Drivers of participant’s choices of monthly mobility bundles: Key behavioural findings from the Sydney Mobility as a Service (MaaS) trial

Abstract

Australia’s first Mobility as a Service (MaaS) trial commenced in April 2019 in Sydney. A key objective of the trial is to assess interest in various MaaS subscription plans through bundling public transport, ride share, car share and car rental with varying financial discounts and monthly subscription fees, in contrast to pay as you go (PAYG). This paper develops a mixed logit choice model to investigate the participants’ choice between PAYG and four subscription plans (or bundles) that were incrementally introduced over a 5-month period. This is the first paper to model real uptake as previous studies are based on stated preference data. New evidence is provided on what role financial savings, estimated using tracking technology embedded in the MaaS app, play in the context of modal offerings and a monthly subscription fee as well as socio-demographic and seasonal effects. Behaviourally, we present evidence on the extent of take up of each bundle relative to PAYG as well as elasticity estimates for all exogenous influences and estimates of willingness to pay and scenario assessment, particularly for how much someone would have to save over a previous month’s cost outlay to be willing to subscribe to a particular bundle in a subsequent month. Within the context of the trial, the findings suggest a substantial market for mobility bundles but PAYG is an option preferred by many, particularly those with varying travel needs. We are, however, not in a position yet to conclude that these choices necessarily align with a contribution to societal sustainability goals.

Full text

1
Drivers of Participant’s Choices of Monthly Mobility Bundles: Key Behavioural
Findings from the Sydney Mobility as a Service (MaaS) Trial
Chinh Q. Ho
David A. Hensher
Daniel J. Reck
Institute of Transport and Logistics Studies (ITLS)
The University of Sydney Business School
Sydney NSW Australia 2006
Chinh.Ho@sydney.edu.au
David.Hensher@sydney.edu.au
Daniel.Reck@ivt.baug.ethz.ch
13 October 2020
Abstract
Australia’s first Mobility as a Service (MaaS) trial commenced in April 2019 in Sydney. A key objective
of the trial is to assess interest in various MaaS subscription plans through bundling public transport,
ride share, car share and car rental with varying financial discounts and monthly subscription fees, in
contrast to pay as you go (PAYG). This paper develops a mixed logit choice model to investigate the
participants’ choice between PAYG and four subscription plans (or bundles) that were incrementally
introduced over a 5-month period. This is the first paper to model real uptake as previous studies are
based on stated preference data. New evidence is provided on what role financial savings, in particular,
play in the context of modal offerings and a monthly subscription fee as well as socio-demographic and
seasonal effects. Behaviourally, we present evidence on the extent of take up of each bundle relative to
PAYG as well as elasticity estimates for all exogenous influences and estimates of willingness to pay
and scenario assessment, particularly for how much someone would have to save over a previous
month’s cost outlay to be willing to subscribe to a particular bundle in a subsequent month. The findings
suggest that there is a market of customers who are interested in selecting a subscription bundle over
PAYG when invited to join a MaaS program and indeed pay more than the expected savings for a
bundle, which suggests a viable business model. We are, however, not in a position yet to conclude that
these choices necessarily align with a contribution to societal sustainability goals.
Keywords: Mobility as a Service (MaaS), MaaS trial, Sydney, mobility bundles, Pay as you go
(PAYG), MaaS subscription, mixed logit model, willingness to pay, elasticities
Suggested citation: Ho, Q. C., Hensher, D.A. & Reck J.D. (2021) Drivers of participant’s choices of
monthly mobility bundles: Key behavioural findings from the Sydney Mobility as a Service (MaaS)
trial, Transportation Research Part C: Emerging Technologies, Volume 124, 102932,
https://doi.org/10.1016/j.trc.2020.102932.
Acknowledgments: The Sydney MaaS Trial is a project of the iMove Cooperative Research Centre
(CRC) Program. The partners in the trial are the Institute of Transport and Logistics Studies (ITLS) at
The University of Sydney Business School, Insurance Australia Group (IAG) as the industry lead
partner, SkedGo as the digital platform developer, and the iMove CRC. We are grateful for the
contributions of other members of the project team, especially Goran Smith, Andre Pinto (ITLS), Sam
Lorimer, Hugh Saalmans, David Duke, and Ivy Lu of IAG.
All Authors can confirm that there is no conflict of interest.

2
1. Introduction
Mobility as a Service (MaaS) is currently at the centre of the popular view on future collaborative and
connected mobility. MaaS operates on a concept that public and private transport services can be
integrated to provide everyday travellers a one-stop access to all services required through a common
interface. While the literature on MaaS is fast growing (see Hensher et al. 2020 for a detailed synthesis
of progress to date), many unknowns exist. These include preferences for specific subscription plans, a
viable business model for MaaS, benefits to society, and the impacts of MaaS on its user’s behaviour
which translate into changes to traffic observed on transport network.
One way to verify these unknowns beyond previous stated preference studies, which are subject to
hypothetical bias (Hensher, 2010), is to undertake a trial. Trial objectives could be many fold. Examples,
expressed in the form of research questions, could be whether integrating multiple complementary
transport services into a MaaS platform improves the travellers’ experience in terms of cost, travel time,
frequency, convenience, health benefits and perceived safety; whether MaaS contributes to
improvements in broader community benefits such as better air quality, reduced congestion and
greenhouse gas emissions; and whether MaaS could be provide a pertinent alternative to owning and
using private vehicles.
In responding to the challenge to test MaaS in a real market, the Sydney MaaS trial, a first in Australia,
was developed and designed to meet the following objectives: (1) To explore appropriate transport
service mixes for early adopters of MaaS; (2) To generate first-hand knowledge of actual MaaS
experiences; (3) to explore commercially viable business model for MaaS as a pay-as-you-go vs.
subscription plans; (4) To advance the understanding of user uptake and preferences for monthly
mobility bundles; (5) To test the ability to influence travel behaviour through introducing MaaS
solutions; and (6) To document the experience in designing, planning and undertaking a MaaS trial.
The achievement of these objectives has been reported on in detail in a number of other papers (notably
Ho et al. 2020a, and Hensher et al. 2020a,b); however we have yet to present the findings from a formal
discrete choice modelling exercise to enable us to predict the uptake of specific bundles compared to
pay as you go (PAYG) as a way to promote the use of specific modes, especially public transport, with
anticipated reductions in the use of less emission friendly modes. This is the focus of the current paper.
The paper is organised as follows. We begin with a brief overview of the literature on MaaS bundling
and previous stated preference studies. We then introduce the setting within which the Sydney MaaS
trial took place, followed by a discussion of the subscription bundles and their take up. This is followed
by a discussion of the financial savings associated with bundle that participants subscribe to. The
modelling approach is then presented followed by a descriptive overview of the data collected over the
five months of the trial, and then model results are set out and discussed, including a number of key
elasticity and willingness to pay indicators that inform us of the behavioural responsiveness of various
influences on choosing a MaaS subscription plan. We conclude with comments on what the evidence
might mean for an ongoing commitment to MaaS.
2. Literature review
The literature on MaaS subscription plans (or “MaaS bundles” to recognize its origin in the economics
and marketing literatures1) is young and quickly growing. It builds on the core idea of MaaS to innovate
access to transportation services by integration them across operational, informational and transactional
dimensions (Hensher et al., 2019; Lyons et al., 2019; Sochor et al., 2018). In fully integrated systems,
it is envisaged that users will be able to choose between ‘pay-as-you-go’ (PAYG) and monthly
subscription plans (“MaaS bundles”). The societal motivation for MaaS in general (and MaaS bundling
in specific) is to change travel behaviour from private car ownership and usage to a service-based, more
sustainable and intermodal way of travelling (Hensher and Mulley, 2020; Jittrapirom et al., 2017;
Kamargianni et al., 2016; Mulley, 2017; Wong et al., 2020).
1 For reviews of these literatures, see Kobayashi (2005) and Stremersch and Tellis (2002).

3
MaaS bundle design has recently become an active area of research due to its centrality to MaaS
business models and its potential to promote sustainable travel behaviour (for a recent review, see Reck
et al., 2020). In lack of sizeable trials, most studies to date are based on stated preference surveys aiming
to identify potential user groups, their willingness to pay, and, in result, estimate the size of the potential
market (e.g., Caiati et al., 2020; Feneri et al., 2020; Guidon et al., 2020; Ho et al., 2018; Ho et al., 2020b;
Matyas and Kamargianni, 2019; Mulley et al., 2020; Polydoropoulou et al., 2020). Results from these
studies indicate a potentially large market for MaaS bundles (e.g., almost half of all participants in the
study conducted by Ho et al., 2018, in Sydney would have bought one of the presented MaaS bundles).
Hypothetical uptake is typically found to vary substantially across the samples and to be correlated with
previous mobility tool usage (Ho et al., 2018; Ho et al., 2020b; Matyas and Kamargianni, 2019;
Polydoropoulou et al., 2020). There is still some ambiguity whether customers are ready to pay for an
integrated travel app per se (Guidon et al., 2020, find a positive willingness to pay in their study in
Switzerland, while Ho et al., 2020, find the opposite for the UK). Social influence variables and socio-
demographic profiles also appear to have an effect on bundle uptake with younger age, higher education
and higher income levels positively impacting hypothetical bundle uptake (Caiati et al., 2020; Matyas
and Kamargianni, 2020).
3. To the best knowledge of the authors, this is the first study to analyse MaaS bundle
uptake based on real purchasing decisions of trial participants and thus unique in its
explanatory power as it is not subject to hypothetical bias as previous stated
preference studies.A high-level overview of the Sydney MaaS trial setting
The Sydney MaaS trial, which commenced in April 2019 as a 2-year project, was designed to obtain
contributing evidence on whether MaaS is a value-added mobility proposition in the presence of the
existing transport options and digital platforms. Partners in the trial are the Institute for Transport and
Logistic Studies (ITLS) at the University of Sydney, The Insurance Australia Group (IAG) and SkedGo.
The trial leveraged off of unique knowledge about potential MaaS users’ preferences acquired through
previous research at ITLS, as well as IAG’s existing relationships with a wide range of transport service
providers in Sydney and a strong customer/value design focus, and SkedGo’s multimodal travel planner
TripGo, which was modified for the trial as Tripi. A graphical representation of the main components
of the trial are given in Figure 1. Details of all aspects of the trial are provided in Ho et al. (2020a) and
Hensher et al. (2020b, forthcoming).
Figure 1. The overall MaaS trial approach
•Nov. 2019 –April 2020
•Sydney, Australia
•IAG employees
•Pre-and post-trial
survey and interviews
•Public transport, Uber,
Taxi, Car share, Car
rental
•Booking data available
during trial
•Called Tripi (supplied
by SkedGo)
•Enabled trip planning,
booking, payment,
invoicing Digital
Channel
App Suppliers
Trial
Period
Customers

4
4. Monthly bundle choice and switching
The MaaS trial in Sydney, Australia, adopted an incremental approach to the design of monthly bundles
for the participants to subscribe to. Every month, cumulative data on trips made by the participants were
analysed and a new monthly bundle was designed to target a specific segment of the participants. This
incremental approach to bundle design is deemed most appropriate for two reasons. First, while some
data on individual participant’s demand for various transport services were collected in the pre-trial
survey, the insights gained from analysing this dataset were limited to the desired features of a MaaS
digital platform and socio-environmental context of the potential participants. Thus, pre-trial data are
not sufficient to form an accurate basis for how much of different transport services each participant
would require for their monthly travel. Second, participants were on-boarded gradually, and thus
introducing all monthly bundles at once would risk missing some emerging segments and create
unnecessary admin and technical work (in the backend to implement the rules) to manage these bundles.
The second reason is evidenced in Figure 2 which shows the progress of on-boarding the participants.
Most of the participants were on-boarded in November and December 2019, with a few more
participants joining the trial in January 2020. By the 20th January 2020, the on-boarding was completed
with a total of 93 participants.
Figure 2. Progress of on-boarding participants to the Sydney MaaS trial
Over the course of the five-month trial, four monthly bundles were designed and incrementally
introduced for subscription, plus a PAYG option that all participants were defaulted to for the first
month they joined the trial, whenever this would be (see the progress of on-boarding the participants
during the trial above). This default rule means that for the first month of the trial (i.e., November 2019),
PAYG was the only option available. This first month of PAYG provided important data for the MaaS
study team to conduct segmentation analysis and design monthly bundles that best suits these segments,
based on their travel demands for the various mobility services the trial offers via a smart-phone app
called Tripi.
Full details of the bundle design process are given in Ho et al. (2020a) and Hensher et al (2020b) with
some explanation included herein to ensure that each bundle’s heritage is clear. Reck et al. (2020)
provides a recent overview and a framework on / for MaaS Bundle Design that was developed based

5
on learnings from Sydney MaaS trial and the Augsburg MaaS trial. Other contributions to the small but
growing literature on bundle design in MaaS are Matyas and Kamargianni (2019), Ho et al. (2018),
Caiati et al. (2020) and Guidon et al. (2020); and in all cases the findings are based on a stated choice
experiment.
In the Sydney trial, the first monthly bundle, called Fifty50, was offered in late November and became
active on the 1st December 2020. As the first bundle, we had limited knowledge of support or otherwise
for bundles. The following logic was used in arriving at a bundle to offer. We assumed $100 worth of
incentives being available on average per person. 2 Given the goal to reduce emissions through
incentivising more sustainable travel, and public transport being the most sustainable available mode,
we suggested spending at least $50 as incentives on public transport use. $50 incentives and the weekly
$50 cap (monthly $200 cap) translates to the following pricing structure for public transport in a bundle:
free public transport would cost $200 - $50 (incentives) = $150, and a 75 % discount on public transport
would cost $200 * 0.75 – $50 (incentives) = $100. The specifics of the December holiday season such
as less working days, inferior public transport service, and the increasing need to get chauffeured rides
after parties, suggest fewer than four weeks of regular public transport and higher than usual Uber/taxi
use, and thus demand for bundles that include these modes. With a remaining budget of $50 worth of
incentives per person, a $3 reduction per Uber/taxi ride is a conservative design that enables 50/3 = ~17
rides per person before the incentive budget runs out.
We thus proposed a conservatively priced first December bundle at $50 that includes a 50% discount
on public transport using Opal cards, which are eligible smart-card tickets for buses, trains, light rails,
ferries and the on-demand transport BRIDJ buses; and a $3 discount on every Uber/taxi ride. This
Fifty50 bundle was well received with 11 subscribers taking up the bundle in December 2019 (see
Figure 3).
Figure 3. Monthly bundle subscription at the aggregate level by month
2 Based on a $50,000 allocation for financial incentives.

6
As more and more participants joined the trial, the PAYG segment becomes larger since all participants
were required to join Tripi as a PAYG user for the first month. Thus, further segmentation analysis was
conducted to identify new segments that could be targeted with new monthly bundles. Each month the
trial introduced one new monthly bundle that targeted a particular segment amongst the PAYG users,
instead of attracting the existing subscribers of other bundles. The Saver25 monthly bundle was first
introduced in late December for subscription and use in January 2020. Likewise, a GreenPass bundle
was introduced in February 2020. It is noted that once introduced, these monthly bundles remained
available for the subsequent months of the trial, except for the Saver25 which was replaced by the
SuperSaver25 in March 2020 (for an overview of all available bundles, see Figure 4).
As can be seen from Figure 3, the design of monthly bundles was quite successful in terms of attracting
PAYG users instead of the existing subscribers as the months progressed (in March 2020, a total of 43
users had subscribed to a bundle), although few switches between bundles can be observed. For example,
the Saver25 bundle was introduced in January 2020, which had 10 participants subscribed, of which
eight subscribers were previously on PAYG while two subscribers switched from the Fifty50 bundle.
Similarly, the GreenPass bundle was introduced in February 2020, which was subscribed to by 12
participants, with seven from PAYG users and five from the Fifty50 bundles. The number of subscribers
to this bundle increased to 18 participants in March 2020 when it was re-offered. Interestingly, all 12
subscribers to the GreenPass bundle in Feb 2020 stayed with this bundle in March, suggesting that these
subscribers find good value in the GreenPass bundle. This monthly bundle also attracted six new
subscribers in March 2020. Finally, SuperSaver25, that replaced the Saver25 in March 2020, had 11
subscribers, of which six were previously on the Saver25 bundle while four came from PAYG and one
from the Fifty50 bundle. It is not surprising that the majority of the SuperSaver25 subscribers were
previously Saver25 subscribers who individually received a personal phone call from the MaaS study
team informing about the changes before the communication email went to each of the participants.

7
Figure 4. Monthly plans when they were (a) first offered to participants and (b) fine-turned in March
2020
In addition to replacing the Saver25 bundle with the SuperSaver25 bundle, another minor revision took
place in March which changed the percentage discount offered in the GreenPass and Saver25 bundles
to the fixed dollar discount for Taxi and Uber trips (i.e., the 15% discount was changed to a $3 discount
for every Uber and Taxi ride). Unlike the replacement of Saver25 which was suggested by data analysis
of the previous trial months, the revision of the percent discount to a dollar discount was driven by
feedback received from the participants who took part in the mid-trial interviews who indicated that
they prefer an absolute dollar amount (Ho et al. 2020a, Hensher et al. 2020b). It became clear that most
Uber and taxi trips are relatively short, and so a $3 incentive is better value that a percentage, where the
latter may be more appealing for long trips. Details of the revisions are summarised in Figure 4 with
the top summarising the monthly bundles when they were first offered and the bottom showing the same
after fine-tuning in March 2020.
Apart from the monthly bundles, two special offers were introduced to test participants’ willingness to
use GoGet car-sharing and to reduce the cost to the environment (i.e., emissions) of their travel. More
specifically, a GoGet car-sharing credit worth $20 was offered to all participants in January 2020,
regardless of which monthly bundle they subscribed to, including PAYG, with the main goal of nudging
first time users towards trying out new modes. This can occur only once any time in January 2020 and
is unrelated to a bundle. We found that the take up of GoGet was essentially existing GoGet trips, and
hence we did not see any benefit linked to the goals of the trial, and removed the incentive associated
with GoGet car-share in all March bundles.

8
Another special offer relates to a group CO2 challenge which was introduced in March 2020 (see
Hensher et al. 2020b for full details), which aimed to test the potential use of gamification in changing
travel behaviour to obtain societal goals of MaaS in reducing emissions associated with travel. This
challenge, was posed as a group effort with an incentive to each participant of $1 for every percentage
reduction in average CO2 emission per km travelled that the entire cohort would obtain by the end of
the March, using February 2020 CO2 emission as the benchmark. In assessing the impact of the emission
challenge, we looked at the data on private car travel, ride share and GoGet but excluded Thrifty car-
rental (which was negligible in its take up). The findings as of March 8 suggest that the entire cohort
had increased CO2 emissions (kg) by 1%, and hence there is no financial reward; but with Covid-19
restrictions after March 12, we cannot conclude anything about the potential impact of such an initiative
to reduce car use. Further analyses excluded the influence of these offers from individual monthly
bundle choices and switching.
While the month-on-month changes to bundle subscriptions presented in Figure 3 are useful for
identifying the size of each travel segment, the aggregate nature of the results means that the participant
segment that each bundle should target cannot be identified. To this end, an individual analysis of
monthly bundle subscription is conducted and the results presented in Figure 5. Based on the dynamics
of bundle subscription at the individual level, key segments of travellers can be identified and their
travel patterns, before and after subscribing to a monthly bundle, can be analysed further for an
improved understanding of why these people subscribed to a certain bundle. For example, the same
dynamics of bundle subscription can be observed for participants P002, P005, P008, and P040
(SuperSaver 25); for participants P037, P072, P082, P083, P089 (Green Pass); and for PAYG users
(P011, P014, P016, P018, P019, P022, P024, etc.). That is, these participants stayed with PAYG since
joining the trial, but took up SuperSaver25 in March 2020, while ignoring its pedigree, the Saver25
bundle, which was available in January and February 2020. The Saver25 bundle is very similar to the
SuperSaver25 offer, except for the latter adding a $5 flat fare for any Uber trip that connects to/from
public transport trips and is up to 5 km. It is therefore reasonable to conclude that these people may be
attracted by the $5 flat fare for Uber trips that connect to/from public transport trips. Similarly, we can
identify the patterns for Fifty50 or GreenPass. More detailed analyses could then be performed on each
group of the participants to identify the commonalities in their travel patterns, environmental settings,
and socio-demographics that lead to the same preferences for the monthly bundle.

9
Figure 5. Temporal transitions of monthly bundle subscription at the participant level

10
5. Exploratory analysis
The initial exploratory analysis aims to identify the key drivers of monthly bundle subscriptions for the
purpose of market segmentation. More specifically, each monthly bundle is likely to serve a particular
group of users better than other groups. It is therefore useful for the business case and potential
commercial proposition to understand the size of each segment and their preferences, including
willingness to pay such that MaaS providers can design monthly bundles and assess their potential in a
given market. The exploratory analysis considers the impact of the following factors on individual
decision on monthly bundle subscription:
• Multimodality
• Monthly Opal outlay (i.e., cost of using train, bus, ferry, and BRIDJ)
• Intra-person week-to-week variation in travel demand/patternsNumber of days active on Tripi
• The bundle the participant was on in the previous month
Figure 6 displays monthly outlay for all travel modes (e.g., bus, car-rental, ferry, …) of individuals who
stayed with PAYG since onboarding. We find that many participants who stayed with PAYG are
unimodal train travellers (P017, P036, P038, P068, P070, P073, P081, P086, P098, P181, P173, P058).
This is evidenced by the fact that the monthly cost by mode for these participants is mostly single-
coloured. By contrast, bundle subscribers, particularly the Saver25 in February3, are most likely to be
multi-modal users (see Figure 7).
3 For exploratory analysis presented herein, February is selected as a regular full month of the MaaS trial when
all participants have been on-boarded and have been back to work after the holiday. The modelling analysis,
however, is able to consider all five months of the trial while accounting for disruptions such as Christmas/New
Year holiday and Covid-19 due to the multivariate nature of the adopted model.

11
Figure 6. Monthly cost breakdown for participants who stayed with PAYG since onboarding
(a) Saver25 users
(b) GreenPass users
Figure 7. Monthly cost breakdown for participants who subscribed to (a) Saver25 bundle and (b)
GreenPass bundle in Feb 2020
To explore the impact of traveller’s multimodal behaviour on monthly bundle subsection, an entropy-
index (Ho and Mulley, 2015, Cervero and Kockelman, 1997) is used as a proxy for multimodality. The

12
entropy is equal to
1
1ln( )
N
ii
i
pp
N=
∑
with pi the proportion of trips made by mode i to the total number
of trips made by each individual, and N the number of modes available in Tripi (i.e., train, bus, ferry,
taxi, Uber, GoGet, and car rental). The index has a mean of 0.349, a standard deviation of 0.256 and a
range from 0 to 0.946. Looking ahead, this index was considered in choice model estimation but was
not found to be statistically significant. Upon closer investigation, this entropy index is higher for the
Saver25 and SuperSaver25 bundles (~0.5) than for other subscription plans, including PAYG, Fiffty50
and GreenPass (~0.32), successfully measuring the multimodality behaviour of these subscribers as
observed in the exploratory analysis; however, the number of participants subscribing the Saver25 and
SuperSaver25 bundles are quite small (22 and 11 monthly bundles respectively). These small sample
sizes may well be the main reason for this index being statistically insignificant in distinguishing
individual preferences for bundles vs. PAYG.
In respect of the monthly public transport outlay (i.e., Opal cost of using train, bus, ferry, and BRIDJ),
since there are very few Uber and taxi trips per person per month, we can assume that an individual
decision to subscribe to each of the four monthly bundles is somewhat limited to the potential savings
on the Opal cost. By definition, when the monthly Opal cost outlay is smaller than $100, it is not worth
paying the membership fee in return for the discount on Opal trips. For example, at $100 per month, a
participant is indifferent between travelling as PAYG users and as Fifty50 users since the cost of
subscription to the latter is $50 and the discount received is 50% × 100 = $50. The same logic applies
for Saver25 when we ignore the benefits associated with taxi and Uber discounts offered under this
bundle (more on this below). As shown in Figure 7b, participants subscribing to the GreenPass bundle
in February 2020 have a high monthly Opal outlay, typically exceeding $125 such that it is worth it to
pay the $125 membership fee to travel free on public transport. It is also noted that these participants
also show multimodal behaviour, which can be seen by the different coloured bars within the same
participant. This is one of the reasons why multivariate analysis (i.e., choice modelling) is required to
separate the impact of potential cost savings from other factors, such as multi-modality, on bundle
subscriptions.
Intra-person week-to-week variation in travel demand/patterns are an informative way of tracking
changes through time in travel activity and what impact this may have on bundle choice. Reck and
Axhausen (2020) analyzed how intra-person week-to-week variation in travel demand impacts the
viability of MaaS bundles using RP data from Denmark. They found that stability in travel demand
indeed is an important criterion for the financial viability of individual modal components in MaaS
bundles and that week-to-week variation in shared modes (i.e., carsharing, bikesharing, taxi) is often
too high for them to be included in recurring bundles. In turn, PAYG is suggested as a more suitable
way of including them in MaaS solutions. Informal feedback from participants in the Sydney MaaS trial
suggests that variations in travel demand from one month to the next indeed has an impact on bundle
subscription. Figure 6 shows that a few PAYG participants (P070, P086, P017, P011, P098) have a
monthly Opal outlay in excess of $125 for some months, while the outlay for other months are smaller
than $100. This variation may well be one of the factors that explains why these participants chose
PAYG, even for the month that their travel cost would exceed the value that they would be financially
better off by subscribing to any bundle on offer rather than using PAYG. Two indices, referred to as
intra-personal week-to-week travel variation (IWTV) that measure the average variation in PT trips and
Uber/taxi trips are investigated in modelling to capture the impact of travel variation on bundle
subscription. The intra-personal day-to-day travel variation (IDTV) described in Stopher (2012) and Li
et al (2018) are adapted to measure IWTV.
The number of days active on Tripi is a special consideration for December and January which were
disrupted by the Christmas shutdown and New Year holiday travel when some participants were not
using Tripi for a part or all of the month for which subscription fee is payable. For the purpose of
computing this metric, a participant is said to be active if they have at least one trip recorded to the Tripi

13
or the Safer Journey4 database. Finally, the previous bundle a participant was on is a relevant metric
designed to capture an inertia effect and/or ‘forgotness’ effect, the latter relating to forgetting to change
a bundle by the first of each month, the date required to allow bundle switching, and hence becoming
stuck with the previous bundle for the current month.
6. The model form – Mixed Logit with correlated random parameters
The mixed logit model differs from the MNL model in that it assumes that at least some of the
parameters are random, following a certain probability distribution (Hensher et al. 2015). These random
parameter distributions are assumed to be continuous over the sampled population. The choice
probabilities of the mixed multinomial logit (MMNL) model,
*
,
n
P
depends on the random parameters
with distributions defined by the analyst. The MMNL model is summarised below in (1).
1
exp( )
Prob( | , , ) exp( )
ns
nsj
ns nsj n n J
nsj
j
V
choice j
V
=
= = ∑
x zv
(1)
where
Vnsj = βn′xnsj
βn = β + Δzn + Γvn
Xnsj = the K attributes of alternative j in choice situation c faced by individual n,
zn = a set of M characteristics of individual n that influence the mean of the taste parameters; and
vn = a vector of K random variables with zero means and known (usually unit) variances and
zero covariances.
The MMNL model embodies both observed and unobserved heterogeneity in the preference parameters
of individual n. Observed heterogeneity is reflected in the term Δzn while the unobserved heterogeneity
is embodied in Γvn. Structural parameters to be estimated are the constant vector, β, the K×M matrix
of parameters Δ and the nonzero elements of the lower triangular Cholesky matrix, Γ, the latter to
account for interdependencies between random parameters. The expected probability over the random
parameter distribution can be written as equation (2).
( )
**
( (|),
nn
EP P f d= Ω
∫
β
β) β) β
(2)
(| )fΩ
β
is the multivariate probability density function of
,β
given the distributional parameters
.
θ
By using a transformation of β such that the multivariate distribution becomes semi-parametrical, we
can write Equation (2) as equation (3).
( )
()
**
( | ) () ,
nn
z
E P P z z dz
βφ
= Ω
∫
(3)
(| )z
β
Ω
is a function of z with parameters
,Ω
and where
()
z
φ
is the multivariate non-parametrical
distribution of z. It is common to use several (independent) univariate distributions5 instead of using a
single multivariate distribution, such that Equation (3) can be written as equation (4).
4 Safer Journeys is a car-based program with GPS tracking technology installed to make car journeys safer, by for
example deploying an ambulance to the accident location if the driver did not answer the phone from the call
centre who recognises some sudden incidents that may have happened with the trip, based on the tracking data. A
by-product of this program is that private car use can be tracked and used as a complementary data source to
assess the success of the MaaS trial in terms of reducing emissions through reduced car kilometres.
5 Note that if one would not like to assume independent random variables, then one can sample directly from the
multivariate distribution. In case of a multivariate normal distribution, this is possible through a Cholesky
decomposition, see e.g., Hensher et al. (2015).

14
( )
()
1
**
11 1 11 1
( | ), , ( | ) ( ) ( ) .
K
n n KK K KK K
zz
E P P z z z z dz dz
β θ β θφ φ
=∫∫
  
(4)
11 1
( |,)zz
β µσ µ σ
= +
with
1
~ (0,1)zN
following a standard normal distribution, as used herein.
Note that a fixed parameter is a special case of a random parameter, such that all equations also hold in
the case that only some of the parameters are considered random. For a fixed parameter
k
β
we take
(| ) ,
kk k k
z
β µµ
=
and
( ) 1.
k
z
φ
=
Models involving multivariate distributions are generally limited to
situations in which all random parameters are assumed to be normally distributed, and involve
correlating the random parameters via a process known as Cholesky decomposition. To calculate the
elements of this matrix given
r
Ω
equations 5a-5d are utilised.
11 11
,1s kl
η
= ∀==
(the first diagonal element) else (5a)
12
,1
k
kl kl kl
k
s s kl
η
−
= − ∀=≠
∑
(all other diagonal elements) else (5b)
( ) , 1,
kl kl kl
s s k il
η
= ∀= ≠
(lower off-diagonal elements in the first column) (5c)
1
( ) , 1,
k
kl kl km ml kk
k
s ss s k k l
η
−
= − ∀≠ ≠
∑
(lower off-
diagonal elements not in the first
column)
(5d)
Once computed, the values for
k
ϖ
are determined as shown in (7), which may be rewritten as (8).
11
11
21 22
22
31 32 33
33
41 42 43 44
44
000
00 ,
0
sz
ss z
sss z
ssss z
ϖ
ϖ
ϖ
ϖ


  


  


  
=

  


  

  


(7)
1 11 1
2 21 1 22 2
3 31 1 32 2 33 3
4 41 1 42 2 43 3 44 4
,
,
,
,
sz
sz sz
sz sz sz
sz sz sz sz
ϖ
ϖ
ϖ
ϖ
=
= +
=++
=+++
(8)
where
kl
s
are parameters to be estimated, and
k
z
are draws from univariate standard Normal
distributions.
k
ϖ
are reported in Table 2 below for the diagonal and off-diagonal components of the L
matrix.
We have allowed for the panel nature of the data. The derivation of the log-likelihood functions of the
panel formulations of the MMNL model differs to those of their equivalent cross sectional forms in that
the choice observations are no longer assumed to be independent within each respondent (although the
independence across respondents assumption is maintained). Mathematically, this means that
( )
12 1 2
( ) ( ),E PP EPEP≠
hence the log-likelihood function of the panel MMNL model may be
represented as

15
( )
1
log ( ) log ,
nsj
n ns
Ny
N nsj
nsS jJ
EL E P
=∈∈

=


∑∏∏
(9)
7. Descriptive overview of the trial
The rich data available from the trial has been integrated into a format suitable for choice modelling.
We have captured data from the pre-trial travel activity survey which included socio-economic details,
residential and work locations, and modal use in terms of frequency of use. The mobility activity
captured through the digital platform Tripi during the trial was used to construct a number of measures
of travel response associated with the take up of specific bundles or the use of PAYG.
Given that we are particularly interested in how participants’ travel activity and costs outlays, especially
cost savings, influence MaaS bundle choices, we have constructed a number of measures designed to
test for candidate influences on selection of a specific bundle or PAYG, as summarised in Table 1 with
the overall set of variables in the data compiled for the analysis herein. Most notably, these are the
estimated monthly saving by subscribing to a specific bundle instead of PAYG, the percent cost outlay
compared to the previous monthly cost for each of the available modes (public transport, ride share and
car share), and the intra-personal week-to-week travel variation (IWTV) for each of the modes. In
recognising that the participant data is collected over a number of months with noticeable seasonal
variation in mobility activity, for example the December and January holiday months compared with
the February and March period, we have controlled for these effects through monthly dummy variables.
Table 1 Descriptive statistics of the sample
Variable
Mean (Standard Deviation)
number of days joined Tripi when the Communications email sent out
46.34 (27.6)
number of days joined Tripi by the end of the month when choice of bundle must be made
56.43 (25.9)
monthly cost estimated in the Communications email ($)
181.2 (127.5)
estimate monthly saving by subscribing to the bundle instead of PAYG ($)
5.93 (26.19)
cost of previous month travel ($)
337.85 (363.0)
percent cost of PT to previous monthly cost
0.727 (0.305)
percent cost of taxi/Uber to previous monthly cost
0.235 (0.270)
percent cost of GoGet to previous monthly cost
0.037 (0.154)
user's own estimate of monthly cost
174.4 (142.8)
user's own estimate of monthly saving if subscribe to a bundle
2.073 (25.45)
average IWTV for PT trips
1.811 (0.998)
standardised index of IWTV for PT trips
0.728 (0.178)
percentage of variation in PT trips to average number of weekly PT trips
0.265 (0.139)
average IWTV for taxi/Uber/GoGet trips
0.039 (0.208)
standardised index of IWTV for taxi//Uber/GoGet trips
0.033 (0.157)
percentage of variation in taxi/Uber/GoGet trips to average number of weekly taxi/Uber/GoGet trips
0.018 (0.089)
number of active days in that month
16.18 (7.81)
number of household adults (adults)
2.187 (0.823)
number of household children
0.798 (0.911)
having access to private car = 1 (caracc)
0.765
number of household drivers (hhdriv)
1.925 (0.772)
number of household cars (hhcar)
1.242 (0.786)
subscribe to Fifty50 in the previous month
0.135
subscribe to GreenPass in the previous month
0.056
subscribe to Saver25 in the previous month
0.090
has driver licence = 1
0.895
age <=24 years dummy
0.039
age 25 - 34 years dummy
0.343
age 35 - 44 years dummy
0.377
age 45 - 54 years dummy
0
age 55 - 64 years dummy
0.054
male participant dummy
0.472
Age in years
28.98 (16.31)
Households with zero cars
0.112
Households with one car
0.611
Households with two or more cars
0.263
Proportion of sample in January 2020
0.231
Proportion of sample in February 2020
0.297
Proportion of sample in March 2020
0.348
Proportion of sample in December 2019
0.123

16
The profile of the sample is summarised by a number of key indicators in Figure 8 to 10 for participants
choosing each bundle and PAYG. The number of active days represents the number of days in a month
with travel activity, and is included in the formal choice model to recognise these difference in temporal
participant rates which may influence choices made. Although the average number of active days in a
month for the trial sample is 16.18 days, the average varies from 13.67 days for PAYG through to 21.43
days for Saver25, with 17.77 days for GreenPass, 18 for SuperSaver25 and 18.75 for Fifty50. It is
interesting to note the much lower average number of days for PAYG, which may be a contributing
influence on participants’ decisions to stay with PAYG, given that the bundles are monthly
commitments. This is an important point since it appears that participants may be reticent about
subscribing to a monthly bundle when they do not anticipate regular travel over most weekdays (at
least) throughout a month. In the post-Covid-19 period, we expect this to be a concern that will require
careful thought on the subscription period for MaaS bundles with increasing working from home (see
Hensher 2020).
The socio-economic profile has some noticeable differences between the chosen bundle and PAYG.
Those who chose GreenPass and SuperSaver25 have on average 0.32 and 0.36 children respectively,
which is much lower than Fifty50 (0.90) and SuperSaver25 (1.07), and PAYG comes in at 0.79 children.
In contrast, the average number of adults in a household associated with participants choosing a bundle
or PAYG are very similar, averaging from 2.0 (SuprSaver25) to 2.4 (Saver25). The most interesting
result is the incidence of access to a car, which is very flat across all alternatives, varying from 0.73 for
GreenPass to 0.82 for Saver25 and, with similar standard deviations, might explain why this was not
statistically significant in the choice model. The number of household cars, however, does vary much
more across the chosen bundles and PAYG, from 1.05 for GreenPass to 1.57 for Saver25. The choice
model results below find the number of household cars to have an influence on the alternatives offered.
Although there are sizeable differences in the number of participants in each age category, there is very
little variation within an age category across the chosen bundles and PAYG. This may explain why, in
the choice modelling, the conversion of age to a continuous variable was the only way to engender
variation in age across the sample that was able to be related to bundle and PAYG choice.
Figure 8. Descriptive profiles of participants
The left hand side of Figure 9 shows the profile of three monthly cost indicators, namely the cost of
travel in the previous month (pmthcost), the user’s own estimate of the monthly cost outlay for the next
month (uestcost) if they were to repeat the travel patterns observed to date, and the user’s own estimate
of the monthly cost savings (uestsave) if they subscribe to a bundle instead of paying-as-they-go. The
profiles show that a participant’s estimate of cost outlays is, on average, greatest for participants who
chose Saver25 ($332.9) and the lowest for those who chose PAYG ($129.7). This is an interesting
finding, suggesting that when one has a lower travel cost outlay per month, the likelihood of choosing
a bundle decreases, possibly because the expectation of a financial savings is going to be smaller
compared to those participants who have a greater financial outlay. While intuitive, the results lend
credit to previous SP studies of Ho et al. (2018, 2020b) in that users’ current travel patterns
(multimodality, transport cost per month) is a good indicator of viability of bundles. This finding

17
provides an important indicator criterion in identifying MaaS segments. The perceived average user
savings is very small in the range of $12 for SuperSaver25 choosers and $39.87 for GreenPass, with
$17.6 for Saver25 and $15.6 for Fifty50.
The right hand side of Figure 9 shows the proportion of the modal cost relative to monthly cost of travel
for participants choosing a bundle or PAYG. As a proportion, this metric is bound between zero and
one, with the value of zero for any particular mode (or modes of the same type such as taxi and Uber)
indicating that the subscribers do not use that mode. The greatest proportion of cost outlay for public
transport is associated with the GreenPass subscribers at 87%, which is expected given that subscribers
to this bundle are heavy public transport users (and hence willing to pay the highest subscription fee of
$125 per month to travel on PT for free). By contrast, the proportion of PT cost to monthly cost is lowest
for Saver25 and its successor, the SuperSaver25 (at around 48%), while that for the Fifty50 and PAYG
users is much higher, at 77% and 72% respectively. Note that the monthly cost for PAYG users is lowest
(see Figure 9), which may explain why they stayed with PAYG even though various bundles offer
generous discounts for PT fares. For taxi and Uber, the greatest proportion of monthly month is for
Saver25 and SuperSaver25 subscribers, on average accounting for about 45% of the monthly transport
cost. This proportion is much smaller, at around 20% the Fifty50 and PAYG users, and 12% for
GreenPass subscribers. For car share (i.e., GoGet), with so few participants selecting to use GoGet, the
results are uninformative. The message in this evidence is that the choice of monthly bundle is likely to
show a significant influence from this indicator, with GreenPass users opt in for PT discounts while
Saver25 users are in for discount to ride share. We test these assumptions in the model estimation, and
results are reported below.
Figure 9. Cost indicators associated with participants choosing each bundle and PAYG
The final descriptive profile, presented in Figure 10, refers to variations in the number of modal trips
that is designed to capture impacts on bundle choice. We have developed a measure to capture the
average (m_), the standardised index (z_) and the proportional (p_) intra-personal week-to-week travel
variation for public transport, and for taxi/Uber (ride-share) and car-share (GoGet) combined. The
greater the variation, the more likely we suggest that participants might be more careful in committing
to a bundle, which in a sense is linked to variety seeking in contrast to habitual mobility behaviour.
Looking at the mean variations (mIWTV), we find on average for public transport, the greatest variation
for choosers of GreenPass and Fifty50, with the other two bundles and PAYG being similar (1.56 to
1.71), and for ride share and car share, the greatest variation is with choosers of Saver25 and
SupeSaver25, both of which were designed to focus on grow use of these modes with a lower entry
barrier (subscription cost of $25 per month). The mean variation for PAYG, Fifty50 and GreenPass for
ride share and car share is so small that it suggests habitual weekly behaviour. What this suggests, to
some extent, is that a greater variation in trip activity between weeks does tend to increase the chance
of choosing some bundles but not all bundles; however when we consider the standardised index
(zIWTV), we find almost no difference associated with public transport across all alternatives; with the
same applying for ride share and car share, except for Saver25 which exhibits sizeable standardised
variation. Finally the proportional variation of modal trips relative to the average number of weekly
modal trips is almost identical for public transport (in the range of 0.24 to 0.28); and for ride share and

18
car share the range is negligible from 0 to 0.06, with again choosers of Saver25 showing the greatest
variation.
Figure 10. Variation in monthly mode-specific trip activity for participants choosing each bundle and
PAYG
In concluding this descriptive assessment, it is important to note that the profiles above are presented at
their average levels, within participants choosing each bundle, whereas in the choice model estimation
in the next section we allow for the full distribution across all participants, which can be expected to
result in a different set of behavioural inferences.
8. Model results
The final MMNL model is summarised in Table 2. The overall goodness of fit is very impressive with
a pseudo R2 of 0.60. In selecting this model form, we also estimated other choice model forms including
nested logit, random regret, latent class, random regret latent class, error components6 and combinations
thereof, but none were found to improve on the mixed logit model with Cholesky decomposition
included to account for correlated random parameters.
Three variables are represented through random parameters, namely the estimated monthly savings by
subscribing to Fifty50, GreenPass and Saver25 (or SuperSaver 25 in March) bundles. 1,000 Shuffled7
Halton draws were used. All are statistically significant at the mean and standard deviation and of the
expected positive sign, although the mean estimate for Saver25/Supersaver25 is marginally significant
at 1.85 but with a significant standard deviation beta well above the 95 percent confidence level, with
t-value at 2.60. The estimated monthly savings to subscribing to a bundle is a very significant source of
preference heterogeneity as captured by the random parameters and their Cholesky decomposition.
This suggests that a higher estimated monthly savings has a positive impact on the probability of
choosing a bundle, which aligns with the role that financial discounts might play in promoting a
6 Including error components was problematic in its impact on other variables of interest which become
statistically non-significant.
7 Train et al. (2004) proposed a refinement of the method of Halton sequences that involves assembling the pools
of draws that are deterministic of a Markov chain, and shuffling them before using them in estimation, as a way
of further reducing imposed correlation.

19
subscription to a bundle offer. fThe Cholesky parameter estimates, accounting for the interdependence
between the random parameters, play an important role in purging the mean and standard deviation
estimates of correlation amongst the estimates. As shown in Table 2, the diagonal and off-diagonal
parameter estimates are in the main statistically significant, and indeed the model in which we assumed
independent random parameter effects resulted in some statistical insignificance for the standard
deviation beta for Saver25/SuperSave25. Without Cholesky, the random parameters have to be a
constrained normal to obtain significant SD betas, essentially constraining the full distribution in
capturing preference heterogeneity8. The pairwise correlations between the random parameter estimates
for these three bundles are 0.647 between the GreenPass and saver25/SuperSaver 25, 0.895 for Fifty50
and Saver25/SuperSaver25 and 0.893 for Fifty50 and GreenPass. The extent of the behavioural
response is discussed later when we report and interpret the elasticity estimates and simulated scenarios.
Turning to PAYG, in addition to controlling for the month of participation, we find that there are some
statistically significant socio-economic influences on the probability of choosing PAYG and to stay
with it over time. Male participants are associated with a statistically significant positive parameter
estimate, suggesting that they have a greater propensity to choose PAYG compared to female
participants, and interestingly, households with one car and with two plus cars, compared to zero car
households, both have a lower probability of choosing PAYG compared to choosing a bundle. This is
an interesting finding given that it was thought, a priori, that households with cars might be less inclined
to choose a bundle which offers mobility options that are likely to reduce the need to use household
cars. The evidence suggests this is not the case, and knowing that having more cars in the household is
not a negative influence on subscribing to a bundle is a very powerful finding.
We also found that participant age has a statistically significant negative influence on the probability of
choosing PAYG, and hence it is expected that older participants have a lower probability of choosing
PAYG compared to young participants. We might speculate that younger participants in general have
greater flexibility in their travel activity needs, described as less fixed in their ways, compared to older
people, and are much more satisfied to pay as they go. However, as shown in Figure 11, we see that the
relatively very young (12 persons in the sample) have a low propensity to choose PAYG, in contrast to
participants in their 30s (13 participants) and 40s (11 participants); however there is overall a stronger
propensity to choose a bundle, particularly participants in the 25 – 34 age bracket. The 16 participants
over 65 also generally preferred a bundle to PAYG. We retained the single age variable after testing a
set of dummy variables for age range (Figure 11), which were found to be statistically insignificant.
8 It is interesting to note that without accounting for the interdependencies between the random parameters, we
may have artificially adopted a constrained distribution to obtain statistically significant parameter estimate while
denying the full representation of the distribution of preference heterogeneity.

20
Figure 11 Choice of monthly bundle by age groups
We also investigated other available socioeconomic variables such as car access and licence holding
(Table 1), and none of them had a statistically significant influence in the utility expressions of any of
the five alternatives.
Table 2. Final Mixed Logit Model of monthly bundle choice
Variables
Units
Alternatives
Parameter
estimates
(t-value)
95%
confidence
interval
Mean of random parameters
Estimated monthly saving by subscribing to a bundle
$/month
Fifty50
0.0680 (3.18)
0.026 to 0.1099
GreenPass
0.1080 (4.38)
0.059 to 0.156
Saver25,
SuperSaver25
0.0657 (1.85)
-0.004 to 0.135
Standard deviation random parameters (normal distribution)
Estimated monthly saving by subscribing to a bundle
$/month
Fifty50
0.0919 (3.73)
0.044 to 0.140
GreenPass
0.0787 (4.31)
0.043 to 0.115
Saver25,
SuperSaver25
0.1479 (2.60)
0.036 to 0.260
Non-random parameters
Male
1,0
PAYG
1.0146 (2.25)
0.129 to 1.900
Age
years
PAYG
-0.0391 (-2.63)
-0.068 to -0.010
1 car in household
1,0
PAYG
-3.0372 (-2.47)
-5.44 to -0.630
2 or more cars in household
1,0
PAYG
-4.326 (3.08)
-7.080 to -1.580
January monthly participation
1,0
PAYG
1.8884 (2.22)
0.218 to 3.558
February monthly participation
1,0
PAYG
1.4454 (1.92)
-0.027 to 2.918
March monthly participation
1,0
PAYG
1.8123 (2.22)
0.214 to 3.411
# days joined Tripi by end of month < 21 days
1,0
All Bundles
1.8581 (2.19)
0.192 to 3.524
Bundle specific constant
Bundle specific constant
Bundle specific constant
Bundle specific constant
Fifty50
-6.5942 (-3.72)
-10.00 to -3.121
GreenPass
-10.824 (-3.48)
-16.92 to -4.730
Saver25
-5.835 (-3.37)
-9.228 to -2.441
SuperSaver25
-4.9977 (-2.87)
-8.391 to -1.584
Number of active days in the month
#days/month
Fifty50
0.0814 (2.07)
0.004 to 0.158
Number of active days in the month
#days/month
GreenPass
0.1591 (2.36)
0.027 to 0.291
Cost of previous month travel
$/month
All bundles
0.0045 (2.24)
0.001 to 0.009
Proportion of cost of PT to previous month cost
0 to 1
GreenPass
2.7481 (1.60)
-0.708 to 6.20
Proportion of Uber/Taxi cost to monthly cost
0 to 1
Saver25,
SuperSaver25
2.3075 (2.10)
0.053 to 4.56
Proportion of carshare cost to monthly cost
0 to 1
Saver25,
SuperSaver25
6.7628 (2.95)
2.274 to 11.251
Proportion of variation in PT trips to average weekly PT
trips
0 to 1
Saver25,
SuperSaver25
-3.5368 (-1.60)
-7.910 to 0.836
Diagonal values in Cholesky matrix, L
Fifty50
Fifty50
0.9186 (3.73)
0.043 to 0.141
GreenPass
GreenPass
0.0354 (1.72)
-0.005 to 0.076
Saver25/SuperSaver25
Saver25
0.0432 (1.40)
-0.019 to 0.106
Below diagonal values in L matrix
Fifty50:GreenPass
0.0703 (3.38)
0.030 to 0.111
Fifty50:Saver25/SuperSaver25
0.1323 (2.65)
0.034 to 0.231
GreenPass:Saver25/SuperSaver25
-0.0499 (-0.96)
-0.152 to 0.052
Covariances of random parameters
Fifty50
0.0065 (1.89)
-0.000 to 0.013
GreenPass
0.01215 (1.70)
-0.002 to 0.026
Saver25/SuperSaver25
0.0075 (1.12)
-0.006 to 0.021
Goodness of Fit
Log-likelihood at convergence
-195.83
Restricted log-likelihood
-489.27
McFadden Pseudo R2
0.600
AIC/sample
1.473
Sample size
304
We wanted to see if the participants follow a suggestion made to them of not taking a bundle if they
joined Tripi less than a certain duration such as two weeks by the end of the month when choice of
bundle must be made. We found that a dummy variable distinguishing up to three weeks from more
than three weeks worked best and was statistically significant in contrast to periods of one or two weeks.
The duration of this effects was extensive, ranging for 3 days to 102 days, with a mean of 56 days.

21
The number of days active in a month is a statistically significant conditioning effect since we expect
and find that the more active days in a month, the higher the probability of choosing Fifty50 and
GreenPass bundles compared to PAYG, Saver 25 and SuperSaver25. The cost of the previous month’s
travel is significant and positive in all of the bundle alternatives, suggesting that the more the participant
spends in a previous month, the more likely they are to take up a bundle. This is a very important
finding, indicating that trial participants are clearly hoping to benefit by bundle subscription, especially
when the monthly transport cost outlay, and hence the potential saving, justifies the subscription fees
of at least one monthly bundle.
An interesting variable investigated relates to the percent cost of a specific mode cost relative to the
previous monthly cost in a bundle. The mean values in the sample are 72.7%, 23.5% and 3.7%
respectively for public transport, ride share and care share. The evidence supporting this expectation
varies across the bundles, which is encouragingly consistent with our effort to ensure that we offered
up, incrementally, various bundles that did not necessarily dilute the take up of other bundles. We found
that the percentage of the cost of public transport relative to the previous month’s cost for the GreenPass
was positive and statistically significant, suggesting that again with the high incidence of public
transport use under free travel offered by this bundle, the cost gain is sizeable and attractive in the
selection of GreenPass. In Saver25 and SuperSaver25, however, we found statistically significant
parameters for the percent cost of taxi/Uber and also GoGet compared to the previous month, with
positive signs and similar magnitudes, suggesting that, ceteris paribus, when the share of monthly costs
associated with the use of flexible modes (taxi, Uber, and GoGet which are generously discounted in
the Saver25 and SuperSaver25 bundles) increases, participants prefer to choose Saver25 and
SuperSaver25 bundles compared to the PAYG, Fifty50 and GreenPass options. This suggests that the
lower entry barrier of the Saver25 and its successor, the SuperSaver25 are appealing to travellers who
rely more on taxi, Uber and GoGet than on public transport and who plan to increase their use of ride
share and car share, relative to public transport use in the previous month. This may potentially be
problematic in the context of promoting sustainable solutions where we would want to see a growth in
public transport use; however since the use of the private car is missing in this comparison, it is not
possible to conclude that this position is supporting or not sustainability objectives. Hensher et al.
(2020a) investigated this matter for the subsample of participants who also subscribed to IAGs Safer
Journeys program which captured car use; however while that research did not distinguish between the
bundles it did conclude that a subscription to a monthly mobility bundle does influence monthly car use
in a statistically significant way. Overall, those participants who use public transport more often have
shown, through modelling evidence, strong preferences for the Fifty50 and the GreenPass bundles,
which offer greater discounts on PT.
The final variable that was statistically significant and associated with Saver25 and SuperSaver 25 is
the percentage of variation in public transport trips around the average number of weekly public
transport trips. This variable, tested for each mode initially, is designed to establish whether variations
in modal trip activity impact on the choice of a bundle compared to PAYG as discussed in a previous
section. For each of public transport, ride share and car share, we considered three specifications: the
average intra-personal day-to-day travel variation (IWTV) for specific modal trips, the standardised
index of IWTV for such trips, and the percentage of variation in modal trips relative to the average
number of weekly trips by that mode. Only one effect was identified as statistically significant,
suggesting that a greater percentage variation around the mean for public transport trips tends to reduce
the probability of choosing Saver25 and SuperSaver25 compared to PAYG and the other two bundles.
Despite the statistical significance of the parameter estimates discussed in Table 2, care must be taken
in interpreting the numerical magnitude of each parameter estimate since they are non-comparable in
this logit non-linear form, and hence below we present elasticities as a way of meaningfully comparing
the impacts of each bundle component. Elasticities are more informative since they combine the
parameter estimates with the levels of the variables and the choice probability outcomes to reveal
behavioural responsiveness of interest in the probability of selecting a bundle. For the logit form, the
point (or marginal) elasticity of the probability is given in equation (10), and the marginal effect in
equation (11).

22
log ( |) (|) marginal effect
log (|) (|)
kk
kk
xx
Eyx Ey x
x Ey x x Eyx
∂∂
= =
∂∂
⋅⋅
(10)
''
'
(|) () () ''
'() f()
()
xx
x
E y x F dF xx
F
xx
d
ββ ββ
β ββ
β
∂∂
= = = =
∂∂
(11)
The point elasticity estimates of interest are summarised in Table 3. All the elasticities are statistically
significant at the 95 % confidence interval or better9. In interpreting Table 3, an elasticity is with respect
to a change of X in row choice on column choice – the probability of choosing a specific bundle or
PAYG; and direct elasticities are shared in grey and non-shaded values are cross-elasticities. As an
example of the interpretation of the elasticity estimates, the estimated monthly saving by subscribing to
a bundle ($/month) is as follows: a 1 percent (or marginal) increase in the estimated savings associated
with choosing the Fifty50 bundle results in a 0.7332 increase in the probability of choosing Fifty50.
The largest switch of another bundle or PAYG is from Saver 25. Likewise a 1 percent increase in
estimated savings associated with choosing the GreenPass bundle results in a 0.8423 increase in the
probability of choosing GreenPass and 1.0045 in the probability of choosing Saver25. Extrapolating to
non-marginal changes is risky and is discussed later in this section where we undertake simulated
scenarios and report the implied direct arc elasticity as a way of accounting for behavioural responses
associated with large non-marginal changes.
In reviewing all the findings in Table 3, with a few exceptions they are relatively inelastic (i.e., less than
1.0). The exceptions are 1.1349 for the number of days a participant joined Tripi by end of month when
choice of bundle must be made for GreenPass, 1.0045 for the estimated monthly saving, and 1.2387
and 1.7778 respectively for the cost of previous month travel associated with Saver25 and
SuperSaver25. The great majority of the direct point elasticities less than unity are in the range between
0.5 and 0.85 regardless of sign. Since we have no extant literature to draw on, given the uniqueness of
this modelling activity, there is no way of establishing the portability and reinforcement of the empirical
evidence. What we can conclude is that there is significant behavioural responsiveness to changes in
the levels of almost all the explanatory variables, which suggests that the choice of a specific bundle
and PAYG can be quite volatile and hence, unlike other modal situations where manipulating travel
costs and times have a far lesser behavioural responsiveness (i.e., sensitivity), this is not the case in
MaaS. This is an important finding for designing and pricing MaaS bundles in practice as it suggests
that modular designs (as trialled in the UbiGo project in Sweden), where each individual / household
can design their own MaaS bundle, might be superior to pre-defined MaaS bundles in exploiting the
multiplicity of different willingness to pay levels for profit maximization.
This noticeably high (relative) behavioural response provides very encouraging signs for developing
MaaS subscription bundles that will appeal to the market. To take one example, if we can identify a
cost saving of 10 percent associated with a bundle compared to what people are currently outlaying,
assuming this can be interpreted as marginal, then the probability of choosing each of the bundles we
have designed and offered increases by 7.3 percent for Fifty50, 10 percent for Saver 25, 8.4 percent for
GreenPass and 8.8 percent for SuperSaver25. Likewise, if we can identify individuals that exhibit a
greater incidence of habitual behaviour in respect of weekly public transport trips, at least reducing
variation by 10%, which is currently on average at 26.5%, the probability of choosing a bundle increases
by 5.2% for Saver25 and 58.3% for SuperSaver25.
Table 3. Summary of direct and cross elasticities
9 These results are available on request, including t-values and 95% confidence intervals. In computing elasticities,
it is complex to compute the asymptotic standard errors. In NLOGIT, used in model estimation, we have developed
approximation method using the Krinsky and Robb method to obtain the standard error estimates, enabling
derivation of t-values and confidence levels. See Hensher et al. (2015) for more details.

23
PAYG
Fifty50
Saver25
GreenPass
SuperSaver25
Estimated monthly saving by subscribing to a bundle ($/month)
Fifty50
-0.0758
0.7332
-0.5136
-0.4332
-0.3437
Saver25
-0.0224
-0.1811
1.0045
-0.0599
-0.4455
GreenPass
-0.0291
-0.2933
-0.1401
0.8423
-0.1957
SuperSaver25
-0.0028
-0.0542
-0.2335
-0.0367
0.8788
Number of days joined Tripi by end of month
Fifty50
-0.1321
0.7189
-0.2884
-0.2728
-0.2616
GreenPass
-0.0796
-0.2993
-0.1309
1.1349
-0.2211
Cost of previous month travel ($/month)
Fifty50
-0.1083
0.7697
-0.4216
-0.3248
-0.5017
Saver25
-0.0428
-0.1553
1.2387
-0.0644
-0.6854
GreenPass
-0.0427
-0.1822
-0.0981
0.6902
-0.1898
SuperSaver25
-0.025
-0.0968
-0.3592
-0.0653
1.7778
Percent cost of Ride Share (Uber, Taxi) to previous month cost
Saver25
-0.0302
-0.0674
0.553
-0.0179
-0.1695
SuperSaver25
-0.0092
-0.027
-0.0889
-0.013
0.4951
Percent cost of Car Share (GoGet) to previous month cost
Saver25
-0.0128
-0.0122
0.2432
-0.0045
-0.1821
SuperSaver25
-0.0059
-0.0039
-0.0954
-0.0038
0.3083
Percent of variation in PT trips to average # weekly PT trips
Saver25
0.0287
0.0637
-0.52
0.0214
0.1401
SuperSaver25
0.0136
0.0321
0.0735
0.02
-0.5825
It is clear, despite the limited sample size, that there is interest in MaaS subscription plans if they are
designed to appeal to the market (Table 4). The sample uptake for each of the bundles and PAYG is,
respectively for PAYG, Fifty50, GreenPass and Saver 25, for February10 as the last full month before
Covid-19 stopped most travel activity, 52.5%, 16.3%, 16.23% and 15%, representing 47.5% preference
for one of the offered bundles. The estimated MMNL model (but not calibrated) has been able to come
close to predicting this relative uptake, with the equivalent mean choice shares being 49.5%, 18.7%,
10.8% and 21%, giving a total of 50.5% preferring to choose a MaaS subscription plan.
Table 4. Probability of choosing PAYG and each bundle in February 2020
Choice of alternative
Months available
Predicted probability
(actual choices)
PAYG
All months
0.495 (0.525)
Fifty50
December to March
0.187 (0.163)
GreenPass
February, March
0.108 (0.163)
Saver25
January, February
0.210 (0.150)
SuperSaver25
March
Revised Saver25 N/A February
The marginal rate of substitution (MRS) between two specific influences of interest are now discussed,
namely monthly cost savings and the previous month’s cost outlay. These are the equivalent of a
willingness to pay or accept measure. We used the delta test (Hensher et al. 2015) to obtain the standard
errors for each MRS estimate. The mean estimate of $25.18 (t-value of 1.89) suggests that, ceteris
paribus, a representative individual is willing to subscribe to a Fifty50 bundle (priced at $50) in a
particular month if it saves, on average, $25.18 per month over the previous months cost, although the
95% confidence interval of -.97 to 51.34 suggests a range in which willingness to subscribe exists as
lower and higher estimates. The equivalent MRS for GreenPass (priced at $125) is $32.55 (t-value of
10 We have chosen February as the last full month given that after March 20, with the onset of restrictions
associated with Covid-19, travel activity reduced substantially and we provided a 50% fee reimbursement of the
subscription fee for the balance of March for participants who had subscribed to a bundle.

24
2.07) and for Saver25 (priced at $25) it is $30.91 (t-value of 1.65). The 95% confidence intervals,
respectively for Fifty50, GreenPass and Saver25, are -$0.97 to $51.3, $.79 to $63.1, and -$5.78 to $67.6.
While the MRS is on value, in order to identify the extent to which changes in monthly financial savings
over the previous month influence the probability of choosing each of the bundles and PAYG, we
simulated a scenario in which we varied the savings from -$200 to $200 with a $10 increment. The
results, shown in Figure 14, show very clearly the extent of switching to a bundle away from PAYG as
the savings increases in the positive range. As might be expected, a negative savings is associated with
a higher probability of choosing or staying with PAYG, although there is a small amount of bundle
uptake, presumably for other reasons which were identified in the mid-trial and end of trial surveys as
including curiosity and potential convenience in using a single app (Hensher et al. 2020b) and the flat
rate effect (Reck et al., 2021). When the savings are positive and increase, we see an increase in the
probability of choosing a bundle, with the greatest probability increase being for Fifty50, followed by
GreenPass, with Saver25 flattening out once the savings reached around $100 per month. Note that the
largest cost saving in the trial is $81 per month (Figure 15), with the greatest incidence of a positive
saving in the range of $10 to $50. There are also a significant number of negative savings which equate
with the dominance of choosing PAYG over that range, and which would have to be reviewed in a
continuing MaaS program. This is an important finding as it indicates that there might be a business
model in MaaS bundle provision when the mean MRS (e.g., $32 for GreenPass) plus provider subsidies
($50) is lower than its market price ($125).
The evidence suggests that financial savings do attract interest and take up of a bundle, and that the
savings do not have to be overly large per month to influence this outcome. An important message from
the trial is the critical importance of providing feedback to subscribers on a very regular basis of the
amount of money outlaid in a month, and the potential savings associated with undertaking the exact
same trips through subscription to a bundle plan. Cost feedback was identified by participants as a very
positive feature of the trial, revealing cost outlays that many had never realised they made.
Figure 12. Simulated scenario impact of estimated monthly savings ($) by subscribing to a bundle vs. PAYG

25
Figure 13. The distribution of monetary savings between months for all bundles
Finally, the simulated scenarios have an underlying implied arc elasticity which can differ from the
point elasticities in Table 3, given the non-marginal numerical magnitude of the changes in monthly
cost savings. The equivalent arc direct elasticities (with point elasticities from Table 3 in brackets) are
0.418 (0.733) for Fifty50, 0.016 (1.00) for Saver25, 0.313 (0.842) for GreenPass, and 0.151 (0.879).
These differences are substantial and are a reminder that point elasticities are valid at the margin for
very small changes, but are unreliable for non-marginal large changes. Specifically, in the case of
changes in monthly cost savings, a point elasticity over-estimates the behavioural response by a
significant amount, and hence must not be used for such large changes.
9. Conclusions
This paper is the first revealed preference study, in contrast to earlier stated preference studies, to
develop a choice model to assess the interest in MaaS subscription bundles compared to PAYG.
Tracking choices made in the Sydney MaaS trial, we have been able to obtain rich data on a monthly
basis to establish the preferences of a sample of trial participants for four incrementally developed
bundles introduced into the market against the fall back or default option of PAYG.
The findings are a transparent contribution to a topic that has garnered huge interest and often been
subject to what is referred to as hype and rhetoric (Hensher et al. 2020). What has been missing is
rigorous evidence on whether MaaS through subscription plans or bundles, has the potential to be of
interest to segments of the population or the population more generally. While no one denies the desire
to find, through innovative mobility services, better ways of supporting sustainability goals, the question
remains as to whether MaaS is a force to do this, in a niche sense or in a scalable manner.
While the challenge is still out there, the Sydney trial has confidently shown that MaaS can develop a
market of followers provided if it has been able to do a behaviourally appealing job in the design and
offer of a range of subscription plans, and that the take up can be sizeable. It is suggested from this
study and the modelling activity, that noticeable savings in travel outlays matter and can influence the
support for MaaS bundles, provided there is enough variation in bundle offers to accommodate the
many and varied needs of the population.

26
In suggesting that MaaS does indeed have real potential to contribute to the broader goals of society
through changing travel behaviour that aligns with sustainability improvements, such as reduced
emission and congestion, there is no guarantee that this will secure a business model that in any sense
might be described as having commercial legs. This will in time be established and may require greater
financial support from government (see Hensher et al. 2020 for a detailed discussion of this broader
MaaS agenda), but we would suggest that the design and introduction of subscription bundle is essential
to the future of MaaS, and that PAYG alone is unlikely to have the capability of growing the market
and delivering sustainable outcomes. In ongoing research we are investigating the influence that specific
modes play in the take up of particular subscription plans.
Authors’ contributions
Both authors designed the MaaS trial approach and contributed equally to the writing of this paper.
Chinh Ho was responsible for the intricate extraction of data from the pre-trial surveys and the in-field
trial digital platform. David Hensher undertook the model estimation and initial interpretation of the
modelling results and simulation application. Daniel Reck contributed to the development of the overall
bundle design framework and the incremental approach. All authors undertook the overall review and
final editing of the paper.
References
CAIATI, V., RASOULI, S. & TIMMERMANS, H. 2020. Bundling, pricing schemes and extra features
preferences for mobility as a service: Sequential portfolio choice experiment. Transportation
Research Part A: Policy and Practice, 131, 123-148.
CERVERO, R. & KOCKELMAN, K. 1997. Travel demand and the 3Ds: Density, diversity, and design.
Transportation Research Part D: Transport and Environment, 2, 199-219.
FENERI, A. M., RASOULI, S. & TIMMERMANNS, H. J. P. 2020. Modeling the effect of Mobility-
as-a-Service on mode choice decisions. Transportation Letters. DOI:
https://doi.org/10.1080/19427867.2020.1730025
GUIDON, S., WICKI, M., BERNAUER, T. & AXHAUSEN, K. 2020. Transportation service bundling
– for whose benefit? Consumer valuation of pure bundling in the passenger transportation
market. Transportation Research Part A: Policy and Practice 131, 91-106. DOI:
https://doi.org/10.1016/j.tra.2019.09.023
HENSHER, D. A. 2010. Hypothetical bias, choice experiments and willingness to pay. Transportation
Research Part B: Methodological, 44, 735-752.
HENSHER, D. A. 2020. What might Covid-19 mean for mobility as a service (MaaS)? Transport
Reviews, 40 (5).
HENSHER, D. A., ROSE, J.M. & GREENE, W.H. 2015. Applied Choice Analysis, Second Edition,
Cambridge University Press, Cambridge.
HENSHER, D. A., MULLEY, C., HO, C. Q., NELSON, J., SMITH, G. & WONG, Y. 2019.
Understanding MaaS: Past, Present and Future. Paper presented at the 2nd International
Conference on Mobility as a Service (ICoMaaS), Tampere, Finland, December.
HENSHER, D. A. & MULLEY, C. 2020. Editorial to Special issue on developments in Mobility as a
Service (MaaS) and intelligent mobility. Transportation Research Part A: Policy and Practice,
131, 1-4.
HENSHER, D. A., HO, C. Q., MULLEY, C., NELSON, J. D., SMITH, G. & WONG, Y. Z. 2020.
Understanding Mobility as a Service (MaaS). Elsevier, New York.
HENSHER, D.A., HO, C. & RECK, D.J. 2020a. Mobility as a Service and private car use: evidence
from the Sydney MaaS trial, submitted to Transportation Research Part A: Policy and Practice.
HENSHER, D.A., HO., C, RECK, D.J., SMITH, G., LORIMER, S & LU, I. forthcoming, 2020b. The
Sydney Mobility as a Service (MaaS) Trial: design, implementation, lessons and the future,
including Appendices.
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.

27
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, 117, 302-318.
HO, C. Q., MULLEY, C. & HENSHER, D. A. 2020b. Public preferences for mobility as a service:
Insights from stated preference surveys. Transportation Research Part A: Policy and Practice
131, 70-90.
HO, C. Q., HENSHER, D. A., RECK, D. J., LORIMER, S. & LU, I. 2020a. MaaS bundle design and
implementation: Lessons from the Sydney MaaS Trial. Submitted to Transportation Research
Part A: Policy and Practice.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 (2), 13-25.
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.
KOBAYASHI, B. H. 2005. Does economics provide a reliable guide to regulating commodity bundling
by firms? A survey of the economic literature. Journal of Competition Law & Economics, 1
(4), 707-746.
LI, W., HOUSTON, D., BOARNET, M. G. & PARK, H. 2018. Intrapersonal day-to-day travel
variability and duration of household travel surveys: Moving beyond the one-day convention.
Journal of Transport and Land Use, 11, 1125-1145.
LYONS, G., HAMMOND, P. & MACKAY, K. 2019. The importance of user perspective in the
evolution of MaaS. Transportation Research Part A: Policy and Practice, 121, 22-36.
MATYAS, M. & KAMARGIANNI, M. 2019. The potential of mobility as a service bundles as a
mobility management tool. Transportation, 46, 1951-1968.
MULLEY, C. 2017. Mobility as a Services (MaaS) – does it have critical mass? Transport Reviews, 37
(3), 247-251.
MULLEY, C., HO, C. Q., BALBONTIN, C., HENSHER, D. A., STEVENS, L., NELSON, J. D. &
WRIGHT, S., 2020. Mobility as a service in community transport in Australia: Can it provide
a sustainable future? Transportation Research Part A: Policy and Practice, 131, 107-122.
POLYDOROPOULOU, A., TSOUROS, I., PAGONI, I., & TSIRIMPA, A. 2020. Exploring Individual
Preferences and Willingness to Pay for Mobility as a Service. Transportation Research Record,
0361198120938054.
RECK, D. J. & AXHAUSEN, K. W. (2020) How Much of Which Mode? Using Revealed Preference
Data to Design Mobility As a Service Plans. Transportation Research Record 2674 (7), 494-
503.
RECK, D. J., AXHAUSEN, K. W., HENSHER, D. A. & HO, C. Q. 2021. Multimodal Transportation
Plans: Empirical Evidence on Uptake, Usage and Behavioral Implications from the Augsburg
MaaS Trial. Submitted for presentation at the 100th Annual Meeting of the Transportation
Research Board, Washington, January.
RECK, D. J., HENSHER, D. A. & HO, C. Q. 2020. MaaS Bundle Design. Submitted to Transportation
Research Part A: Policy and Practice.
SOCHOR, J., ARBY, H., KARLSSON, I. C. M. & SARASINI, S. 2018. A topological approach to
Mobility as a Service: A proposed tool for understanding requirements and effects, and for
aiding the integration of societal goals. Research in Transportation Business & Management,
27, 3-14.
STOPHER, P. 2012. Collecting, Managing, and Assessing Data Using Sample Surveys, Cambridge,
Cambridge University Press.
STREMERSCH, S. & TELLIS, G. J. 2002. Strategic Bundling of Products and Prices: A New Synthesis
for Marketing. Journal of Marketing, 66 (1), 55-72.
TRAIN, K., HESS, S., & POLAK, J. 2004. On the use of randomly shifted and shuffled uniform vectors
in the estimation of a mixed logit model for vehicle choice, Paper 04-433, 83rd Annual Meeting
of the Transportation Research Board, Washington DC.
WONG, Y. Z., HENSHER, D. A. & MULLEY, C. 2020. Mobility as a Service (MaaS): Charting a
future context. Transportation Research Part A: Policy and Practice, 131, 5-19.