Conference PaperPDF Available

Modeling the Propensity to Join Carsharing Using Hybrid Choice and Latent Variable Models and Mixed Internet-Paper Survey Data.

Authors:

Figures

Content may be subject to copyright.
1"
MODELING THE PROPENSITY TO JOIN CARSHARING 1"
USING HYBRID CHOICE AND LATENT VARIABLE MODELS 2"
AND MIXED INTERNET/PAPER SURVEY DATA 3"
4"
5"
6"
7"
Dimitrios Efthymiou, MSc* 8"
PhD Candidate, Laboratory of Transportation Engineering 9"
National Technical University of Athens 10"
Visiting Scholar, Transport and Mobility Laboratory (TRANSP-OR) 11"
École Polytechnique Fédérale De Lausanne 12"
EPFL ENAC INTER, Station 18, Lausanne, CH-1015 13"
Tel: 0041-21-6939329; Fax:0041-21-6938060 14"
Email: defthym@mail.ntua.gr, dimitrios.efthymiou@epfl.ch 15"
16"
17"
18"
Constantinos Antoniou 19"
Assistant Professor, Laboratory of Transportation Engineering 20"
School of Rural and Surveying Engineering 21"
National Technical University of Athens 22"
9, Iroon Polytechniou St., Zografou Campus, 15780 23"
Tel: 0030-210-7722783; Fax:0030-210-7722629 24"
Email: antoniou@central.ntua.gr 25"
26"
27"
28"
29"
30"
31"
32"
33"
34"
35"
36"
37"
38"
39"
40"
*Corresponding author 41"
Word Count: 5570 words + 5 tables + 2 figures = 7370 42"
Submitted on August 1st, 2013 43"
Submitted for presentation in the 93rd Annual Meeting of the Transportation 44"
Research Board and publication in Transportation Research Record: Journal of the 45"
Transportation Research Board. 46"
2"
MODELING THE PROPENSITY TO JOIN CARSHARING 1"
USING HYBRID CHOICE AND LATENT VARIABLE MODELS 2"
AND MIXED INTERNET/PAPER SURVEY DATA 3"
4"
5"
6"
7"
ABSTRACT 8"
9"
Carsharing is a sustainable, alternative transportation mode, easily implemented at 10"
local level. Its service combines characteristics from both private and public 11"
transportation, rendering it attractive to specific groups of commuters. In order to 12"
increase the demand for carsharing and optimize its operation, the characteristics of 13"
the potential users should be investigated. 14"
In this paper, the propensity of people to join a carsharing scheme is being 15"
modeled using mixed Internet and paper survey data. An ordered logit model is 16"
developed, with dependent variable the stated propensity of young Greeks to join 17"
carsharing and explanatory demographic characteristics and travel attributes. The 18"
satisfaction of the commuters about the current travel patterns is considered as latent 19"
variable in the modeling framework. The aim of combining the two datasets is to 20"
measure the difference of the variance of the error term generated from the paper and 21"
Internet respondents, who are prone to be positive in their responses in order to satisfy 22"
the interviewer. This is verified in the current research by the positive sign of the 23"
scale parameter applied at the utility function of the Internet-based data. The use of 24"
latent classes enhances the model estimation, by measuring the parameters that 25"
determine the respondents’ unobserved, underlying behaviors. 26"
The results show that people who use taxi for social activity, are of medium to 27"
low income and environmental conscious, are more willing to joint the scheme. The 28"
results are compared with a 2013 study, in order to identify the advantages of using 29"
this advanced modeling framework. 30"
31"
Keywords: Carsharing, discrete choice models, latent classes, mixed internet/paper 32"
data, Athens 33"
34"
35"
36"
37"
38"
39"
40"
41"
42"
43"
44"
45"
46"
47"
48"
49"
3"
INTRODUCTION 1"
2"
The always increasing generalized travel cost and the current economic difficulties 3"
that many households are experiencing due to the crisis, has rendered car ownership a 4"
luxury good that not many people can afford. Carsharing users enjoy the privacy of 5"
the car without incurring the cost of purchase. Moreover, the exact cost of their trip is 6"
known at the same time, avoiding the possible miscalculations that occur in private 7"
transport. The users pay a combination of a registration fee, a monthly amount and a 8"
cost per distance unit driven or time spent using the service. 9"
In this paper, the research that begun by Efthymiou et al. (1), where the 10"
authors analyzed the factors that affect the willingness to join car -or bike-sharing 11"
systems in Greece, is being extended. In Efthymiou et al. (1), the authors begin with 12"
an extensive literature review of the available car and bike-sharing systems over the 13"
world, then analyze the factors that affect car and bike ownership, and conclude in 14"
measuring the factors that affect the potential to join car and bike sharing. The 15"
satisfaction of the respondents to their current travel patterns is, first, separately 16"
estimated, using the traditional framework of the ordered logit model. Then, the 17"
potential to join these schemes in three different time frames (immediately, within the 18"
first few years and eventually) is modeled. 19"
The objective of this paper is to model the propensity of people to join 20"
carsharing, using mixed Internet/paper survey data and hybrid choice and latent 21"
variable model. In order to measure the difference of the variance of the error term 22"
between the Internet and paper surveys, every coefficient in the utility function of the 23"
ordered logit model for the Internet data is multiplied by the same scale parameter. 24"
Moreover, the satisfaction of the respondents about their current travel patterns has 25"
been included as latent behavioral characteristic in the main utility function. The 26"
reason behind this structure is to reveal and measure the characteristics that determine 27"
the latent behavior, leading to a more detailed and accurate join model. The 28"
estimation of all the utility functions will be made simultaneously. 29"
The remainder of this paper is structured as follows: the next section describes 30"
carsharing schemes around the world focusing on studies that aim to identify the 31"
characteristics of carsharing users. The following Section describes the hybrid choice 32"
and latent variable model, and the mixed Internet/paper based survey data. The next 33"
Section analyzes the methodology that was followed, beginning with the data 34"
collection and then describing the model specification and the estimated coefficients. 35"
Finally, in the last Section, conclusions and recommendations for further research are 36"
presented. 37"
CARSHARING SYSTEMS AROUND THE WORLD 38"
39"
The carsharing system in Zurich, Switzerland, in 1987 was the first attempt to 40"
implement such a service in high scale (2). Since then, the same company has been 41"
developed to one of the most successful carsharing providers, operating more than 42"
1950 vehicles and serving about 77100 members (3). In North America, in 2006, 43"
Zipcar was serving about 180,000 members, more than 50% of the total users of the 44"
area, that are about 280,000 according to Shaheen et al. (4). 45"
In 2006, carsharing was counting more than 350,000 members in 600 cities 46"
and 18 countries around the world (4), while the number is expected to have been 47"
highly increased nowadays. The demographic characteristics of the carsharing users 48"
4"
have been investigated by a number of studies until now. In the recent literature, it is 1"
documented that the composition of the carsharing members differs from city to city. 2"
Carsharing systems act as complementary good of the public transport (5), since their 3"
users usually become the first members of the systems. Carsharing combines the 4"
positive characteristics of public transport and car, but it also attracts bike users and 5"
people traveling on foot. The variability of carsharing users’ travel patterns, and the 6"
need to understand the factors that determine their decision to join, maintains the 7"
research interest around this subject high. 8"
Burkhardt and Millard-Ball (6) found that in North America the majority of 9"
the users are between 25 and 35 years old, while there are not many members below 10"
the age of 21, because of driving license restrictions. The members are usually highly 11"
educated and environmental aware, while 50% of them has relatively high income 12"
(>60,000$). Finally, the great majority (72%) live in households without any available 13"
car. Carsharing system providers suggest the use to persons who drive between 14"
10,000 to 16,000 km per year, while the service remains a choice of students and low-15"
income households (6). Zhou et al. (7) concluded that car owners with high income 16"
are not interested in the service, and opposite to Burkhardt and Millart-Ball (6), the 17"
education level is not a significant determinant of the willingness to join the scheme. 18"
Another study (8) found that the users are mainly males, members of low-size 19"
households, between 30 and 40 years old, of high education level. Stillwater et al. (5) 20"
concluded that the users are members of households owning only one car, while solo 21"
drivers are not willing to join the scheme. 22"
Millard-Ball et al. (9) found that many members of carsharing schemes either 23"
cancel the purchase of their car or they sold it after joining the scheme. It has been 24"
estimated that in North America, carsharing resulted to the removal of 90,000 to 25"
130,000 cars from the streets (10), resulting to increasing the number of parking lots 26"
and reducing the air pollution and traffic congestion (11). Muheim and Partnet (12) 27"
found that in Europe the 15.6% to 31.6% of carsharing users sold their private car, 28"
and about 13% to 16.2% cancelled the purchase. Another study by Ryden and Morin 29"
(13) that was conducted in Belgium and Berm, found that 21% to 34% of people sold 30"
their car after joining a carsharing scheme. Other studies in Northern America 31"
concluded that the equivalent percentage reaches 68%, which can be explained since 32"
car ownership plays more important role in daily transport there, and as a result it is 33"
expected that the margins of car usage reduction are higher (14). According to the 34"
same research, only 8% of the American households do not own car, which is the 35"
main travel mode for short and long trips. However, these findings are based on stated 36"
preference survey, which are vulnerable to biases, either because of the difference 37"
between the stated and the final decision of the respondents, or because of the mis-38"
execution of the experiment (15, 16). Stasko et al. (17) found that carsharing has lead 39"
to a reduction od 15.3 personal vehicles for every carsharing vehicle in Ithaca, a 40"
finding that is similar to researched performed in major US cities. 41"
Carsharing companies invest on the development of new user-interface 42"
technologies in order to increase the flexibility of the users and attract more members. 43"
The results of a study that was conducted by Nerenberg et al. (18) in San Francisco 44"
between 1996 and 1998, shows that the women who are attracted by electric car-45"
sharing are mainly driven by environmental incentives, while men because they found 46"
interesting the technical perspective of the service, revealing that the system should 47"
not only be functional, but have technological and environmental aspects. 48"
Known carsharing providers around the world are “Greenwheels”, in 49"
Netherlands and Germany, “CityCarClub” in Sweden, Finland and “Zipcar”, in USA, 50"
5"
Canada and UK. Zipcar and Flexcar, the two major US companies, were merged in 1"
2007. The business model of these companies is usually different. CityCarShare is a 2"
non-profit company and returns the revenues to the community, while on the other 3"
hand Zipcar is for-profit. Moreover, they use vehicles of different type in their fleets. 4"
The basic idea of this scheme is that the users are members of a system where 5"
they can utilize vehicles provided by a fleet, by paying an amount of money per time 6"
or distance of usage. The user can book a vehicle on-line, or via a smartphone 7"
application, and can then get access to it by using a personal reservation password. In 8"
its conventional form, the service requires the return of the vehicle to the parking lot 9"
from where it was hired (two-way carsharing) while in its more flexible form, the user 10"
can leave the car to any parking lot reserved for the company, which can be different 11"
from the one that he begun his trip (one-way carsharing). 12"
While carsharing usage is currently mainly addressed to the general public, it 13"
is believed than in the future the majority of users will be enterprises (13). The 14"
companies will provide their staff with the opportunity to use the service, in order to 15"
stop using their private cars for business trips. As a result, carsharing could 16"
complement of substitute part of the leasing usage on the business domain (19). 17"
Concerning the environmental impact of carsharing usage, Shaheen et al. (4), 18"
found that the number of vehicle kilometers driven has been reduced, and carsharing 19"
has definitely contributed to that. Douma and Gaug (20) suggest that people are 20"
turning to more environmental friendly travel patterns, either for their transition to the 21"
carsharing stations, or in general. 22"
The information of the people about carsharing systems is of major 23"
importance for the promotion of the scheme. Clavel et al. (2) found that in France, 24"
60% of the participants of a survey were not aware of the service, even in Paris area, 25"
where carsharing is operating since 10 years. Moreover, a 28% was confounding the 26"
term carsharing with carpooling, considering that the first refers to the second. 27"
Shaheen and Martin (21) performed a survey in Beijing, when the carsharing 28"
scheme was still in early stages of operation, and found that the 40% of the 29"
respondents were not familiar with the scheme. In a similar research in Greece (1), 30"
where the system is still absent, 52.4% of the respondents declared ignorance. 31"
Other initiatives like Getaround (www.getaround.com) offer peer-to-peer 32"
solutions through websites and smartphone applications. In these systems, car-owners 33"
can rent their vehicles for the time they don’t use it. 34"
35"
METHODOLOGY 36"
37"
Ordered logit model 38"
39"
A usual form of alternative response options in a survey is using a rating scale (22). 40"
The respondents can select between ordered alternatives, usually 5 or 7, such as: very 41"
satisfied / satisfied / neutral / dissatisfied / very dissatisfied. Instead of using a 42"
multinomial logit model, the ordered logit can be estimated when these kinds of 43"
responses are available. In ordered logit, the error terms for each alternative are 44"
independent (23). The alternatives closer to each other are more similar in the ordered 45"
response. The ordered model estimates the threshold values between the choices, 46"
apart from the coefficients of the alternatives. 47"
6"
Figure 1 shows an illustration of the distribution of the preferences. In case of 1"
four alternative ordered responses, there are three thresholds (or critical values) that 2"
separate the choices. When the utility of the respondent is between k1 and k2, he 3"
selects the alternative ‘Not Good’, while when it is more than k3, the alternative ‘Very 4"
Good’ is selected. 5"
6"
7"
8"
9"
FIGURE 1. Distribution of the respondents’ preference 10"
(adapted from 24) 11"
12"
13"
14"
Hybrid choice and latent variable model 15"
16"
In order to reveal the latent behaviors of the respondents, Structural Equation Models 17"
(SEM) are used. The interest around the application of SEM in transportation research 18"
is increasing. SEMs have been applied in modeling travel behavior (25) and modal 19"
choice (26). Golob (27) makes an extensive review of the SEM applications in 20"
transport until 2003. 21"
Walker (28) introduced the hybrid choice and latent variable models, which 22"
integrate the structural equation models in the discrete choice. In these models, the 23"
latent explanatory variables are developed by one or more discrete indicators. 24"
Recently, behaviors such as the convenience, environmental consciousness, safety, 25"
comfort and flexibility have been introduced in mode choice modeling (e.g. 29). The 26"
modeling framework of the hybrid choice and latent variable modes is as follows: 27"
28"
Structural equation 29"
30"
𝑍!"
=𝑋!𝜆!+𝜔!",𝜔!~𝑁(0,𝛴!) (3) 31"
32"
where l is the index number of the latent variable, X is the explanatory variable,!λ is 33"
the coefficient of the explanatory variables, Ζ is the latent variable and 𝜔 is the error 34"
term 35"
36"
𝑈!=𝑋!𝛽!+𝑍!
𝛽!+𝜀!,𝜀!~𝑠𝑡𝑎𝑛𝑑𝑎𝑟𝑑!𝑙𝑜𝑔𝑖𝑠𝑡𝑖𝑐 (4) 37"
38"
Antoniou"and"Tyrinopoulos"
" 9"
"
more desirable. One key question associated with factor analysis is the determination of the 1"
appropriate number of factors. The approach followed in this research is to incrementally 2"
increase the number of factors and check the amount of the variance that each additional 3"
factor explains. This process stops when no additional factor can explain more than 10% of 4"
the variance of the data. This process is similar to the scree plot approach and is demonstrated 5"
in the factor analysis results presented in the next section. 6"
7"
Ordered probit model 8"
Respondents in surveys are often asked to express their preferences in a rating scale. Such 9"
scales are often called Likert scales (17, 18). A multinomial logit model could be specified 10"
with each potential response coded as an alternative. However, the ordering of the 11"
alternatives violates the independence of the errors for each alternative, and therefore the 12"
Independence for Irrelevant Alternatives (IIA) assumption of the logit model. Nested or 13"
cross-nested models are one approach to overcoming this issue (19), while multinomial probit 14"
models also do not suffer from this limitation. Ordered logit and probit models provide 15"
another approach that estimates parameter coefficients for the independent variables, as well 16"
as intercepts (or threshold values) between the choices. 17"
Figure 1 shows the distribution of the choice probability P as a function of the utility 18"
U. Assuming a ranking scale with four levels (like the one used in the questionnaire in this 19"
study), there are three thresholds or critical values (k1 through k3) that separate the four 20"
choices (“Unacceptable”, “Not good”, “Good”, “Very good”). For example, respondents 21"
choose the alternative “Unacceptable” if the utility is below k1, alternative “Not good” if the 22"
utility is between k1 and k2, and so on. 23"
24"
25"
FIGURE 1. Distribution of the respondents' preference (adapted from Train, 20). 26"
27"
In the ordered probit models developed in this research, the ordered response is used 28"
directly as the dependent variable. In each model, the response variable takes numerical 29"
values between 1 and 4, with 1 indicating that the respondent is stating that he considers the 30"
public transport system in Kos as unacceptable and 4 indicating that the respondent would 31"
evaluate the system as very good. 32"
If Y is the response factor with K levels, the model can be written as: 33"
!!!!!=!!!!
!=!!!
!!!=Φ!
!!!
where: 34"
P
U
k1 k2 k3
Unacceptable Very good Not good
How would you evaluate the public transport system of Kos?
Good
7"
U is the utility, X is the explanatory variable, β is the coefficient of the xplanatory 1"
variable, Z is the latent variable and ω is the error term 2"
3"
Measurement model 4"
5"
𝐼!" =𝑍!
𝛼!+𝑢!",!!!!𝑤𝑒𝑟𝑒!𝑟!𝑖𝑠!𝑡𝑒!𝑡𝑜𝑡𝑎𝑙!𝑛𝑢𝑚𝑏𝑒𝑟!𝑜𝑓!𝑖𝑛𝑑𝑖𝑐𝑎𝑡𝑜𝑟𝑠 (5) 6"
"7"
"8"
Questionnaire design and dissemination 9"
10"
The data used in this research were collected by on-line and paper questionnaires. The 11"
on-line dataset has been used in the past by Efthymiou et al. (1) in a preliminary 12"
attempt to identify the characteristics of the forthcoming sharing-vehicle systems’ 13"
users in Athens, Greece. Apart from carsharing, that survey was also containing 14"
questions about the perception of people about bikesharing schemes. In that study, the 15"
respondents were asked to rate several factors that would affect their decision to join 16"
car/bike-sharing schemes, and their willingness to join them with electric or 17"
conventional vehicles. 18"
However, on-line questionnaires are vulnerable to biases (30), since the 19"
respondents usually tend to be more positive in questions related with the core 20"
subject. For that reason, the authors consider that it is essential to enrich the sample 21"
with paper responses. These were collected in Athens during the same period, within 22"
the framework of a Master’s Thesis at NTUA (31). The structure of the paper survey 23"
is similar to the on-line, focusing particular on carsharing (and not bikesharing). More 24"
specifically, the questionnaire is divided in three parts. The first part includes 25"
questions about the current travel patterns of the respondents, such as which mode 26"
they use for trips to different activities (e.g. work, school, social activities), how much 27"
time they travel per day, how many times per day they used the car last week, if they 28"
have a car driving license, their perception about the advantages and disadvantages of 29"
car ownership systems (costs, environment, parking, etc.) and finally it asks them to 30"
rate their satisfaction about their current travel patterns. The second part begins with a 31"
description of the carsharing system and a number of relevant questions then follows, 32"
such as how important they rate characteristics of its operation (distance of stations 33"
from home/job, ability to return the vehicle to other stations than the one that they 34"
begun their trips, way of reservation). The survey ends with demographic questions, 35"
such as age, gender, income, size of household, municipality of residence and job. 36"
Wherever it was applicable, the answers were structured in ordered scale, ranging 37"
between “absolutely no” and “absolutely yes” (e.g. the answers of the question “how 38"
possible do you think it is that you will join carsharing in the short-term”), or “totally 39"
unimportant” to “very important” (e.g. about the car-ownership characteristics). When 40"
the individual preference is scaled, the collected information is richer and more useful 41"
than the binary. This rating system in known as Likert scales (22, 32). This scaled 42"
type of answers is ideal for the estimation ordered logit models. 43"
The on-line survey was developed in Google forms, and was disseminated via 44"
mailing lists and social media, while the paper was filled by individual interviews, 45"
that were performed at the metro stations, Malls, and local neighborhoods in Athens. 46"
These locations are suitable for data collection, because they are used by people of 47"
random demographic characteristics living in different areas around Attica, leading to 48"
a good classified sample. In a period of three months (April to July 2012) 194 49"
completed paper questionnaires were collected (118 of which are used in this 50"
8"
research, due to the age subgroup 18-35 that has been used). Although this survey led 1"
to a more representative sample of the population -in terms of age, education level 2"
and household size distribution- than with the on-line, in this paper we focus on the 3"
young respondents (18 to 35 years old), in order to be consistent with both datasets. 4"
One difference between the two datasets was the range of “satisfaction about 5"
the current travel patterns” question, which on the paper survey was taking values 6"
from 1 to 7, while on the on-line questionnaire from 1 to 5. Therefore, in order to 7"
make the responses compatible, the paper responses 2, 3 and 5, 6 were grouped (in 8"
two categories). 9"
The results of the surveys are expected to differ in terms of systematically 10"
different behavior, understanding of the subject, and honesty of the participants when 11"
responding. It is expected that the respondents of the on-line questionnaires tend to be 12"
more positive in their responses, while it is also possible that some of them skip 13"
reading the introductory paragraph of the carsharing systems, and were probably 14"
confusing it with the carpooling. This bias was avoided during the paper survey, 15"
because the interviewer explained the carsharing scheme in detail, making sure that 16"
the respondent was not confusing it with carpooling before completing. Table 1 shows 17"
the summary statistics of our joined dataset: 18"
19"
20"
21"
22"
23"
24"
25"
26"
27"
28"
29"
30"
31"
32"
33"
34"
35"
36"
37"
38"
39"
40"
41"
42"
43"
44"
45"
46"
47"
48"
49"
9"
TABLE 1: Summary statistics 1"
2"
3"
4"
5"
6"
7"
8"
9"
10"
11"
12"
13"
14"
15"
16"
17"
18"
19"
20"
21"
22"
23"
24"
25"
26"
27"
28"
29"
30"
31"
32"
33"
34"
35"
36"
37"
38"
39"
40"
41"
42"
43"
Notes: 1: percent of sample, 2: percent of full population, N/A: not available 44"
45"
46"
Paper: 118
Internet: 233
Survey
(ages 18-35)
2001 Census
(ages 20-35)
Age
18-25
52.1%1
7.6%2
26-35
47.9%1
15.7%2
Gender
Male
46.4%
51.5%
Female
53.6%
48.5%
Marital status
Single
94.0%
67.0%
Married
6.0%
29.6%
Employment status
Full-time
43.9%
61.6%
Part-time
11.1%
Student
35.6%
10.2%
Unemployed - looking for
a job
8.5%
4.5%
Unemployed not
looking for a job
0.6%
Homemaker
0.3%
12.7%
Other
-
11.0%
Education level completed
High school
12.5%
45.4%
College
5.4%
14.5%
Higher education -
University
47.1%
11.8%
Masters
29.3%
1.1%
Doctorate
5.1%
0.2%
Other (less than high
school)
0.3%
26.8%
Survey
(ages 18-35)
2001 Census
(All ages)
Household size
1
29.6%
7.0%
2
16.8%
20.0%
3
18.2%
22.5%
4
27.9%
29.3%
5 or more
7.4%
21.0%
Household income
<10
18.5%
N/A
10-15
17.3%
N/A
15-25
22.2%
N/A
25-50
25.3%
N/A
50-75
8.5%
N/A
75-100
4.0%
N/A
>100
4.0%
N/A
10"
Model Development 1"
2"
In this study, an ordered logit model is developed, to model the willingness of young 3"
Greeks (18 to 35 years old) to join a possible, future carsharing scheme in the mid-4"
term future (1 to 5 years). The dependent variable takes values from 1 (absolutely no) 5"
to 5 (definitely yes). Moreover, the model contains a latent variable with ordered 6"
responses as nests, aiming to capture the latent satisfaction about the current travel 7"
patterns of the respondents. The two different datasets, on-line and paper responses, 8"
were used jointly. The method used is similar to the combination of SP and RP data, 9"
introduced by Ben-Akiva and Morikawa (33) in order to overcome the shortcoming of 10"
the one to the other. Louviere et al (34) makes a review of studies combining 11"
different datasets for model estimation. 12"
The hybrid model is composed by three utility functions: 1) the willingness to 13"
join using the on-line dataset, where a common scale is applied to every explanatory 14"
variable; 2) the willingness to join using the paper dataset and 3) the satisfaction. The 15"
explanatory variables of the utilities to join the carsharing scheme for both datasets 16"
are the same. 17"
The authors tried to retain the model consistent with Efthymiou et al. (1), 18"
during the development. The estimation begun by including all the available 19"
variables, and then gradually eliminate the insignificant (one by one) until reaching 20"
the final specification, without considering the latent classes at first step. Then, the 21"
latent class of the satisfaction was developed, similar to Efthymiou et al. (1). 22"
However, not all the variables including in (1) remained significant and as a result 23"
many of them were eliminated. The coefficients of the utility functions were 24"
estimated simultaneously using the maximum likelihood estimation in BIOGEME 25"
(35). 26"
27"
The variables of the main model (willingness to join) are: 28"
29"
social_taxi: the respondent uses taxi for trips to social activity. The model 30"
examines this case over the alternatives, which are the car, bicycle, public 31"
transport, and other 32"
income3: a dummy variable that takes the value 1 for household income 33"
between 15 to 25 thousand euros, else 0. The alternatives are incomes <10, 10-34"
15, 25-75, 75-100 and >100 thousand euros 35"
envcon: respondents’ perception of how environmental conscious they think 36"
they are; takes values from 1 to 7 37"
satisfaction: satisfaction of the respondents about the current travel patterns; it 38"
is the latent variable 39"
40"
41"
The variables of the satisfaction latent class model are: 42"
43"
age: a dummy variable that takes the value 1 if the respondent is within the 44"
age 26 and 35, else 0 45"
time: a dummy variable that takes the value 1 if the respondent spends more 46"
than 45 minutes travelling to work/school per day, else 0 47"
socialbus: a dummy variable that takes the value 1 if the respondent uses the 48"
bus for trips to social activities, else 0 49"
11"
married: a dummy variable that takes the value 1 if the respondent is married, 1"
else 0 2"
carown: a dummy variable that takes the value 1 if the respondent owns a car, 3"
else 0 4"
omega: the error term of the latent variable 5"
6"
The functions of the main models are presented below. 7"
8"
V
! is the utility function applied on the on-line dataset 9"
V
! is the utility function applied on the paper dataset 10"
satisfaction is the formulation of the latent satisfaction 11"
12"
13"
14"
𝑉
!=𝛽!"#$%!×!𝛽!"#_!"#!×!𝑒𝑛𝑣𝑐𝑜𝑛 +!𝛽!"#!×!𝑠𝑎𝑡𝑖𝑠𝑓𝑎𝑐𝑡𝑖𝑜𝑛 +𝛽!"!!!×!𝑖𝑛𝑐𝑜𝑚𝑒!
+!𝛽!"_!𝑎!"!×!𝑠𝑜𝑐𝑖𝑎𝑙_𝑡𝑎𝑥𝑖
15"
(1) 16"
17"
𝑉
!=𝛽!"#_!"#!×!𝑒𝑛𝑣𝑐𝑜𝑛 +!𝛽!"#!×!𝑠𝑎𝑡𝑖𝑠𝑓𝑎𝑐𝑡𝑖𝑜𝑛 +𝛽!"!!!×!𝑖𝑛𝑐𝑜𝑚𝑒!
+!𝛽!"_!"#$!×!𝑠𝑜𝑐𝑖𝑎𝑙_𝑡𝑎𝑥𝑖
18"
(2) 19"
20"
𝑠𝑎𝑡𝑖𝑠𝑓𝑎𝑐𝑡𝑖𝑜𝑛 =𝛽!"#!"#$!×!𝑎𝑔𝑒 +!𝛽!"#$!×!𝑡𝑖𝑚𝑒 +!𝛽!"_!"#!×!𝑠𝑜𝑐𝑖𝑎𝑙𝑏𝑢𝑠
+!!𝛽!"##$%&!×!𝑚𝑎𝑟𝑟𝑖𝑒𝑑 +!!𝛽!"#$%&!×!𝑐𝑎𝑟𝑜𝑤𝑛 +!𝛽!"#$%!×!𝑜𝑚𝑒𝑔𝑎
21"
(3) 22"
23"
24"
Figure 2 shows the full path diagram of the Willingness to Join model with Latent 25"
Variables, in an form that is similar to (28). 26"
27"
28"
29"
30"
31"
32"
33"
12"
1"
2"
FIGURE 2: Full Path Diagram for Willingness to Join Model with 3"
Latent Variables 4"
5"
6"
where X1-8 are the explanatory variables of the structure and the latent models, U1on-
7"
line is the utility of the on-line survey respondents, U2paper is the utility of the paper 8"
survey respondents, and y is the Propensity to Join. 9"
Tables 2 and 3 present the model estimation results (Table 3 separates the 10"
ordered logit models’ threshold estimates for clarity), while Tables 4 and 5 present the 11"
models of the previous research (1). Comparing the model with Efthymiou et al. (1), it 12"
is observed that household size, high education level (doctorate), using public 13"
transport to work/school, and age are not significant in the new models. On the other 14"
hand, the income (15-25K Euros) and the environmental consciousness are significant 15"
in both models. The more environmental conscious the respondent is, the more likely 16"
it is that he will join the scheme, while the more satisfied he is with his current travel 17"
patterns, the less likely it is (to join the new scheme). Respondents with middle 18"
income (between 15 and 25 thousand euros per year) are more likely to join. This 19"
probably indicates that people with lower income expect carsharing to be expensive, 20"
and they will still prefer to travel by public transit or foot. On the other hand, those 21"
with higher incomes will still find more attractive the choice of using their own 22"
vehicles. Moreover, employees, and those who use taxi for their trips to social 23"
activities (taxi is very popular for social trips in Greece) are more likely to join. 24"
Concerning the satisfaction model, variables that were found significant in 25"
Efthymiou et al. (1), where the model was estimated separately and are not significant 26"
here, are: high education level (doctorate), number of trips to work/school, mode for 27"
food shopping drive with others or bike and travel time to work/school in the range of 28"
30-40 minutes. On the other hand the car ownership became significant. More 29"
specifically: young respondents between 26 to 35 years old are less satisfied; the more 30"
time they spend for trips to work/school the less satisfied they are; those using public 31"
transport for trips to work/school are less satisfied; satisfaction is higher for married 32"
respondents and those that own a car. 33"
U1on(line"
X1:"Income"15(25K"
X2:"Use"taxi"for"trips"to"
social"ac9vity(
X3:(Environmental"
consciousness(
X4:(Age"26(35(
X5:"Time"to"work/school">"
45min"(
X6:"Use"bus"for"trips"to"
social"ac9vity(
X7:"Married(
X8:"Owns"car(
Sa9sfac9on"
ω1"
y"
Scale(
U2paper"
β#X,1"
β#X,2"
β1"
τon"
τpap"
Ι1:"How"sa9sfied"are"you"with"
your"current"travel"paZerns?"
λ#J1"
13"
TABLE 2: Joined on-line and paper model estimation with latent variables 1"
2"
Variable
Value
t-test
𝛽!"#$%
0.797
2.17
𝛽!"#_!"#
0.340
2.84
𝛽!"#!
0.630
2.04
𝛽!"_!"#$
1.77
2.27
𝛽!"#
-0.456
-1.80*
𝛽!"#$
-0.696
-2.67
𝛽!"#$
-0.959
-3.29
𝛽!"_!"#
-1.46
-3.07
𝛽!"##$%&
0.751
1.49*
𝛽!"#$%&
0.850
3.05
σ
1.02
2.18
Data sample n=351
Log-Likelihood =-908.459
𝜌 = 0.638
* not significant at 95% confidence interval 3"
4"
5"
TABLE 3: Intercepts of joined on-line and paper model estimation with latent 6"
variables 7"
8"
Model
Intercept**
Value
t-test
Main (on-
line)
1|2
-1.02
-1.40*
2|3
1.05
9.02
3|4
2.54
9.46
4|5
4.38
6.95
Main
(paper)
1|2
-0.76
-1.14*
2|3
1.03
5.49
3|4
2.78
6.75
4|5
5.18
5.19
Satisfaction
latent
1|2
-4.40
-7.07
2|3
-1.29
6.24
3|4
0.54
6.39
4|5
4.00
6.92
not significant at 95% confidence interval 9"
**Intercepts are the “cutpoints” or “thresholds” of the ordered logit model 10"
11"
12"
13"
"14"
"15"
"16"
"17"
"18"
14"
TABLE 4: Model of satisfaction for current travel patterns (N=233) 1"
Adapted from Efthymiou et al. (2013) 2"
3"
Ordered Logit Satisfaction Model
Variable
Value
t-test
Education: Doctorate
0.92
1.52
Trips per week to School: 3
1.59
2.79
Trips per week to School: 4
1.18
2.28
Trips per week to School: 5 or more
0.87
2.04
Mode for Work/School trips: Drive with others / Bicycle
1.02
1.83
Mode for Groceries Shopping: Drive with others
0.91
2.39
Mode for Social Activity: Bus/Trolley/Tram
-1.33
-2.54
Time spent per day to Work/School: 30-45min
-1.52
-3.97
Time spent per day to Work/School: more than 45min
-2.62
-4.71
Employment status: Working full/part – time
1.06
3.00
Age: 26-35 (reference level: 18-25)
-0.85
-2.60
Intercepts*
1|2
-4.56
-5.64
2|3
-2.28
-4.94
3|4
-0.75
-1.84
4|5
0.82
2.03
5|6
2.07
4.81
6|7
4.29
7.71
Residual Deviance: 532.09
AIC: 566.09
*Intercepts are the “cutpoints” or “thresholds of the ordered logit model
4"
5"
6"
7"
8"
9"
10"
11"
12"
13"
14"
15"
16"
17"
18"
19"
20"
21"
22"
23"
15"
TABLE 5: Intention to Join Carsharing 1"
Adapted from Efthymiou et al. (2013) 2"
3"
4"
Indicators
Carsharing (n=233)
Within the first few years
Value
Error
t-test
Environmental Consciousness
0.46
0.13
3.58
Household Income is 15-25KE
0.93
0.32
2.92
100-150Km per day last week
-1.80
0.59
-3.04
Take taxi for Social Activities
1.74
0.54
3.22
Walk to Work/School
0.76
0.38
1.99
Household Size is equal to 2
0.69
0.33
2.07
>60min per day to Work/School
1.05
0.48
2.19
Intercepts
Definitely no | Probably no
0.47
0.71
0.66
Probably no | Maybe
2.66
0.72
3.68
Maybe | Probably yes
4.15
0.75
5.53
Probably yes | Absolutely yes
6.08
0.82
7.45
Residual Deviance
568.986
AIC
590.986
5"
"6"
CONCLUSIONS 7"
8"
Carsharing forms an alternative, sustainable transportation mode that has met wide 9"
acceptance within the last few years. Many researchers have attempted to identify the 10"
characteristics of the users and predict the demand of forthcoming schemes using 11"
Stated Preference survey data; however, the analysis is often based on simple models 12"
and assumptions. 13"
In this paper, the propensity to join carsharing using hybrid choice and latent 14"
variable models is being modeled. The dependent variable is in ordered form, and the 15"
explanatory variables are discrete. Satisfaction of the respondents about their current 16"
travel patterns is included in the form of a latent variable, structured by one discrete-17"
response indicator. The datasets used for the analysis were collected via on-line and 18"
paper surveys. Two different utility functions were developed for each dataset, and a 19"
unique scale parameter was applied on the variables of the on-line utility. The reason 20"
behind this scale is to measure the distance between the responses of the participants 21"
of the different datasets. The structural and latent models were estimated 22"
simultaneously. 23"
The results show that, as expected, the Internet survey respondents are prone 24"
to give more positive answers, probably in order to satisfy the interviewer. This is 25"
verified by the positive scale variable into the model. Young Greeks with income 15-26"
25 thousand euros and those who use taxi for their trips for social activity, are more 27"
willing to join a possible carsharing system in the future. The more satisfied people 28"
are with their current travel patterns, the less possible it is that they will join a 29"
16"
scheme. Young respondents between 18 and 25 years old, married and car owners are 1"
more satisfied with their current travel patterns. On the other hand, those who use bus 2"
for their trips to social activity and those who spend much time travelling, are less 3"
satisfied. Comparing these findings with Efthymiou et al. (1) we observe that 4"
variables that were significant in the Internet-data model (1) are not significant in the 5"
current research, and vise versa. For example, the current model shows that the 6"
household size, the education level and the age, do not determine the willingness to 7"
join. Moreover, the results show that the responses of the on-line data that were used 8"
in the 2013 study are over-rated, probably leading to biases. 9"
The authors are concerned that the work –in some extend- is being redone, 10"
however, other dimensions of the research are being examined here. The results might 11"
differ from one study to the other, but the aim is to show that when applying different 12"
modeling methodologies and/or using different data, this is possible (we are not able 13"
to make a direct conclusion that the results of the first –or second- research are biased 14"
or not). Researchers and policy makers should be aware of these potential differences 15"
during transport policy planning. 16"
With the proposed modeling methodology for measuring the propensity of 17"
people to join carsharing, the authors aim to contribute to the literature of behavioral 18"
analysis on this topic. Although the results are initial and are based on a stated 19"
preference survey, the methodology could be applied to extensive revealed preference 20"
data if available. 21"
22"
23"
24"
25"
26"
27"
28"
29"
30"
31"
32"
33"
34"
35"
36"
37"
38"
39"
40"
41"
42"
43"
44"
45"
46"
47"
48"
49"
17"
REFERENCES 1"
2"
1. Efthymiou, D., Antoniou, K. and Waddell, P. (2013). Factors Affecting the 3"
Adoption of Vehicle Sharing Systems by Young Drivers. Transport Policy, 29, pp. 4"
64-73. 5"
6"
2. Clavel, R., Marioto, M. and Enoch, M.P.(2009). Carsharing in France: Past, 7"
present and future. Proceedings of the 88th Annual Meeting of the Transportation 8"
Research Board, Paper No. 09-2007, Washington D.C., January 2009. 9"
10"
3. Heling (2009). User Characteristics and Responses to a Shared-use Station Car 11"
program: An Analysis of ZEV NET in Orange Country, CA. Proceedings of the 12"
88th Annual Meeting of the Transportation Research Board, Washington D.C., 13"
January 2009. 14"
15"
4. Shaheen, S., Cohen, A. and J. D. Roberts. Carsharing in North America: Market 16"
Growth, Current Developments and Future Potential. In Transportation Research 17"
Record: Journal of the Transportation Research Board, No. 1986, Transportation 18"
Research Board of the National Academies, Washington, D.C., 2006, pp. 116–124. 19"
20"
5. Stillwater, T., Mokhtarian P. L, Shaheen, S. A. (2008). CarSharing and the built 21"
environment: A GIS –based study of one U.S. Operator. Institute Of 22"
Transportation Studies, 2110, pp. 27-34, November 2008. 23"
24"
6. Burkhardt, J. and A. Millard-Ball. Who’s Attracted to Car-Sharing? In 25"
Transportation Research Record: Journal of the Transportation Research Board, 26"
No. 1986, Transportation Research Board of the National Academies, Washington, 27"
D.C., 2006, pp. 98–105. 28"
29"
7. Zhou, B., Kockelman K. and R. Gao. Opportunities for and Impacts of Carsharing: 30"
A Survey of the Austin, Texas Market. International Journal of Sustainable 31"
Transport 5 (3), 2011, pp. 135-152. 32"
33"
8. Transit Cooperative Research Program (2005). Carsharing: How and where it 34"
succeeds, Project B-26, Transit Cooperative Research Program, Federal Transit 35"
Administration, Washington D.C., July 2005. 36"
37"
9. Millard-Ball, A., G. Murray and J. ter Schure. Car-Sharing as a Parking 38"
Management Strategy. Proceedings of the 85th Annual Meeting of the 39"
Transportation Research Board, January 2006, Washington, D.C. 40"
41"
10. Elliot, M., Shaheen, S. and L. Jeffrey. Carsharing’s impact on household vehicle 42"
holdings: results from a North American shared-use vehicle survey. Institute of 43"
Transportation Studies, UC Davis, 2010. 44"
45"
11. Rodier, C. and S. Shaheen. Carsharing and Carfree Housing: Predicted Travel, 46"
Emission, and Economic Benefits. A case study of the Sacramento, California 47"
region. Proceedings of the 83th Annual Meeting of the Transportation Research 48"
Board, January 2004, Washington, D.C. 49"
18"
1"
12. Muheim Peter and Partner, Car Sharing Studies: An Investigation. Prepared for the 2"
Graham Lightfoot Study, Ireland, 1996, which cites Conrad Wagner, ATG- 3"
UMFRAGE 1990. ATG, Stans. Germany, 1990. 4"
5"
13. Rydén, C., Morin, E. Mobility Services for Urban Sustainability. Environmental 6"
Assessment. Report WP 6. Trivector Traffic AB. Stockholm, Sweden, January, 7"
2005. 8"
9"
14. Bureau of Transportation Statistics, National Household Transportation Survey 10"
2001, Highlights Report. 11"
12"
15. Bonsall, P. (2002). Car share and car clubs: Potential impacts, Department of 13"
Transport (Great Britain), Local Government and the Regions and The Motorists 14"
Forum, London, UK 15"
16"
16. Wardman, M. (1988): A Comparison of Revealed Preference and Stated 17"
Preference Models of travel Behavior, Journal Of Transport Economics And 18"
Policy, vol. 22, issue 1, pp. 71-91, January 1988. 19"
20"
17. Stasko, T. J. and Buck, A. B. and H. O. Gao (2013). Carsharing in a University 21"
setting: Impacts on vehicle ownership, parking demand, and mobility in Ithaca, 22"
NY. Transport Policy, Vol. 30, pp. 262-268 23"
24"
18. Nerenberg, V., Bernard, M.J. and Collins, N.E. (1999) Evaluation Results of San 25"
Francisco Bay Area Station-Car Demonstration. In Transportation Research 26"
Record: Journal of the Transportation Research Board, No. 1666, pp. 110-117, 27"
Transportation Research Board of the National Academies, Washington, D.C. 28"
29"
19. Ciari, F., Balmer, M. and Axhausen, K. W. (2009). Concepts for large scale car- 30"
sharing system: Modelling and evaluation with an agent-based approach. 31"
Proceedings of the 88th Annual Meeting of the Transportation Research Board, 32"
1888, Washington, D.C., January 2009. 33"
34"
20. Douma, F., Gaug, R., (2009). Carsharing in the Twin Cities: Measuring Impacts on 35"
Travel Behaviour and Automobile Ownership. Proceedings of the 88th Annual 36"
meeting of the Transport Research Board, 2140, Washington D.C., January 2009. 37"
38"
21. Shaheen, S. and E. Martin. Assessing Early Market Potential for Carsharing in 39"
China: A Case Study of Beijing. Proceedings of the 86th Annual Meeting of the 40"
Transportation Research Board, January 2007, Washington, D.C. 41"
42"
22. Likert, R. (1932). A technique for the measurement of attitudes. Arch. Psychol. 43"
(Frankf), 200 1932, pp. 140, 55. 44"
45"
23. Ben-Akiva, M., Lerman, S.R., 1985. Discrete choice analysis. MIT Press, 46"
Cambridge, MA. 47"
48"
24. Train K. (2009) ‘Discrete Choice Methods with Simulation’. Second Edition, 49"
Cambridge University Press, ISBN 0 521 74738 4. 50"
19"
1"
25. Scheiner, J. and Holz-Rau, C. (2007). Travel mode choice: affected by objective or 2"
subjective determinants?, Transportation 34: 487–511. 3"
4"
26. Tyrinopoulos, Y. and Antoniou, C. (2014). Analysis of passengers' perception of 5"
public transport quality and performance. International Journal of Operations 6"
Research and Information Systems (IJORIS) (in press) 7"
8"
27. Golob, T. F. (2003). Structural equation modeling for travel behavior research, 9"
Transportation Research Part B: Methodological 37(1): 1–25. 10"
11"
28. Walker, J. (2001). Extended Discrete Choice Models: Integrated Framework, 12"
Flexible Error Structures and Latent Variables, Doctoral dissertation, Department 13"
of Civil and Environmental Engineering, Massachusetts Institute of Technology. 14"
15"
29. Johansson, R. and D. Wichern, (1992). Multivariate Statistical Analysis, 3rd ed. 16"
Prentice- Hall, Englewood Cliffs, NJ. 17"
18"
30. Dillman, D., Smyth J. D. and M. Christian. Internet, Mail and Mixed-Mode 19"
Surveys: The Tailired Design Method. 3rd edition, John Wiley & Sons, Inc., New 20"
York, 2009. 21"
22"
31. Derebouka (2012). Investigation of the Perception of Greek Travelers on the New 23"
Transportation Modes, with Usage of Econometric Models. National technical 24"
University of Athens, Master Thesis 25"
26"
32. Richardson, A. J. (2003). Simulation study of estimation of individual specific 27"
values of time using adaptive stated-preference study. In Transportation Research 28"
Record: Journal of the Transportation Research Board, Transportation Research 29"
Board of the National Academies, No 1804, pp. 13–20, Washington, D.C., January 30"
2003. 31"
32"
33. Ben-Akiva, M. and Morikawa, T. (1990). Estimation of travel demand models 33"
from multiple data sources, in M. Koshi (ed.), Transportation and Trac Theory, 34"
Elsevier, New York, pp. 461–476 35"
36"
34. Louviere, J. J., Meyer, R. J., Bunch, D. S., Carson, R., Dellaert, B., Hanemann, W. 37"
M., Hensher, D. and Irwin, J. (1999). Combining sources of preference data for 38"
modeling complex decision processes, Marketing Letters 10(3): 205–217. 39"
40"
35. Bierlaire, M. (2003). BIOGEME: A free package for the estimation of discrete 41"
choice models, Proceedings of the 3rd Swiss Transportation Research Conference, 42"
Ascona, Switzerland. 43"
44"
20"
1"
... That is, the framework of the HCM has been developed to enrich the behavioral realism of the DCM by accounting for latent factors such as perceptions and attitudes, and employing more flexible error structures. The framework of the HCM has been applied in various transportation contexts, such as mode choice (Johanson et al., 2006;Polydoropoulou et al., 2013;Abou-Zeid & Ben-Akiva, 2010), vehicle purchasing (Bolduc et al., 2008), and route choice (Efthimiou and Antoniou, 2014;Tsirimpa et al., 2007). ...
Thesis
Full-text available
Substantial changes in lifestyles, urban environments and transportation systems have led to changed physical activity patterns, especially among underage people. Although we may know the demographic and economic characteristics of underage students’ families and the communities where they live and attend school, we have little scientific evidence of the individual teenagers’ (12 to 18 years old) activities, travel behavior and attitudes. In traditional societies there was comparatively little discrepancy between adolescents and adults because they grew up in comparable worlds. However, with rapid change and social media bringing the outside world into teenagers’ lives, larger generational differences are emerging. This thesis contributes to the understanding of various factors that affect teenagers’ travel behavior. As a first step, we analyze teenagers’ activity patterns and time use in school days and in Saturday and the transport mode that they use, in order to identify their travel needs. The results indicate that teenagers conduct a number of trips, especially after-school trips, without the supervision of their parents, while the mode use patterns significantly differ among the trip purposes and among distinct geographical areas. As current teenagers spend significant amount of time on online social networking (OSN), we further analyze how much, why, and how teenagers utilize social media, and how its usage affects their travel behavior. Latent Class Poisson Regression models are developed in order to identify teenagers’ trip making behavior for social purposes of various OSN usage styles, while the results indicate that those who use OSN in a rational or addictive way, conduct more social trips than those who are indifferent to OSN, thus OSN does not substitute face-to-face communication. The developed framework offers significant insights to researchers for the data required in order to model the relationship between OSN and trip making behavior. The thesis is also concerned with investigating the effect of social influence on decision making and more specific the effect of parents’ walking patterns on teenagers’ attitudes towards walking and mode choice behavior. We present a methodological framework that incorporates social interaction effect into Hybrid Choice Models (HCM) and provide the required mathematical equations. The model estimation results indicate that, if the teenagers perceive that their parents are walking-lovers, then this increases their probability of loving walking too. Even though the application focus on teenagers, the framework is general and can be applied to modeling adults’ behavior as well. Moreover, this thesis contributes to the understanding of how teenagers perceive various built-environment characteristics and which of them work as constraints to active transport. We use their perceptions of built-environment characteristics, the actual built-environment characteristics of their routes from home to school and weather conditions in order to capture their effect on mode-to-school choice behavior. A latent variable model is developed for each urban environment to further investigate the differences among urban, rural and insular areas. The results show that the presence of wide pavements, greenery and traffic lights at major intersections affects positively the choice of active transport to school, while rain and bad weather conditions affect negative the choice of active transport. The most significant walkability constraint for urban teenagers seems to be the safety issue, while for rural and insular ones it is the absence of sidewalks, along with poor lighting. For the analyses and model estimation we use data that are collected directly from teenagers. The survey took place in two countries; Greece and Cyprus, while in Greece the survey took place not only in urban, but rural and insular areas as well. The sample from Greece consists of 3,293 adolescent students, while the sample from Cyprus consists of 10,093 adolescent students, covering the 21% of the total high-school population of the country. The contributions and innovation of this research cover several topics. First of all, to our knowledge it is the first time that such a large-scale survey on travel behavior, focusing only on teenagers, has taken place. Second, the questionnaire used for the data collection was designed specifically to investigate teenagers’ perceptions of travel behavior, not only by transport engineers but also by psychologists and economists, with the aim of approaching the multidimensional nature of transportation problems in depth. Third, this thesis contributes to the modeling of social influence effect on the decision making process by proposing an extension to HCM. Fourth, the Latent Class models that are developed contribute to the understanding of the relationship between OSN and trip making behavior. The findings of this thesis offer guidelines as to the types of transport policies that could promote active transport and increase environmental consciousness. Finally, the interventions at this age could develop the desired behaviors that could be retained in adulthood.
... Environmental awareness is also an important determinant, confirming the results obtained both studying the actual demand (Burkhardt and Millard-Ball, 2006;Loose, 2010;Costain et al., 2012;Kim et al., 2015;Lang, 2015) and on the potential demand (Zheng et al., 2009;Efthymiou and Antoniou, 2014). On these bases it is possible to foresee that initiatives aimed at promoting and describing the lower environmental impact of CS in the FVG region (fairs, conferences, info-point with central nodes of the transport network such as train stations or airports), including real test via fleets temporarily available at urban traffic generators and attractors (universities, hospitals, shopping centers, stations, airports), would increase the willingness to use this mode of transport as an alternative or as a complement to a private car. ...
Article
The recent strong growth of carsharing (CS) has taken place worldwide mainly in the large cities. We discuss whether there is a potential role for CS in medium to small-sized towns and in less-densely populated rural areas. In large cities CS is provided by large profit oriented firms, capable of offering efficient and flexible services, using the most advanced technologies with highly differentiated and customized prices. In lower density areas and small-to-medium size cities the challenges are much greater. The lower demand resulting from a complex of unfavorable factors (higher car ownership level, larger parking availability, low public transport service, fewer or absent taxi service and so on) requires a different business model for the provision of CS: more socially oriented, with a greater involvement of local municipalities and public transport operators to offer a service, most likely at favorable prices. The paper reports the results drawn from a survey on the potential demand for CS conducted in the Friuli Venezia Giulia region, Italy, a region characterized by small-sized towns and less-densely populated rural areas where CS is not currently offered. It is found that there is a not negligible demand (3.7% of the population holding a driving license) and that the likely users would be the ones that are better informed about CS, environmentally conscious and young. They would be more frequent among students or unemployed people, rather than among professionals, as in the more urban environments. It is also found that CS would mostly be used for non-commuting and longer trips.
Article
Full-text available
In this paper, socio-demographic factors influencing the demand for different car-sharing forms are examined. An in-depth review of such factors is provided based on the type of shared car service, geographic area, and specific travel demand aspect being considered. Conclusions highlight the differences between car sharing operational schemes. The number of males, young individuals, and people with above-average income among free-floating members is higher than in other car-sharing services. Also, although round-trip car-sharing users appear to be less educated than other car-sharing services users, car-sharing members may follow a more efficient and sustainable lifestyle than the one-way shared car system members. Besides, some suggestions are recommended for future studies. A research gap has been identified regarding the direction of causation between vehicle ownership levels and car-sharing demand. Most studies have worked on the impact of vehicle ownership on car-sharing or simply noted a correlation between the two. However, clarifying any reverse effect would help in better assessing the sustainability of car-sharing services. This overview can guide policymakers, planners, and other stakeholders to enhance the car-sharing program's effectiveness and opt for the best kind of service according to their goals.
Article
Full-text available
Carsharing involves the communal ownership and use of a fleet of vehicles, typically on an hourly basis (Millard-Ball et al. 2005). Austin CarShare (ACS) was launched in the fall of 2006, making Austin the first city in Texas with carsharing services. While many studies have discussed the positive impacts of carsharing, few have examined widespread public opinion of carsharing. This study undertook the challenge of investigating traveler preferences during ACS's service launch, in order to anticipate latent demand for such services. The survey provides rich information on public opinion of different aspects of the ACS program, as well as the expected demand on the service and possible changes in travel patterns. Supplementing the survey results with spatial data, membership models of two pricing plans reveal that households with higher vehicle ownership and income-to-adults ratios are less likely to join the program, while level of education exhibits a convex relationship with the probability of joining the Freedom Plan, ceteris paribus. Although potential carsharing users share similar characteristics, the two plans serve slightly different customer sets and have the potential to supplement one another.
Book
Full-text available
This book describes the new generation of discrete choice methods, focusing on the many advances that are made possible by simulation. Researchers use these statistical methods to examine the choices that consumers, households, firms, and other agents make. Each of the major models is covered: logit, generalized extreme value, or GEV (including nested and cross-nested logits), probit, and mixed logit, plus a variety of specifications that build on these basics. Simulation-assisted estimation procedures are investigated and compared, including maximum simulated likelihood, method of simulated moments, and method of simulated scores. Procedures for drawing from densities are described, including variance reduction techniques such as anithetics and Halton draws. Recent advances in Bayesian procedures are explored, including the use of the Metropolis-Hastings algorithm and its variant Gibbs sampling. No other book incorporates all these fields, which have arisen in the past 20 years. The procedures are applicable in many fields, including energy, transportation, environmental studies, health, labor, and marketing.
Article
Full-text available
We review current state-of-the-art practices for combining preference data from multiple sources and discuss future research possibilities. A central theme is that any one data source (e.g., a scanner panel source) is often insufficient to support tests of complex theories of choice and decision making. Hence, analysts may need to embrace a wider variety of data types and measurement tools than traditionally have been considered in applied decision making and choice research. We discuss the viability of preference-stationarity assumptions usually made when pooling data, as well as random-utility theory-based approaches for combining data sources. We also discuss types of models and data sources likely to be required to make inferences about and estimate models that describe choice dynamics. The latter discussion is speculative insofar as the body of literature on this topic is small.
Article
This article describes the statistical models commonly used to study discrete choices. It concentrates on the 'basic' discrete choice models, and the theoretical choice situations that lead to these models. Specifically the choice situation addressed include: the ordered choice situation and the unordered choice situation. In addition, the article discusses two extensions of the basic discrete choice models commonly seen in political science research - models allowing for heteroskedasticity in the choices made across political agents (such as the heteroskedastic probit), and models that estimate substitution patterns across choice alternatives (such as the multinomial probit and mixed logit). Suggestions for further reading are also given.
Article
Carsharing is growing rapidly in popularity, often backed by government and private partners, such as universities and developers. While reduced parking demand is frequently cited as a reason to promote carsharing, virtually no quantitative analysis has been done on the impact of carsharing on parking demand. Instead, prior studies focus on vehicle ownership, which has an implied connection to parking demand. This paper analyzes the impact of carsharing on parking demand in a university setting (with just over 1000 carsharing users) using a member survey and parking permit sales data. Changes in parking demand are broken down by geographic area and parking type. Members report the vast majority (over 76%) of forgone vehicles would be parked in the same area as the member's household on most weekdays, nights, and weekends. Roughly 30% would be parked on the street at most times, with the percentage parked in personal driveways and garages peaking at roughly 40% on nights and weekends and dropping to 26% on weekdays. Members reported an increase in shopping trips made by car or truck (statistically significant at 1% level), leading to a small increase in parking demand at stores, but this increase was much smaller than the reductions seen elsewhere. The paper also assesses other impacts which have so far been nearly exclusively measured in relatively large cities. For example, the survey revealed a reduction of 15.3 personal vehicles for every carsharing vehicle, roughly equivalent to findings from major cities.
Article
Transportation patterns in big cities are redefined by the growing trend of car ownership and usage costs (e.g. the initial cost of buying a car, the constantly growing fuel prices, additional maintenance and insurance costs and the increased parking demand and time of travel). Under these circumstances, the demand for alternative vehicle-sharing transportation modes, such as carsharing or bikesharing, increases. Companies and authorities planning to develop such schemes need to know the factors driving their adoption, so that they can optimally position these services in a cost-effective way that will maximize their use. In this paper, the results of an on-line survey that was conducted in Greece, a country where carsharing is effectively non-existent and bikesharing is just emerging, are presented and analyzed. Given the nature of the survey and the fact that younger people are the most likely target audience for these services, the analysis focuses on data from the age group 18–35 years old.