Modeling of a mode choice behavior toward agent-based Mobility as a Service simulation

Abstract

The concept of Mobility as a Service (MaaS), which means integrating multiple transportation modes, not only fixed transportation but also on-demand service, and providing them as a single service, is attracting attention. For the efficient operation of the MaaS system, appropriate number of on-demand service vehicles and pricing are required. Agent-based simulation is suitable for analyzing the relationship between such transportation setting and user behavior. Therefore, we construct a simulation tool to quantitatively analyze and evaluate a MaaS system that consists of users mode choice model as demand side, on-demand shared service allocation method and railway/bus operational information as supply side, and traffic simulator, SUMO. As a case study, we analyze the impact on an actual MaaS demonstration where on-demand shared services were introduced into public transportation modes such as railways, fixed-route buses. Through the cost-benefit analysis using simulation, we found that if the number of demands is sufficiently larger, the benefit of introducing on-demand share service can be expected. The proposed simulation is useful for making decisions on the number of vehicles and appropriate pricing required when introducing new mobility services and MaaS services.

Full text

Modeling of a mode choice behavior toward
agent-based Mobility as a Service simulation
Ryo Nishida1, 3, Ryo Kanamori2, and Itsuki Noda3
1Tohoku University
2Nagoya University
3National Institute of Advanced Industrial Science and Technology
1ryo.nishida.t4@dc.tohoku.ac.jp
Abstract: The concept of Mobility as a Service (MaaS), which means integrating multiple transportation modes, not only fixed
transportation but also on-demand service, and providing them as a single service, is attracting attention. For the efficient
operation of the MaaS system, appropriate number of on-demand service vehicles and pricing are required. Agent-based
simulation is suitable for analyzing the relationship between such transportation setting and user behavior. Therefore, we
construct a simulation tool to quantitatively analyze and evaluate a MaaS system that consists of users mode choice model as
demand side, on-demand shared service allocation method and railway/bus operational information as supply side, and traffic
simulator, SUMO. As a case study, we analyze the impact on an actual MaaS demonstration where on-demand shared services
were introduced into public transportation modes such as railways, fixed-route buses. Through the cost-benefit analysis using
simulation, we found that if the number of demands is sufficiently larger, the benefit of introducing on-demand share service can
be expected. The proposed simulation is useful for making decisions on the number of vehicles and appropriate pricing required
when introducing new mobility services and MaaS services.
Keywords: Mobility as a Service, On-demand shared service, Mode choice modeling, Agent-based simulation, SUMO
1 INTRODUCTION
Mobility as a Service (MaaS) is based on the concept to
integrate multiple transportation modes as one service, such
as railways, fixed-route buses, taxis, and new on-demand ser-
vices like ride hailing. MaaS is proposed by Heikkilä [1] and
Hietanen [2] in Finland in 2014. In Helsinki, Finland, traffic
congestion between a city and suburbs has become a problem,
and the concept of MaaS was born as a solution to improve the
convenience of public transportation that is necessary to reduce
the dependence on private cars.
To improve the convenience of public transportation, only
railways and buses that are operated with fixed timetables and
routes are not enough, so on-demand services are an important
key for MaaS. On-demand "shared" services such as Via and
UberPool are becoming a popular. The on-demand shared ser-
vices have the characteristic of riding people together who travel
by the similar route. For users, the fare is cheaper than taxis,
and it is possible to move more flexibly than buses, and for the
provider, the demand can be processed with a small number of
vehicles, which improves driving efficiency.
When introducing such mobility service, service providers or
city authorities require a tool that can analyze cost-benefit. For
example, when they plan to deploy on-demand shared services,
they want to know answers of questions such as "how many ve-
hicles do we need?," "how much is the appropriate price?," and
"will the convenience of users improve?" Based on the concept
of MaaS, such questions should be answered considering not
only on-demand shared services but also railways and buses.
Moreover, in a scientific field, the necessity of quantitatively
analyzing integrated mobility service is also pointed out. As a
technical review for MaaS, Jittrapirom et al. [3] review model-
ing approach of demand side (travelers behavior), supply side
(transport operation), and business model (demand and sup-
ply matching). In addition, Kamargianni et al. [4] proposed
agent-based modeling and simulation framework including such
models considering MaaS concept. However, since most of the
MaaS services in each region are in the early stages of in-
troduction and demonstration experiments, few studies models
and simulates MaaS system which cover both fixed and flexi-
ble transportation, using actual travelers’ behavior data and real
map scenario.
In our study, therefore, we produce an agent-based simula-
tion framework which can analyze cost-benefit of introducing
an on-demand shared service based on MaaS concept. We
model mode choice behavior by Nested Logit model (demand
side) and on-demand shared services operations by an existing
vehicle allocation optimization method (supply side), and inte-
grate demand and supply model with traffic simulator, SUMO
[5] to model demand-supply interaction. Proposed simulation
framework can evaluate the impact of introducing new mobility
services. In order to quantitatively evaluate the impact, we de-
fine a various indicators from users’ and providers’ perspectives
according to the cost-benefit analysis approach of economics.

Mode Choice Model
Parameters
Number of vehicles
Base fare of SAVS
SUMO
Results
User benefit
Provider benefit
Map data
(OpenStreetMap)
OD
SAVS
service
level
choosed
mode
SAVS system
(Allocation method)
vehicles’ positions
vehicles’ routes
Railway/Bus
service level
Figure 1: overview of simulation
As a case study, we analyze the impact on an actual MaaS
demonstration. The demonstration was conducted in Shizuoka
City in Japan, in November 2019 where on-demand shared ser-
vices were introducing into public transportation modes such as
railways, fixed-route buses.
The remainder of the paper is organized as follows. Section
2 describes our proposed simulation framework. Section 3
reports results of analyzing cost-benefit of introducing the on-
demand shared service considering various scenarios. Section
4 concludes the paper with our future work.
2 AGENT-BASED SIMULATION FRAMEWORK
We propose a simulation framework that consist of the trans-
portation mode choice model, on-demand shared services vehi-
cle allocation optimization system, and traffic simulator SUMO.
This simulation framework treats users and vehicles as agents.
In this paper, SAVS (Smart Access Vehicle Service) means a on-
demand shared service for convenience of explanation. SAVS is
provided by Mirai share [6] and is used for the vehicle allocation
and route optimization system of the on-demand shared service.
Figure 1 shows an overview of a simulation. SUMO is used to
simulate the movement status of SAV (Smart Access Vehicle).
The "SAVS system" communicates with SUMO at each simu-
lation step and gets the position information of the SAV. When
a demand is generated, vehicle allocation optimization is per-
formed and the SAV route and service level (price, travel time)
are calculated. Then user calculates the choice probability of
each modes with transportation mode choice model based on
the calculated service level of SAVS and the service level of the
railway / bus at the demand OD, and choose one mode accord-
ing to the choice probability. If the choosed mode is SAVS, the
calculated route information is sent to SUMO, and the vehicle
moves according to that route on the SUMO. This is repeated
to simulate the traffic of one day.
2.1 Mode choice model
We model users’ mode choice behavior using Nested Logit
(NL) model [7]. NL model is based on the principle of utility
maximization and is derivation of a logit model. The logit
model models the discrete choice behavior as the behavior of
Figure 2: Nested tree
selecting the option that maximizes the utility obtained. The
utility function consists of two terms, a deterministic term 𝑉
and a probability term 𝜀. If 𝜀is assumed a Gumbel distribution,
the choice probability can be calculated from the 𝑉. The logit
model does not consider the correlation between choices and
may overestimate the selection probability. The NL model
prevents this problem by nesting similar options.
In our case, the nested tree is as shown in Figure 2. The
choice probabilities of the main mode 𝑚and the route 𝑟are
defined by the following equation (1), and is the product of the
route choice probability and mode choice probability, like
𝑃(𝑚, 𝑟 )=𝑃(𝑟|𝑚)𝑃(𝑚)
=
exp (𝑉𝑚𝑟)
𝑟′exp (𝑉𝑚𝑟′)
exp 𝜂𝑚𝑉𝑚+𝑉′
𝑚
𝑚′exp 𝜂𝑚′𝑉𝑚′+𝑉′
𝑚′,(1)
where 𝜂𝑚is a scale parameter of a Gumbel distribution of a
mode choice.
𝑉𝑚𝑟 is the deterministic term of the route 𝑟whose main mode
is 𝑚, and is defined as the linear sum of the price, travel time
and number of transfers, and their weights, such as
𝑉𝑚𝑟 =𝛽price ∗𝑥pricemr +𝛽time ∗𝑥timemr +𝛽transfer ∗𝑥transfermr.(2)
And 𝑉𝑚represents the utility peculiar to transportation mode.
𝑉′
𝑚is called a logsum variable and represents the expected value
of 𝑉𝑚𝑟 for each mode of transportation, and is defined as (3)
𝑉′
𝑚=
1
𝜂𝑟
ln 
𝑟
exp (𝜂𝑟𝑉𝑚𝑟),(3)
where 𝜂𝑟is a scale parameter of a Gumbel distribution of a
route choice.
Generally, the parameters of model such as the weights 𝜷,𝜂𝑚
and 𝜂𝑟can be estimated by maximum likelihood estimation.
2.2 SAV allocation method
SAV allocation and route are optimized based on the sequen-
tial insertion method called successive best insertion proposed
by Noda et al [8]. When a new demand is generated, SAVS
allocate the vehicle according to the following process.
1. Try insert a new demands’ pick-up / drop-off points into a
SAV’s current waypoint queue (an allocated sequence of
existing demands’ pick-up / drop-off points) while main-
taining the order of a sequence.

2. Calculate a lost time the existing demand will lose due to
the insertion of a new demand. Also, for new demand, the
drop-off time is calculated. At that time, if the deadline
time is exceeded for either existing or new demand, the
insertion candidate is omitted.
3. For all SAV, find the insertion candidate that minimizes
the lost time and the drop-off time, and then use that as the
final allocation.
4. If all the insertion candidate are rejected, vehicle allocation
is not performed and send rejecte message to a user.
2.3 Evaluation indicators
In order to quantitatively evaluate the impact of the introduc-
ing on-demand share services, we define a various indicators
from users and providers perspectives according to the cost-
benefit analysis approach of economics.
Social Cost
•User cost: a detail in Social Benefit
•Provider cost:
–Railways or fixed-route buses provider: Since the
operation of railways/buses are fixed, the operation
cost is assumed to be constant and is not considered.
–On-demand shared service provider: We define the
cost for each vehicle. Costs include driver labor costs,
vehicle management costs, and so on.
𝐶𝑜=
𝑣
𝐶𝑣(4)
Social Benefit
•User benefit: We define the amount of cost reduction as a
benefit like
𝐵𝑢=𝐶𝑊
𝑢−𝐶𝑊 𝑂
𝑢,(5)
where 𝐶𝑊
𝑢is user cost with on-demand shared services
and 𝐶𝑊 𝑂
𝑢is user cost without one. User cost is defined
using a logsum of mode utility in other words an expected
utility. Use 𝛽price to convert utility value to generalized
cost, user cost is defined as
𝐶𝑢=−1
𝛽price
1
𝜂𝑚
ln 
𝑚
exp 𝜂𝑚𝑉𝑚+𝑉′
𝑚.(6)
•Provider benefit: We define the difference of profit
with/without on-demand shared service as benefit like
𝐵𝑝=
𝑖
𝑃𝑊
𝑖−
𝑖
𝑃𝑊 𝑂
𝑖,(7)
where 𝑃𝑖is price of travel of user 𝑖.
Note that general cost-benefit analysis also deals external
costs such as traffic congestion, traffic accident, CO2 emissions,
and pollution, but not considered here. From above definitions,
we define social costs as 𝑆𝐶 =𝐶𝑜and social benefits as 𝑆𝐵 =
𝐵𝑢+𝐵𝑝. If a cost-benefit rate 𝑆𝐵/𝑆𝐶 is higher than 1, an
introducing the on-demand shared service is effective.
MaaS Demonstration Experiment
Train Bus On-demand
shared service
Route search & Reservation & Payment
MaaS App
Mode Choice Modeling
Social Simulation
App Usage Data
Mode Choice Behavior Data
# transfers
Price Travel time
Mode
Choice
Model
Choice prob.
0.1
0.6
0.3
<Demand Side>
OD, Choice Model
<Supply Side>
Timetable,
Allocation Model SUMO
MaaS Simulator
Cost
Benefit
Figure 3: Flow image from real world to modeling and simula-
tion in case of MaaS
3 CASE STUDY
As a case study, we analyze the impact of an actual MaaS
demonstration. The demonstration was conducted in Shizuoka
City in Japan, in November 2019 where on-demand shared
services were introduced into public transportation modes such
as railways, fixed-route buses. Figure 3 shows process from a
real world experiment to a social simulation. From here, we
describe the details of each process.
3.1 MaaS demonstration
A MaaS demonstration covers railways, fixed-route buses,
and on-demand shared service. For this demonstration, we de-
veloped an MaaS App that enables route search of such multiple
transportation. The demonstration aims at the following four
purposes
•Verification of social acceptability for paid operation of
on-demand shared services
•Verification of the operability of the MaaS App
•Revealed preference survey of transportation mode choie
in MaaS.
•Stated preference survey of MaaS schemes usage intentions
The MaaS App is a combined application of a "complex route
search service" by Val Laboratory and "SAVS" by Mirai Share.
With the MaaS App, users can search for multimodal routes
such as railway, fixed-route buses, and SAVS, and only make
reservations for a sole and immediate use of SAVS.
Figure 4 shows examples of App screenshots when a user is
searching routes. The user can compare routes based on the
service levels such as price, travel time, and number of transfer.
As with a general route search service, once a destination is
specified, the route from the origin to the destination will be
displayed. The transportation modes are displayed as a icon
at the bottom of the App screen. Press the blue button on the
upper right to display the details of the option. The user can
check the service level of the route. For a SAVS option, press
the "Start reservation" button at the bottom to proceed to the
reservation screen. The scheduled boarding time, scheduled
getting-off time, and fare for SAVS are displayed. When the

Figure 4: MaaS App screen examples
reservation is confirmed, the SAVS vehicle will be dispatched1.
The user can check the current position of the vehicle on the
App.
In this demonstration, to use the APP, participants were re-
quired to register a LuLuCa membership authorized by the
Shizuoka Railway. The LuLuCa membership card offers the
passengers a point card function, a credit card function, and a
traffic IC card function. Traffic IC card records when and where
traveler use railways or fixed-route buses.
The main purpose of this demonstration for this study is
to acquire real travel behavior data, in particular mode choice
behavior data, in a MaaS scenario. As mentioned above, a
SAVS usage history can be specified from SAVS system’s log
of the MaaS App. However, the MaaS App doesn’t provide the
usage history of railways or fixed-route buses because the App
does not link to the information of reservation and payment for
those modes. Therefore, we try to specify the travelers’ usage
using traffic IC card history.
3.2 Usage of App and transport modes
This section describes the basic usage results such as the
number of users, user characteristics, and changes in the num-
ber of searches, and then describes the usage results of each
transportation.
App usage
This demonstration attracted 255 MaaS App users for a
month. The anonymized user information was obtained via
the registration data of LuLuCa card membership because the
IDs of the App and the card are linked. Figure 5 shows the
number of users classified by gender and age. More than 70%
of users are male. In particular, among males, 45-54 years old
are the most used (35%), while among females, 25-34 years old
are the most used (36%). Only 3% of the users were elderly (65
years old or older), and only 5% were young people (24 years
old or younger). There were no users under the age of 19.
1Note that in this demonstration reservations can only be made for immediate
use of SAVS, so it is not possible to make advance reservations for SAVS for
routes that transfer from railways or fixed-route buses to SAVS.
Figure 5: Number of MaaS App users by gender and age
Figure 6: Number of daily searches
In addition, Figure 6 shows the number of daily searches. 899
accesses to the route search function and the maximum number
of searches was 73. More users used the route search function
on weekdays rather than weekends and holidays. In the latter
half duration of this demonstration, the number of searches has
increased specially in the last week, whereas the users, during
these two weeks earned LuLuCa reward points every access to
the route search.
Each transportation mode usage
Next, we describe the each transportation mode usage. From
the App log, we calculated that 179 people used SAVS (315
times). On the other hand, when comparing the route search
results of the App with the usage history by LuLuCa, only 44
out of 899 searches were confirmed to be used (75 cases were
used during the same day, and 273 cases were using on any day
during the experiment).
From data analysis point of view, there are some limitations
to modeling and simulation from only demonstration data. It
seems that some user just used App as trial for route search.
In addition, some user who using public transportation as com-
muting has become a habit, are thought that use transportation
without searching with the App. From these reasons, it was
found more difficult than expected at this point to identify the
route that searched with App and actually used.

Table 1: Estimation result
variables estimate std error t-value p-value
𝛽price [100yen] -1.144 0.373 -3.066 0.002 **
𝛽time[10min] -2.035 0.595 -3.419 0.001 ***
𝛽transfer -1.374 0.820 -1.677 0.094 *
ASC𝑆 𝐴𝑉 𝑆 11.089 3.830 2.895 0.004 **
𝜂𝑚0.239 0.087 2.922 0.004 **
Initial log-likelihood -400.144
Final log-likelihood -208.343
¯𝜌20.467
3.3 Mode Choice modeling
In this Shizuoka MaaS demonstration, we collected 390 mode
choice behavior data that consist of mode/route choice set and
choosed result. Of the 390 cases, 315 cases were identified
that SAVS was used. Since the number of samples that can be
specified that railways or fixed-route buses were used is small,
we adopted 75 cases that could be identified as using the same
route (at different times) as the recommended route.
Although the number of samples was small, we modeled
the mode choice behavior using the NL model. As mentioned
section 2.1, the explanatory variables 𝒙were price, travel time,
and number of transfers. In addition, we define 𝑉𝑚for SAVS
as a Alternative-specific Constants (ASC) in order to consider
selection bias that can be occured when new mobility service is
introduced. Therefore, parameters to be estimated are weights
𝜷for price, travel time and number of transfer, constant to SAVS
𝐴𝑆𝐶SAVS, and a scale parameter of Gumbel distribution 𝜂𝑚(we
set 𝜂𝑟=1). Table 1 is the estimation result. In the tables,
∗ ∗ ∗,∗∗, and ∗, represent significance levels of 0.001, 0.01, and
0.1, respectively.
From the statistical values of each parameter and the value of
the modified coefficient of determination of McFadden [9], ¯𝜌2,
the estimation results are valid to some extent.
We find some knowledge from estimation result as follows.
•The weights 𝜷were all negative values, and it was found
that the sensitivity for each explanatory variable was rea-
sonably valid even under the MaaS demonstration experi-
ment.
•Since the SAVS constant (ASC𝑆𝐴𝑉 𝑆) is a positive value,
it was found that SAVS was used more positively than
railways and fixed-route buses. This is thought to reflect
the concern that "in the MaaS demonstration experiment,
there may be a bias toward using transportation that is not
normally used." pointed out by Durand et al. [10].
•The time of value obtained from the parameter ratio of
price and travel time was 17.8 [yen / minute], and the
value is a generally reasonable value.
3.4 Simulation and results
We analyze cost-benefit of introducing SAVS and the differ-
ence of them on various scenario. In this demonstration, the
SAVS fare was set to be the base fare up to 1.473km and then
added according to the distance. In addition, the discount was
25% compared to regular taxis regardless of whether or not they
were shared. We set scenario where the number of SAVS are
changed from 1 to 21 (default is 21), and the base fare of SAVS
are changed from 0 yen to 1400 yen at 100 yen intervals (default
690 yen). We analyze some scenario with both small and large
demands setting. As small demands setting, we use the demand
of the day with the highest demand, and the number is 73. On
the other hand, as large demands setting, we use all demand
which generated through MaaS App in this demonstration, and
the number is 899. This demand includes similar demands of
the same person but different dates, so to deal with these demand
as a day’s demand is not realistic. However, we treated these de-
mands as one day’s demands assuming that there are potentially
multiple people with similar attributes and they generate same
demand. In these simulation, we set vehicle cost 𝐶𝑣in equation
(4) to 10000 yen as driver labor cost, and it is unified for all
vehicles. From the result of mode choice modeling, users of
MaaS App have selection bias to SAVS, so we omit 𝐴𝑆𝐶𝑆 𝐴𝑉 𝑆
from utility definition.
Figure 7 and 8 show the cost-benefit with small / large de-
mands with various base fare of SAVS, respectively. The gray
dotted line is the line that cost-benefit rate is 1. Note that in
these simulation the social cost corresponds to the cost of SAVS,
which is proportional to the number of vehicles.
From Figure 7, we found that the benefits do not change when
the SC is more than 20,000 or 30,000, that is, when the number
of vehicles is more than 2 or 3. This indicates that if the number
of demands is small, increasing the number of vehicles will not
produce any effect. On the other hand, when the number of
demands is large, as shown in Figure 8, the benefits increase in
proportion to the cost (number of vehicles), although there are
differences depending on the base fare.
In this demonstration, the maximum number of vehicles was
21, and the base fare was 690 yen. From Figure. 8 (a), it can
be seen that the cost-benefit rate is about 1 if the demand was
close to 900. In other words, in this demonstration experiment,
if 900 demands using the MaaS App were generated in daily,
the benefits commensurate with the cost can be obtained. In
addition, from Figure 8 (b) and (c), for users, the lower the price,
the greater benefit, but for the providers, the benefit cannot be
increased like users, and an appropriate price is required.
4 CONCLUSION
In this study, we produced an agent-based simulation frame-
work which can analyze cost-benefit of introducing an on-
demand shared service based on MaaS concept. As a case study,
we analyze the impact on an actual MaaS demonstration where
on-demand shared services were introducing into public trans-
portation modes such as railways, fixed-route buses. Through
the cost-benefit analysis using simulation, we found that if the
number of demands is sufficiently larger, the benefit of intro-
ducing on-demand share service can be expected. The proposed
simulation is useful for making decisions on appropriate pricing
and the number of vehicles required when introducing a new
mobility service.

(a) Social cost - Social benefit (b) Social cost - User benefit (c) Social cost - Provider benefit
Figure 7: Cost-benefit with 73 demands
(a) Social cost - Social benefit (b) Social cost - User benefit (c) Social cost - Provider benefit
Figure 8: Cost-benefit with 899 demands
Our simulation is not a complete MaaS analytical simulation.
To improve our simulation framework, we need to go stepfurther
in demand side and supply side modeling. In demand side
modeling, we need a model that takes into account the long-
term value of being able to handle the choice of owning or
letting go of a private car. On the other hand, in supply side
modeling, it is necessary to introduce a method for optimizing
routes that combine multiple modes of transportation, such as
the "Integrated Dial a Ride Problem", and create a situation
in which on-demand services becomes a branch line of fixed
transportation. Moreover, we will add external costs such as
traffic congestion, traffic accident, CO2 emissions, and pollution
to cost-benefit analysis. These reductions of external costs are
indicators at MaaS integration level 4 defined by Sochor et al.
[11], and are important indicators for quantitative analysis of
MaaS.
ACKNOWLEDGEMENTS
The data used in this research was provided by the Shizuoka
MaaS (Shizuoka MaaS Core Business Demonstration Project
2020) Consortium. We would like to express our gratitude
here.
REFERENCES
[1] Sonja Heikkilä. Mobility as a service - a proposal for
action for the public administration, case helsinki. MSc
dissertation, Aalto University., 2014.
[2] Sampo Hietanen. “mobility as a Service”—The new trans-
port model? Eurotransport, 12(2), 2–4, 2014.
[3] Peraphan Jittrapirom, Valeria Caiati, A-M Feneri, et al.
Mobility as a service: A critical review of definitions,
assessments of schemes, and key challenges. Urban Plan-
ning, 2:13–25, 2017.
[4] Maria Kamargianni, Lampros Yfantis, Jakub Muscat, et al.
Incorporating the mobility as a service concept into trans-
port modelling and simulation frameworks. In Transporta-
tion Research Board Annual Meeting, 2019.
[5] Pablo Alvarez Lopez, Michael Behrisch, Laura Bieker-
Walz, Jakob Erdmann, Yun-Pang Flötteröd, Robert
Hilbrich, Leonhard Lücken, Johannes Rummel, Peter
Wagner, and Evamarie Wießner. Microscopic traffic simu-
lation using sumo. In The 21st IEEE International Confer-
ence on Intelligent Transportation Systems. IEEE, 2018.
[6] Mirai share. http://www.miraishare.co.jp/en/
savs/.
[7] H C W L Williams. On the formation of travel demand
models and economic evaluation measures of user benefit.
Environment and Planning A, 9(3):285–344, 1977.
[8] Itsuki Noda, Masayuki Ohta, Kosuke Shinoda, et al. Eval-
uation of usability of dial-a-ride systems by social simu-
lation. In Multi-Agent-Based Simulation III. 4th Interna-
tional Workshop, MABS 2003 (LNAI-2927), pages 167–
181, 2003.
[9] Daniel McFadden. Quantitative methods for analyzing
travel behaviour of individuals: Some recent develop-
ments. Technical Report 474, 1977.
[10] Anne Durand, Lucas Harms, Sascha Hoogendoorn-
Lanser, et al. Mobility-as-a-Service and changes in travel
preferences and travel behaviour: a literature review. 2018.
[11] Jana Sochor, Hans Arby, Marianne Karlsson, et al. A
topological approach to mobility as a service: A proposed
tool for understanding requirements and effects, and for
aiding the integration of societal goals. In International
Conference on Mobility as a Service 2017, pages 187–201,
2017.