Potential uptake and willingness-to-pay for Mobility as a Service (MaaS): A stated choice study

Abstract

Mobility as a Service (MaaS), which uses a digital platform to bring all modes of travel into a single on-demand service, has received great attention and research interest. Different business models have emerged in which travellers can either pre-pay for their mobility services bundled into a MaaS plan, or pay-as-they-go using a smart app linked to the service. This study aims to understand how large the potential market of MaaS would be if travellers are offered this one-stop access to a range of mobility services, and how much potential users might value each item included in a MaaS plan. A stated choice survey of 252 individuals administered via a face-to-face method is conducted in Sydney, Australia and a state of the art preference model is estimated to address the research questions. Results indicate that almost half of the sampled respondents would take MaaS offerings, and the potential uptake levels vary significantly across population segments, with infrequent car users being the most likely adopters, and car non-users the least. On average, Sydney travellers are willing to pay $6.40 for an hour of access to car-share, with one-way car-share valued more than station-based car-share. Estimated willingness-to-pay for unlimited use of public transport is $5.90 per day which is much lower than the current daily cap. These findings suggest a careful segmentation of the market and a cross-subsidy strategy is likely to be required by MaaS suppliers to obtain a commercially viable uptake level.

Full text

1
Potential Uptake and Willingness-to-Pay for Mobility as a Service (MaaS): A Stated
Choice Study
Chinh Q Hoa,b, David A Hensherb, Corinne Mulleyb and Yale Z Wongb
a Corresponding author
b Institute of Transport and Logistics Studies (ITLS), The University of Sydney Business School, NSW 2006,
Australia; +61 418 992 227; chinh.ho@sydney.edu.au, david.hensher@sydney.edu.au,
corinne.mulley@sydney.edu.au, yale.wong@sydney.edu.au
Abstract
Mobility as a Service (MaaS), which uses a digital platform to bring all modes of travel into a single on-
demand service, has received great attention and research interest. Different business models have emerged in
which travellers can either pre-pay for their mobility services bundled into a MaaS plan, or pay-as-they-go
using a smart app linked to the service. This study aims to understand how large the potential market of MaaS
would be if travellers are offered this one-stop access to a range of mobility services, and how much potential
users might value each item included in a MaaS plan. A stated choice survey of 252 individuals administered
via a face-to-face method is conducted in Sydney, Australia and a state of the art preference model is estimated
to address the research questions. Results indicate that almost half of the sampled respondents would take
MaaS offerings, and the potential uptake levels vary significantly across population segments, with infrequent
car users being the most likely adopters, and car non-users the least. On average, Sydney travellers are willing
to pay $6.40 for an hour of access to car-share, with one-way car-share valued more than station-based car-
share. Estimated willingness-to-pay for unlimited use of public transport is $5.90 per day which is much lower
than the current daily cap. These findings suggest a careful segmentation of the market and a cross-subsidy
strategy is likely to be required by MaaS suppliers to obtain a commercially viable uptake level.
Suggested citation: Ho, C.Q., Hensher, D.A., Mulley, C. & Wong Y.Z. (2018), Potential uptake and
willingness-to-pay for Mobility as a Service (MaaS): A stated choice study, Transportation Research Part A:
Policy and Practice, Volume 117, Pages 302-318, https://doi.org/10.1016/j.tra.2018.08.025.
Keywords
Mobility as a Service (MaaS), choice experiment, service bundles, willingness-to-pay, nonlinear choice model
Acknowledgment: This paper contributes to the research program of the Volvo Research and Education Foundation Bus
Rapid Transit (BRT+) Centre of Excellence, as well as the TfNSW program in ITLS associated with the Chair in Public
Transport. This research was partially funded by the University of Sydney Business School Pilot Research Scheme.
Comments of the two anonymous reviewers are also appreciated.

2
1. Introduction
Imagine a city landscape in which you could travel seamlessly from a shared car, stationed close to your home,
to a train after leaving the car at a designated parking lot at the platform's doorstep, to an express bus operating
on a designated corridor, and then to a taxi which takes you to your final destination. Imagine that you can use
any combination of these transport modes without the need to own a car or public transport tickets, check bus
and train timetables, or pre-book a taxi, since they are all available via an app on your smartphone linked to a
mobility service subscription plan. This mobility service gives you access to all modes and real time
information for each journey, as well as providing instant journey planning and booking. This kind of door-to-
door service, powered by disruptive transport technologies, has previously been a vision but is now starting to
emerge in cities around the world under the name ‘Mobility as a Service’ or MaaS.
MaaS is no longer a theoretical concept to bring all modes of transport into a single mobility plan that travellers
can subscribe and pre-pay for the use of all transport modes, in much the same way as people choose a mobile
phone plan that meets their need in terms of calls, text and internet access. MaaS has recently been
commercialised in Helsinki, Finland through the smartphone app called Whim and the service will soon be
available in the West Midlands in the UK. Other countries such as Sweden, Austria, Germany, and the US
have tested mobility services on real people, real networks with real mobility plans. Australia and New Zealand
are also joining the MaaS trend with the establishment of MaaS Australia, SkedGo, and other players actively
working to bring the concept to market.
With a mobility plan customised to each subscriber, MaaS has a real potential to shift the traditional car
ownership paradigm away from outright ownership, thereby changing the overall modal share given that car
use starts from car ownership (Maat and Timmermans, 2007, Ho and Mulley, 2015). A shift in ownership is
already being observed, accelerated by a myriad of mobility options such as Uber, GoGet (car-sharing), Car
Next Door (peer-to-peer car-sharing), bike-sharing, and ride-sharing schemes. A move from car ownership to
shared membership will no doubt be affected by the increasing deployment of self-driving vehicles promoted
by several major automakers and technology giants moving to make self-driving vehicles commercially
available by 2020 (Muoio, 2016). The commercial release of fully self-driving vehicles opens new markets for
car-sharing (as existing non-drivers will be able to travel in a fully self-driving vehicle on their own),
suggesting the effect of car-sharing on car ownership will be substantial. Recent research evidence suggests
that the mobility services offered by Uber and GoGet have resulted in deferring their users’ decision to
purchase a car (SGS Economics & Planning, 2012, Newberg, 2015). This is a sign that shared self-driving
vehicles and MaaS could deepen the reduction of private vehicle ownership, especially among the younger
generations (Delbosc and Currie, 2013, Goodwin and Van Dender, 2013) since mobility will increasingly be
achieved without the need to own a car or even a valid driving licence.
Among the most important questions for local governments, transport modellers, planners and economists, is
whether the arrival of the repackaging of transport modes will trigger changes to short-term travel patterns
(e.g., transport mode choice) and long-term choices, for instance home and work locations. Just like Uber and
car-sharing services, MaaS and in due course, self-driving vehicles are initially expected to become dominant
in dense urban areas, providing a good opportunity for cities to reduce the role of private cars and their negative
consequences on the liveability of the city from air pollution and emissions. Once MaaS and self-driving
vehicles spread beyond urban centres, the boredom of car commutes may disappear as commuters can do more
productive activities such as eating, working, reading, and even sleeping. This means that travel time can be
used more productively in a self-driving car than in a conventional car, and thus car users may not value travel
time savings as highly as they do now. The distinction between working at the office and working while
travelling may start to blur with travellers becoming ‘passengers’. This possibility is supported by Ho et al.
(2016) and other national guidelines, such as the New Zealand Economic Evaluation Manual (2013) and the
Netherlands (Significance et al., 2012), which evaluate the value passengers place on travel time savings
(VTTS) as a proportion of the car driver’s values. Such a reduction in VTTS means that people are likely to
be more tolerant to a long commute and this may impact on longer term decisions such as residential location,
possibly moving further away from the cities to benefit from lower dwelling prices or more land for a given
expenditure. This would have a significant impact on urban planning, traffic management, greenhouse gas
emissions, congestion, and the viability of conventional public transport services.

3
The impact of shared mobility on our cities and our lives is manifold and the question of how transport
technology innovations might disrupt or alter urban transport systems and, in turn travel behaviour, is being
highly debated with much speculation but little substantive insight. This is partly due to the lack of relevant
behavioural data and models that can provide guidance on the potential uptake of new mobility opportunities
and how emerging transport options will change travel choices in the short and long terms. This paper aims to
set a benchmark in identifying mobility service packages that align with the preferences of travellers, how they
may take these services up, and how drivers may question the necessity to own vehicles if and when offered
one-stop access to a range of transport mobility options. This study is timely in informing MaaS providers as
to the business model to follow and how best to package, cost and market mobility plans to end users to obtain
sustainable goals by way of designing MaaS plans that are likely to have a high take-up rate.
The remainder of the paper is organised as follows. The next section reviews the literature on MaaS with a
focus on the business model and the way in which different transport options are packaged into MaaS plans
for end users to subscribe. The section that follows sets out the design of a choice experiment to capture the
data necessary to establish preferences for new mobility opportunities as mixtures of transport options under
varying plans, followed by a summary of the empirical setting and sampling strategy required to obtain
geographical coverage and representativeness of the sample. Empirically, a nonlinear logit model is estimated
to provide quantitative evidence of mobility-cost trade-offs, taking into account current travel needs, and
establishing the WTP for the various elements of a MaaS plan. The paper concludes with the implications the
evidence has for the demand for MaaS and the development of MaaS plans that are attractive to achieve
sustainability outcomes.
2. Literature on MaaS
Mobility as a Service (MaaS), also known as Transportation as a Service (TaaS), describes a personalised,
one-stop travel management platform digitally unifying trip creation, purchase and delivery across all modes.
For customers, it offers total integration across public, intermediate (ridesourcing, microtransit and taxi) and
private (through car-sharing or cycle hire) modes of transport. MaaS provides user benefits in terms of true
competition with vehicle ownership and a seamless customer experience, and benefits service providers by
improving the capacity utilisation of their vehicles and opening up new opportunities for forward thinking
businesses (as mobility brokers). For society, MaaS can circumvent some of the potential urban efficiency
issues (e.g., autonomous vehicle externalities related to deadheading, traffic congestion, land use and the urban
form) associated with new transport technologies and trends (Wong et al., 2017).
At present, there exists a small but growing academic literature specific to MaaS—investigating government
interest (Heikkilä, 2014), impacts on land use (Rantasila, 2015), customer expectations (Sochor et al., 2015b),
intelligent transport systems (Hu et al., 2015, Brendel and Mandrella, 2016, Giesecke et al., 2016), integration
opportunities (Kamargianni et al., 2016), institutional requirements (Mukhtar-Landgren et al., 2016), end user
demand (Sochor et al., 2016, Matyas and Kamargianni, 2017), impacts on public transport contracts (Hensher,
2017), scalability (Mulley, 2017), potential business models for service delivery (Kamargianni and Matyas,
2017) and the societal imperative for MaaS (Wong et al., 2017). The vast majority of this work constitute think
pieces and literature reviews rather than empirical research, and the range of topics relate to three core areas
in technological requirements, the customer interface and implementation challenges. MaaS is also fast
becoming mainstream across the grey literature, with new publications from popular media (The Economist,
2016), consultancy (Atkins, 2016, Datson, 2016a, Hazel, 2017), think tanks (Datson, 2016b, Dotter, 2016,
RethinkX, 2017), trade associations (TravelSpirit, 2017), transport suppliers (Cubic Transportation Systems,
2016), transport operators (e.g., Keolis Downer) and transport regulators (e.g., Transport for New South Wales,
Transport for London). However, many reports by consultancies, think tanks and industry bodies are generally
based around expert opinion and (potentially) driven by commercial intent.
A number of trials are currently underway or have recently been concluded in a range of European cities to
market test various MaaS models, with Kamargianni et al. (2016) pointing to 15 mobility schemes which range
from partial to advanced integration. Partial integration describes systems with only ticketing integration or
information and communication technology integration (but not both)—e.g., transport smartcards (Opal in the
empirical context of Sydney, Australia) or journey planning apps extended to encompass intermediate and
private modes. With advanced integration, such as Hannovermobil (Germany), customers can receive an

4
integrated mobility monthly bill, but cannot purchase a bundled package of mobility. Smile (Vienna) is another
such scheme innovative in not only brokering cooperation between transport providers, but also other
interested parties like software companies, engineers and environmental groups. Only the proposed Helsinki
(Finland) model (subsequently commercialised as Whim) has reached an advanced level of integration with
mobility packages, receiving a 10/10 mobility integration in the index score proposed by Kamargianni et al.
(2016). An in-depth review of these schemes and the broader literature cited above helped inform the
development of mobility packages used in the experimental survey described in the subsequent section.
A major theoretical contribution in the design and implementation of MaaS may be related to the three Bs—
bundles, budgets and brokers (Hensher, 2017). The key innovation in MaaS is the ability for customers to
purchase ‘bundles’ of mobility granting them a defined volume of access to each mode, with a specified level
of service (e.g., pickups within five minutes). These mobility packages may be sold as subscriptions, or as pay-
as-you-go options, and may be tailored by age, occupation or location to suit different market segments, and
enable providers to cross subsidise between mobility options or practise price discrimination. Budgets refer to
end user preferences and service provision possibilities. Understanding behaviour is a recurring theme and will
help discern willingness-to-pay for various mobility packages, as well as forecast demand and mode shares for
new transport propositions, to ensure commercial viability and societal optimality. Finally, brokers describe
broadly the business models around which MaaS will be delivered, including the potential for new entrants
and implications on existing public transport contracts (Hensher, 2017).
Bundles and budgets form the core focus of this paper. Bundling is a mechanism to repackage existing services
together with new services to create a more attractive way for people to access mobility. Indeed, bundling is
common amongst service industries and can stimulate demand to achieve cost economies (Guiltinan, 1987)—
as is the case with transport modes which are interdependent and characterised by a high ratio of fixed to
variable costs. A major unknown is how to design these mobility bundles, and understanding the budgets, or
consumer preferences, including people’s willingness-to-pay and trade-off propensity between attributes like
convenience, comfort and price. Of course, other authors are beginning to fill this research gap, some with
stated choice (SP) experiments as in this paper. However, this paper provides the first empirical evidence for
Australia (as compared to Europe) using an efficient pivot designs to customise MaaS offers to each respondent
and state of the art sophisticated choice modelling techniques to contribute to this emerging body of
knowledge1
Much of the existing literature has assumed that the key innovations behind MaaS in integrating public,
intermediate and private modes, and linking customer information with fare integration sold as subscriptions
are sufficiently attractive enough in their own right for consumers. Whilst this may well hold true, it is
important to investigate empirically the prospects for people to switch out of conventional transport services
to MaaS packages. This study aims to shed light on a number of key unknowns around:
• What commercially-viable mobility plans can be created and marketed to transport customers?
• How large is the market for MaaS if a mobility broker (like MaaS Australia) offers its customers one-stop
access to a range of travel services through a smartphone-based app?
• How large must travel cost savings be for consumers to take up MaaS plans and what are the implications
for car ownership (including consideration of households disposing their second car)?
• How much would people be willing to pay for each mobility offering (hours/kms of car-sharing, days of
unlimited public transport use, or kms of taxi/Uber)?
• How might mobility service subscribers alter their mix of travel between public, intermediate, private and
active modes of transport and in what form will this take (e.g., first/last mile vs. point-to-point)?
1 Stated preference studies are being undertaken in all parts of the world, but there are significant differences either in the
design or in the analysis or both and, indeed, by offering insights to behavioural responses in different settings. For
example, Matyas and Kamargianni (2017) in their stated choice (SP) investigation on the potential demand for MaaS
packages in London was aided by a smartphone-based revealed preference component on existing travel behaviour
whereas this study uses recall for recent travel patterns, Matyas and Kamargianni (2017) has a relatively small sample
size (n=80 for full survey completion), provides an empirical setting of Europe as opposed to Australia where there is a
considerable difference in urban form and higher private vehicle mode share and uses a random rather than efficient pivot
design.

5
To address these questions, this research employs the experimental study approach which offers a random
sample of respondents a set of ‘customised mobility plans’ and asks them to choose their preferred plans over
multiple games based on their current travel needs. The next section describes the method in more detail.
3. The Choice Experiment Survey
The centrepiece of this study is a choice experiment designed to understand: (a) the potential uptake of MaaS
with the Sydney Metropolitan area serving as a laboratory environment, (b) how much people are willing to
pay for various items included in a mobility plan, and (c) how a MaaS plan will likely change car ownership
models and travel patterns. The use of a choice experiment is required because MaaS as a package of modal
options is not yet available in the Sydney market (even though we observe single modes such as UberPOOL
and Car Next Door), and thus behavioural data can only be collected within the setting of a stated preference
(SP) method. In operating such an experiment where the respondent faces options pivoted around their current
behaviour, the experiment seeks from each respondent their current circumstance and travel pattern for a
typical two-week period through a set of background questions.
The survey instrument has five major parts. The first part seeks socio-demographic information with questions
relating to the respondent’s home postcode, age group, employment status, commuting mode and frequency,
gender, ability to drive, disabilities, daily access to car, smartphone ownership, internet use, car-share
membership (and pod distance from home), number of household cars and drivers, and household structure.
The second part asks the respondent to describe their current travel patterns for a typical week in terms of the
number of one-way trips undertaken by different modes (public transport, taxi/Uber and car) for every day of
the week, daily public transport (PT) fare, daily taxi/Uber cost, daily distance and time travelled by car, daily
parking cost, as well as typical access mode and access time if PT is used. This information is used, as noted
above, to pivot the experiment around the respondent’s experience. Each respondent is then introduced to the
concept of MaaS and MaaS plans including how to interpret each transport component of the MaaS plan.
Based on the information provided in the first two parts, especially how much the respondent uses each mode
of transport (including PT, taxi/Uber and car) and their driving licence status, we offer each respondent a
number of mobility plans presented in the form of a choice task, and ask them to indicate which plan they
would prefer and whether they would definitely consider subscribing if it were available. A binary responding
mechanism (i.e., yes vs. no) is used to verify how certain the respondent would make the choice that they did
if the selected mobility option were to be available in a real market. Adopting a binary response instead of a
certainty scale (e.g., 0 to 10) for this supplementary question removes the need to locate a threshold value for
a certainty index should such information be used in the modelling (Rose, Beck and Hensher, 2015).
In terms of plan design, bundling was based on the case study of the Sydney Metropolitan area from which
respondents were sought. From concept to implementation, a range of considerations were made in relation to
which mobility services (i.e., attributes) ought to be included in the MaaS plans and how best to measure them
(i.e., attribute levels) to reflect best practice in survey design and the range of mobility options available in the
study area. At the outset, only existing travel modes (though including some creative variations) were included
as part of the MaaS plans, as a trade-off between providing a comprehensive set of mobility services and
minimising the cognitive burden for the respondent. Four modes of transport were included in each mobility
plan. These included PT, car-share, taxi and UberPOOL (microtransit).2 The choice was made to alternate
between the use of generic mode labels and existing operator brands to maximise community recognition of
each product, informed in part by the pilot studies. Existing PT modes in metropolitan Sydney include buses,
trains, light rail, and ferries, with ticketing (though not fares) already integrated via the use of the Opal
smartcard, introduced in 2013. Given this, all PT options were incorporated under one umbrella mode for
simplicity.
As for car-sharing, two additional mode-specific attributes were included to refine the product offering and
look to future service possibilities. This included selecting the advance booking time for the share-car and
indicating whether it operated on a return-to-base or one-way service format. The one-way car-share option
2 Bike-share was not included in the plan design partly because it was not available in Sydney at the time and partly
because cycling as a transport mode accounts for a very limited market share.

6
served as a proxy for investigating preferences for autonomous vehicles operating under a shared model,
without introducing the concept at this stage as we believe the inclusion of autonomous technologies could
add psychological factors (for instance around safety and comfort) clouding respondents’ decision process.
Station-based or round-trip car-share was branded as GoGet whilst the one-way service was called Car2Go to
maintain branding consistency (the former is available in Sydney whilst the latter is not yet available).
Mobility package cost and the volume of access defined for each mode are expected to constitute two of the
most important criteria for respondent choices. For PT, volume was defined as days with unlimited access to
the system, with existing travel costs based on daily ($15) and weekly ($60) fare caps under the adult Opal
used as inputs to determine final package price. Unlike other jurisdictions, fare caps in Sydney are system-
wide and there are no zonal issues to complicate cost determination. Further, eligibility for free and subsidised
travel were not considered in the experiment. Access volume for car-share was defined as hours or days per
fortnight, reflecting the current practice of many car-sharing companies operating in Sydney and elsewhere.
Car-share costs were pivoted off existing GoGet prices with a premium/discount in place for one-way car-
share and booking time reductions. The attribute levels for taxi and UberPOOL were designed as discounts off
an unspecified pay-as-you-go rate even for bundled subscriptions. The rationale for this approach is to test
different forms of integration and the recognition that a more primitive implementation of integration with the
mobility broker may be more realistic in the short term. Also, an analysis of the Sydney Household Travel
Survey, reported in the next section, informs a low incidence of frequent taxi/UberPOOL users being sampled
for this study, and thus to keep the offers relevant to every respondent, it was deemed best to offer percentage
discounts instead of kms or hours of taxi/UberPOOL entitlement.
Each respondent faced four choice tasks, each involving four options (alternatives) including two customised
mobility plans, a pay-as-you-go plan (PayG) and an option of not choosing any plan on offer, which, when
selected caused an open-ended question to pop up seeking the respondent’s reason for not purchasing any plan.
Where the respondent chose any one of the mobility plans, they were asked how the chosen mobility plan
would impact their PT use. Figure 1 provides an illustrative choice screen for a respondent who reported PT
use 4 days per fortnight and spent 10 hours driving a total distance of 242 kms during this period. The current
travel record provides a reference point (i.e., existing travel behaviour) so that the respondents can relate
themselves to the scenarios presented and is included as a choice in each of the choice tasks. The two pre-
defined plans represent a subscription business model where different services are integrated and
commercialised under a MaaS plan granting defined volume of access to each mode. The PayG option offers
modal discounts and integration benefits of the journey planner, representing another possible business model
whereby app-developers such as Moovit or SkedGo commercialise a digital platform that integrates multiple
modes of transport into a smart app that also offers journey planning, ticketing, billing, whilst being
supplemented with various discounts provided by mobility partners. These three alternatives to the status quo
or current travel record exist in each of the choice tasks. Finally, each of the choice tasks maintain the full set
of the mobility services (i.e., the MaaS offerings) with the levels of mobility offerings being changed across
the tasks so as to give the maximum information for model estimation.

7
Figure 1. An illustrative choice screen
Sitting behind the SP experiment are three D-efficient designs (see Hensher et al., 2015) of which one was
assigned to the respondent based on their ability to drive (i.e., having a driving licence or not) and to use PT
(i.e., having any physical or cognitive difficulties in using PT). For example, respondents holding a valid
driving licence and having no physical/mental disabilities were offered mobility plans that included both car-
share and PT modes (Design D1). In contrast, those holding no valid driving licence were offered mobility
plans that contained no car-share option (Design D2). Similarly, mobility plans excluding a PT component
were presented to respondents who currently encounter some physical or cognitive difficulty preventing them
from using PT (Design D3). In addition, the designs were further customised to each respondent by the use of
three sub-designs within each design (e.g., Design D1.1 to D1.3). The sub-designs were implemented by using
different reference points and pivot levels for respondents with different transport needs, defined by their
typical weekly record of PT and car use. Table 1 shows the pivot levels for each attribute, the mobility service
included in each design, and the rules employed to assign the designs to respondents. The choice experiment
was designed using Ngene (Choice Metrics, 2012) with 12 choice tasks, blocked into three sets of four. Priors
for the main survey were obtained from a pilot survey of 20 respondents.

8
Table 1: Pivot levels of the SP experimental designs and assignment rules
Mode Attribute Attribute level
[Reference: Pivot level] Design Apply for respondents w/
current travel record of …
Public transport Days with unlimited use of PT [0 : +1,+2,+4] D1.1 Low level of PT use
(0 or 1 day/week)
Car-share Hours use of car-share in bundle [10 : ±1, ±3, ±5] D1.1
Public transport Days with unlimited use of PT [4 : +0,+2,+4] D1.2
Medium level of PT use
(2-4 days/week)
Car-share Hours use of car-share in bundle [6 : ±1, ±2, ±3] D1.2
Public transport Days with unlimited use of PT [8 : +0,+2,+4] D1.3
High level of PT use
(5-7 days/week)
Car-share Hours use of car-share in bundle [6 : ±1, ±2, ±3] D1.3
Public transport Days with unlimited use of PT [0 : +1,+2,+4] D2.1 Low level of PT use
(0 or 1 day/week)
Public transport Days with unlimited use of PT [4 : +0,+2,+4] D2.2 Medium level of PT use
(2-4 days/week)
Public transport Days with unlimited use of PT [8 : +0,+2,+4] D2.3 High level of PT use
(5-7 days/week)
Car-share Hours use of car-share in bundle [4 : +1,+2,+3,+4,+5,+6] D3.1 Low level of car use
(<3 hours/week)
Car-share Hours use of car-share in bundle [6 : +1,+2,+3,+4,+5,+6] D3.2 Medium level of car use
(3 - 5 hours/week)
Car-share Hours use of car-share in bundle [10 : +1,+2,+3,+4,+5,+6] D3.3
High level of car use
(>5 hours /week)
Car-share Car-sharing scheme Round-trip, One-way D1, D3 All respondents w/ licence
Car-share Advance booking time 15, 30, 60 minutes D1, D3 All respondents w/ licence
Car-share Hourly rate if PayG $6, 6.50, 7, 7.50, 8, 8.50 D1, D3 All respondents w/ licence
Taxi % discount off every taxi bill 10%, 20% All All respondents
UberPOOL % discount off UberPOOL bill 5%, 10% All All respondents
Unused credit Roll-over, Lost All All respondents
Plan price Formulae All All respondents
After completing the four choice tasks, the respondent was offered an opportunity to create their own mobility
plan. This ‘Create It Yourself’ or CIY plan serves as a ‘safety net’ when all pre-designed plans offered in the
choice tasks fail to fully satisfy the respondent’s travel needs. In this part, the respondent has a chance to
bundle different mobility services into a fortnightly plan and the survey instrument then prices the CIY plan
and asks if the respondent would be willing to subscribe.
The fifth and final part of the survey randomly selected one of the mobility plans which the respondent said
they would buy if it were available in the local market, and asked a series of debrief questions as to how this
would change their travel patterns (in terms of PT trips, car hours, car kms, taxi/Uber kms) and whether they
would put their car up for sale if they had one. This was followed by a series of attitudinal questions on MaaS
evaluated on a 7-point Likert scale from strongly disagree to strongly agree.
4. Sampling and Sample Profile
The survey was conducted with assistance from Taverner Research who helped recruit participants for the
Computer-Assisted Personal Interview (CAPI) accessed via desktop computers. Before starting the interview,
each respondent was asked to watch a two-minute video explaining the concept of MaaS and its potential
benefits.3 Interviewers sat with the respondents and provided any explanation that was required in working
through the survey, whilst not offering answers to any of the questions. Interviews were conducted between
23rd March and 09th April 2017 at Chatswood, Marrickville and Campsie shopping centres where shoppers
3 URL to the video is https://vimeo.com/96486671

9
were randomly recruited to participate in the study. These three shopping centres were selected on the basis
that potential respondents from these areas would have access to a number of transport options included in the
MaaS plan.
A sample of 200 valid interviews was contracted with quotas split evenly across the three shopping centres.
All people aged 18 and above were eligible with no other screening criteria or quotas applied. A sample of 252
valid interviews was obtained. On average, each interview took 17 minutes with a standard deviation of 5
minutes. Figure 2 shows the distribution of respondents by their home postcodes.
Figure 2. Distribution of respondents by home postcode
Table 1 provides a profile of the sample and compares this against the equivalent population in the Sydney
Metropolitan area. The sample has an average age of 39 years, with a standard deviation of 14 years. Of the
252 respondents that make up the sample, fulltime workers account for about half (48%), followed by part-
time workers (17%), full-time students (13%) and the unemployed (8%). Compared to the Sydney
Metropolitan area, the sample represented the working population well; however, students and unemployed
individuals were over-represented while retirees and home-keeper were under-represented. In terms of
household structure, the sample included 28% couples with children, 26% couples with no children, 9% single
person, 7% single parent, with the balance (29%) being other household types such as group households and
multiple generation households. The sampled respondents hailed from households with fewer cars than average
(1.57 car per household in the sample vs. 1.85 per household across the Sydney Metro), while the number of
household drivers (i.e., people with valid driving licences) in the sample were similar to that of the Sydney
Metropolitan area.

10
Table 1 Descriptive profiles of the sample and the Sydney Metropolitan area
Sample
( td d )
Sydney Metro
( td d )
Respondents age (year)
39 (14)
46 (17)
Male (1/0) .45 .49
Full-time worker (1/0) .48 .47
Part-time worker (1/0) .17 .16
Unpaid voluntary worker (1/0) .01 .00
Fulltime student (1/0) .13 .07
Part-time student (1/0) .03 .01
Pensioner (1/0) .02 .04
Unemployed (1/0)
.08
.03
Retiree (1/0) .05 .16
Look after home (1/0)
.02
.06
Use internet every day (1/0) .97 N/A
Use internet few days/week (1/0)
.03
N/A
Holding valid driving licence (1/0)
.85
.86
Daily access to car (1/0)
.74
N/A
Having a smartphone (1/0)
1.00
N/A
Membership of GoGet car-share (1/0)
.08
N/A
Number of household cars 1.57 (0.977) 1.85 (1.134)
Number of household driving licences 2.10 (1.009) 1.93 (0.966)
Person living alone (1/0) .09 .11
Couple with children <15 and 15+ (1/0) .03 .06
Couple only (1/0) .26 .23
Single parent with child(ren) <15 (1/0) .00 .02
Couple with child(ren) <15 years (1/0) .12 .21
Single parent with child(ren) 15+ (1/0) .06 .08
Couple with child(ren) 15+ (1/0) .13 .20
Single parent with children <15 and 15+ (1/0) .00 .01
Other household types (1/0)
0.29
0.08
Sample size (number of people aged 18+)
252
2,948,289
Note: N/A information is not available from the Sydney Household Travel Survey
Table 2 summarises the profile of the respondents with their travel demand in a typical week. On a typical
Monday, the respondent on average undertook 1.12 PT trips, 0.03 trips by taxi and 1.25 trip by car (both as a
driver and as a passenger) with an average car mileage of 12.5 kms and car travel time of 29 minutes. These
travel statistics appear to be stable throughout the weekdays but change significantly on the weekend,
especially on Saturday when the average number of PT trips per person reduces to 0.72 while taxi and car trips
increase to 0.20 and 1.65, respectively.

11
Table 2 Travel demand in a typical week of sampled respondents
Sample
Sydney metro
Date
Travel demand
Mean
Std. Dev
Mean
Std. Dev
Monday
PT trips per person
1.12
1.24
0.34
0.80
Taxi trips per person
0.03
0.21
0.02
0.18
Car trips per person
1.25
2.01
3.06
2.62
Car kms per person
12.56
22.55
31.87
44.05
Car minutes per person
28.60
51.65
63.80
66.21
Sample (#people aged 18+)
252
352,554
Tuesday
PT trips per person
1.11
1.22
0.38
0.83
Taxi trips per person
0.04
0.24
0.02
0.20
Car trips per person
1.29
2.10
3.21
2.80
Car kms per person
12.84
22.17
30.97
39.98
Car minutes per person
28.52
48.20
62.37
62.06
Sample (#people aged 18+)
252
354,991
Wednesday
PT trips per person
1.09
1.33
0.42
0.86
Taxi trips per person
0.04
0.24
0.02
0.18
Car trips per person
1.36
2.16
3.17
2.86
Car kms per person
13.24
22.42
30.35
38.60
Car minutes per person
30.43
52.94
64.21
67.42
Sample (#people aged 18+)
252
366,244
Thursday
PT trips per person
1.11
1.30
0.41
0.87
Taxi trips per person
0.03
0.22
0.03
0.23
Car trips per person
1.42
2.09
3.21
2.95
Car kms per person
13.93
23.04
31.90
44.45
Car minutes per person
30.91
49.64
65.69
69.52
Sample (#people aged 18+)
252
372,559
Friday
PT trips per person
1.11
1.21
0.37
0.81
Taxi trips per person
0.15
0.47
0.05
0.29
Car trips per person
1.33
2.10
3.39
2.84
Car kms per person
12.93
23.23
35.42
48.81
Car minutes per person
29.93
53.79
70.08
71.82
Sample (#people aged 18+)
252
362,455
Saturday
PT trips per person
0.72
1.13
0.17
0.58
Taxi trips per person
0.20
0.53
0.06
0.33
Car trips per person
1.62
1.77
3.53
2.56
Car kms per person
15.44
24.68
38.81
49.05
Car minutes per person
32.04
40.89
72.36
65.54
Sample (#people aged 18+)
341,685
Sunday
PT trips per person
0.66
1.03
0.14
0.50
Taxi trips per person
0.05
0.28
0.02
0.16
Car trips per person
1.39
1.64
3.01
2.21
Car kms per person
13.41
37.51
34.51
48.99
Car minutes per person
25.13
35.24
61.07
60.58
Sample (#people aged 18+)
252
322,123
It should be noted that the sampled respondents exhibit a different travel demand to that of Sydney residents
which is shown in the last two columns of Table 2. These latter statistics were derived from the 5-year pooled
Sydney Household Travel Survey data weighted to the population of the Sydney Metropolitan area.
Specifically, compared to the average Sydney resident, the sampled respondents use PT and taxi more and are
less reliant on use of a private car. This difference is expected since the samples were sourced from shopping

12
centres with good PT services and a majority of the respondents have access to train services at their home
(see Figure 2). It is important to take account of the difference in travel behaviour and socio-demographic
characteristics between the sample and the population when generalising the findings reported in the next
section.
5. Descriptive Analysis and Model Specification
This section provides a descriptive analysis of the empirical data to provide a basic understanding of factors
that drive a user’s subscription to MaaS and the stated share associated with different MaaS plans in
comparison to the status quo (i.e., conventional transport services). The descriptive analysis also serves as a
precursor to modelling the stated choice responses of the MaaS plans in the presence of conventional transport
services.
5.1 Descriptive analysis
Figure 3 shows a histogram of car usage, defined by the number of days (a) and hours (b) of car use in a typical
week by the sampled respondents. These two measures of car use are selected as the MaaS plans offered in the
SP experiment include a car-share mode whose costs are calculated based partially on the numbers of days and
hours of use. As can be seen in Figure 3a, for a typical week, a large proportion of the respondents used car
either every day (75/252 respondents) or not at all (69/252), with the balance using car between one day to six
days per week. In terms of car hours, an overwhelming majority of respondents used car less than 15 hours per
week with 51.6% of the sample (130/252 respondents) using car two hours or less in their typical week.
(a)
(b)
Figure 3. Observed car use in a typical week of the sampled respondents
Figure 4 shows the stated shares of MaaS plans in comparison to the status quo, segmented by car user type
(defined by the number of days the respondents were observed to use car). The last panel of Figure 4 shows
that the stated shares of the MaaS plans across the sample are 47.2% of which 36.2% would subscribe to pre-
defined MaaS plans tailored to their current travel patterns and a further 11% would use MaaS as PayG. The
stated shares of MaaS plans relative to the status quo do vary substantially across car user types with car non-
users showing the lowest likelihood of buy-in to MaaS plans (i.e., 62% of car non-users stay with conventional
transport services) whilst infrequent car users are most likely to subscribe. Also, amongst car users, those who
use car more frequently are less likely to subscribe. Interestingly, allowing respondents to create their own
mobility plan does not increase the take-up rate. Indeed, this reduces to 32.1% (from 36.2% if counting MaaS
packages or from 47.2% if including PayG). This suggests two possibilities. First, pre-defined MaaS plans
tailored to the respondent’s current travel needs suit their travel patterns well. The second possibility is that
having an opportunity to create their own plans leads respondents to try to replicate their current travel patterns,

13
such as the number of days using car regardless of whether they use the car for the whole day or not, and were
then not prepared to take the plan because of its price tag (a full day of access to car is priced at 10 hours of
car-share and hence quite expensive). Further investigation showed that 36% of respondents included the same
number of full days access to car-share in their own plan, while up to 57% of respondents included fewer car
days than what they were currently observed to use. In contrast, 59% of respondents included exactly the same
number of PT days in their own plan, with 30% including one to four days more and only 11% of respondents
including fewer days than currently used.
Figure 4. Stated shares of MaaS plans by type of car user
The empirical data were analysed in a similar way but this time the frequency of car use was replaced by
frequency of PT use, taxi/Uber use, and GoGet car-share membership. Similar effects on MaaS subscriptions
were observed, albeit to a lesser extent. These observations have a significant implication for modelling. That
is, the effect of current travel patterns should be taken into account in modelling the potential uptake of MaaS
when travellers are offered this one-stop access to a range of mobility services. Given a strongly negative
correlation between PT use and car use observed in the empirical data (r = -0.554), respondents were classified
into four groups based on the number of days in a typical week they used car (see Figure 4) and included these
variables in the model specification to recognise that their likelihood of subscription to MaaS plans was
conditioned on their current travel patterns.
5.2 Model specification
Beginning with the standard utility expression associated with alternative j contained in a choice task of J
alternatives, we assume that an index, In defining the current travel pattern of individual q, conditions the utility
expression. Specifically, the utility that the traveller q derives from an alternative j, be it the status quo or the
MaaS plans on offer, can be written as (see Swait and Adamowicz, 2001, Hensher and Ho, 2016):
*
( ), 1,...,
jq q jq q jq jq
UIUIV jJ
ε
== = +
(1)
where
*
jq
U
the utility that the traveller q derives from an alternative j, which is specified as a function of current
travel patterns Iq and how attractive the mobility plan j is to the respondent q, denoted as Uiq (or how much
utility the respondent could obtain from the potential offer). Equation (1) is a form of heteroscedastic
conditioning where Iq recognises individual-specific circumstances, proxied by some metric such as their
current usage of different transport modes and socio-demographic characteristics.
Assuming that the random variables
q jq
I
ε
are iid Gumbel with unit scale factors, it follows that the probability
expression associated with Equation (1) takes the same form as the standard multinominal logit model (MNL)
on the condition that the conditioning function Iq is non-negative:
( ) ( )
**
Pr U U Pr ( ) ( ) Pr ( ) ( ) Pr U U , , , 0
jq iq q jq iq q iq jq jq iq iq jq jq iq q
IVV I VV ijJI
εε εε
  
≥ = −≥ − = −≥− = ≥ ∀∈ ≥
  
(2)
53%
17%
20%
11%
25% 50% 25% 50% 25% 50% 25% 50% 25% 50% 
Not subscribe (status quo)
Customised Plan A
Customised Plan B
Pay-As-You-Go

Non-user
(0 day/week)
Infrequent user
(1 or 2 days/week)
Frequent user
(3 or 4 days/week)
Very frequent user
(5 -7 days/week)
All user type
(0 -7 days/week)
Data source: MaaS survey (this study)

14
The conditioning function Iq is specified as a parametric function of respondent’s characteristics and current
travel patterns on the recognition that the same MaaS plan may be more attractive to some people than to
others. The specific functional form of heteroscedastic conditioning, implemented herein, is given in Equation
(3).
11
1
LK
q ll k k
lk
I zy
γϕ
= =
=++
∑∑
(3)
where
l
z
is a set of L variables describing the current travel patterns (such as type of car users or whether the
respondents is a member of GoGet car-share) and
k
y
is a set of socio-demographic characteristics (e.g., age,
gender or employment status), and γl, φk are parameters to be estimated, together with the vector of parameters
β representing respondent’s preferences for various mobility services, X included in the MaaS plan. Note that
when all γl, φk are not statistically different from zero, the conditioning function receives that value of 1 and
the heteroscedastic conditioning model collapses to the standard MNL. Note also that the conditioning function
Iq varies across individuals (subscript q) but not across alternatives within a choice task. It is therefore
necessary to normalise the conditioning function of one alternative (e.g., the status quo) to 1.0 and allow others
(e.g., MaaS plans) to be freely estimated.
The probability that individual q chooses an alternative j in a choice task containing J alternatives can then be
written in an MNL-like model shown in Equation (4):
11
11 11
exp 1 ( )
exp( )
Pr
exp( ) exp 1 ( )
LK
l l k k qj
lk
q qj
jq JJ LK
q qi l l k k qi
ii lk
zy
IV
IV zy
γϕ
γϕ
= =
== = =


++ ×


×

= = 

×++ ×




∑∑
∑∑ ∑∑
X
X
β
β
(4)
Preference heterogeneity may be layered on top of model (4) in the form of random parameters (5), or error
components (6), or scaled MNL (7), or all of these features:
qm m m qm
v
β βσ
= +
(5)
1
C
qj jc qj qjt
c
dE
εη
=
= +
∑
(6)
2
exp( ) where ~ [0,1]
2
qq q
w wN
τ
στ
= −
(7)
where
m
β
is the population mean of the individual parameter
qm
β
associated with attribute m,
qm
v
is the
individual specific heterogeneity with mean zero and standard deviation one,
m
σ
is the standard deviation of
the distribution of
qm
β
around
m
β
,
q
E
are error components which are alternative specific random individual
effects that activate when the dummy variable
jc
d
=1,
q
σ
is the scale factor which is choice invariant but varies
across individuals q by a scale parameter τ and a random variable
q
w
, and
qjt
η
is iid extreme value.
The overall model with all components is:
11 1
1 11 1
exp 1 ( )
Pr
exp 1 ( )
LK C
q l l k k qj jc qj
lk c
jq J LK C
q l l k k qj jc qj
i lk c
z y dE
z y dE
σ γϕ
σ γϕ
= = =
= = = =




++ ×+







=



++ ×+







∑∑ ∑
∑ ∑∑ ∑
β
β
X
X
(8)

15
Models (4) to (8) extend the utility functions beyond the linear specification by a specification that combines
a heteroscedastic conditioning function with random parameters, error components, and the scaled multinomial
logit model.
6. Estimation Results and Willingness-to-Pay for MaaS
Table 3 presents the empirical model results for the stated choice of MaaS plans in the presence of the status
quo (i.e., continued use of conventional transport services). This model was adopted after an extensive number
of non-linear specifications were estimated using NLOGIT version 6 (for details see Ch 20 Hensher et al.,
2015). The best model, presented in Table 3, suggests two sources of preference heterogeneity. The first one
links to the heteroscedastic conditioning function Iq where preferences for MaaS plans were found to vary
systematically across different types of car user and socio-demographic groups. The second component relates
to the utility function itself in which preferences for the entitlement to the number of days with unlimited PT
use and the number of car-share hours were found to vary randomly across the respondents. No statistical
evidence was found to support the presence of preference heterogeneity in the form of scaled MNL and error
components. The final adopted model is a heteroscedastic non-linear random parameter model with the two
random parameters following the constrained triangular distribution where the spread (i.e., half the range or
the standard deviation times √6) was constrained to equal the mean.
The non-linear random parameter model fits the data reasonably well with the McFadden adjusted R2 of 0.467.
This model rejects the non-linear MNL-like model (4) at any level of confidence based on a log-likelihood
test. All the parameter estimates have the expected sign with the kernel density plots shown in Figure 5
assisting the interpretation of model parameters. Figure 5 shows that the standard utilities (Viq) of all MaaS
plans are in the negative domain. Thus, a positive parameter for individual-specific variables included in the
conditioning function Iq suggests a lower likelihood of subscription because the negative standard utilities are
scaled by a factor larger than 1 (see equation (3)), resulting in more negative (i.e., higher) disutility, and hence
less preferred. By contrast, a negative parameter for the conditioning function suggests the opposite. That is,
people in these groups are more likely to buy into MaaS plans in the presence of the status quo. Note also that
the conditioning function is in the positive domain, satisfying the non-negative condition required by Equation
(2).

16
Table 3. Estimation results for the stated choice of MaaS plans in the presence of the status quo, Sydney
Metropolitan Area 2017
Description
Para
Sig
t-value
Heteroscedastic conditioning function (Iq)
Car non-user (1/0/-1, base = infrequent user)
0.306
***
6.87
Car frequent user (3-4 days/week) (1/0/-1, base= infrequent user)
-0.147
***
-4.06
Car very frequent user (5-7 days/week) (1/0/-1, base= infrequent user)
0.021
0.65
Age between 35 and 44 (1/0/-1, reference = age 18-24)
-0.200
***
-7.46
Age 55 or over (1/0/-1, reference = age 18-24)
0.326
***
8.61
Household with 2+ children (1/0/-1, base = up to one child)
0.096
***
4.22
Car-negotiating household (1/0/-1, base = car-sufficient household)
-0.012
-0.76
Member of GoGet car-share (1/0/-1, base = not a member)
0.003
0.08
Standard utility function (Vjq)
Fortnightly fee of pre-defined MaaS plans ($)
-0.069
***
-12.43
Fortnightly fee of CIY MaaS plan ($)
-0.083
***
-13.33
Fortnightly fee of PayG MaaS plan ($)
-0.117
***
-3.19
Number of days entitled to unlimited PT use, mean = spread
a
0.447
***
11.76
Number of hours entitled to car-share use, mean = spread
a
0.411
***
12.2
Hourly rate of car-share if PayG ($/hour)
-0.062
*
-1.9
One-way car-sharing (1/0/-1, base = return-to-base or round-trip)
0.252
***
4.47
Advance booking time for car-share (minutes)
-0.005
-1.25
Entitled to taxi discount (% off every taxi bill)
0.026
**
2.22
Entitled to ride-share discount (% off every UberPOOL bill)
0.050
***
4.81
Unused credit lost (1/0/-1), reference = roll-over
0.009
0.14
Constant of MaaS Plan A
-1.871
***
-5.61
Constant of MaaS Plan B
-1.768
***
-5.63
Constant of PayG MaaS Plan
-1.858
***
-2.79
Constant of CIY MaaS Plan
0.307
0.77
Average fortnightly cost of car ownership and use ($)
-0.006
***
-10.15
Number of days using PT in a typical fortnight (day)
-0.037
*
-1.79
Number of hours using car in a typical fortnight (hour)
-0.023
***
-2.68
Model summary statistics
Number of choice tasks (#interviewees)
1,260 (252)
Number of model parameters
23
Log likelihood at convergence
-1,058.3
McFadden adjusted pseudo-R
2
0.467
AIC/N
1.755
Note: a Random parameters follow a constrained triangular distribution with mean = spread
***
Significance at 99%,
**
95%, and
*
90% level of confidence.

17
Figure 5. Kernel density of the conditioning (Iq) standard utility (Vqj) and nonlinear utility (

∗) functions
Table 3 shows that the effects coded ‘car non-user’ variable has a significantly positive parameter, indicating
that car non-users are less likely than average to take up MaaS.4 Conversely, those who typically use a car
three to four days per week (i.e., car frequent users) are more likely than average to subscribe to MaaS, as
evidenced in the significantly negative parameter associated with this variable. The parameter for the
infrequent car users (the base) can be calculated as the negative sum of the ‘car non-user’, ‘car frequent user’,
and ‘car very frequent user’ parameters (−0.306 + 0.147 − 0.021= −0.180). That is, those who use car one or
two days per week (i.e., car infrequent users) are more likely than average to take up MaaS plans. The relative
differences in the magnitude of these parameters suggest that car non-users (0.306) are least likely to subscribe,
followed by very frequent car users (0.021), whilst infrequent car users (−0.180) and frequent car users
(−0.147) are most likely adopters. The modelling evidence confirms the effect of current travel patterns on
MaaS potential uptake described in section 5.1, especially Figure 4. As car non-users are very likely to be
frequent users of PT (pairwise correlation of 0.406), these findings suggest that frequent PT users are happy
with the Opal card that allows them to pay for the use of all PT modes at a cap of $15 per day or $60 per week.
Thus, the experimental MaaS plans suggested for these people, which includes mainly PT days and an
entitlement to few hours of car-share and discounts of taxi/Uber bills, are less attractive even with a heavy
discount to PT cap fares (fixed at $10 a day or $50 a week in the SP experiment).
4 Effects coding is an alternative to dummy coding in which an attribute with L levels is transformed into L – 1 variables
with the reference level being coded as –1 instead of 0. Each effects coded variable is set equal to 1 when the attribute is
present, equal to –1 if the reference case is present, and equal to 0 if otherwise. When a categorical variable is effects
coded, the comparison would be with the grand mean (i.e., average across the entire sample) as opposed to the reference
group in the case of dummy coding.
Kernel Density of Standard Util ity (Viq)
STDUTIL
.11
.22
.33
.44
.56
.00
-15.65 -11. 74 - 7.83 -3. 92 .00
-19.57
PlanA Pla nB PayG Own(CI Y)
Density
Kernel density of conditi oning funct ion (Iq)
CONDZ
.33
.67
1.00
1.33
1.66
.00
.60 .80 1. 00 1.20 1.40 1.60
.40
Kernel density est imate for CONDZ
Density
Kernel Density of Utility (V*) by Mobility Option
UTIL
.15
.30
.46
.61
.76
.00
-15.92 - 9.57 -3.22 3. 13 9. 48-22.27
SQ PlanA Pl anB PayG Own(CI Y)
Density

18
Amongst the individual and household characteristics tested with the model specification, age and number of
children in the household were found to impact MaaS subscription whilst gender, car-sharing membership
status, household structure, household size, household car ownership (defined directly as number of household
cars or as relative to number of household drivers such as car-negotiating and car-sufficient households)
exhibited no impact on MaaS subscription propensity. Specifically, households with two or more children were
significantly less likely to subscribe to MaaS than households with up to one child. This is reasonable as the
demand for a private car and its convenience (e.g., available at any time and able to install child car restraints
on it) increases with household children numbers.
Turning to the utility of different MaaS plans, Table 3 shows that the parameters associated with the costs of
MaaS plans are significantly negative, as is the parameter associated with the average fortnightly cost of car
ownership and use. Although it is difficult (and debatable) to compare the two costs of different natures (one
links to the subscription fee of a MaaS plan where users are entitled to a number of transport services vs.
another linked to vehicle ownership, maintenance and use), it may be safe to interpret, from the parameters
associated with the costs and their value ranges, that travellers are much less sensitive to the long-term costs
of vehicle ownership (and use) than the on-going cost of MaaS subscription. This finding supports the
hypothesis expounded within the literature that car drivers consider only direct out of pocket costs at the point
of travel (Scott and Axhausen, 2006).
With respect to different mobility services the MaaS subscribers are entitled to, all else being equal, people
prefer more days with unlimited PT use and more hours of car-share use, but the preference varies significantly
across the respondents. This is evidenced by the significantly positive random parameters associated with these
two mobility elements. As for car-sharing, a one-way car-share format is preferred to return-to-base car-share
(i.e., round-trip car-share). Also, a MaaS plan that requires longer advance booking time for the use of a shared
car is seen as less attractive than the one that guarantees access to car-share within a shorter notice period.
Adding taxi and ridesharing discount entitlements to MaaS plans is found to improve their attractiveness, but
this positive effect applies only to regular Taxi/Uber users (24% of the sample), defined in this study as people
who use these modes at least once in their typical week. Surprisingly, the roll-over of unused credits (i.e.,
entitlements to car hours and PT days) does not exhibit any effect on the attractiveness of a MaaS plan. One
possible explanation for this insignificant parameter is that the MaaS plans suggested by the experiment were
customised to each individual’s travel needs by pivoting around their current travel patterns. This probably
results in respondents seeing most credits being used up by the end of the period, making the need to roll-
over/transfer unnecessary.
Alternative-specific constants (ASCs) were estimated for all MaaS plans with the status quo serving as the
reference. Since all categorical variables included in the model are effects coded, these ASCs reflect their pure
effect in reproducing the sample shares (as opposed to when dummy codes are used, the constants partly
capture the effect of the reference group—see Bech and Gyrd-Hansen (2005)). The ASCs estimated for the
pre-defined MaaS plans and PayG were significantly negative, suggesting that all else being equal, respondents
are more likely to stay with the status quo. Although not shown, t-tests of statistical differences between these
ASCs were conducted with no differences found. This is also evidenced in the kernel density of the final utility
shown in Figure 5 where the utility of the two unlabelled MaaS plans (Plan A and Plan B) are almost identical.
Together, the evidence suggests no preference bias toward a particular MaaS plan due to its position in the SP
experiment (note that the PayG plan is always placed to the rightmost of each choice task as per Figure 1).
Finally, the constant of the CIY Plan is positive, suggesting that allowing travellers to create their own MaaS
plans increases the likelihood of them subscribing to MaaS, although the gain is small (due to a small constant,
not its statistical insignificance which is immaterial because the constant should always be included to
reproduce the market share).
The signs and the significance levels of the parameters presented in Table 3 identify the potential adopters of
MaaS when it is introduced, but this does not tell us how much they are willing to pay for different mobility
services/entitlements included in a MaaS plan, referred to as MaaS components. To this end, Table 4 presents
the sample average willingness-to-pay (WTP) for different MaaS components. Before going into any detail, it
is important to highlight that these WTPs were derived from the sampled respondents and not the population
of the Sydney Metropolitan area. Thus, caveats must be taken into account for the use of these estimated WTPs
in bundling and pricing MaaS plans for practical release as bundles for purchase.

19
Table 4 shows that on average, Sydney travellers are willing to pay $6.40 for an hour of access to car-share or
about $64 per day if a full day of access is priced at the rate of 10 hours—a rule employed in this experimental
study and also by Sydney-based car-sharing companies such as GoGet. Options for car-sharing also include
the car-sharing formats (one-way or round-trip) and advance booking time which, on average, the respondents
are willing to pay an extra $7.30 per fortnight for one-way car-sharing and about $1.05 less for every 15
minutes increase in advance booking time. The average WTP for a day of unlimited PT use to be included in
a fortnightly MaaS plan is $5.90 but this varies across the sample with a standard deviation of $2.40 and a
maximum of $11.85 (min = 5 cents).
Table 4. Estimated respondent’s WTP for different mobility entitlements of MaaS plan
MaaS component
WTP
($/fortnight)
An hour access to car-share
$6.39
A full day access to car-share (10 hours)
$63.85
One-way car-share
$7.27
Round-trip car-share
$0.00
Every 15 minutes increase in advance booking time
−$1.06
A day of unlimited PT use
$5.92
10% discount to every taxi bill
$3.68
10% discount to every ride-sharing bill
$7.18
Placing these WTPs into context for how MaaS is likely to work in practice, Figure 6 shows the distribution
of respondents’ WTP for two example fortnightly MaaS plans. Plan 1 includes two full day’s use of car-share
(e.g., one day for each week’s social outing), 10 hours of one-way car-share with a 60 min advance booking
time to cover the remaining days of the fortnight (e.g., three days per week doing 30 min one-way drop-
off/pick-up children at a day care centre en route to/from work and two hours weekly shopping), four days of
unlimited PT use per fortnight (e.g., commuting by PT two days per week when children do not attend day
care), and a 10% discount off every taxi and ridesharing (e.g., UberPOOL) bill. The average WTP for this
example plan is $185 per fortnight. Plan 2 offers more days with unlimited PT use and more car hours on a
round-trip car-sharing format with shorter notice than Plan 1. The average WTP for this plan is estimated at
$231 per fortnight.
Entitlement per
fortnight
Plan 1
Plan 2
Car days
2
2
Car hours
10
15
Car-sharing scheme
One-
way
Round-
trip
Advance notice
60 mins
30 mins
Taxi discount
10%
20%
Ridesharing discount
10%
10%
PT days
4
6
Average WTP
$185
$231
Figure 6. Distribution of WTP for example MaaS plans
Distribution of WTP for example MaaS Plans, dollar/fortni ght
<- Xi - >
.001075
.002088
.003101
.004114
.005127
.000062
95 191 286 382 4770
WTPPLAN1 WTPPLAN2
Density

20
7. Conclusions and Ongoing Research
This paper is an initial contribution to understanding what types of MaaS subscription plans might appeal to
potential users. Using state of the art stated choice methods to reveal preferences as of today in the context of
Sydney, Australia, provides a baseline of candidate drivers of interest in MaaS packages. The packages
investigated include what we believe are likely to be the main inputs into the design and offering of a range of
app-based subscription plans.
There are many innovative features in this study, in particular the design of the choice experiment which tends
to be more complex than what one would typically develop for a single mode alternative (even with the choice
set allowing for more than one labelled mode); however, complexity must be aligned with behavioural validity
of the potential subscription plans and hence an extensive amount of work was undertaken to ensure that the
choice experiment was comprehensive and comprehendible. One way this is ensured is through an investment
in the survey design customised to each individual and administered face-to-face where well-trained
interviewers sat with each respondent to answer any question they might have had about MaaS.
MaaS and how it might occur in the market is an area where many researchers are devoting time. With the
same goals in mind, researchers in this domain will have overlapping ideas even when they are located on
other sides of the world. The main contributions of this paper were developed when the ideas of MaaS were
still being developed, and the development of an SP experiment for the Australian context was a natural
extension of the descriptive information available at the time. With increasingly more researchers undertaking
SP experiments and choice modelling, or working in the MaaS domain more generally, we will all benefit
from work on common themes particularly with SP studies when they are executed using state of the art choice
methods, many developed by ITLS—notably Hensher, Rose and Bliemer. 5
This study highlights some important questions for policy-makers in the public transport arena. The MaaS
plans were not particularly attractive to existing public transport users. To move towards the sustainable future
craved almost universally by governments all over the world, building public transport patronage with choice
riders is essential. This study suggests that there may be a need for lowering public transport fares, especially
the daily and weekly cap, to attract more choice riders and to retain current travellers. This does not necessarily
mean that greater subsidy is required since disruptive transport technologies could make public transport
provision cheaper (through reducing operating costs or offering individual plans that consider subscriber’s
travel patterns including the number of trips made and fare bands).
Another feature of this research that we have shown to be increasingly important in other studies unrelated to
MaaS is experience with the alternatives (Hensher and Ho, 2016). While MaaS in its full definition is not yet
available in Australia (and indeed in most geographic jurisdictions), various modal elements are clearly well
established (e.g., conventional public transport) or entering fast into various markets (e.g., Uber taxi-based
services and car-sharing). The findings in this study reinforce the importance of accounting for experiences in
use or awareness of specific modal inputs into MaaS packages, conditioning the utility expressions defining
the attributes of each MaaS and pay-as-you-go alternative. This is particularly important in a preference model
that will be updated over time on a regular basis as people get exposed to these new opportunities which will
modify experience and hence can be expected to have a significant impact on preference towards MaaS. What
may be a meaningful preference regime today may very likely be unreliable in the near and far future as more
and more MaaS plans are rolled out. Ongoing research must involve regular surveys in order to capture this
changing experience setting, guiding ongoing assessment of the changing demand for MaaS plans and how it
might impact on the future demand for choosing to use and pay for a single mode outside of a MaaS plan.
5 In this relatively new research space where empirical research is emerging, we need to acknowledge the important
parallel contributions of Kamargianni and Matyas at UCL (e.g., Kamargianni et al 2016 and Karmargianni and Matyas
2017) UK; Nelson at Aberdeen University, UK (e.g., Ramazzotti et al.); Kaarlson, Smith, Sochor, and Stromberg at
Chalmers University, Sweden (e.g., Sochor et al., 2015a); Timmermans, Caiati at Eindhoven University, The Netherlands
(Ebrahimi et al., 2018); Meurs, Jittrapirom, Marchau at Rambould University, The Netherlands (Jittrapirom et al., 2017 )
and Alonso-Gonzalez at TU Delft, The Netherlands (Jittrapirom et al., 2017); and Fijigaki, Takami at Tokyo University,
Japan (Fujigaki et al., 2018) and the many others who have contributed to this knowledge. Many of these authors have
contributed papers to the Special Issue on MaaS and Intelligent Mobility, to appear in Transportation Research Part A in
2019.

21
As a first detailed effort of studying a sample of individual preferences for MaaS subscription plans over the
status quo and pay-as-you-go, we have learnt a lot about what can be done differently in the next phase of
research. Specifically, we will allow for the MaaS packages to be subscribed to by households since, for
example, the ownership of a car is more likely to be a household-based decision. It might also be important to
allow groups of any form to be the basis of a MaaS plan (families, friends, etc.) rather than just individuals,
but maybe with a limit on the number of registered users (with the limit being one aspect that is tested). A
closer look at the future role of the car will take on greater interest, especially as we move to autonomous
vehicles, which is expected to change the cost of accessing a car in the MaaS plan either as a shared vehicle
(in the sense of sharing with others) or as a use alone option. We need to consider whether the first move out
of the private car to the non-owned car available in a MaaS plan might be through the second car and not all
cars in a current private fleet. In addition to individuals and groups of individuals, we need to look at prospects
MaaS subscriptions targeted at organisations. It is also worth noting that the SP experiment follows the MaaS
plans that are in place elsewhere in the world in assuming that people are willing to share, and that car-sharing
and ride-sharing (e.g., UberPOOL) would be acceptable. It is clear that the population at large needs to be
more prepared to share resources in the future and that this would likely have a detectable difference on the
uptake of MaaS plans.
As the concept of MaaS plans become more widespread and understood, it will become easier to run large
scale surveys where the travel behaviour of the sample is closer to population behaviour. Whilst the current
work is based on convenience sampling, only time will reveal whether differences in the travel behaviour of
the sample will be mirrored in the population or whether differences will be carried through to the take up of
MaaS plans. In the short term, however, the willingness-to-pay estimates will act as a guide to suppliers
designing MaaS bundles for market.
The focus of this paper has been on the preferences of users; however, we have commenced an investigation
into the preferences of suppliers of modal services into MaaS subscription plans. These plans will be provided
by what we call brokers but also known (in the UK) as aggregators. A stated preference survey is presently in
field studying the preferences of potential suppliers to MaaS schemes, be they established transport operators
or non-mobility providers including entrepreneurs interested in managing and securing modal supply. We also
want to understand the extent to which particular types of MaaS plans might be delivered under an
economically deregulated market model or whether they may be controlled to some extent by government
through competition for the market using competitive tendering. While we currently believe that many
governments would desire to encourage private enterprise to deliver greater mobility services to the market,
time will tell whether the MaaS models on offer align with the objectives of government in representing the
broader interests of society, which might include using MaaS as a way of providing improved public transport
for a lower subsidy outlay than at present, and avoiding the growth in car-based travel, with its implications
on congestion and urban efficiency. Clearly the latter phenomenon will be less likely to occur with shared cars
(and reduced private ownership) accessed through MaaS but it may place the use of mass transit at risk in a
way that government is not prepared to allow to occur.
References
Atkins 2016. Journeys of the Future: Introducing Mobility as a Service.
Bech, M. & Gyrd-Hansen, D. 2005. Effects coding in discrete choice experiments. Health Economics, 14,
1079-1083.
Brendel, A. B. & Mandrella, M. 2016. Information Systems in the Context of Sustainable Mobility Services:
A Literature Review and Directions for Future Research.
Cubic Transportation Systems 2016. Mobility as a service: Moving towards demand-based travel.
Datson, J. 2016a. Developing Intelligent Mobility and exploring Mobility as a Service.
Datson, J. 2016b. Mobility as a Service: Exploring the Opportunity for Mobility as a Service in the UK.
Delbosc, A. & Currie, G. 2013. Causes of Youth Licensing Decline: A Synthesis of Evidence. Transport
Reviews, 33, 271-290.
Dotter, F. 2016. Mobility-as-a-Service: A New Transport Model.
Ebrahimi, S., Sharmeen, F. & Meurs, H. Innovative Business Architectures (BAs) for Mobility as a Service
(MaaS)—Exploration, Assessment, and Categorization Using Operational MaaS Cases. Transportation
Research Board 97th Annual Meeting, 7-11/01/2018 2018.

22
Fujigaki, Y., Takami, K., Parady, G. T. & Harata, N. The Multicycle Model: a Modeling Framework to
Evaluate MaaS Contract Schemes. Transportation Research Board 97th Annual Meeting, 7-11/01/2018
2018.
Giesecke, R., Surakka, T. & Hakonen, M. Conceptualising Mobility as a Service. Ecological Vehicles and
Renewable Energies (EVER), 2016 Eleventh International Conference on, 2016. IEEE, 1-11.
Goodwin, P. & Van Dender, K. 2013. ‘Peak Car’—Themes and Issues. Transport Reviews, 33, 243-254.
Guiltinan, J. P. 1987. The price bundling of services: A normative framework. The Journal of Marketing, 74-
85.
Hazel, G. 2017. Smart Mobility Network Integrator Phase 2
Heikkilä, S. 2014. Mobility as a Service-A Proposal for Action for the Public Administration, Case Helsinki.
Hensher, D. A. 2017. Future bus transport contracts under a mobility as a service (MaaS) regime in the digital
age: Are they likely to change? Transportation Research Part A: Policy and Practice, 98, 86-96.
Hensher, D. A. & Ho, C. Q. 2016. Experience conditioning in commuter modal choice modelling—Does it
make a difference? Transportation Research Part E: Logistics and Transportation Review, 95, 164-176.
Hensher, D. A., Rose, J. M. & Greene, W. H. 2015. Applied Choice Analysis, Cambridge, Cambridge
University Press.
Ho, C. & Mulley, C. 2015. Intra-household Interactions in tour-based mode choice: The role of social,
temporal, spatial and resource constraints. Transport Policy, 38, 52-63.
Ho, C., Mulley, C., Shiftan, Y. & Hensher, D. 2016. Vehicle value of travel time savings: Evidence from a
group-based modelling approach. Transportation Research Part A: Policy and Practice, 88, 134-150.
Hu, X., Giang, N. K., Shen, J., Leung, V. C. & Li, X. Towards mobility-as-a-service to promote smart
transportation. Vehicular Technology Conference (VTC Fall), 2015 IEEE 82nd, 2015. IEEE, 1-5.
Jittrapirom, P., Caiati, V., Feneri, A.-M., Ebrahimigharehbaghi, S., Alonso-González, M. J. & Narayan, J.
2017. Mobility as a Service: A Critical Review of Definitions, Assessments of Schemes, and Key
Challenges. Urban Planning, 2, 13-25.
Jittrapirom, P., Marchau, V. A. W. J. & Meurs, H. Dynamic adaptive policymaking for implementing mobility
as a service (MaaS). European Transport Conference, 4-6/10/2017 2017 Barcelona, Spain.
Kamargianni, M., Li, W., Matyas, M. & Schäfer, A. 2016. A critical review of new mobility services for urban
transport. Transportation Research Procedia, 14, 3294-3303.
Kamargianni, M. & Matyas, M. 2017. The Business Ecosystem of Mobility-as-a-Service. 96th Transportation
Research Board (TRB) Annual Meeting. Washington DC, United States.
Maat, K. & Timmermans, H. Household Car Ownership in Relation to Residential and Work Location.
Transportation Research Board. 86th Annual Meeting, CD-ROM, 21-25 January 2007 Washington D.C.
Matyas, M. & Kamargianni, M. 2017. Stated Preference Design for Exploring Demand for “Mobility as a
Service” Plans. International Conference of Choice Modelling. Cape Town, South Africa.
Mukhtar-Landgren, D., Karlsson, M., Koglin, T., Kronsell, A., Lund, E., Sarasini, S., Sochor, J. L. & Wendle,
B. 2016. Institutional conditions for integrated mobility services (IMS): Towards a framework for
analysis.
Mulley, C. 2017. Mobility as a Services (MaaS)—Does it have critical mass? Transport Reviews, 37, 247-251.
Muoio, D. 2016. Here are all the companies racing to put driverless cars on the road by 2020. Business Insider.
Newberg, M. 2015. Uber may be encouraging delay of car purchases: Survey. CNBC.
New Zealand Transport Agency, 2013. Economic Evaluation Manual. Available at
http://nzta.govt.nz/resources/economic-evaluation-manual/economic-evaluation-manual/docs/eem-
manual.pdf
Planning, S. E. 2012. Benefit-Cost Analysis of car share within the City of Sydney. Available at
http://www.cityofsydney.nsw.gov.au/__data/assets/pdf_file/0012/122502/CarShareEconomicAppraisa
lFINALREPORT.pdf
Ramazzotti, D., Ambrosino, G., Nelson, J. D. & Boero, M. From the Concept of Flexible Mobility Services to
the ‘Shared Mobility Services Agency’. Paratransit: Shaping the Flexible Transport Future.
Rantasila, K. 2015. The impact of mobility as a service concept to land use.
RethinkX 2017. Rethinking Transportation 2020-2030: The Disruption of Transportation and the Collapse of
the Internal-Combustion Vehicle and Oil Industries.
Rose, J. M., Beck, M. J. & Hensher, D. A. 2015. The joint estimation of respondent-reported certainty and
acceptability with choice. Transportation Research Part A: Policy and Practice, 71, 141-152.
Scott, D. M. & Axhausen, K. W. 2006. Household Mobility Tool Ownership: Modeling Interactions between
Cars and Season Tickets. Transportation, 33, 311-328.

23
Significance, University, V., Bates, J., TNO, NEA, TNS, NIPO & PanelClix 2012. Values of time and
reliability in passenger and freight transport in The Netherlands.
SGS Economics & Planning 2012, Benefit-cost analysis of car share within the city of Sydney, Sydney,
TCOCO, Sydney, Australia.
Sochor, J., Karlsson, I. M. & Strömberg, H. 2016. Trying Out Mobility as a Service: Experiences from a Field
Trial and Implications for Understanding Demand. Transportation Research Record: Journal of the
Transportation Research Board, 57-64.
Sochor, J., Strömberg, H. & Karlsson, I. C. M. 2015a. Implementing Mobility as a Service. Transportation
Research Record: Journal of the Transportation Research Board, 2536, 1-9.
Sochor, J., Strömberg, H. & Karlsson, I. C. M. 2015b. Implementing Mobility as a Service: Challenges in
Integrating User, Commercial, and Societal Perspectives. Transportation Research Record: Journal of
the Transportation Research Board, 2536, 1-9.
Swait, J. & Adamowicz, W. 2001. Choice Environment, Market Complexity, and Consumer Behavior: A
Theoretical and Empirical Approach for Incorporating Decision Complexity into Models of Consumer
Choice. Organizational Behavior and Human Decision Processes, 86, 141-167.
The Economist. 2016. Transport as a service: It starts with a single app. September 29.
TravelSpirit 2017. Open or closed? The case for openness in Mobility as a Service.
Wong, Y. Z., Hensher, D. A. & Mulley, C. 2017. Emerging transport technologies and the modal efficiency
framework: A case for mobility as a service (MaaS). 15th International Conference on Competition and
Ownership in Land Passenger Transport (Thredbo 15). Stockholm, Sweden.