Mobility as a service and private car use: Evidence from the Sydney MaaS trial

Abstract

Australia’s first Mobility as a Service (MaaS) trial commenced in April 2019 in Sydney, running for two years. The objective of the trial is at least twofold – to assess interest in various MaaS subscription plans through bundling public transport, rideshare, car share and car rental with varying financial discounts and monthly subscription fees, in contrast to pay as you go (PAYG); and to assess the extent to which the use of the private car might change following a subscription to a monthly mobility bundle. This paper assesses the second objective by investigating the potential for changes in monthly car use in the presence of a MaaS program. There is no previous research that we are aware of that has tested the relationship between MaaS bundle uptake and private car use. The paper develops a joint discrete-continuous model system to explain the choice between monthly bundles and PAYG, and subsequently, the total monthly car kilometres. Controlling for monthly differences due to other influences such as seasonal travel activity, the findings suggest that the offered bundles do have an encouraging impact on private car use. Within the limits of what was tested under the Sydney MaaS trial, indicative evidence suggests that MaaS has the potential to change travel behaviour in a way aligned with sustainability objectives, although this evidence should not be taken as suggesting that MaaS is a commercially viable mobility strategy.

Full text

1
Is there a relationship between private car use and subscribing to
mobility as a service (MaaS) bundles?
David A. Hensher
Chinh Q. Ho
Daniel J. Reck
Institute of Transport and Logistics Studies (ITLS)
The University of Sydney Business School
Sydney NSW Australia 2006
David.Hensher@sydney.edu.au
Quoc.Ho@sydney.edu.au
Daniel.Reck@ivt.baug.ethz.ch
April 29 2020
Abstract
Australia’s first Mobility as a Service (MaaS) trial commenced in April 2019 in Sydney, running for
two years. The objective of the trial is at least twofold – to assess interest in various MaaS subscription
plans through bundling public transport, rideshare, car share and car rental with varying financial
discounts and monthly subscription fees, in contrast to pay as you go (PAYG); and to assess the extent
to which the use of the private car might change following a subscription to a monthly mobility bundle.
This paper assesses the second objective by investigating the potential for changes in monthly car use
in the presence of a MaaS program. The paper develops a joint discrete-continuous model system to
explain the choice between monthly bundles and PAYG, and subsequently, the total monthly car
kilometres. Controlling for monthly differences due to other influences such as seasonal travel activity,
the findings suggest that the offered bundles do have an encouraging impact on private car use. Within
the limits of what was tested under the Sydney MaaS trial, indicative evidence suggests that MaaS has
the potential to change travel behaviour in a way aligned with sustainability objectives, although this
evidence should not be taken as suggesting that MaaS is a commercially viable mobility strategy.
Keywords: Mobility as a Service (MaaS), MaaS trial, Sydney, mobility bundles, Pay as you go
(PAYG), MaaS subscription, Discrete-continuous model, Poisson regression, elasticities, marginal
effects
Suggested citation: Hensher, D.A, Ho, Q.C., & Reck, J. D. (2021) Mobility as a service and private
car use: Evidence from the Sydney MaaS trial, Transportation Research Part A: Policy and Practice,
Volume 145, Pages 17-33, https://doi.org/10.1016/j.tra.2020.12.015.
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

2
contributions of other members of the project team, especially Andre Pinto (ITLS), Sam Lorimer and
Ivy Lu of IAG.
Introduction
Mobility as a Service (MaaS) has generated a huge amount of interest as a prospective way to garner a
greater commitment to mobility activity that aligns with achieving sustainability objectives such as
reducing road congestion and emissions. At the same time, MaaS gives travellers greater choices
through targeted information on travel planning with the support of a digital platform (Sochor et al.
2016, Smith and Hensher 2020, Wong et al. 2020). In contrast to an almost-daily-commentary on the
virtues of MaaS and a growing number of researchers questioning whether MaaS is scalable or niche,
very few MaaS schemes present in real markets. There appears to be a plethora of digital platforms
promoted as MaaS despite the fact that most are trip planners with a pay as you go (PAYG) option.
These enhanced digital platforms, typically in a smart-phone app, enable users to obtain information
about travel options as well as booking of mobility services available on the platform. While this all
sounds very appealing, we have yet to see a MaaS product that is a successful business model and which
offers various multimodal bundles through a subscription plan, despite a number of applications such
as Whim in Finland, Ubigo in Sweden, and swa Augsburg (see Hensher et al. 2020, Chapter 3 for
details).
The Sydney MaaS trial, which commenced in April 2019 as a two year project, was designed to obtain
contributing evidence on whether MaaS is a value added mobility proposition. It is the first MaaS trial
in Australia, and has the following objectives: (1) To explore appropriate transport service mixes and
subscription plans for early adopters of MaaS; (2) To generate first-hand knowledge of actual MaaS
experiences; (3) To advance the understanding of user uptake and willingness to pay for MaaS; (4) To
test the ability to influence travel behaviour through introducing MaaS solutions; and (5) To document
the experience in designing, planning and undertaking a MaaS trial.
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 (Ho et al. 2018,
2020), 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 the entire trial are set out in Hensher et al. (2020a).
Figure 1. The overall MaaS trial approach
•PAYG, Month 1
survey and
Subscription
period plans
(months 2 to 6)
•Pre-trial survey
•Promotion,
participant
selection, and
onboarding
•Modal Suppliers
•Supplier
integration
•App is called Tripi
•Trips, booking,
transactions,
payments,
invoicing
Digital
Channel
App
Suppliers
Trial
Period
Customers

3
This paper explores the relationship between subscription to a MaaS bundle in contrast to staying with
PAYG and what influence, if any, the decision to subscribe to a bundle through choosing one of the
offered bundles has on private car use. Tripi, the digital platform used in the trial did not capture car
use1; however, a complementary program called Safer Journeys, run by IAG, provided car use data for
a subset of participants who also subscribed to this complementary program. Participants were asked
whether they were involved in the Safer Journeys trial run and whether they would be interested in
joining, and if they consent to sharing their Safer Journeys data with the MaaS Trial. The Safer Journeys
trial was car-based with tracking technology installed so that car use kilometres can be measured, a
crucial piece of data given the objective on reducing emissions through reduced car use.
The MaaS trial in Sydney adopted an incremental approach to the design of monthly bundles for the
participants to subscribe to. Although the trial was officially started in November 2019, all participants
were on-boarded as PAYG users. Monthly bundles were gradually introduced in December 2019
(Fifty50), January 2020 (Saver25), February 2020 (GreenPass) and March 2020 (SuperSaver25). Figure
2 summarises these monthly bundles. Once introduced, the monthly bundles remained available for
subsequent months of the trial, except for the Saver25 bundle which was replaced by the SuperSaver25
bundle. The latter aims to encourage greater use of public transport through a financial incentive
associated with the first and last mile (access and egress) part of a door-to-door public transport trip.
Specifically, in addition to the Saver25 offers, the SuperSaver25 added a $5 Uber flat fare for the
subscribers to connect to/from public transport trips. Free first and last mile trips were considered but
were rejected due partly to the available incentive budget, and partly to a concern about the impact of
this offer on existing bus services in accessing and egressing a rail station. The compromise was to
introduce a financial incentive for Uber only (determined also by the way Uber is integrated into Tripi)
with a distance cap option in order to provide absolute certainty to participants that they would not face
different Uber fares for the same trip, for example, from their home to a local train station, regardless
of the time of the day these trips were undertaken. Using distance (cf. fare) as a cap to define eligible
Uber trips for the flat fare is important since Uber has a surge price (i.e., high demand price), which
may result in a situation where users pay different fares for the same distance travelled. Analysis of data
collected prior to the introduction of the SuperSaver25 bundle in March 2020 suggested that 75 percent
of participants live within 5 kms of a train station, with PAYG users generally living slightly further
from a train station than bundle subscribers.
In addition to this change to Saver25, renamed as SuperSaver25, we also changed the 15% on Taxi and
Uber to be a flat $3 reduction, given feedback that participants preferred an absolute dollar amount. 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. We stayed with the subscription
fee of $25/month and the 25 percent discount on all public transport trips. Car-based options were
provided through GoGet and Car rental2. 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. In the
current paper, we take the bundles as given, with details on how they were designed given in Hensher
et al. (2020a).
1 Although the possibility of undertaking a journey by car was shown in the options, the use of the car was not
tracked.
2 In a Webinar hosted by Global MaaS Transit on April 17, 2020, by Sampo Hietanen, Founder & CEO, MaaS
Global titled ‘Mobility-as-a-Service - The End of Car Ownership?’, in response to a question, Sampo said that
‘'the profitable part [of MaaS] is having access to a car on weekend otherwise MaaS is just a utility service.' The
Sydney trial accommodated this feature through GoGet and car rental. This is also a position supported by research
in Belgium by Storme et al. (2020).

4
Figure 2. PAYG and Bundles for the PAYG and four bundle offer months of the trial
The Joint Model System of Bundle Choice and Car Use
Formally, we have two models, one representing the choice between taking up a MaaS bundle and
choosing to stay with PAYG, and the other the monthly kilometres travelled by a private car. This is a
discrete-continuous choice model (simultaneous equation) system, where the discrete component is
either a logit or probit form, and the continuous component is a count model such as a zero inflation
Poisson regression. The binary choice model for PAYG versus a bundle is defined by a binary outcome
yi taking the values 0 (for PAYG) and 1 (for bundle) with the probability of choosing a bundle defined
as:
Prob[yi=1] = F(β'xi) such that F'(β'xi) ≥ 0 and 0 < F(β'xi) <1. (1)
We first estimate the binary choice model as logit by maximum likelihood to obtain estimated
parameters for influences on a bundle vs PAYG choices, and then use the estimated model to compute,
for each participant, a predicted probability of choosing a bundle. This probability is then fed into a
Poisson regression model for monthly car kilometres (see below). The estimation at both steps is
consistent; however we still need to correct the estimated asymptotic covariance matrix for the estimator
at step 2 for the randomness of the estimator carried forward from the binary choice model. The standard
Murphy and Topel (1985) correction is implemented, so that the standard errors and hence the t-values
of the Poisson model are asymptotically efficient.
The amount on monthly car kilometres travelled by each participant in the trial is obtained from their
participation in the Safer Journey’s program. Monthly kilometres is a positive number compliant with
a count model such as zero inflation Poisson (ZIP) with latent heterogeneity. It is connected to the

5
binary choice model by the method described above. As a non-negative continuous count value, with
truncation at zero, discrete random variable, Y, observed over a period of length Tn (i.e., a month) and
observed kilometres, yn, (n observations), the Poisson regression model is given as equation (2).
Prob(Y = yn|xn) =
yn
exp(- )
λλ
nn
!
yn
, yn = 0,1,...; logλn = β′xn. (2)
In this model, λn is both the mean and variance of yn; E[yn|xn] = λn. We allow for unobserved
heterogeneity as well as consider the ZIP form for count data (see Greene 2000) to recognise the
possibility of partial observability if data on monthly kilometres being observed for any period within
the four months exhibits no car use3. Specifically, the answer ‘zero’ could arise from two underlying
responses. If we were unable to capture any car use, we would only observe a zero; however, the zero
may be due to the measurement period (i.e., a particular month) and the response might be some positive
number in other periods. We define z = 0 if the response would always be 0, 1 if a Poisson model
applies; y = the response from the Poisson model; then zy = the observed response. The probabilities of
the various outcomes in the ZIP model are:
Prob[y = 0] = Prob[z = 0] + Prob[z = 1]×Prob[y = 0 | Poisson] (3a)
Prob[y = r > 0] = Prob[z = 1] × Prob[y = r | Poisson]. (3b)
The ZIP model is given as (Greene 2012) Yn = 0 with probability qn and Yi ~ Poisson (λn) with probability
1 – qn so that
Prob[Yn = 0] = qn + [1 – qn]Rn(0), and
Prob[Yn = r > 0]= [1 – qn]Rn(r) (4)
where Rn(y) = the Poisson probability = e -λn λn yn / yn! and λn = eβ′xn . We assume that the ancillary,
state probability, qn, is distributed normal; qn ~ Normal [vn]. Let F[vn] denote the normal CDF. Then,
vn = τlog[λn] = τβ′xn (5)
Equation (5) defines a single new parameter (which may be positive or negative). If there is no (or little)
evidence of zero kilometres in any observations, then we do not expect the τ parameter to be statistically
significant, and we can default to the Poisson form with normal latent heterogeneity.
The two models are estimated using the combined data from the Safer Journeys program (monthly car
use) and the data obtained from the Tripi app and the pre-trial survey. The former provided details of
the bundles chosen each month and the latter socio-demographic information of each participants.
Descriptive Profile
Of the 92 participants of the Sydney MaaS trial, 33 participants were also the Safer Journeys Program
subscribers. Car use of this subset of the MaaS participants, together with the MaaS monthly bundle
subscription dataset, form the core datasets for this paper. It is worth mentioning that car use data prior
to the MaaS trial was also available since the safer journey scheme was launched before the MaaS trial
was conducted. However, for the purpose of assessing the impact of MaaS bundle subscription on
private vehicle kilometre travelled, only safer journey data between the time when a participant joined
the trial and when s/he left were extracted and used for analysis. This is because not every participant
joined the MaaS trial in November 2019, and not everyone continued active up to the end of the trial in
3 Fitting a simple Poisson model would overstate (‘inflate’) the theoretical probability of zero in the Poisson
model. The ZIP model involves a joint estimation of a count data model and a binary probit (or logit) model where
the latter tests for whether the response will always be zero or otherwise (up to a probability).

6
March 2020. Also, few participants were active in the MaaS program (i.e., made trips using Tripi app)
but did not undertake any car trips in specific months. These are included to avoid biasing the data. In
total, the dataset represents 171 participant months. A summary of monthly MaaS bundle subscription
and the switching between PAYG and bundles for these 33 participants is summarised in Figures 3 (as
absolute takeup) and 4 (as percentage take up). Over the four months between December 20219 and
March 2020 when monthly bundles were available for subscription, 52 bundle offers were accepted.
This includes a participant staying with or switching away from a bundle, including moving between
bundles but excluding moving from a monthly bundle to a PAYG option. Overall, for the 171 participant
months, the Fifty50 bundle offered in December through to March was chosen 36.54% of the time;
Saver 25 introduced in January, was selected by 15.38% of participant months; and in February when
we introduced the GreenPass bundle, its popularity in February and March resulted in the highest
participant month uptake of 38.45% of all bundles. The SuperSaver25 bundle that replaced the Saver25
bundle in March represented 9.26%. PAYG was overall 30.41% of participant month’s choice.
Figure 3. A High level summary of the absolute take up of each bundle and PAYG each month for
Safer Journey’s participants
The evidence on the acceptance of monthly bundles is very encouraging, with the 91.87% for PAYG in
December dropping to 81.64% in January and then 44.73% in February before increasing marginally to
46.2% in March. In the final month when all bundles are available (although SuperSaver25 replaced
Saver25), we have a bundle take up of 53.85 percent4. The percentages for each bundle in March are
12.82%, 12.82% and 28.21% respectively for Fifty50, SuperSaver25 and GreenPass. This aggregate
share from real preference evidence is within the range of what has been found in stated preference
studies (30-55%) such as Ho et al. (2018) and is the first tangible evidence of the external validity, at
4 This compares with 36.5 percent for all trial participants.

7
least at an aggregate level, of the stated preference survey responses5. What we are seeing is some
learning of bundle experience, in part influenced by changing monthly travel needs. Given the sample
size, however, care is taken in generalising this evidence.
Figure 4. The percentage distribution of PAYG and bundles subscription for each month (note: each
month sums to 100%)
Figure 5 shows the monthly car kilometres at the participant level and the monthly bundle they were on
for each month. Figure 5 also identifies, on the x-axis, the month each safer journey user joined the
MaaS trial and the month they left the trial (a couple of participants resigned their position at IAG
during the trial and hence they must leave the trial). On the other hand, the vertical axis shows private
car kms travelled on a log10 scale while the colours identify the participant’s mobility agreement, be it
PAYG user or monthly subscriber. Overall, a lot of variation in monthly car kms can be observed at the
individual level, with December being most substantial. This is expected in the holiday season when
many travellers would typically use car more, either for shopping or for road trips (e.g., P042 drove
from Sydney to Melbourne and back) or do not use their private car at all (e.g., P178, P082).
Without an intervention of monthly mobility bundles, car use after the holiday season was expected to
go back to normal, and the total kms travelled in January and February 2020 should be comparable with
that in November 2019 for PAYG users. This expectation is observed in several PAYG participants
such as P002, P004, P007 (see Figure 5). After taking up a bundle, these participants appear to reduce
their monthly car kilometres, while very little change in monthly car kilometres was observed for those
who continued to use MaaS as a PAYG user (P031, P099, P173). While month-to-month variation in
car kilometres is an issue that needs to be acknowledged in this descriptive analysis, the evidence is that
MaaS subscriber’s car kilometres in February are generally much lower than those individuals who
continued with PAYG. The average kilometres for February are 658, 266, 477 and 222 respectively for
PAYG, Fifty50, Saver25 and GreenPass (an average of 284 kilometres for the all three bundle
subscribers). This is a very important result suggesting that MaaS bundles do attract interest by active
5 We must recognise that studies such as Ho et al. (2018) include respondents who do not have access to a car,
and so a direct comparison of samples must be cautioned.

8
car users, and that these appear to be participants who rely less on the car for their mobility needs. We
are not, however, able to conclude that subscription to a MaaS bundle has reduced car kilometres
compared to what car usage would have been if the bundle subscription was not available. The only
months that have similar periods for travel activity comparison, that are not a special month like
December and January6, are October and November, where the average monthly kilometres are
respectively 513 and 474, still much more than the average of 284 for the bundle subscribers in
February. In the formal modelling below, we investigate this matter.
Figure 5. Profile of monthly car km travelled by Safer Journey users in the Sydney MaaS trial
Model Results for Trial Months with Mobility Bundles
The model system estimated for the choice between PAYG and a bundle, as well as monthly car
kilometres, considered a number of variables that describe the socioeconomic status of the participants,
the incentives offered with each bundle including the change in the metric for ride share (from a
percentage to dollars) and the amount of money saved per month compared with PAYG. Table 1
summarises the data items that were considered in various models, resulting in the preferred model
summarised in Table 2. Specifically, we found that a series of dummy variables associated with each
specific bundle and the associated trial month (e.g., December Fifty50 bundle), relative to PAYG, did
not provide as good a behaviourally and statistically significant explanation of the choice between
PAYG and bundles in contrast to modal trip activity by each mode during the trial, month-specific
dummy variables, estimated financial savings each month associated with bundle selection7 and
socioeconomic characteristics.
6 In Australia, December is very much a party month with an elevated use of rideshare and January is the main
holiday month with reduced metropolitan travel (hence local public transport) and increased use of air and car for
long distance travel.
7 Defined as the estimated saving associated with a subscribed bundle (i.e., the bundle the participant actually
subscribed to for that month) compared to the cost outlay under PAYG. A negative value indicates that the
subscribed bundle is more expensive than PAYG, and a positive estimate indicates a saving. We set PAYG to $0.

9

10
Table 1. Descriptive Statistics of Data (Sample size = 149 participant months)
Variable
Units
Mean (standard deviation)
Range (min, max)
Monthly kilometres
kilometres
433.9 (587)
6.5, 3708
Pay as you go
proportion
0.651
0,1
Fifty50 bundle
1,0 (Dec-Mar)
0.127
0,1
Saver25 bundle
1,0 (Jan-Feb)
0.054
0,1
SuperSaver25 bundle
1,0 (March 1-20)
0.033
0,1
Green Pass
1,0 (Feb-March 1-20)
0.134
0,1
Gender
Male = 1, Female = 0
0.597
0,1
Age
Years
40.87 (8.54)
30,60
Adults in household
Number
2.27 (0.76)
1,5
Children in household
Number
1.13 (0.84)
0,2
Car licenced drivers in household
Number
2.13 (0.70)
1,5
Number of cars in household
Number
1.62 (0.75)
1,4
Access to a car
Access to a car = 1
0.893
0,1
Subscribe to a bundle
1,0
0.349
0,1
Sample month participation - December
1,0
0.228
0,1
Sample month participation - January
1,0
0.255
0,1
Sample month participation - February
1,0
0.255
0,1
Sample month participation - March
1,0
0.262
0,1
December Fifty50 bundle
1,0
0.022
0,1
January Fifty50 bundle
1,0
0.034
0.1
February Fifty50 bundle
1,0
0.040
0.1
March Fifty50 bundle
1,0
0.034
0,1
January Saver25
1,0
0.013
0,1
February Saver25
1,0
0.040
0,1
February GreenPass
1,0
0.060
0,1
March GreenPass
1,0
0.074
0,1
March SuperSaver25
1,0
0.033
0,1
Weekly modal travel activity:
Car driver trips
Number
44.13 (29.4)
2,131
Car passenger trips
Number
0.041 (0.26)
0,2
Public transport trips
Number
22.59 (16.19)
2,76
Rideshare (taxi and Uber) trips
Number
3.31 (4.11)
1,25
Estimated monthly financial saving on the
subscribed bundle compared to PAYG
$ per month
6.97 (16.42)
-28,72
Subscription bundle fee and discounts:
Subscription fee
$ per month
25.33 (42.91)
0,125
Public transport discount
%
21.44 (35.25)
0,100
Ride share discount
$
0.483
0, 3
Ride Share discount
%
1.933
0,15
Car share discount
%
0.805
0,15
In estimating the models, given that the unit of analysis is a participant month and there is more than
one observation per participant, the data structure is like a panel (repeated observation for each
respondent) and hence there exist observations in a group that are likely to be correlated through
common latent heterogeneity across four months. This sequential time period of data defined by the
month, is accommodated through a cluster algorithm that is similar to a random effect. The parameter
estimator is unchanged, but an adjustment is made to the estimated asymptotic covariance matrix (see
Greene 2012) to correct the standard errors. We tested for fixed and random effects; however the fixed
effect model did not work due to sample size, and the random effects model failed to converge. A
random parameter form (normally distributed) for monthly kilometres was investigated, but it was
found to be statistically non-significant due, we suspect again, to sample size.

11
Table 2. Model results
Parameter
estimates
Clustered
standard errors
z-value
Discrete Choice: Binary Logit Bundle (1) vs PAYG (0)
Constant
-0.7098
3.272
-0.22
February dummy variable (1,0)
2.2856
0.943
2.42
March dummy variable (1,0)
2.7440
1.277
2.15
Monthly car passenger trips
-28.096
2.354
-11.93
Monthly public transport trips
0.0702
0.019
3.57
Monthly savings in costs for a bundle cf. PAYG (log$)
41.572
1.203
34.57
Male (1,0)
-1.7499
0.962
-1.82
Car licenced drivers in household
-3.7002
2.261
-1.64
Restricted log likelihood
-22.170
McFadden pseudo R2
0.780
Continuous model of Monthly Car Kilometres: Poisson
model with normal heterogeneity
Constant
6.1751
0.0272
227.3
Access to a car (1,0)
1.1223
0.0209
53.52
Age of participant (years)
-0.0339
0.00052
-43.72
Gender of Male (1,0)
-0.2834
0.0088
-32.38
Number of children in household
0.1174
0.0055
21.44
Predicted probability of choosing a monthly bundle
-0.6690
0.0114
-60.05
Sigma
0.584
0.012
47.90
Log likelihood
-36,637.4
McFadden pseudo R2
0.134
Vuong statistic vs Poisson (favours the extended model)
6.147
The Choice between take up of a bundle and PAYG
We begin by discussing the binary choice logit model. The overall goodness of fit is very impressive
with a McFadden pseudo R2 of 0.780. Except for the two socioeconomic characteristics (male and
number of licensed car drivers in a household) and the constant, the variables are statistically significant
at the 95 percent level of confidence. The socioeconomic characteristics are statistically significant at a
slightly lower level of statistical confidence.
Initially, we had anticipated that we might be able to introduce a series of variables to represent the
subscription fee and the mode-specific discounts, since although they do not vary within a particular
bundle offer, they do vary across the offered bundles. However, these variables are highly correlated
and result in a very unstable bundle choice model. The amount of variance is not sufficient to capture
the role of such discounts, and indeed is a reason why revealed preference data like that in the trial
creates challenges in model estimation, and is one of the reasons why stated preference data is
appealing. Until there is sufficient variation in the incentives and subscription fees associated with real
market offerings of MaaS bundles, there will be limitations to using such data in studying bundle choice.
This may require the pooling of many MaaS products to be able to obtain sufficient variation. However,
we found that the variation in the financial savings associated with each bundle (Table 1) relative to
PAYG for each participant enables this influence to be tested. All other influences remaining constant,
we find that as the financial savings increases, the probability of choosing a bundle increases. The extent
of the change is presented below as a semi-elasticity.

12
Two dummy variables for the months of February and March were statistically significant and positive.
What this suggests, and reaffirms what we know about take up of bundles as we offer additional bundles
in a monthly sequence, is that relative to the previous months of December and January, the probability
of choosing a bundle increases as we move monthly through the trial. This is very reassuring and
supports the way in which we assessed new bundles given the experience with previously introduced
bundles. The way the trial was designed provides strong clues as to how bundles can be designed
through time. A caveat must be mentioned here, namely Covid-19, which has had a massive impact on
travel in general after March 128, with the greatest impact on public transport and ride share patronage.
Car use has continued, but in general it also dropped (see Figure 5). What we can say, however, is that
car use was greater in December and January than in February and March up to the 12th March.
Finally, we found that two socioeconomic characteristics, gender and the number of licensed drivers in
a household, were statistically significant negative influences on the probability of choosing a bundle,
suggesting that male participants and those in households with more driving licences are less likely,
holding all other influences constant, to choose a bundle over PAYG. The other socioeconomic
characteristics in Table 1 were not found to be statistically significant.
Care must be taken in interpreting the numerical magnitudes of each parameter estimate since they are
non-comparable in this logit non-linear form, and hence below we present partial effects and elasticities
as a way of meaningfully comparing the impacts of each bundle component. The behavioural sensitivity
of the probability of choosing a bundle compared to PAYG for each of the explanatory variables can
be given by an elasticity and a partial effects indicator. For the logit form, the elasticity of the probability
is given in equation (6) and the partial (or marginal) effect in equation (7).
log ( |) ( |) marginal effect
log (|) ( |)
kk
kk
xx
Eyx Ey x
x Eyx x Ey x
∂∂
= =
∂∂
⋅⋅
(6)
''
'
(|) () () ''
'() f()
()
xx
x
E y x F dF xx
F
xx
d
ββ ββ
β ββ
β
∂∂
= = = =
∂∂
(7)
The direct elasticity of the probability of choosing a bundle compared to PAYG with respect to the
number of monthly public transport trips is 0.820 (with a t-value of 3.86 and a 95 percent confidence
interval of 0.042 to 1.24 using the Delta method). The direct elasticity of the probability of choosing a
bundle compared to PAYG with respect to the number of monthly car passenger trips is -1.25 (with a
t-value of 3.38 and a 95 percent confidence interval of -1.978 to -0.527). Thus, all other influences
remaining unchanged, a one percent decrease in monthly car passenger trips will result in a 1.25 percent
increase in the probability of choosing a bundle over PAYG; and for public transport trips a one percent
increase in monthly public transport trips will result in a 0.820 percent increase in the probability of
choosing a bundle over PAYG. As an example, if we work with the average bundle choice share of
0.349 (Table 1) and if we can achieve a 10 percent increase, on average, in monthly public transport
trips, then we predict an increase in the probability of choosing a bundle of 8.2 percent, or an increase
from 0.349 to 0.378. The equivalent change for a reduction in car passenger trips is 12.5 percent or an
increase to 0.393. If this evidence was scalable given the mix of MaaS bundles offered in the trial, we
can expect a significant improvement in traffic congestion if we are able to reduce car kilometres by 6
to 10 percent, which is equivalent to returning the road environment to school holiday levels of
congestion.
8 On March 12, all participants as employees of IAG were advised to work at home where possible and all non-
local domestic and international travel was to cease, which aligned with the Stage 2 restrictions imposed on
Society on March 20.

13
For the average monthly financial saving, since it is transformed as a natural logarithm, we have to use
a semi-elasticity formula (equation 8), interpreted as the change in probability for a 1% change in x.
That is, a semi elasticity formula measures the relationship between a percentage change in X and an
absolute (not percentage) change in Y, and hence we refer to a unit increase in the explanatory variable
(not percentage, but change in percentage points).
Prob Prob(1 Prob)
100
100* x
x
β
∆−
=
∆
(8)
The model obtains a semi-elasticity of 1.853 (t value of 4.4) for average monthly financial savings
relative to PAYG. A 1 percent change in the average monthly financial savings will result in a 1.853
change in the probability of choosing a bundle over PAYG. The average change in the probability of
choosing a bundle when the average savings increases by 10% (noting a change in the log of .1 is about
a 10% increase in the average savings), is .185 or 18.5 probability of bundle choice points. Thus, given
the average bundle choice share of 0.349, this amount to a probability of bundle choice of 0.414 (=
0.349 × 1.853) for an additional 10 percent financial savings over PAYG.
For the remaining variables, we have calculated the partial or marginal effects. The average marginal
effect provides an effect on the probability. It is the average change in the probability when an
explanatory variable increases by one unit. When a variable is a dummy variable, we have to take the
exponential. The marginal effect parameters (with t values in brackets) for the February dummy, the
March dummy, male and number of household members with a drivers licence are respectively 0.129
(3.83), 0.177 (2.61), -0.076 (1.56) and -0.289 (1.12). For example, for the February dummy variable
influence, we obtain exp(0.1294)=1.138 suggesting that we are 11.38% more likely to choose a bundle
in February compared to December and January; the equivalent percentage for March is
exp(0.177)=1.194 or 11.94% more likely to choose a bundle in March compared to December and
January. The main implication of all of these findings is that they provide suggestions for ways of
harnessing MaaS as a policy instrument supporting sustainable transport outcomes.
The Relationship between monthly private car kilometres and bundle take up
Turning to the Poisson regression model (in Table 2), with monthly kilometres defined as an integer for
the Poisson count model, the overall goodness of fit (as pseudo R2) is 0.134. We found that the tau (τ)
parameter (equation 5) associated with the zero inflated Poisson model with normal heterogeneity was
not statistically significant and so we opted for the Poisson model with normal heterogeneity, where the
sigma (σ) parameter, the standard deviation of heterogeneity, is statistically significant at the 1 percent
level. The Vuong statistic of 6.147 suggests that the estimated extended Poisson model in Table 2 is
favoured over an unaltered Poisson model.
Four socioeconomic characteristics have a statistically significant influence on monthly car kilometres,
namely access to a car, participant age, gender and the number of children in the household. All other
influences remaining constant, having an access to a private car increase monthly car kilometres, but
older participants tend to drive less than younger participants, with male respondents having fewer
monthly car kilometres than female participants. Households with more children tend to use their cars
much more, as might be expected. These findings are sample-specific and so cannot be generalised,
although it does suggest that there are socioeconomic segments associated with car use which translates
into differences in the propensity to take up a bundle compared with PAYG if someone decides to
participate in MaaS.
Finally, we have included the predicted bundle choice as a probability measure obtained from the binary
logit model of bundle vs PAYG. It is a statistically significant and negative effect which indicates that,
ceteris paribus, if the probability of choosing a bundle increases, then there is an expected reduction in

14
monthly car kilometres. The parameter estimates of a Poisson regression may be interpreted as semi-
elasticities as discussed above in equation 8.9 The parameter of -0.669 for the probability of choosing a
bundle indicates that, ceteris paribus, if we increase the probability by 1 percent, we expect to have a
66.9 absolute reduction in monthly car kilometres. At the mean monthly kilometres of 434, a 1 percent
increase in the probability of choosing a bundle (from 0.349 to 0.353) is predicted to reduce monthly
kilometres from 434 to 367 kilometres. If scalable over a large population of MaaS subscribers, this is
a significant reduction in car kilometres.
The overall findings reinforce a position that a well-designed suite of subscription plans under MaaS
can influence the use the car in a positive and sustainable way, contributing to a reduction in emissions.
Conclusions
Part of the remit for MaaS is to offer a more attractive way for individuals (and groups of any
denomination) to be better informed about mobility options. This in turn opens up opportunities to
‘entice’ changing patterns of travel behaviour that not only provide direct benefits to the travelling
public, but also support achieving broader societal goals. It is these societal goals that are increasingly
being promoted through what many governments are calling mobility frameworks (Hensher et al. 2020).
One of the overarching themes in the MaaS ecosystem is the desire to reduce private car use, and as a
consequence contribute to the reduction in emissions and other negative externalities such as traffic
congestion.
The Sydney MaaS trial is well placed to investigate how MaaS may appeal to these personal and societal
agendas. In this paper, we have focussed on the trial activities of the sub sample who have provided
data on their car use over the months in which subscription bundles have been offered, starting with a
single bundle in the first month after a PAYG familiarisation period with the digital platform (the Tripi
App), and then incrementally adding a bundle each month. The sequential enhancement of bundles
offers an innovative way of learning by doing with a sample of participants, such that the opportunity
to grow interest in a bundle may be increased through analysis of each months travel activity and bundle
choices. This was indeed an appropriate strategy since it resulted in a slow but noticeable move away
from PAYG to bundles (see Figures 3 and 4), even if PAYG remained the dominant way of using Tripi
and participating in the trial.
With the growing preferences favouring specific bundles, we wanted to know if there was an impact on
monthly car use, and what features of the bundles in particular might be the main triggers for changing
patterns of car use. Using a joint discrete and continuous choice model system to study the influences
on the choice between a bundle and PAYG (the discrete choice model), and the influences on monthly
private car kilometres (the count model), we have been able to show that a subscription to monthly
mobility bundle does influence monthly car use in a statistically significant way.
Importantly, it is the combination of a subscription fee and a suite of mode-specific financial discounts
that will ultimately determine the appeal of MaaS bundles and indeed Maas more generally. We would
argue that having monthly mobility bundles for subscription will be the key influence on whether MaaS
is to grow in a scalable way or remain a niche construct. Without subscription options, MaaS seems be
nothing more than a potentially attractive trip planner which may change travel behaviour to some
degree as a result of better information, but there is no guarantee. The in depth qualitative interviews
with trial participants (Wong and Hensher 2020) suggest that many individuals are looking for a limited
multimodal subscription plan (maybe two or three modes at the most), and that a digital platform under
PAYG alone is unlikely to be of much interest to the majority of improved mobility aspirants. It is clear
9 Suppose x is not a log variable, then β associated with x can be interpreted as a semi-elasticity, meaning that a
one unit change in x will change E [y |x] by 100β (Wooldridge 2002).

15
that money talks loudly, despite the added recognition of the importance of good quality public transport
and rideshare services, in terms of convenience, reliability safety, transfers and travel times. Thus,
without financial incentives, especially where there are no service enhancements only available to MaaS
subscribers, the likelihood of MaaS achieving noticeable outcomes aligned with sustainability
objectives may be unlikely to be achieved.
This then is the challenge going forward for prospective commercial MaaS products. What subscription
plans will attract users, how will the financial incentives be funded, and can we deliver a digital platform
that is sufficiently multimodal, easy to use, and providing information of the users travel activity that
can inform their future choices in a way that has private and social benefits? (Hensher et al. 2020,
2020a) The current paper has shown, through the Sydney trial, that there is an appetite for such a
product, and that it can contribute to achieving sustainability goals; however, a lot of ongoing research
is required to integrate the constituent parts and feature a business case that is attractive to both private
interests, users and government. The trial has indeed commenced that journey.
References
Greene, W.H. (2000) Econometric Analysis. Prentice Hall, Englewood Cliffs.
Greene, W.H. (2012) Limdep Version 10, Econometric Modelling Guide. Econometric Software Inc.,
New York and Sydney.
Hensher, D.A., Rose, J.M. and Greene, W.H. (2015) Applied Choice Analysis, Second Edition,
Cambridge University Press, Cambridge.
Hensher, D.A., Mulley, C., Ho, C., Nelson, J., Smith, G. and Wong, Y. (2020) Understanding Mobility
as a Service (MaaS) - Past, Present and Future. Elsevier Science UK., June.
Hensher, D.A., Ho., C, Reck, D., Smith, Wong, Y., Sarkar, N., G., Lorimer, S and Lu, I. (2020a) The
Sydney Mobility as a Service (MaaS) Trial: design, implementation and lessons (for Thredbo 17,
Kobe, September 2021), latest full draft 9 April 2020.
Ho C, Hensher DA, Mulley C and Wong YZ (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., Mulley, C. and Hensher, D.A. (2020) Public preferences for mobility as a service: insights from
stated preference surveys, prepared for special issue of Transportation Research Part A on MaaS
and Intelligent Mobility Guest Edited by David Hensher and Corinne Mulley, 131, 70-90.
https://doi.org/10.1016/j.tra.2019.09.031
Jittrapirom P, Ciati V and Feneri AM (2017) Mobility as a Service: A critical review of definitions,
assessments of schemes and key challenges Urban Panning 2. 13-25.
Lambert, D (1992) Zero-inflated Poisson regression, with an application to defects in manufacturing,
Technometrics, 34, 1-14.
Murphy, K. and Topel, R. 1985) Estimation and Inference in Two-Step Econometric Models Journal of
Business & Economic Statistics 3 (4), 370-379.
Smith, G. and Hensher, D.A. (2020) Towards a framework for Mobility as a Service policies, Transport
Policy, 89, 54-65. https://doi.org/10.1016/j.tranpol.2020.02.004
Sochor J, Karlsson ICM and Stromberg 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 2542, 57-64.
Reck, D.J., Hensher, D.A. and Ho, C. (2020) MaaS bundle designs, submitted to Transportation
Research Part A, 10 February 2020.
Storme, T., De Vos, J. De Paepe, L. and Witlox, F. (2020) Limitations to the car-substitution effect of
MaaS. Findings from a Belgian pilot study, Transportation Research Part A, 131, 196–205.
Wooldridge, J.M. (2002) Econometric Analysis of Cross Section and Panel Data, Cambridge: MIT
Press.

16
Wong, Y. and Hensher, D.A. (2020) MaaS mid-trial qualitative interviews, submitted to Transport
Policy.
Wong, Y., Hensher, D.A. and Mulley, C.M. (2020) Emerging transport technologies and the modal
efficiency framework: a case for mobility as a service (MaaS), in special issue of Transportation
Research Part A on MaaS and Intelligent Mobility Guest Edited by David Hensher and Corinne
Mulley, 131, 5-19. https://doi.org/10.1016/j.tra.2019.09.030