Content uploaded by Oded Cats
Author content
All content in this area was uploaded by Oded Cats on Feb 22, 2019
Content may be subject to copyright.
The Impact of Automated Vehicles on Travel Mode Preference for Different Trip
1
Purposes and Distances
2
3
To be cited as:
Ashkrof P., Correia G., Cats O. and van Arem B. (2019). The Impact of Automated Vehicles on Travel
Mode Preference for Different Trip Purposes and Distances. Transportation Research Record, in press.
4
5
Peyman Ashkrof, Corresponding author
6
Department of Transport and Planning
7
Delft University of Technology
8
P.O. Box 5048, 2600 GA Delft, The Netherlands
9
Phone number: +31 15 27 85279; Email: P.ashkrof@tudelft.nl
10
11
Gonçalo Homem de Almeida Correia
12
Department of Transport and Planning
13
Delft University of Technology
14
P.O. Box 5048, 2600 GA Delft, The Netherlands
15
Phone number: +31 15 27 813 84; Email: G.correia@tudelft.nl
16
17
Oded Cats
18
Department of Transport and Planning
19
Delft University of Technology
20
P.O. Box 5048, 2600 GA Delft, The Netherlands
21
Phone number: +31 15 27 81384; Email: O.cats@tudelft.nl
22
23
Bart van Arem
24
Department of Transport and Planning
25
Delft University of Technology
26
P.O. Box 5048, 2600 GA Delft, The Netherlands
27
Phone number: +31 15 27 86342; Email: B.vanarem@tudelft.nl
28
29
30
31
32
33
34
35
36
37
38
Ashkrof, Correia, Cats, van Arem 1
ABSTRACT
1
Due to the technology penetration in the transportation system, the Automated Vehicle (AV) is
2
foreseen to be available as a mode of transport in the future. Given the major potential
3
implications of AVs, the investigation of the impact of these vehicles on travel behavior is vital
4
for a wide range of purposes, especially for policymaking. In this study, we report the results of a
5
stated preference survey distributed in the Netherlands where respondents had to choose between
6
conventional cars, public transportation, and AVs for different travel distances and trip purposes.
7
Having collected the information from 663 respondents we conducted an integrated study
8
incorporating classic trip attributes (such as travel time and travel costs), attitudinal factors, and
9
socioeconomic variables to understand people’s choices. We study a particular form of AVs
10
denoted as Automated Driving Transport Service (ADTS) which we define as a door-to-door
11
transport service provided by a vehicle similar to a conventional car albeit driverless, controlled
12
automatically. Results suggest that travelers’ mode preference varies significantly for different
13
travel distances and purposes. We found that conventional cars and public transportation are
14
perceived as being the least attractive alternatives in terms of in-vehicle travel time in short
15
distance commuting trips and long distance commuting trips, respectively. Preferences towards
16
ADTS lie in between, neither the best nor the worst alternative in all scenarios. Our findings
17
suggest that ADTS adopters are likely to prefer this mode for long distance leisure trips rather
18
than short distance commuting trips.
19
20
21
22
23
Keywords: Automated vehicles, discrete choice model, travel mode preference, travel behavior,
24
trip purpose, trip distance
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
Ashkrof, Correia, Cats, van Arem 2
1
1. INTRODUCTION
2
Advanced technologies have revolutionized many aspects of humans’ life in various sectors
3
including the transportation system. In fact, the transportation system is one of the essential areas
4
which can be tangibly exposed to individuals, that has been significantly affected by
5
digitalization and technology development. Technological advancement in transportation, in
6
particular automation, aims at making trips safer, faster, sustainable, and more efficient.
7
In recent years, the concept of automated driving has been introduced as an outstanding
8
platform for the next generation of the driving system which is expected to improve safety,
9
traffic flow efficiency, capacity, accessibility and reduce congestion through the application of
10
some new technologies such as Vehicle to Vehicle and Vehicle to Infrastructure
11
communications. Using cameras, sensors, Global Positioning System, Adaptive Cruise Control,
12
Light Detection and Ranging, Advanced Driver Assistance System, Automated Vehicles(AVs)
13
can steer the vehicle and drive it automatically while passengers delegate the control to a
14
computer. Ultimately, by replacing the driver role with an automated driving system, automated
15
vehicles are able to totally free up the passengers under level 4 and 5 automation levels. (1). In
16
other words, AV users can mostly behave like passengers inside the vehicle which implies that
17
they will be able to multitask and be productive by allocating the travel time to do other activities
18
such as sleeping, working, reading, eating, drinking, watching movies, monitoring the
19
environment, etc. Thus, the mentioned capabilities might be the strong motivations to shift from
20
conventional (humanly-drive) cars to automated vehicles (2, 3).
21
There is a growing body of literature that recognizes the importance of travel behavior
22
studies that allow positioning AVs in the future mobility market, to investigate how travelers
23
respond to this technologically advanced mode of transport. AVs can potentially change the
24
market composition and travelers’ mode choice (2, 4–11). Given that the travel mode choice and
25
the built environment are interlinked (12), AV adoption can also influence the land use and
26
urbanization patterns. Therefore, a clear understanding of the potential travel behavior changes
27
might assist policymakers in managing the positive and negative implications of AVs.
28
Early studies focused on the classic alternative-specific attributes including travel time
29
and travel costs for capturing travelers’ preference amongst different AVs forms including
30
Private Automated Vehicle (PAV) and Shared Automated Vehicle (SAV), when compared to
31
other modes (5–8). Arentze et al. (13) believe that although travel time and travel costs are
32
initially assessed by travelers when establishing a trade-off between the alternatives, travel
33
distance can play an important role in their final choice. Whereas most of the studies have only
34
considered short distance trips (4, 5), automated vehicles might be more attractive in long
35
distance trips because users can utilize their travel time more efficiently and be more productive
36
in longer trips (8). Moreover, travel purpose is another influential factor which may affect mode
37
preference and travel behavior (14, 15). In addition, surveys such as those conducted by Yap et
38
al. (4), Zmud et al. (8), and Haboucha et al. (11) have shown that psychological factors and
39
attitudes might have a significant impact on whether an individual is inclined to choose AVs as a
40
mode of transport. However, previous research has established that substantial uncertainties exist
41
about the AV adoption as well as the possible effect of automation on the travel behavior (2, 11).
42
A large body of travel behavior research asserts that travelers value different travel time
43
components including in-vehicle time, walking time, waiting/searching time, and also travel
44
costs constituents such as ticket cost, fuel price and parking cost. However, insofar only a few
45
studies have integrated all the mentioned attributes in a survey related to AV (4, 5, 11).
46
Ashkrof, Correia, Cats, van Arem 3
To the best of our knowledge, this is the first study that attempts to simultaneously examine the
1
impact of travel distance (long distance and short distance trips) and travel purpose (commuting
2
and leisure trips) on travelers’ mode preference towards Automated Driving Transport Service
3
(ADTS). ADTS is a drivererless vehicle similar to a conventional car albeit controlled
4
automatically by a computer or a remote operator as a door-to-door transport service, in
5
competition with conventional cars and public transportation. This is a centrally controlled fleet
6
of driverless vehicles which transport travelers on demand. This definition is presented so that
7
respondents can easily imagine and relate to the ADTS thus aiming at minimizing the risk of
8
capturing confused perceptions of how these vehicles look like. Additionally, classic attributes
9
including travel time components and travel costs as well as attitudinal factors and the social
10
demographic characteristics of the respondents are included in order to comprehensivly identify
11
choice determinants.
12
Since ADTS has not been introduced to the market yet, a Stated Preference (SP)
13
experiment has been designed in order to present this hypothetical alternative to respondents.
14
The next section presents the details of the study specifications, survey design, and model
15
formulation followed by the results of the models and conclusion.
16
17
2. METHODOLOGY
18
2.1. Study specifications
19
Inspired by the work of Azari et al. (14) and Shiftan et al. (15) who suggest that selecting the
20
mode of transport and a place to park is dependent on the trip purpose, we incorporate the trip
21
purpose in the study as a context variable to explore its effect on travelers’ mode choice. To
22
minimize the number of choice sets, we consider commuting and leisure trips which are the
23
majority of the trip purposes in the Netherlands (16).
24
As mentioned, travel distance might also affect the travelers’ mode choice (4, 13, 17).
25
Hence, various scenarios are considered based on the trip distance to investigate its influence on
26
travel behavior. In order to decrease the complexity of the choice sets, we categorize the trips
27
into two groups, short distance of 10 kilometers and long distance trips of 40 kilometers long.
28
The modal alternatives, attributes, and attribute levels are specified accordingly.
29
Conventional car, public transportation (i.e., bus, tram, and train), and Automated
30
Driving Transport Service (ADTS) as a door-to-door service provided by an AV are presented as
31
the modal alternatives. Statistics show that 93 percent of trips are done by car, train, and bicycle
32
while travelers mostly use bicycle for trips up to 5 km in the Dutch context (16). Consequently,
33
due to the travel distance assumptions which are 10 and 40 km in this study, the bicycle is not
34
included in the choice sets as an alternative. The access and egress legs of public transport trips
35
are assumed to be made by walking to prevent an increase in the number of alternatives and also
36
attributes.
37
The automated driving system could be introduced in different forms such as Private Automated
38
Vehicle (PAV) and Shared Automated Vehicle (SAV). Through the literature we see that PAV
39
would increase the distance traveled which is opposing sustainability policies despite the
40
probability that AV prices might be much higher than conventional cars. Thus policymakers
41
might not be enthusiastic to support this type of motorized alternative widely. Moreover, SAV
42
may not be popular amongst travelers due to the lack of privacy caused by sharing the system.
43
We, therefore, decided to present another form of the automated vehicle, which is the Automated
44
Driving Transport Service. One of the advantages of ADTS adoption might be the possibility of
45
shifting mobility from an asset like a private car to an on-demand service. In fact, ADTS can be
46
easily and impartially presented to the respondents who have not experienced an AV.
47
Ashkrof, Correia, Cats, van Arem 4
Acknowledging the fact that the survey respondents have not experienced automated vehicles, a
1
Stated Preference (SP) experiment is selected as the data collection stryategy. It is desirable that
2
attributes and attribute levels in the SP experiment are based on real circumstances to which
3
respondents can easily relate. Travel time is decomposed into in-vehicle, walking, waiting, and
4
searching time for parking for the car to explore the difference between time spent inside and
5
outside of the vehicle. Attribute levels are pivoted on the statistical data for the mobility
6
characteristics of travelers in the Netherlands (16) as well as data obtained by some trip planners,
7
for the hypothesized trips per distance category. The attribute levels of ADTS are specified
8
identically to those of conventional car, except for walking time which is set to null. This is
9
because ADTS is defined as a door-to-door transport service, therefore, walking time for this
10
alternative is approximately perceived to be zero while car drivers should park their cars in
11
parking spaces and then walk to the final destination.
12
The instruction was written with care to ensure that the respondents can imagine the
13
conditions as much as possible given the nature of an SP experiment. Table 1 provides an
14
overview of all attributes and the corresponding attribute levels per trip distance category.
15
16
TABLE 1 Overview of Attributes of the Choice Alternatives Used in the SP Experiment
17
Alternatives
Attributes
Conventional Car
Public Transportation (PT)
ADTS
Short Distance
Long Distance
Short Distance
Long Distance
Short Distance
Long Distance
In-vehicle travel
time
15, 20, 25
50, 55, 60
20, 25, 30
55,60,65
15, 20, 25
50, 55, 60
Waiting time /
Searching time for
parking
0, 5, 10
0, 5, 10
5, 10, 15
10, 15, 20
0, 5, 10
0, 5, 10
Walking time
0, 4, 8
0, 4, 8
4, 8, 12
8, 12, 16
NA
NA
Travel costs
4, 6, 8
12, 14, 16
2, 3, 4
8, 10, 12
4, 6, 8
12, 14, 16
18
Besides classic attributes, travelers’ preference might be affected by attitudinal factors
19
such as ‘trust in AVs’, ‘concern about the environment’, ‘AV functionality’, etc. (4, 11). Since
20
attitudes as latent variables cannot be directly observed, 22 indicators are created and adapted
21
from other studies to capture the level of agreement of the respondents by using a five-point
22
Likert scale. Table 2 presents the statements as well as the source of each of them.
23
24
2.2. Survey design
25
An online SP survey designed was distributed among a sample of the Dutch population by a
26
survey panel provider in the Netherlands. The data of 663 respondents was used after checking
27
the screening questions and a minimum time of completion. Several questions were embedded at
28
the beginning of the survey to filter respondents who do not fill the requirements: older than 18
29
years old, own a driver license, and use car at least once a month. As an additional screening
30
measure, several contradicting statements were introduced in order to examine whether
31
respondents paid attention to the survey or not. For example, subjects that had given the same
32
answer to the contradicting questions, for instance, “I enjoy driving” and” I do not enjoy
33
Ashkrof, Correia, Cats, van Arem 5
driving”, have been screened out. Finally, if the survey was completed within less than 7:30 min,
1
the corresponding questionnaire was not incorporated into the dataset.
2
The questionnaire consists of four parts: (i) transport-related attitudes which included the
3
mentioned indicators; (ii) current individual mobility behavior, for instance, the frequency of
4
using existing modes and driving short and long distance trips; (iii) 6 choice sets per respondent,
5
and; (iv) socio-economic questions at the individual and household levels such as gender, age,
6
education level, household size.
7
8
TABLE 2 The List of Indicators Used in the EFA.
9
No.
Statement
Source
1
I believe an automated vehicle would drive on populated streets better than conventional
cars.
(18)
2
I would be comfortable entrusting the safety of a close family member on an automated
car.
(18)
3
I am concerned that an automated vehicle is not reliable enough to make a safe takeover.
(19)
4
Using an automated vehicle could make me nervous at night time.
Created for the present
study
5
I would rather take manual control of the automated vehicle on some occasions.
(18)
6
I believe that an automated vehicle would have a better fuel efficiency than regular ones.
(18)
7
My productivity will be increased by using an automated vehicle
(18)
8
I am afraid that there will be no car available when I request and I will have to wait for a
while.
(19)
9
I am concerned to share my location and phone number for taking an automated vehicle.
Created for the present
study
10
It is more fun to take an automated vehicle compared to a conventional car.
(19)
11
I do what I can to contribute to reducing global climate changes, even if it costs more and
takes time.
(19)
12
I rarely worry about the effects of pollution on myself and my family.
(20)
13
It is important for me to follow technological development.
(19)
14
New technologies create more problems than they solve.
(19)
15
I often purchase new technology products, even though they are expensive.
(19)
16
I believe that an automated vehicle might produce fewer pollutant emissions.
(18)
17
I like the sound and power of a conventional car engine.
(19)
18
I prefer to drive myself rather than others driving me.
(21)
19
I enjoy driving.
(22)
20
It makes me uncomfortable to ride on public transit with strangers.
(23)
21
I believe that people use public transit when they don’t have any other choice. (last
priority)
(23)
22
I feel safe taking public transit.
(24)
10
The choice-sets presented in section (iii) of the survey are orthogonally designed using
11
the software package NGENE (25) with the aim at decreasing the task effort from the
12
respondents. The orthogonal design is selected due to the lack of in-depth and knowledge about
13
the AVs acceptance to determine priors for an eventual D-efficient design and also the fact that
14
orthogonal design is the most widely used design (26).
15
In total, 54 treatment combinations are arranged in 9 different blocks of 6 choice sets,
16
which are then randomly distributed amongst the respondents. The choice sets consist of 3
17
scenarios for short distance and 3 for long distance trips. Additionally, each respondent is faced
18
Ashkrof, Correia, Cats, van Arem 6
with selecting two choices, one for commuting and one for leisure trips per choice task. To
1
illustrate the set-up, an example of a choice set is depicted in Figure 1.
2
Regarding the order of the different parts of the survey, we preferred to embed the choice
3
sets in the third section since the first and second parts could shed light on the automated driving
4
so that the subjects are already conscious of the prospects of automated driving prior to making
5
decisions on their mode choice.
6
7
8
FIGURE 1 Example of a Choice Set for a Short Distance Trip.
9
10
2.3. Discrete Choice Models
11
Discrete Choice Models (DCM), in particular, Random Utility Maximization (RUM) framework
12
is one of the most applicable approaches to study the travelers’ preference towards transport
13
mode choice. It hypothesizes that alternative is chosen by individual when the associated
14
utility is the highest compared to the other options (27, 28). The utility functions for all
15
alternatives are defined using Eq. 1.
16
17
(1)
18
19
Where the first component of the utility function corresponds to the classic alternative-
20
specific attributes including in-vehicle travel time, waiting/searching time, walking time, and
21
travel costs depicted in table 1. is a vector of coefficients that indicate the importance of the
22
Ashkrof, Correia, Cats, van Arem 7
exploratory variables . Alternative specific coefficients are estimated for each travel time
1
component while the travel costs coefficient is assumed generic.
2
The second component is associated with the factors that are intrinsic to the individuals
3
including attitudes as well as socio-economic variables. Previous studies found that individual-
4
related variables are also vital determinants of mode choice. is a vector of coefficients that
5
implies the importance of various socio-economic variables and attitudinal factors which are
6
derived from an Exploratory Factor Analysis (EFA).
7
The EFA is applied to reduce the number of variables and classify them under the
8
minimum required number of components to explain most of the variance. Based on the parallel
9
analysis run in SPSS the number of components is determined and then the factor scores are
10
incorporated in the discrete choice model. This type of model is also known as a hybrid choice
11
model because it estimates the Integrated Choice and Latent Variable (ICLV) either in a
12
sequential or simultaneous way. It should be noted that sequential estimation postulates that the
13
latent variables are error-free. While a joint estimation is able to reflect the complexity of the
14
relationship between psychological factors and socio-economic variables, it requires a substantial
15
increase in estimation time and a more complex modeling. We, therefore, opted for the
16
sequential method as it was found sufficient to investigate whether travelers’ attitudes influence
17
the choices (4, 11).
18
The last component in the utility function in Eq. 1, the error term, represents unexplained
19
variation. The assumption regarding the distribution of the error term results in different model
20
specifications. With restricting all covariances to be zero, the simplest logit model is the
21
Multinomial Logit (MNL) model (27, 28). It assumes that the random variables are
22
independently and identically distributed (IID) following Extreme Value type 1. It hence
23
neglects the taste heterogeneity amongst individuals as well as the correlation between choices
24
made by the same individual across time. In order to overcome these restrictions, more
25
complicated with higher degrees of freedom models including nested logits (NL), cross-nested
26
logit (CNL), mixed logit (ML), and hybrid mixed logit models have been estimated in this study.
27
The hybrid mixed logit model with panel effects resulted in the best model fit. Following an
28
iterative process, two final models are reported for long distance and short distance trips. The
29
sample size and attributes are the same for both of them, therefore, enabling direct comparison.
30
In order to gain a deeper insight into the impact of trip purpose, we checked whether the trip
31
purpose interacts with in-vehicle travel time. That is why we incorporated the coefficients of the
32
interactions between alternative-specific parameters of in-vehicle travel time and trip purpose
33
which are commuting and leisure trips in the utility function.
34
The software package PythonBiogeme (29) is used to perform the maximum likelihood
35
estimation of the discrete choice model, whereas the exploratory factor analysis was conducted
36
in SPSS. In the next section, the results of the models are presented and discussed.
37
38
3. RESULTS
39
3.1. Exploratory Factor Analysis (EFA)
40
The software SPSS was used for conducting the EFA in order to reduce the number of variables
41
by categorizing highly correlated statements. Various methods including eigenvalue greater than
42
1 (K1), scree plot, and parallel analysis were applied to determine the minimum number of
43
required components. Although K1 which is the most widely used method is the default setting
44
of many software packages, the Parallel Analysis (PA) is one of the most strongly recommended
45
techniques and has shown greater accuracy in results (30–33). This process resulted with the
46
extraction of five components, bundling in total 16 out of 22 indicators by PA with the following
47
Ashkrof, Correia, Cats, van Arem 8
specifications. Direct Oblimen rotation method was initially tested to investigate any strong
1
correlations between the components. However, the results indicated that an orthogonal design
2
could be applied and thus the Varimax method was utilized. Indicators with communality > 0.4
3
as well as factor load > 0.50 are included in the final EFA.
4
Table 3 presents the latent variables, associated indicators, and the respective factor
5
loads. The latent attitudinal factors can be denominated as trust in AVs, public transport interest,
6
driving interest, positive viewpoints towards AV efficiency, and environmentally friendly
7
attitudes. The corresponding factor scores are incorporated in the utility function detailed in the
8
next section.
9
10
TABLE 3 Results of the Exploratory Factor Analysis
11
1
2
3
4
5
Using an automated vehicle could make me nervous at night time.
0.802
I am concerned that an automated vehicle is not reliable enough to make a safe
takeover.
0.764
I believe that a conventional car would drive on populated streets better than
automated vehicles.
0.713
I would be comfortable entrusting the safety of a close family member on an
automated car.
-0.681
I would rather take manual control of the automated vehicle in some occasions.
0.574
It is more fun to take an automated vehicle compared to a conventional car.
-0.536
It makes me uncomfortable to ride on public transit with strangers.
0.789
I feel safe taking public transit.
-0.734
I believe that people use public transit when they don’t have any other choice. (last
priority)
0.684
I enjoy driving.
0.810
I like the sound and power of a conventional car engine.
0.723
I prefer to drive myself rather than others driving me.
0.568
I believe that an automated vehicle would have a better fuel efficiency than regular
ones.
.857
I believe that an automated vehicle might produce fewer pollutant emissions.
.765
I do what I can to contribute to reduce global climate changes, even if it costs more
and takes time.
0.819
I rarely worry about the effects of pollution on myself and my family.
-0.751
12
3.2. DCM Model Estimation Results
13
The results of the final two models are summarized in table 4 including the model fit,
14
coefficients estimates, standard deviation of random variables, and the level of significance (t-
15
value). Overall, 36 and 33 variables were estimated for long distance and short distance trips
16
models, respectively, by the hybrid mixed logit model with panel effects and 2000 Halton draws.
17
In the table, ASC represents the alternative specific coefficients which translate the mean
18
unobserved preference towards the associated alternatives. B and SIGMA indicate the estimated
19
mean and the standard deviation of each random parameter, respectively. Finally, the suffix
20
CAR, PT and ADTS indicates in the utility function of which alternative the variable has been
21
placed.
22
Results show that the signs and values of all estimated coefficients are plausible. As
23
expected, travel time components and travel costs lead to disutility with the associated
24
coefficients’ having negative signs in both models. In-vehicle travel time denoted as “INV” in
25
the table, which is defined by the time spent inside the vehicle is hypothesized to interact with
26
the travel purpose. In general, the marginal value of in-vehicle travel time is less negative for
27
Ashkrof, Correia, Cats, van Arem 9
leisure trips than commuting trips for the same travel mode regardless of the travel distance. This
1
is arguably because leisure trips are more pleasant for travelers. With the same trip purpose and
2
travel mode, in-vehicle travel time is valued more negatively in short distance than long distance
3
travels. It shows that the disutility of in-vehicle travel time might be generally lower when
4
traveling further away, which is in line with Zmud et al. (8) findings.
5
6
TABLE 4 Results of Exploratory Factor Analysis
7
Long Distance Trips
Short Distance Trips
Coefficient
Estimated value
t-value
Estimated value
t-value
ASC_AV
-0.507
-0.39
-0.751
-0.91
ASC_PT
-2.27
-1.17
-0.729
-0.73
B_INVADTSCOM
-0.0876
-5.66
-0.139
-6.7
B_INVADTSLEI
-0.0599
-3.55
-0.13
-5.76
B_INVCARCOM
-0.087
-5.62
-0.149
-8.33
B_INVCARLEI
-0.0496
-2.98
-0.108
-6.18
B_INVPTCOM
-0.0948
-4.06
-0.129
-6.31
B_INVPTLEI
-0.0495
-2.13
-0.0939
-4.87
B_WAITADTS
-0.0944
-6.37
-0.128
-7.99
B_SEARCHPARKCAR
-0.0969
-6.53
-0.099
-6.86
B_WAITPT
-0.0931
-4.05
-0.0609
-3.51
B_WALKCAR
-0.0751
-4.05
-0.0917
-4.64
B_WALKPT
-0.13
-2.29
-0.175
-4.4
B_TC
-0.351
-10.25
-0.328
-8.8
B_TRUST_ADTS
0.969
8.23
0.881
7.99
B_AVEFF_ADTS
0.593
4.96
0.495
3.09
B_CARINT_ADTS
-0.335
-2.63
-0.419
-3.77
B_ENVFRIEND_ADTS
0.148
1.43
0.294
2.31
B_CARUSER_CAR
0.212
1.97
0.253
2.32
B_PTUSER_PT
0.495
5.92
0.159
3.06
B_TAXIUSER_CAR
-0.27
-2.38
-0.246
-1.88
B_LDPASSANGERINTEREST_CAR
-0.737
-3.43
-
-
B_GENDER_MALE_CAR
-0.519
-2.42
-
-
B_AGE_ADTS_PT
-0.553
-2.09
-0.423
-2.18
B_EMPLOYED_ADTS
0.939
4.01
0.665
2.21
B_CARHH_CAR
0.0195
0.11
-
-
SIGMA_ADTS
0.0607
0.16
1.98
1.79
SIGMA_CAR
2.19
9.27
-0.361
-1.42
SIGMA_INVAVCOM
-0.012
-2.42
-0.103
-4.61
SIGMA_INVAVLEI
0.0362
5.07
-0.0852
-4.05
SIGMA_INVCARCOM
-0.0229
-5.05
0.0791
6.42
SIGMA_INVCARLEI
0.0352
7.21
0.141
6.94
SIGMA_INVPTCOM
0.0243
3.9
0.0733
7.01
SIGMA_INVPTLEI
0.0341
4.88
0.081
6.96
SIGMA_PT
3.53
7.73
2.09
9.95
SIGMA_TC
0.388
8.47
0.408
6.05
Number of observations
663
663
Number of draws
2000
2000
Number of estimated parameters
36
33
Final log likelihood
-3027.26
-3191.876
Rho-Square
0.34
0.24
8
Ashkrof, Correia, Cats, van Arem 10
In the case of short distance trips, the conventional car has the most negative value for in-
1
vehicle travel time in commuting trips (-0.149) amongst all modes. Car transport related issues
2
such as traffic congestion and finding a parking space might be the sources of the intense dislike
3
when performing short distance commuting trips which normally are made in dense urban areas.
4
On the other hand, public transportation has the highest disutility (-0.0948) for travel time in
5
long distance commuting trips which is presumably caused by the higher likelihood of possible
6
delays, line changing, seat unavailability and longer walking distance. However, public
7
transportation has the lowest in-vehicle travel time disutility for leisure trips for both short and
8
long distance trips. These results seem to indicate that the combination of different travel
9
distances and purposes can significantly influence travelers’ preferences.
10
Moreover, ADTS is positioned in between car and public transportation since it is
11
perceived as being neither the best nor the worst alternative in all scenarios. ADTS is evaluated
12
favorably in terms of in-vehicle travel time for long distance leisure trips (-0.0599) while it is
13
perceived less attractive for short distance commuting trips (-0.139).
14
Comparing the estimated coefficients of in-vehicle travel time, we can conclude that in-
15
vehicle travel time for long distance commuting trips in ADTS is perceived to be around 8% less
16
negative than in public transportation while no difference is derived between ADTS and
17
conventional car for this trip category. In contrast, in short distance commuting trips, travelers
18
might experience more pleasant time inside an ADTS than conventional cars, whilst time spent
19
in public transportation is preferred over ADTS.
20
Waiting/searching and walking time play essential roles in determining utilities. That is
21
why we incorporated them in the experiment with alternative specific coefficients. The results
22
show that waiting and walking time are indeed significant for all alternatives in both models. It is
23
worth emphasizing that ADTS is assumed to be a door-to-door service, so walking time is
24
irrelevant for this mode. In both long distance and short distance trips, walking time for public
25
transportation which is access/egress time, is valued more negativly than the duration of walking
26
from the parking of the conventional car to the final destination.
27
In long distance trips, waiting/searching time is perceived approximately equal for all
28
alternatives. Conversely, in short distance trips it is about 20% and 50% more negative for
29
ADTS than conventional car and public transportation, respectively. It implies that travelers are
30
more sensitive to waiting for the envisioned AV service than for other alternatives.
31
The cost of travel is most of the times a great determinant of travel mode choice. The
32
estimated generic coefficients, unsurprisingly, are negative for both models owing to the intrinsic
33
disutility of cost. Moreover, the parameter value is roughly equal in both models which shows
34
that passengers are equally sensitive to the costs in the two travel distance categories.
35
As discussed in the previous section, the factor scores of latent variables were
36
incorporated in the final models. The results show that three and four out of five latent variables
37
are significant in the models estimated for long distance and short distance trips, respectively.
38
Trust in AVs is the most significant latent variable in terms of t-value and the magnitude in
39
comparison to the other factors. This is in line with the study conducted by Molnar et al. (34)
40
which concludes that trust in AVs is the strongest component in explaining potential AV
41
adoption. Our findings suggest that trust in AVs, positive view towards AV efficiency, and
42
environmentally friendly attitudes can decrease the disutility of ADTS whereas having an
43
interest in driving leads to an increase in disutility of AV in both models.
44
Regarding the estimated coefficients of the current mobility characteristics of travelers,
45
we conclude that individuals who currently use their car at least 2-3 times per month (labeled as
46
Ashkrof, Correia, Cats, van Arem 11
car users in table 4) are more likely to select the conventional car as the mode of transport,
1
especially for short distance trips. Interestingly, public transport users (at least 2-3 times per
2
month) are around 70% more willing to choose this mode when performing a long distance trip
3
than a short distance one. Furthermore, taxi users and the ones who prefer to be a passenger in
4
long distance trips are more likely to use either ADTS or public transportation while these
5
variables are not significant for short distance trips.
6
Table 4 also indicates that only a few socio-economic factors are found to be significant
7
determinants, namely employment status, age and gender. People aged 18 to 40 are more willing
8
to use the conventional car while working people prefer ADTS. However, men are more likely to
9
favor ADTS for long distance trips while gender is not a significant factor in short distance trips.
10
11
4. CONCLUSIONS
12
This study aimed at investigating traveler mode choice in an era where AVs are an integral part
13
of the mobility market. More specifically, we examine how mode choice determinants differ for
14
different travel distance and trip purpose categories. Discrete choice models have been estimated
15
using data collected in an SP experiment concerning alternative-specific and individual-specific
16
classic attributes, socioeconomic factors, mobility patterns and attitudes.
17
In summary, the results show that the various scenarios with different travel distances and
18
purposes can significantly affect travelers’ mode preferences. The conventional cars and public
19
transportation are perceived the least attractive alternatives in terms of in-vehicle travel time in
20
short distance commuting trips and long distance commuting trips, respectively. Furthermore,
21
prospective ADTS usage is more favorable for long distance leisure trips while it is valued less
22
attractive for short distance commuting trips.
23
The importance of trust in AVs and other attitudinal factors as determinants of modal
24
choice, and therefore for AV adoption, is reaffirmed in this study. Our findings suggest that trust
25
in AVs is the most significant component amongst other attitudes for travelers to use AVs. Other
26
factors are having positive views towards AV efficiency, and environmentally friendly attitudes,
27
both of which can reduce the disutility of ADTS while having an interest in driving yields an
28
increase in the disutility of AV.
29
As an aside, we find out that regardless of the travel distance, in-vehicle travel time is
30
valued less negativly for leisure trips than commuting trips using the same travel mode. At the
31
same time, it is perceived more negatively for short distance trips than for long distance trips,
32
independently of the travel purpose. The reasons can be attributed to the intrinsic value of leisure
33
trips rather than commuting trips and long distance trips in contrast with short distance trips.
34
This study can provide several policy implications. In general, adoption of ADTS is
35
highly dependent on the travel purpose, travel distance, travelers’ attitudes and their
36
sociodemographic characteristics. Market penetration might be higher amongst middle-aged
37
male users who have environmentally friendly attitudes and intend to perform longer distance
38
trips to do leisure activities. Therefore, ADTS operators can focus on catering the demand of this
39
group at the early stages of the operations for ensuring to attract the potential early-adopters.
40
Furthermore, if policymakers intend to support this mode of transport, they could offer several
41
financial incentives and regulatory initiatives including increasing the visibility and
42
dissemination of information on the functionality of AVs in order to improve travelers’ trust in
43
AVs as a critical factor for AV adoption. If people using their private car adopt such services like
44
ADTS, public space use in urban areas would be revolutionized by replacing large parking
45
spaces, particularly, in the city center with other facilities due to the fact that ADTS is able to
46
transport itself from the users’ destination to the specified places.
47
Ashkrof, Correia, Cats, van Arem 12
There are two notable limitations in this research. The first one is inherent to SP
1
experiments. Since the respondents have not experienced riding AVs, their answers rely on how
2
they envisaged the ADTS alternative. The second issue refers to the hybrid choice model
3
structure. The latent variables can be incorporated either in a sequential or simultaneous way in
4
the utility function. A joint estimation (simultaneous) allows investigating the correlation
5
between attitudinal factors and socio-economic variables through a complex calculation. This has
6
been left for further research since the sequential estimation was sufficient to address the
7
research’s questions.
8
This research raises several interesting avenues for further research. Firstly, travel
9
purpose can be expanded in more categories including shopping trips, maintenance trips (e.g.
10
taking children to/from school, visiting doctor, going to the bank, etc.), and also the possibility of
11
changing the daily activity pattern following the AV adoption might be explored. Secondly, the
12
effect of AVs on other aspects of the travel behavior including the route and destination choice
13
can be investigated. Thirdly, considering other forms of automated vehicles including Private
14
Automated Vehicles (PAVs) and Shared Automated Vehicles (SAVs) as the alternatives
15
competing with ADTS might be an interesting research topic.
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
44
45
46
47
48
49
Ashkrof, Correia, Cats, van Arem 13
AUTHOR CONTRIBUTION STATEMENT:
1
The authors confirm contribution to the paper as follows: study conception and design: B. Arem,
2
G. Correia, O. Cats, P. Ashkrof; data collection: P. Ashkrof; methodology: P. Ashkrof, analysis
3
and interpretation of results: P. Ashkrof, G. Correia, O. Cats; draft manuscript preparation: P.
4
Ashkrof; review and edition: G. Correia, O. Cats, P. Ashkrof. All authors reviewed the results
5
and approved the final version of the manuscript.
6
7
REFRENCES
8
1. SAE International. Taxonomy and Definitions for Terms Related to On-Road Motor Vehicle
9
Automated Driving Systems. https://saemobilus.sae.org/content/j3016_201401.
10
2. Correia, G., D. Milakis, B. van Arem, and R. Hoogendoorn. Vehicle Automation and Transport
11
System Performance. 2016.
12
3. Fagnant, D. J., and K. Kockelman. Preparing a Nation for Autonomous Vehicles: Opportunities,
13
Barriers and Policy Recommendations. Transportation Research Part A: Policy and Practice, Vol.
14
77, 2015, pp. 167–181.
15
4. Yap, M. D., G. Correia, and B. van Arem. Preferences of Travellers for Using Automated Vehicles
16
as Last Mile Public Transport of Multimodal Train Trips. Transportation Research Part A: Policy
17
and Practice, Vol. 94, 2016, pp. 1–16. https://doi.org/10.1016/j.tra.2016.09.003.
18
5. Winter, K., C. Oded, K. Martens, and B. van Arem. Stated Choice Experiment on Mode Choice in
19
an Era of Free-Floating Carsharing and Shared Autonomous Vehicles: Raw Data. Transportation
20
Research Board, 96th Annual Meeting, No. 1, 2017, pp. 1–17.
21
https://doi.org/10.4121/uuid:4ac4d7b7-c8b0-42ec-a096-55a4f1837585.
22
6. Krueger, R., T. H. Rashidi, and J. M. Rose. Preferences for Shared Autonomous Vehicles.
23
Transportation Research Part C: Emerging Technologies, Vol. 69, 2016, pp. 343–355.
24
https://doi.org/10.1016/j.trc.2016.06.015.
25
7. Payre, W., J. Cestac, and P. Delhomme. Intention to Use a Fully Automated Car: Attitudes and a
26
Priori Acceptability. Transportation Research Part F: Traffic Psychology and Behaviour, Vol. 27,
27
No. PB, 2014, pp. 252–263. https://doi.org/https://doi.org/10.1016/j.csda.2004.06.015.
28
8. Zmud, J., I. N. Sener, and J. Wagner. Consumer Acceptance and Travel Behavior Impacts of
29
Automated Vehicles Final Report. 2016, p. PRC 15-49 F.
30
9. Silberg, G., M. Manassa, K. Everhart, D. Subramanian, M. Corley, H. Fraser, V. Sinha, and A. W.
31
Ready. Self-Driving Cars: Are We Ready? Kpmg Llp, 2013, pp. 1–36.
32
10. Bansal, P., K. M. Kockelman, and A. Singh. Assessing Public Opinions of and Interest in New
33
Vehicle Technologies: An Austin Perspective. Transportation Research Part C: Emerging
34
Technologies, 2016. https://doi.org/10.1016/j.trc.2016.01.019.
35
11. Haboucha, C. J., R. Ishaq, and Y. Shiftan. User Preferences Regarding Autonomous Vehicles.
36
Transportation Research Part C: Emerging Technologies, Vol. 78, 2017, pp. 37–49.
37
https://doi.org/10.1016/j.trc.2017.01.010.
38
12. Ding, C., D. Wang, C. Liu, Y. Zhang, and J. Yang. Exploring the Influence of Built Environment
39
on Travel Mode Choice Considering the Mediating Effects of Car Ownership and Travel Distance.
40
Transportation Research Part A: Policy and Practice, Vol. 100, 2017, pp. 65–80.
41
https://doi.org/10.1016/j.tra.2017.04.008.
42
13. Arentze, T. A., and E. J. E. Molin. Travelers’ Preferences in Multimodal Networks: Design and
43
Results of a Comprehensive Series of Choice Experiments. Transportation Research Part A:
44
Policy and Practice, Vol. 58, 2013, pp. 15–28. https://doi.org/10.1016/j.tra.2013.10.005.
45
14. Azari, K. A., S. Arintono, H. Hamid, and S. R. Davoodi. Evaluation of Demand for Different Trip
46
Purposes under Various Congestion Pricing Scenarios. Journal of Transport Geography, Vol. 29,
47
2013, pp. 43–51. https://doi.org/10.1016/j.jtrangeo.2013.01.001.
48
15. Shiftan, Y., and R. Burd-Eden. Modeling Response to Parking Policy. Transportation Research
49
Record: Journal of the Transportation Research Board, Vol. 1765, No. 01, 2001, pp. 27–34.
50
Ashkrof, Correia, Cats, van Arem 14
https://doi.org/10.3141/1765-05.
1
16. Centraal Bureau voor de Statistiek (CBS). Transport and Mobility in the Netherlands, 2016.
2
https://www.cbs.nl/en-gb/publication/2016/25/transport-and-mobility-2016.
3
17. Munshi, T. Built Environment and Mode Choice Relationship for Commute Travel in the City of
4
Rajkot, India. Transportation Research Part D: Transport and Environment, Vol. 44, 2016, pp.
5
239–253. https://doi.org/10.1016/j.trd.2015.12.005.
6
18. Casley, S. V., A. . Jardim, and A. M. Quartulli. A Study of Public Acceptance of Autonomous
7
Cars. Wpi, 2013, pp. 1–146.
8
19. Jensen, A. F., E. Cherchi, and J. de D. Ortúzar. A Long Panel Survey to Elicit Preferences for
9
Electric Vehicles. Transportation, Vol. 41, No. 5, 2014, pp. 973–993.
10
https://doi.org/https://doi.org/10.1007/s11116-014-9517-6.
11
20. Ewing, G., and E. Sarigöllü. Assessing Consumer Preferences for Clean-Fuel Vehicles: A Discrete
12
Choice Experiment. Journal of Public Policy & Marketing, Vol. 19, No. 1, 2000, pp. 106–118.
13
https://doi.org/10.1509/jppm.19.1.106.16946.
14
21. Kim, J.-H., J.-H. Chung, and T. Kim. The Effect of Psychological Traits on Mode Choice
15
Behaviour: An Application to a New Water Transit System in Seoul, Korea. Transportation
16
Planning and Technology, Vol. 36, No. 6, 2013, pp. 547–566.
17
https://doi.org/10.1080/03081060.2013.830891.
18
22. Devarasetty, P. C., M. Burris, W. Arthur, J. McDonald, and G. J. Muñoz. Can Psychological
19
Variables Help Predict the Use of Priced Managed Lanes? Transportation Research Part F:
20
Traffic Psychology and Behaviour, Vol. 22, 2014, pp. 25–38.
21
https://doi.org/10.1016/j.trf.2013.10.006.
22
23. Rubin, J. A. Choosing Transit : The Influence of Past Travel Behavior, Attitudes and Habits on
23
Present Choices. 2012.
24
24. Abou-zeid, M., M. Ben-akiva, M. Bierlaire, C. Choudhury, and S. Hess. Attitudes and Value of
25
Time Heterogeneity. Annual Meeting of the Transportation Research Board (TRB) Compendium
26
of Papers, No. January, 2011, pp. 1–27.
27
25. ChoiceMetrics. Ngene 1.1.1 User Manual & Reference Guide. ChoiceMetrics Pty Ltd. 2012.
28
26. Hensher, D. A., J. M. Rose, and W. H. Greene. Applied Choice Analysis. 2015.
29
27. McFadden, D. Conditional Logit Analysis of Qualitative Choice Behaviour. In Frontiers of
30
econometrics.
31
28. Ortúzar, J. D., and L. G. Willumsen. Modelling Transport. 2011.
32
29. Bierlaire, M. BIOGEME: A Free Package for the Estimation of Discrete Choice Models. Swiss
33
Transport Research Conference, 2003, pp. 1–27.
34
https://doi.org/10.1017/CBO9781107415324.004.
35
30. Zwick, W. R. Comparison of 5 Rules for Determining the Number of Components to Retain
36
Comparison of Five Rules for Determining the Number of Components to Retain. No. September,
37
2015. https://doi.org/10.1037//0033-2909.99.3.432.
38
31. Peres-Neto Pedro R., Jackson Donald A., and S. K. M. "How Many Principal Components?
39
Stopping Rules for Determining the Number of Non-Trivial Axes Revisited.” British Journal of
40
Statistical Psychology 49: 974–97. Vol. 49, 2005, pp. 974–997.
41
32. Henson, R. K., and J. K. Roberts. Use of Exploratory Factor Analysis in Published Research.
42
Educational and Psychological Measurement, Vol. 66, No. 3, 2006, pp. 393–416.
43
https://doi.org/10.1177/0013164405282485.
44
33. Garrido, L. E., F. J. Abad, and V. Ponsoda. A New Look at Horn’s Parallel Analysis with Ordinal
45
Variables. Psychological Methods, Vol. 18, No. 4, 2013, pp. 454–474.
46
https://doi.org/10.1037/a0030005.
47
34. Molnar, L. J., L. H. Ryan, A. K. Pradhan, D. W. Eby, R. M. St, and J. S. Zakrajsek. Understanding
48
Trust and Acceptance of Automated Vehicles : An Exploratory Simulator Study of Transfer of
49
Control between Automated and Manual Driving. Transportation Research Part F: Psychology
50
and Behaviour, Vol. 58, 2018, pp. 319–328. https://doi.org/10.1016/j.trf.2018.06.004.
51
