus ensuring that the solution obtained was a global
maximum (Vij and Walker, 2014).

4.1. Structural model

Table 5 exhibits the estimates and the robust t-values for the structural model we formed through the positive attitude towards
crowding and two alternative-specific latent variables. We found that housing attributes, i.e. living in apartment and living space
area, influence the positive attitude towards crowding. This is in line with the seminal findings of Sundstrom (1978) and Sundstrom

459
L. Márquez, et al. Transportation Research Part A 130 (2019) 452–465

Table 5
Estimated parameters for the structural model.
Latent variable Explanatory variable Estimate Rob. t-test

Positive attitude towards crowding Living in apartment 0.545 1.97


Larger living space 0.425 1.96

Perceived discomfort Standing passenger density 0.318 5.43


High education level 0.419 2.93
Female gender 0.296 1.71

Perceived insecurity Standing passenger density 0.240 3.77


Female gender 0.341 2.11
People aged over 28 0.231 1.61
High-density locality −0.305 −1.80

et al. (1975) who had proposed that physical antecedents were one of the four categories that may have an impact on the perception
of crowding. In our case, the findings show that people who live in apartments, unlike those who live in houses, have a greater
positive attitude towards crowding. It is possible that the different styles of life associated with apartments and houses explain this
behavior. Normally, houses suit a family lifestyle instead of apartments and other living spaces that are more suited to singles,
couples and most prominently to university students, who seem to be more willing to share space with others.
Likewise, we found that people who live in larger spaces have a greater positive attitude towards crowding. This finding enables
us to affirm that people who live in larger spaces have a greater tolerance of crowding which is supported in the findings of Evans
et al. (2000) which show us that residential crowding has a negative effect in terms of psychological distress. Then, we can hy-
pothesize that the psychological distress of the people living in small spaces translates into a decrease in the positive attitude towards
crowding in contrast to those who inhabit larger spaces. Because the culture also plays a role in the positive attitude towards
crowding (Evans et al., 2000), we consider that this finding is valid for the study context and we suggest a degree of caution when
generalizing this pattern of behavior across different cultures.
As expected, in keeping with the work of Evans and Wener (2007), we found that standing passenger density is related to the
perceived discomfort. The positive sign of the coefficient indicates that as the standing passenger density increases, the perceived
discomfort increases. We also found that the perceived discomfort is influenced by socioeconomic characteristics such as education
level and gender. Specifically, we found that users with a higher educational level have a higher perceived discomfort than others.
This result seems reasonable to the extent that people with a higher level of education could have higher expectations than others as
we have seen in studies developed in Colombia in other contexts (Márquez et al., 2014).
The heterogeneity captured through the perceived discomfort shows that gender has a significant effect on this latent variable.
This finding coincides with that observed in the overall results of Tirachini et al. (2017) who demonstrated that gender, income, and
age are significant variables in explaining heterogeneity in crowding disutility. As compared to the findings of Tirachini et al. (2017),
we found that females might be more negatively affected by a large standing passenger density than males. In contrast, we did not
find a significant effect of the income level: neither in the perceived discomfort nor in the perceived insecurity. However, our results
point in the same direction when taking into account that, in Colombia, the highest education levels are reached mainly by people
with higher incomes.
We find the feeling of insecurity to increase with standing passenger density, which is in line with the findings of previous
research (Tirachini et al., 2017). In our case study, age appears to be a significant variable in explaining perceived insecurity. In this
respect, people aged over 28, about 54.04% of the sample, perceive greater insecurity than younger users which is in line with
previous results in the context of BRT systems in Colombia (Márquez et al., 2018b). Unlike other studies, we find that a physical
antecedent, in our case the locality density, also affects the perceived insecurity. Users living in localities with a density greater than
161 inhabitants/ha have a lower perception of insecurity in comparison with people who live in lower density localities. Our finding
is reasonable since inhabiting denser localities could cause people to be more familiarized to crowding and its consequences.
Although the coefficients of measurement equations are not required for evaluating the HDC model in a predictive fashion, we
examined the overall adequacy of the results we presented in Table 6. We found the orientation of indicators to be consistent with the
expectations. The threshold parameters also were significant, which indicated that the ordered models adequately represented an
individuals’ opinions on the statements that we captured through the observed responses of indicators. The A3 indicator estimates
were not significant in the measurement model and for that reason were excluded from the final model estimation. The stem
statement of the A3 indicator was “I like shopping when the stores are not crowded”. Based on the information obtained from a focus
group, we thought this indicator to be affected by the positive attitude towards crowding but modeling results did not support our
premise. In addition, as no answers were obtained for all points of indicators, some threshold parameters were not estimated, as
shown in Table 6.
The results of the measurement model revealed that perceived discomfort was more strongly manifested in the C3 indicator (I feel
I can maintain a comfortable distance from other passengers). This means that proximity to others strongly reflects the perception of
discomfort, as Haywood et al. (2017) have suggested. The measurement model also showed that perceived insecurity was more
strongly manifested in the I2 indicator (I feel safe carrying my belongings or packages inside the vehicle). This suggests that one of the
biggest user concerns in terms of safety has to do with the feeling of risk of theft when carrying their belongings in vehicles. This

460
L. Márquez, et al. Transportation Research Part A 130 (2019) 452–465

Table 6
Estimated parameters for the measurement model.
Latent variable Indicator Estimate Threshold

1 2 3 4

Positive attitude towards crowding A1 Coefficient 0.730 −1.314 0.286 1.520 2.792
Rob. t-test 2.15 −6.50 1.90 7.02 8.40
A2 Coefficient 0.644 −1.458 0.036 1.915
Rob. t-test 2.61 −7.30 0.24 8.86

Perceived discomfort C1 Coefficient 1.037 1.402 3.273


Rob. t-test 6.06 2.55 5.16
C2 Coefficient 1.470 0.628 1.528 4.189
Rob. t-test 5.93 0.96 2.26 5.22
C3 Coefficient −1.753 −6.771 −4.365 −3.203 −0.952
Rob. t-test −6.81 −6.71 −5.36 −4.50 −1.68
C4 Coefficient −1.215 −4.414 −1.610 −0.657 0.463
Rob. t-test −6.33 −7.29 −3.43 −1.52 1.03

Perceived insecurity I1 Coefficient 0.906 −1.080 1.384


Rob. t-test 3.70 −2.14 2.81
I2 Coefficient −1.463 −1.090 0.304 1.481 4.029
Rob. t-test −4.51 −1.90 0.54 2.38 4.91
I3 Coefficient 0.567 −1.830 −0.127 0.878 2.420
Rob. t-test 3.34 −4.93 −0.41 2.95 7.03

result seems logical given the high levels of robberies that occur in Bogotá’s public transportation system.

4.2. Choice models

Table 7 presents the estimated parameters for the choice models. Beside the name of each parameter, in curly brackets, is the

Table 7
Estimated parameters for the choice model.
Parameter MNL1 MNL2 HDC1 HDC2 HDC3 HDC4

ASC BRT {1} −0.0950 −0.1455 −0.0955 −0.0917 0.3239 0.2915


(−1.30) (−1.93) (−1.32) (−1.26) (1.94) (1.84)
ASC Metro {2} 0 0 0 0 0 0
(Fixed) (Fixed) (Fixed) (Fixed) (Fixed) (Fixed)
Travel time {1, 2} −0.07636 −0.06404 −0.0772 −0.0664 −0.0616 −0.0638
(−13.41) (−11.25) (−13.47) (−11.59) (−10.38) (−10.61)
Wait time {1, 2} −0.0742 −0.0744 −0.07347 −0.07362 −0.0751 −0.0796
(−6.63) (−6.27) (−6.54) (−6.55) (−6.47) (−7.12)
Fare {1, 2} −1.0956 −1.1881 −1.0979 −0.9632 −1.1075 −1.0354
(−8.86) (−9.15) (−8.87) (−7.76) (−8.72) (−8.24)
Panel effect {1, 2} 0.4899 0.5286 0.4868 0.1713 0.0551 0.0581
(9.21) (9.58) (9.15) (3.22) (0.68) (0.71)
Travel time * Standing passenger density 6 {1, 2} −0.01435
(−5.05)
Travel time * Standing passenger density 8 {1, 2} −0.0284
(−9.24)
Travel time * Positive attitude towards crowding {1, 2} 0.0207 0.0292 0.0209 0.0229
(1.97) (2.05) (1.96) (2.03)
Travel time * Positive attitude towards crowding * Standing passenger density 6 −0.0123 −0.0101
{1, 2} (−1.98) (−1.64)
Travel time * Positive attitude towards crowding * Standing passenger density 8 −0.0201 −0.0179
{1, 2} (−2.99) (−2.79)
Perceived discomfort {1} −0.6270
(−9.79)
Perceived insecurity {1} −0.3204
(−4.03)
Travel time * Perceived discomfort {1} −0.00464
(−1.96)
Travel time * Perceived insecurity {1} −0.002373
(−2.01)
Log-likelihood for choice component −1602.68 −1571.38 −1580.19 −1553.82 −1539.74 −1562.6
Likelihood ratio test with respect to MNL1 62.60 44.98 97.72 125.88 80.16

461
L. Márquez, et al. Transportation Research Part A 130 (2019) 452–465

number of the utility function in which we specified the corresponding variable, as follows: {1} BRT and {2} Metro. In parentheses,
below each estimate, it is reported the robust t-test. At the end of Table 7, there are indicators of goodness of fit of the models,
specifically the Log-likelihood for the choice component and the Likelihood ratio test with respect to the simplest model.
Models in Table 7 are described as follows:

– MNL1: Multinomial logit model with panel effect and no standing passenger density’s parameter.
– MNL2: Multinomial logit model with panel effect and interactions between travel time and standing passenger density’s levels
– HDC1: Hybrid discrete choice model with interaction between travel time and the positive attitude towards crowding
– HDC2: Hybrid discrete choice model with interactions between travel time and the positive attitude towards crowding for each
standing passenger density’s level
– HDC3: Hybrid discrete choice model with perceptual variables and interactions between travel time and the positive attitude
towards crowding for each standing passenger density’s level, as we have shown in Fig. 2.
– HDC4: Hybrid discrete choice model with interactions between travel time and both the positive attitude towards crowding and
perceptual variables.

In all the models, we estimated the BRT specific constant and, due to identification issues, the Metro specific constant was set to
zero. The BRT specific constant was obtained with a negative sign in all the models, except in the HDC3 model. We consider this
result to be reasonable since the HDC3 model included the effect of two perceptual latent variables, whose coefficients were negative
and significant. Ceteris paribus, based on the negative sign of the BRT specific constant in the MNL1, MNL2, HDC1, and HDC2
models, we can only affirm that users expect the new Metro service to be better than the current BRT service. However, the positive
sign of the BRT constant and the negative signs of the coefficients of the perceptual latent variables in the HDC3 and HDC4 models,
suggest that, in reality, users perceive the current BRT service as uncomfortable and insecure. It is clear that including the perceptual
latent variables in the HDC3 and HDC4 models helps in better explains users’ behavior.
Unsurprisingly, the HDC3 model fitted the data better than the less complex models (Vij and Walker, 2016). All the models
considered the panel effect produced by having nine different choices per respondent, however, this effect was significant in the
simplest models but not in the HDC3 and HDC4 models. We consider this result to be reasonable since the error terms’ simulation of
the perceptual latent variables also considered the panel effect of the data. This is explained partly because, in the HDC3 and HDC4
models, the error terms of these perceptual variables vary across individuals, according to their assumed probability density func-
tions, but are constant over the choice situations for each individual. Similar results can be found in other empirical studies that show
how the introduction of latent variables decreases the panel effect's coefficient. For example, in the choice context between BRT and
motorcycle taxis, Márquez et al. (2018b) showed that the panel effect’s coefficient decreased considerably by introducing the latent
variables in the choice model.
The estimated parameters in all the models have the expected signs; travel time, wait time and fare are in the expected direction:
delays, additional travel times or fares that are more expensive reduce the utility of alternatives. Furthermore, the signs of the latent
variables’ coefficients indicate that these perceptions affect the BRT utility negatively. In particular, the signs of the perceived
discomfort and the perceived insecurity suggest that, as users perceive more discomfort and more insecurity, they will be less likely to
use the BRT when the Metro service is available. Unlike the other models, the inclusion of these latent perceptual variables in the
HDC3 and HDC4 models allows evaluating how changes in the BRT service may affect the perceptions, and thus the choices.
Although HDC3 and HDC4 models cannot be compared using the likelihood ratio test, goodness of fit indicators suggest that the
HDC3 model is better. It is important to note how the specification of the two models produces different marginal utilities of travel
time. On the one hand, in HDC3 model, the interaction between travel time and the positive attitude towards crowding, which, at the
same time, interacts with density, makes it possible for the standing passenger density to be part of the marginal utility of travel time.
In contrast, in the HDC4 model, marginal utility is a function of the standing passenger density due to interactions between travel
time and perceptions of insecurity and discomfort. As these perceptions are a function of density, their interaction with travel time
allows the marginal utility to depend on the standing passenger density through their structural equations.
Except for the MNL1 model, all the other ones allow evaluating the effect of the standing passenger density on the valuation of in-
vehicle time savings (Wardman and Whelan, 2011). Nonetheless, to establish a reference value, according to the estimated para-
meters in the MNL1 model, we found that the value of in-vehicle time savings is 69.70 Colombian pesos per minute, or around 1.44
USD/hour, which is consistent with past findings on Bogotá (Lleras, 2003) when indexing for inflation. As seen in Fig. 3, which shows
the effect of the standing passenger density on the valuation of in-vehicle time savings for the estimated models, the crowding cost
makes users willing to pay more to reduce their travel time. For example, the MNL2 model shows that the crowding multiplier
increases from 1.0 to 1.22 as the density of standing passengers increases from 4 to 6 passengers per square meter, whereas from 6 to
8 passengers per square meter these figures are 1.22 to 1.44. Both time multipliers values are in line with the reported ones from
previous studies, which have been found ranging from 1.0 to 3.0 (Haywood and Koning, 2015; Tirachini et al., 2016).
According to the estimated parameters in the HDC1 model, the value of in-vehicle time savings is 70.32 Colombian pesos per
minute, for users with the least positive attitude towards crowding. However, it is interesting to observe that this figure is 52.03 for
users with the greatest positive attitude towards crowding. This means that the crowding multiplier increases from 1.0 to 1.35 as the
positive attitude towards crowding goes from the greatest to the lest. Based on the results of the HDC2 model, for people with the
greatest positive attitude towards crowding, we found that the crowding multipliers increase from 1.0 to 1.31 as the density of
standing passengers increases from 4 to 6 passengers per square meter, whereas from 6 to 8 passengers per square meter these figures
are 1.31 to 1.51. This shows that, although users with the greatest positive attitude towards crowding are willing to pay less to reduce

462
L. Márquez, et al. Transportation Research Part A 130 (2019) 452–465

Fig. 3. Value of in-vehicle time savings.

their travel time, which seems logical, they are more sensitive to changes in density levels, which needs further research.
The HDC4 model produces smaller multipliers in comparison to the HDC3 model. While the multipliers of the HDC4 model are
1.24 and 1.42, respectively, the multipliers derived from the HDC4 model are 1.08 and 1.16. In this regard, one pattern we noticed in
Fig. 3 is that, as the models became more complex, the valuations of in-vehicle time savings decreasing. In effect, the valuations from
the HDC3 model were the lowest compared to the remaining models. However, the crowding multipliers were almost unchanged. For
example, the crowding multipliers of the MNL2 model are almost equal to those of the HDC3, obtained for users with the greatest
positive attitude towards crowding. Notwithstanding the above, it is important to notice that, given the specification of the positive
attitude towards crowding, in addition to the individuals’ socioeconomic characteristics reported in the literature (Wardman and
Whelan, 2011; Haywood and Koning, 2015; Björklund and Swärdh, 2017), we can also assert that the users’ housing attributes affect
the valuation of crowding reductions.

5. Conclusions and policy implications

We demonstrated that in-vehicle crowding emerges in three dimensions: tangible, through the standing passenger density; atti-
tudinal, through the positive attitude towards crowding, and perceptual, through the perceived discomfort and the perceived in-
security. The modeling results showed that standing passenger density affects the perceived discomfort and the perceived insecurity
of the BRT service. Furthermore, we found that the positive attitude towards crowding affects the perception of the standing pas-
sengers’ density level. These findings imply that transportation authorities need to make further efforts to enhance traditional as-
sessment by including information on attitudes, the perceived discomfort, and the perceived insecurity of the users to better un-
derstand whether tangible density benchmarks are in line with the individual perceptions.
An important finding in this study was the demonstration that the positive attitude towards crowding depends on non-trans-
portation information. We verified the hypothesis that the positive attitude towards crowding is influenced by housing attributes.
Specifically, we found that people who live in larger spaces, as well as people who live in apartments, have a greater positive attitude
towards crowding. From a public policy perspective, it is interesting to recognize that the users’ housing attributes influence the
positive attitude towards crowding. Understanding this may be useful for authorities since users with a greater positive attitude
towards crowding are willing to pay less to reduce crowding. This knowledge may have application in areas such as service quality
and tariff optimization in which the users’ housing characteristics can be incorporated.
The impact that the positive attitude towards crowding, perceived discomfort, and perceived insecurity have on transportation
demand should be considered from the early stages of public projects. In our case study, the construction of the first metro line could
have an effect in reducing BRT crowding that could also be considered in supporting public transportation investments that struggle
to compete in benefit-cost terms with road investments (Tirachini, et al., 2013). In addition, if the levels of overcrowding decrease
then the perceptions of discomfort and insecurity of the users would also be positively affected, as we demonstrated. These effects on
perceptions could also be considered in order to support public policies aimed at improving the levels of public transportation service.
Another important contribution was to have calculated the crowding multiplier in regarding the users’ attitude. To the best of our

463
L. Márquez, et al. Transportation Research Part A 130 (2019) 452–465

knowledge that finding had not been reported in the literature. We found that that the crowding multiplier increases from 1.0 to 1.35
when the positive attitude towards crowding goes from the greatest to the least. We also found the crowding multipliers for people
with the greatest positive attitude towards crowding for each density level. Knowing that in-vehicle time valuation depends not only
on the standing passenger density level but also on the individual attitude towards crowding is relevant for policy recommendations.
For instance, the cost-benefit analysis could be favorable for projects reducing the travel time of overcrowded services or for metro
systems, which are faster than buses. In addition, these analyses can be improved to the extent that authorities acknowledge that
individual attitudes have an effect on valuations.
Urban transportation policies often neglect the adverse effects of crowding in public transportation services when arguing that
modal shift eliminates the social burden of congestion (Hörcher et al., 2017). In the case of Bogotá, where authorities promote
bicycles as an alternative mode of transportation, the potential of using the bicycle to relieve overcrowded public transportation
services for short-distance travel could be studied. Some papers have explored the potential of bicycling as a replacement for short
distance public transportation trips (Sun and Zacharias, 2017). Bicycle use in Bogotá, a predominantly flat city, increased from
around 0.5% of daily trips in 1996 to 6% in 2014, after the construction of the first bikeways in 1997. These conditions seem
appropriate to address the role of the bicycle in relieving overcrowding in the BRT and Metro for short-distance trips. In any case, we
strongly recommend integrating tangible attributes, attitudes, and perceptions in the analysis.
A successful public transportation system requires improvements in reduced crowding across all important stages of the passenger
service chain: in-vehicle, platform and access-way crowding. However, as Li and Hensher (2011) have cautioned, there is still a need
for studies focused on valuing crowding on the platform, and the access-ways, because the experience in these locations is also
important in influencing individual preferences, and consequently modal choice and project design. This is an important insight for
further research to study how the users perceive crowding in the different trip stages. We consider that the works of Qin (2014),
Duduta and Subedi (2015), and Jiao et al. (2017) can be used as references for that purpose.
Finally, in accord with recommendations of Björklund and Swärdh (2017), we considered it important to study whether per-
ceptions of discomfort and insecurity differ among urban areas in Colombia, where the preferences might be relatively similar, but
still not identical. Even though Colombia’s first BRT was built in Bogotá, the national government has made important efforts in
building strategic public transportation systems and to expand BRT to major cities across the country. As the crowding levels differ
from one system to another across the country, it is interesting to adapt and replicate our methodology to study the BRT systems in
other cities to acquire a better understanding of crowding in the three dimensions that we identified in the present work: tangible,
attitudinal and perceptual.

References

Alexander, E.R., 1993. Density measures: a review and analysis. J. Arch. Plann. Res. 10 (3), 181–202.
Ashok, K., Dillon, W.R., Yuan, S., 2002. Extending discrete choice models to incorporate attitudinal and other latent variables. J. Mark. Res. 39 (1), 31–46. https://doi.
org/10.1509/jmkr.39.1.31.18937.
Bahamonde-Birke, F.J., Kunert, U., Link, H., Ortúzar, J.D., 2017. About attitudes and perceptions: finding the proper way to consider latent variables in discrete choice
models. Transportation 44 (3), 475–493. https://doi.org/10.1007/s11116-015-9663-5.
Bahamonde-Birke, F., Ortúzar, J.D., 2014. Is sequential estimation a suitable second best for estimation of hybrid choice models? Transport. Res. Record: J. Transport.
Res. Board 2429, 51–58. https://doi.org/10.3141/2429-06.
Bahamonde-Birke, F., Raveau, S., Yáñez, M.F., Ortúzar, J.D., 2010, The role of tangible attributes in hybrid discrete choice models. European Transport Conference,
Glasgow.
Basu, D., Hunt, J.D., 2012. Valuing of attributes influencing the attractiveness of suburban train service in Mumbai city: A stated preference approach. Transport. Res.
Part A: Policy Pract. 46 (9), 1465–1476. https://doi.org/10.1016/j.tra.2012.05.010.
Batarce, M., Muñoz, J.C., Ortúzar, J.D., Raveau, S., Mojica, C., Rios, R.A., 2015. Use of mixed stated and revealed preference data for crowding valuation on public
transport in Santiago, Chile. Transport. Res. Record 2535, 73–78. https://doi.org/10.3141/2535-08.
Batarce, M., Muñoz, J.C., Ortúzar, J.D., 2016. Valuing crowding in public transport: Implications for cost-benefit analysis. Transport. Res. Part A: Policy Pract. 91,
358–378. https://doi.org/10.1016/j.tra.2016.06.025.
Ben-Akiva, M., Walker, J., Bernardino, A.T., Gopinath, D.A., Morikawa, T., Polydoropoulou, A., 2002. Integration of choice and latent variable models. In:
Mahmassani, H.S. (Ed.), In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges. Emerald, Bingley, United Kingdom, pp.
431–470.
Björklund, G., Swärdh, J.-E., 2017. Estimating policy values for in-vehicle comfort and crowding reduction in local public transport☆. Transport. Res. Part A: Policy
Pract. 106, 453–472. https://doi.org/10.1016/j.tra.2017.10.016.
Bollen, K.A., 2002. Latent variables in psychology and the social sciences. Annu. Rev. Psychol. 53, 605–634. https://doi.org/10.1146/annurev.psych.53.100901.
135239.
Bouscasse, H., de Lapparent, M., 2018, Perceived Comfort and Values of Travel Time Savings in the Rhône–Alpes Region. In: Transportation Research Board 97th
Annual Meeting, 2018-1-7 to 2018-1-11, Washington DC, United States. https://trid.trb.org/View/1496545.
Cheng, V., 2010. Understanding Density and High Density. In: Ng, E. (Ed.), Designing High-Density Cities for Social and Environmental Sustainability. Routledge,
London, pp. 3–18.
ChoiceMetrics, 2012. Ngene 1.1.1 User Manual & Reference Guide, Australia.
Cox, T., Houdmont, J., Griffiths, A., 2006. Rail passenger crowding, stress, health and safety in Britain. Transport. Res. Part A: Policy Pract. 40 (3), 244–258. https://
doi.org/10.1016/j.tra.2005.07.001.
Daly, A., Hess, S., Patruni, B., Potoglou, D., Rohr, C., 2012. Using ordered attitudinal indicators in a latent variable choice model: a study of the impact of security on
rail travel behavior. Transportation 39 (2), 267–297. https://doi.org/10.1007/s11116-011-9351-z.
Das, S., Pandit, D., 2013. Importance of user perception in evaluating level of service for bus transit for a developing country like India: a review. Trans. Rev. 33 (4),
402–420. https://doi.org/10.1080/01441647.2013.789571.
Dell’Olio, L., Ibeas, A., Cecin, P., 2011. The quality of service desired by public transport users. Transp. Policy 18 (1), 217–227. https://doi.org/10.1016/j.tranpol.
2010.08.005.
Duduta, N., Subedi, A., 2015. Understanding platform overcrowding at bus rapid transit stations. Transp. Res. Rec. 2533, 118–123. https://doi.org/10.3141/2533-13.
Eboli, L., Mazzulla, G., 2007. Service quality attributes affecting customer satisfaction for bus transit. J. Public Transport. 10 (3), 21–34. https://doi.org/10.5038/
2375-0901.10.3.2.
Evans, G.W., Lepore, S., Mata-Allen, K., 2000. Cross-cultural differences in tolerance for crowding: Fact or fiction? J. Pers. Soc. Psychol. 79 (2), 204–210. https://doi.
org/10.1037/0022-3514.79.2.204.
Evans, G.W., Wener, R.E., 2007. Crowding and personal space invasion on the train: Please don’t make me sit in the middle. J. Environ. Psychol. 27 (1), 90–94. https://

464
L. Márquez, et al. Transportation Research Part A 130 (2019) 452–465

doi.org/10.1016/j.jenvp.2006.10.002.
Feife, Q., 2014. 2014, Investigating the in-vehicle crowding cost functions for public Transit modes. Math. Probl. Eng. 5, 1–13. https://doi.org/10.1155/2014/502708.
Fletcher, G., El-Geneidy, A., 2013. Effects of fare payment types and crowding on dwell time: fine-grained analysis. Transport. Res. Record: J. Transport. Res. Board
2351, 124–132. https://doi.org/10.3141/2351-14.
Guerra, G., Bocarejo, J.P., 2013. Congestion cost in mass transit systems; Pricing and investment policy implications. Case Study: Bogotá’s BRT system. In: 13th WCTR,
July 11-15, 2013, Rio do Janeiro; Brazil. http://www.wctrs-society.com/wp-content/uploads/abstracts/rio/selected/3292.pdf.
Guzman, L.A., Bocarejo, J.P., 2017. Urban form and spatial urban equity in Bogota, Colombia. Transport. Res. Procedia 25, 4491–4506. https://doi.org/10.1016/j.
trpro.2017.05.345.
Haywood, L., Koning, M., 2015. The distribution of crowding costs in public transport: New evidence from Paris. Transport. Res. Part A: Policy Pract. 77, 182–201.
https://doi.org/10.1016/j.tra.2015.04.005.
Haywood, L., Koning, M., Monchambert, G., 2017. Crowding in public transport: Who cares and why? Transport. Res. Part A: Policy Pract. 100, 215–227. https://doi.
org/10.1016/j.tra.2017.04.022.
Hidalgo, D., Lleras, G., Hernández, E., 2013a. Methodology for calculating passenger capacity in bus rapid transit systems: Application to the TransMilenio system in
Bogotá, Colombia. Res. Transport. Econ. 39 (1), 139–142. https://doi.org/10.1016/j.retrec.2012.06.006.
Hidalgo, D., Pereira, L., Estupiñán, N., Jiménez, P.L., 2013b. TransMilenio BRT system in Bogota, high performance and positive impact – Main results of an ex-post
evaluation. Res. Transport. Econ. 39 (1), 133–138. https://doi.org/10.1016/j.retrec.2012.06.005.
Hörcher, D., Graham, D.J., Anderson, R.J., 2017. Crowding cost estimation with large scale smart card and vehicle location data. Transport. Res. Part B: Methodol. 95,
105–125. https://doi.org/10.1016/j.trb.2016.10.015.
Imre, S., Celebi, D., 2017. Measuring comfort in public transport: a case study for Istanbul. Transp. Res. Procedia 25, 2441–2449. https://doi.org/10.1016/j.trpro.
2017.05.261.
Jiao, L., Shen, L., Shuai, C., Tan, Y., He, B., 2017. Measuring crowdedness between adjacent stations in an urban metro system: a Chinese case study. Sustainability 9,
2325. https://doi.org/10.3390/su9122325.
Kim, K.M., Hong, S.-P., Ko, S.-J., Kim, D., 2015. Does crowding affect the path choice of metro passengers? Transport. Res. Part A: Policy Pract. 77 (C), 292–304.
https://doi.org/10.1016/j.tra.2015.04.023.
Li, Z., Hensher, D.A., 2011. Crowding and public transport: A review of willingness to pay evidence and its relevance in project appraisal. Transp. Policy 18 (6),
880–887. https://doi.org/10.1016/j.tranpol.2011.06.003.
Li, Z., Hensher, D.A., 2013. Crowding in public transport: a review of objective and subjective measures. J. Public Transport. 16 (2), 107–134. https://doi.org/10.
5038/2375-0901.16.2.6.
Li, H., Gao, K., Tu, H., 2017. Variations in mode-specific valuations of travel time reliability and in-vehicle crowding: Implications for demand estimation. Transport.
Res. Part A: Policy Pract. 103, 250–263. https://doi.org/10.1016/j.tra.2017.06.009.
Lleras, G.C., 2003, Bus rapid transit: impacts on travel behavior in Bogotá. Thesis (M.C.P. and S.M.), Massachusetts Institute of Technology, Dept. of Urban Studies and
Planning, 2003. https://dspace.mit.edu/handle/1721.1/39777.
Mahudin, N.M., Cox, T., Griffiths, A., 2011. Modelling the spillover effects of rail passenger crowding on individual well being and organisational behaviour. In: Urban
Transport XVII: Urban Transport and the Environment in the 21st Century, WIT Transactions on the Built Environment. 227–238. DOI: 10.2495/UT110201.
Mahudin, N.M., Cox, T., Griffiths, A., 2012. Measuring rail passenger crowding: scale development and psychometric properties. Transp. Res. Part F 15 (1), 38–51.
https://doi.org/10.1016/j.trf.2011.11.006.
Márquez, L., Cantillo, V., Arellana, J., 2014. How are comfort and safety perceived by inland waterway transport passengers? Transp. Policy 36, 46–52. https://doi.
org/10.1016/j.tranpol.2014.07.006.
Márquez, L., Cantillo, V., Arellana, J., 2018a. Assessing the influence of indicators’ complexity on hybrid discrete choice model estimates. Transportation 45. https://
doi.org/10.1007/s11116-018-9891-6.
Márquez, L., Pico, R., Cantillo, V., 2018b. Understanding captive user behavior in the competition between BRT and motorcycle taxis. Transp. Policy 61, 1–9. https://
doi.org/10.1016/j.tranpol.2017.10.003.
Pel, A.J., Bel, N.H., Pieters, M., 2014. Including passengers’ response to crowding in the Dutch national train passenger assignment model. Transport. Res. Part A:
Policy Pract. 66, 111–126. https://doi.org/10.1016/j.tra.2014.05.007.
Qin, F., 2014, Investigating the in-vehicle crowding cost functions for public transit modes. Math. Probl. Eng. 2014 (502708), 13 p. DOI: 10.1155/2014/502708.
Sun, L., Tirachini, A., Axhausen, K.W., Erath, A., Lee, D.-H., 2014. Models of bus boarding and alighting dynamics. Transport. Res. Part A: Policy Pract. 69, 447–460.
https://doi.org/10.1016/j.tra.2014.09.007.
Sun, G., Zacharias, J., 2017. Can bicycle relieve overcrowded metro? Managing short-distance travel in Beijing. Sustain. Cities Soc. 35, 323–330. https://doi.org/10.
1016/j.scs.2017.08.010.
Sundstrom, E., 1978. Crowding as a sequential process: Review of research on the effects of population density on humans. In: Baum, A., Epstein, Y.M. (Eds.), Human
Response to Crowding. Hillsdale Erlbaum, New York.
Sundstrom, E.B., Busby, P., Asmus, C., 1975. An experimental study of crowding: Effects of room size, intrusion, and goal blocking on nonverbal behavior, self-
disclosure, and self-reported stress. J. Pers. Soc. Psychol. 32 (4), 645–654.
Tirachini, A., Hensher, D.A., Rose, J.M., 2013. Crowding in public transport systems: Effects on users, operation and implications for the estimation of demand.
Transport. Res. Part A: Policy Pract. 53, 36–52. https://doi.org/10.1016/j.tra.2013.06.005.
Tirachini, A., Sun, L., Erath, A., Chakirov, A., 2016. Valuation of sitting and standing in metro trains using revealed preferences. Transp. Policy 47, 94–104. https://doi.
org/10.1016/j.tranpol.2015.12.004.
Tirachini, A., Hurtubia, R., Dekker, T., Daziano, R.A., 2017. Estimation of crowding discomfort in public transport: Results from Santiago de Chile. Transport. Res. Part
A: Policy Pract. 103, 311–326. https://doi.org/10.1016/j.tra.2017.06.008.
Varghese, V., Adhvaryu, B., 2016. Measuring overcrowding in Ahmedabad buses: costs and policy implications. Transp. Res. Procedia 17, 145–154. https://doi.org/10.
1016/j.trpro.2016.11.070.
Vij, A., Walker, J.L., 2014. Hybrid choice models: the identification problem. In: Hess, S., Daly, A. (Eds.), Handbook of Choice Modelling. Edward Elgar Publishing,
Cheltenham.
Vij, A., Walker, J.L., 2016. How, when and why integrated choice and latent variable models are latently useful. Transport. Res. Part B: Methodol. 90, 192–217.
https://doi.org/10.1016/j.trb.2016.04.021.
Vuchic, V.R., 2005. Urban Transit Operations Planning and Economics. John Wiley & Sons, Hoboken, New Jersey.
Wardman, M., Whelan, G., 2011. Twenty years of rail crowding valuation studies: evidence and lessons from British experience. Trans. Rev. 31, 379–398. https://doi.
org/10.1080/01441647.2010.519127.

Luis Márquez is a Transportation and Highways Engineering (1993) and M.Sc. in Transportation Engineering (2009), both from the Universidad Pedagógica y
Tecnológica de Colombia, and Ph.D. in Civil Engineering (2017) from the Universidad del Norte. He is an Associate Professor at the Universidad Pedagógica y
Tecnológica de Colombia since 2006. His research interests include transportation planning, transportation economics, transportation modeling, and logistics. He is
currently the head of i-MOVyT (Institute of Research and Development in Mobility and Transport) of the same university.

Julieth V. Alfonso V. is a Transportation and Highways Engineering (2017) from the Universidad Pedagógica y Tecnológica de Colombia. Nowadays, she is a younger
researcher in the transportation research group at the Universidad Pedagógica y Tecnológica de Colombia. Her research interests include transportation planning,
pedestrian modeling, transportation modeling, and urban planning.

Juan C. Poveda is a Transportation and Highways Engineering (1989) from the Universidad Pedagógica y Tecnológica de Colombia and M.Sc. in Informatics (2001)
from the Universidad Industrial de Santander. He is an Associate Professor at the Universidad Pedagógica y Tecnológica de Colombia since 1994. His research interests
include logistics, optimization, and transportation planning.

465

You might also like