Modeling the effect of Mobility-as-a-Service on mode choice decisions

ABSTRACT

Mobility-as-a-Service (MaaS) can be considered the latest innovative technological solution to sustainable transportation. It offers travelers access to a pre-paid bundle of transportation modes on a single app with the additional convenience of the integration of planning, booking, and payment. Capturing the impact of this new technology on travel behavior remains a challenge, given the scarce number of comprehensive MaaS pilots. In this study, a stated adaptation experiment was designed to determine travelers’ propensity toward using this new service. Such knowledge sheds light on the potential of MaaS to alter daily travel patterns.

KEYWORDS:

• Mobility-as-a-Service
• Stated Adaptation Experiment
• Travel Behavior
• Mode choice

Introduction

Concepts such as shared and on-demand mobility as well as electric and autonomous vehicles are expected to change transportation mode choice. Recently, Mobility-as-a-Service (MaaS) is viewed as another flexible, innovative mobility solution that can be tailored to users’ different tastes and needs, and forms a convenient alternative compared to the usage of private vehicles. Although the definition of the concept differs widely, in the European context it tends to refer to a subscription to a prepaid bundle of different (on demand) transportation modes with various pricing schemes, organized on a digital platform that allows planning, booking, and payment of transportation services (Hietanen 2014; Finger, Bert, and Kupfer 2015; Kamargianni et al. 2016; Jittrapirom et al. 2017).

Recently, there have been attempts to model the adoption of MaaS and estimate the demand for the service as a function of pricing schemes and bundling options (Ho et al. 2018; Matyas and Kamargianni 2018; Caiati, Rasouli, and Timmermans 2019) since obtaining insights from involved stakeholders and future users (Polydoropoulou, Pagoni, and Tsirimpa 2018) is of the utmost importance for a successful implementation of MaaS. To complement these studies, the focus of this paper is on modeling the effect of MaaS on changing travel patterns, particularly on the choice of transportation mode. Empirical evidence on the effect of MaaS is relevant for various parties; for policymakers to coordinate incentives toward sustainable mobility (mode choice), for transportation operators to refine strategies and avoid cannibalization and for network providers to optimize their service offerings.

The number of studies regarding the impact of MaaS is very limited. Few insights were provided based on small-scale pilots in Sweden and Austria regarding the potential effect of MaaS on travel behavior. Based on a small number of registered users of the six-month Swedish pilot (Sochor, Strömberg, and Karlsson 2015a; Sochor, Strömberg, and Karlsson 2015b; Sochor, Karlsson, and Strömberg 2016) report the effects on the travel patterns at the household level, mentioning a reduction in private car use, a modal shift toward public transportation and car-sharing services as well as minor changes in trip destination and route choice. Karlsson et al. (2017) remark the positive effect of the Austrian pilot (SMILE) on user’s mobility behavior in terms of the use of sharing services and electric mobility, however, stress that the available data obtained from MaaS or MaaS-related schemes are inadequate for a thorough impact assessment of such service. This has also been highlighted by Durand et al. (2018), who conducted an explorative and systematic literature review and pinpointed potential behavioral impacts of MaaS including its possible implications for environmental and social sustainability. Storme et al. (2019) question the possibility of MaaS to substitute car-ownership based on a quite small-scale Belgian MaaS-pilot, since despite the noted decrease of privately owned cars during the intervention, more evidence is needed from larger, in content and sample size, pilot-studies. Interesting insights are also given by Ho et al. (2018) about the effect of current travel patterns on MaaS uptake and the possibility to disrupt car-ownership models in the Australian context.

Considering the largely unknown impact of MaaS, the aim of this paper is to better understand the intended behavioral change (change of transportation mode) as a result of the availability of MaaS, for people that consider themselves likely to adopt such a system by acquiring a subscription to the service. To that end, a transportation mode choice model is estimated.

The next sections describe the data collection process, the design of the stated adaptation experiment, followed by the results of the model examining the propensity of mode-choice adaptations as the result of a subscription to MaaS.

Experiment

Similar to traditional stated choice surveys, stated adaptation experiments attempt to capture individual choice processes. The specific focus concerns the question whether individuals change their current behavior given a new (hypothetical) context and/or choice alternatives. Respondents are asked to complete a set of choice tasks that systematically varying a set of attribute levels by indicating whether under the new hypothetical circumstances they would adapt their behavior and if so how (e.g. D’Arcier, Andan, and Raux 1998; Bladel et al., 2008). The stated adaptation experiment was part of a larger questionnaire, distributed through a market research company in Rotterdam, Amsterdam, and Utrecht in the Netherlands. The questionnaire consisted of 4 parts:

Part I: Introduction and Screening Question

Part II: Baseline Behavior (4-day Travel Diary)

Part III: Stated Adaptation Experiment

Part IV: Personal/Household information and Attitudes

First, to familiarize respondents with the new concept, an animated video explaining MaaS was shown. It was followed by a screening question, asking respondents to indicate their level of interest in a subscription to MaaS, choosing one of three options: uninterested/low chance of subscription; interested/good chance of subscription; very interested/high chance of subscription. From the initial sample of 2143 respondents, 37% was not interested in MaaS and were excluded from the main survey, since we are interested into obtaining data from potentially future users of MaaS. Those interested (53%) and very interested (10%) were asked to proceed and invited to report their typical travel patterns for 2 trips per day across any three weekdays, and one non-weekday. The third part concerned the stated adaptation experiment. Four different subscription schemes (bundles) were configured for the purpose of our study (Figure 1).

Figure 1. The four MaaS bundles (Feneri, Rasouli, and Timmermans 2019).

Respondents were assigned a single randomly selected bundle to avoid they had to mentally process different bundles during the experiment. They were then asked, imagining they subscribed to that bundle, to indicate for the reported trips whether they would continue to use the same transportation mode (Status Quo) or choose one of the modes included in the assigned MaaS bundle.

The constructed design was an orthogonal fractional factorial design, pivoted around the respondent’s reported information for each documented trip to create realistic choice situations. A common problem of such designs using travel time information is that unlikely combinations of travel times by transportation mode are created (e.g. bus being faster than car). To avoid this problem, for each current transportation mode, realistic proportional speed profiles for the modes included in the Maas bundle were assumed (Feneri, Rasouli, and Timmermans 2019). Consequently, the created design is orthogonal in attribute differences with the main potential advantages of equidistant coverage of attribute space and attribute level balance. This design strategy resulted in 128 profiles that were blocked into 16 orthogonal subsets and ultimately led to 8 choice sets per respondent.

In the last part of the main questionnaire, socio-demographic characteristics and personal attitudes toward multimodality, environmental awareness, and innovation value were collected. After the screening question and the data cleaning process, 8080 observations collected from a sample of 1010 respondents were used for analysis and model estimation.

Methodology

In order to examine the determinants of mode choice as a result of the introduction of MaaS, an Error Components Logit (ECLogit) model was estimated. This model estimates extra parameters with respect to the correlation that may exist between the choice alternatives (Train 2009; Hensher, Rose, and Greene 2013). It assumes that the probability of choosing a mode within a MaaS bundle depends on the monthly price of the bundle, the pricing schemes of various transportation modes within the bundle, and few other relevant attributes such as travel time, parking time, and waiting time (depending on the specific transportation mode), while also considering covariances that may exist across the choice alternatives.

As shown in Equation 1(1) $\begin{array}{rl}& U_{n\left(S{Q}_{1-m}\right)t}=AS{C}_{SQ}+{\beta }_{SQ}\ast I_{S{Q}_{1-m}}+{\sum }_k{\beta }_k\left(x_{nkSQ}\right)+{\sum }_z{\beta }_z\left(x_{nz}\right)\\ & +{\sum }_h{\beta }_h\left(x_{nh}\right)+{\sum }_{k^{\prime }\subset k}{I}_{S{Q}_{1-m}}\ast x_{n{k}^{\prime }}+{\sum }_{z^{\prime }\subset z}{I}_{S{Q}_{1-m}}\ast x_{n{z}^{\prime }}\\ & +{\sum }_{h^{\prime }\subset h}{I}_{S{Q}_{1-m}}\ast x_{n{h}^{\prime }}+{\sum }_{k^{\prime }\subset k}{\sum }_{z^{\prime }\subset z}{x}_{n{k}^{\prime }}\ast x_{n{z}^{\prime }}+{\epsilon }_{njt}\end{array}$(1) , a linear relationship is assumed. $U_{n\left(S{Q}_{1-m}\right)t}$is the utility of alternative SQ_1-m chosen in choice task t, given Status Quo 1 (Drive Alone) to m (EBike) for individual n, x_nkj is alternative-specific attribute k, β_k is its corresponding coefficient, $x_{nz}$ is socio-demographic characteristic z of respondent n with its corresponding coefficient ${\beta }_z$, ${\mathrm{x}}_{nh}$ is decision context variable h (such as trip purpose, level of interest in MaaS), and ε_njt is an error term for choice j in choice task t, assumed to be identically and independently Gumbel distributed (IID). ASC is an alternative-specific constant and $I_{S{Q}_{1-m}}$ is an index identifying the type of reference mode, consisting of Drive Alone (DA), Passenger (PASSENG), Walk, Bike and Ebike and ${\beta }_{SQ}$ the corresponding coefficient. Equation 2(2) ${\eta }_j={\sigma }_{jp}{\upsilon }_{np}\sim N\left(0,{\sigma }_{Jp}^2\right)$(2) shows that $U_{n\left(j\right)t}$ is the utility of alternative j for individual n in choice task t and expresses the relative utility for each one of the four MaaS alternatives (MaaS-PT, MaaS-CS, MaaS-BS, MaaS-TAXI). The reasons to separate the utility of Status Quo and the one of MaaS alternatives in two equations are two-fold. First, while in Equation 1(1) $\begin{array}{rl}& U_{n\left(S{Q}_{1-m}\right)t}=AS{C}_{SQ}+{\beta }_{SQ}\ast I_{S{Q}_{1-m}}+{\sum }_k{\beta }_k\left(x_{nkSQ}\right)+{\sum }_z{\beta }_z\left(x_{nz}\right)\\ & +{\sum }_h{\beta }_h\left(x_{nh}\right)+{\sum }_{k^{\prime }\subset k}{I}_{S{Q}_{1-m}}\ast x_{n{k}^{\prime }}+{\sum }_{z^{\prime }\subset z}{I}_{S{Q}_{1-m}}\ast x_{n{z}^{\prime }}\\ & +{\sum }_{h^{\prime }\subset h}{I}_{S{Q}_{1-m}}\ast x_{n{h}^{\prime }}+{\sum }_{k^{\prime }\subset k}{\sum }_{z^{\prime }\subset z}{x}_{n{k}^{\prime }}\ast x_{n{z}^{\prime }}+{\epsilon }_{njt}\end{array}$(1) separate constants have been defined and estimated for the type of mode in Status Quo (to enable us to understand the intensity of inertia for different modes in Status Quo), only a generic constant has been included in the utility function of MaaS options (Equation 2(2) ${\eta }_j={\sigma }_{jp}{\upsilon }_{np}\sim N\left(0,{\sigma }_{Jp}^2\right)$(2) ). The second reason is the treatment of the error structure in the utility functions of MaaS alternatives (which does not exist for Status Quo; Equation (1)(1) $\begin{array}{rl}& U_{n\left(S{Q}_{1-m}\right)t}=AS{C}_{SQ}+{\beta }_{SQ}\ast I_{S{Q}_{1-m}}+{\sum }_k{\beta }_k\left(x_{nkSQ}\right)+{\sum }_z{\beta }_z\left(x_{nz}\right)\\ & +{\sum }_h{\beta }_h\left(x_{nh}\right)+{\sum }_{k^{\prime }\subset k}{I}_{S{Q}_{1-m}}\ast x_{n{k}^{\prime }}+{\sum }_{z^{\prime }\subset z}{I}_{S{Q}_{1-m}}\ast x_{n{z}^{\prime }}\\ & +{\sum }_{h^{\prime }\subset h}{I}_{S{Q}_{1-m}}\ast x_{n{h}^{\prime }}+{\sum }_{k^{\prime }\subset k}{\sum }_{z^{\prime }\subset z}{x}_{n{k}^{\prime }}\ast x_{n{z}^{\prime }}+{\epsilon }_{njt}\end{array}$(1) ).

In order to capture any unobserved similarity among alternatives and the panel effect, m alternative and individual specific error terms are introduced. $I_{jp}$ is an index which is equal to 1 if effect p appears in alternative j and zero otherwise. ${\upsilon }_{np}$ is the effect p for individual n (with standard normal distribution). Interactions between socio-demographic variables, attributes of the choice alternatives and type of Status Quo modes were also estimated. The most significant ones which contributed most to model performance were selected for the final model, and are reported in the table of results (Table 1).

Table 1. Results of Error Components Logit Model

	Coeff.	Std.Er	p-val.
Error components
MaaS.PT–MaaS.CS	1.287***	0.073	0.000
MaaS.PT–MaaS.BS	0.981***	0.073	0.000
MaaS.PT–MaaS.Taxi	0.456***	0.141	0.001
MaaS.CS-MaaS.BS	1.916***	0.097	0.000
MaaS.CS-MaaS.Taxi	1.132***	0.145	0.000
MaaS.BS-MaaS.Taxi	0.984***	0.131	0.000
Parameters in utility functions
Utility for SQ (No MaaS)
ASC_SQ (No MaaS)	5.446***	0.537	0.000
Individual characteristics
Age 18-35	-0.975*	0.547	0.074
Age 35-55	-0.608	0.528	0.249
Age 55-65	0.730	0.748	0.329
Older than 65	0.854
Female	0.117	0.281	0.678
Education Middle	0.585	0.549	0.287
Education High	0.538	0.565	0.341
Education Basic	-1.123
Not empl/Unempl/search	-0.867*	0.499	0.083
Student	0.578	0.763	0.449
Employed (Full/part time)	0.894	0.589	0.129
Retired	-0.604	1.341	0.652
Working from home (Yes)	-0.002
Income 0-1250€/mo	0.067	0.458	0.884
Income 1250-2500€/mo	-0.265	0.374	0.479
Income >2500€/mo	-0.703	0.492	0.153
“I don’t want to say”	0.901
Flexible Working hours(Yes)	-0.091	0.270	0.736
Trip related & Decision context
High Interest in MaaS (Yes)	-0.166	0.255	0.516
Work-related purpose (Yes)	-0.162	0.284	0.570
Weekday (Yes)	0.196	0.259	0.449
Time pressure (Yes)	0.030	0.287	0.917
Travel Time	-0.047***	0.005	0.000
Travel Cost	-1.525***	0.042	0.000
Waiting time 0 min	0.003	0.347	0.993
Waiting Time 5min	0.181	0.316	0.567
Waiting Time 10min	-0.294	0.351	0.403
Waiting Time >10min	0.109
Parking Time 0-5 min	0.038	0.088	0.662
Parking Time 5-10 min	-0.003	0.099	0.974
Parking Time 10-15 min	-0.134	0.100	0.182
Parking Time >15 min	0.099
Parking Cost 0€	0.299***	0.094	0.001
Parking Cost 1.10€/hour	0.037	0.094	0.694
Parking Cost 2.20€/hour	-0.171*	0.092	0.062
Parking Cost 3.30€/hour	-0.164
Access Time 0-3 min	-0.130	0.093	0.161
Access Time 3-6 min	0.128	0.099	0.195
Access Time 6-9 min	-0.052	0.096	0.591
Access Time 9-12 min	0.054
Egress Time 0-3 min	-0.052	0.090	0.565
Egress Time 3-6 min	-0.009	0.095	0.924
Egress Time 6-9 min	0.029	0.090	0.750
Egress Time 9-12 min	0.032
SQ_DA	-1.315***	0.149	0.000
SQ_Passeng	-1.383***	0.227	0.000
SQ_Walk	-0.004	0.174	0.982
SQ_Bike	0.526***	0.172	0.002
SQ_Ebike	2.177
Interactions
Female x Travel Time	-0.001	0.003	0.872
Female x SQ_Walk	0.159	0.099	0.110
x SQ_Bike	0.279***	0.093	0.003
x SQ_DA	0.169**	0.075	0.024
x SQ_Passeng	0.324***	0.125	0.009
Interest x SQ_Walk	-0.345***	0.124	0.005
x SQ_Bike	-0.529***	0.117	0.000
x SQ_DA	-0.640***	0.087	0.000
x SQ_Passeng	-0.382**	0.162	0.018
Work Purp. x SQ_Walk	-0.642***	0.133	0.000
x SQ_Bike	0.001	0.102	0.989
x SQ_DA	0.135	0.083	0.107
x SQ_Passeng	-0.269**	0.124	0.031
Travel Time x SQ_Walk	0.013**	0.006	0.039
x SQ_Bike	-0.002	0.006	0.700
x SQ_DA	0.045***	0.005	0.000
x SQ_Passeng	0.031***	0.007	0.000
Travel Cost x SQ_DA	1.489***	0.042	0.000
x SQ_Passeng	1.517***	0.043	0.000
Utility for MaaS-PT
ASC_MaaS PT	3.719***	0.550	0.000
Individual characteristics
Age 18-35	-0.221	0.580	0.703
Age 35-55	-0.344	0.555	0.535
Age 55-65	0.511	0.751	0.496
Older than 65	0.054
Female	0.024	0.288	0.934
Education Middle	0.474	0.566	0.402
Education High	0.225	0.578	0.697
Education Basic	-0.700
NotEmpl./Unempl/search	-0.874*	0.520	0.092
Student	0.501	0.803	0.533
Employed (Full/part time)	0.616	0.589	0.295
Retired	0.094	1.341	0.944
Working from home (Yes)	-0.337
Income 0-1250€/mo	0.141	0.489	0.773
Income 1250-2500€/mo	-0.094	0.385	0.808
Income >2500€/mo	-0.835	0.513	0.104
“I don't want to say”	0.788
Flexible Working hours (Yes)	-0.003	0.280	0.993
Trip related & Decision context
High Interest in MaaS (Yes)	-0.371	0.267	0.166
Work-related purpose (Yes)	-0.091	0.276	0.742
Weekday (Yes)	0.269	0.258	0.297
Time pressure (Yes)	0.048	0.287	0.868
Travel Time	-0.004	0.004	0.227
Travel Cost	-0.013	0.011	0.255
Waiting time 0 min	-0.034	0.061	0.576
Waiting Time 5min	0.028	0.059	0.631
Waiting Time 10min	0.001	0.059	0.983
Waiting Time >10min	0.004
No Transfers	0.162***	0.032	0.000
Access Time 0-3 min	-0.036	0.059	0.545
Access Time 3-6 min	0.036	0.062	0.561
Access Time 6-9 min	0.016	0.059	0.785
Access Time 9-12 min	-0.016
Egress Time 0-3 min	0.053	0.061	0.388
Egress Time 3-6 min	-0.010	0.059	0.868
Egress Time 6-9 min	-0.085	0.061	0.166
Egress Time 9-12 min	0.042
MaaS Bundle A	-0.224*	0.121	0.066
MaaS Bundle B	-0.102	0.121	0.400
MaaS Bundle C	0.261**	0.117	0.025
MaaS Bundle D	0.064
Interactions
Female X MaaS A	0.049	0.119	0.678
X MaaS B	-0.072	0.121	0.551
X MaaS C	-0.008	0.117	0.945
X Travel Time	0.006*	0.003	0.095
Utility for MaaS-CS
ASC_MaaS CS	1.621**	0.652	0.013
Individual characteristics
Age 18-35	-0.021	0.664	0.975
Age 35-55	-0.365	0.640	0.568
Age 55-65	0.617	0.803	0.443
Older than 65	-0.231
Female	0.038	0.298	0.899
Education Middle	0.901	0.584	0.123
Education High	0.680	0.601	0.258
Education Basic	-1.581
Not empl/Unempl/search	-0.514	0.592	0.385
Student	0.303	0.843	0.720
Employed (Full/part time)	0.535	0.646	0.408
Retired	0.175	1.483	0.906
Working from home (Yes)	-0.499
Income 0-1250€/mo	0.225	0.520	0.666
Income 1250-2500€/mo	-0.008	0.407	0.985
Income >2500€/mo	-0.818	0.568	0.150
“ I don’t want to say”	0.601
Flexible Working Hours	0.099	0.291	0.734
Trip related & Decision context
High Interest in MaaS (Yes)	-0.054	0.296	0.855
Work-related purpose (Yes)	-0.188	0.279	0.500
Weekday (Yes)	0.208	0.264	0.432
Time pressure (Yes)	0.183	0.289	0.528
Travel Time	-0.004	0.005	0.391
Travel Cost	-0.024**	0.011	0.029
Waiting time 0 min	0.053	0.084	0.528
Waiting Time 5min	-0.100	0.082	0.224
Waiting Time 10min	0.152*	0.083	0.068
Waiting Time >10min	-0.105
Parking Time 0-5 min	0.121	0.081	0.134
Parking Time 5-10 min	0.006	0.079	0.938
Parking Time 10-15 min	-0.062	0.084	0.458
Parking Time >15 min	-0.065
Access Time 0-3 min	-0.037	0.082	0.651
Access Time 3-6 min	-0.017	0.088	0.848
Access Time 6-9 min	0.066	0.086	0.444
Access Time 9-12 min	-0.012
Egress Time 0-3 min	0.010	0.082	0.899
Egress Time 3-6 min	0.000	0.083	0.996
Egress Time 6-9 min	-0.003	0.087	0.975
Egress Time 9-12 min	-0.007
MaaS Bundle A	-0.231	0.174	0.185
MaaS Bundle B	0.129	0.175	0.459
MaaS Bundle C	0.357**	0.173	0.039
MaaS Bundle D	-0.255
Interactions
Female X MaaS A	0.263	0.168	0.117
X MaaS B	-0.041	0.164	0.800
X MaaS C	-0.274	0.179	0.125
X Travel Time	0.005	0.004	0.163
Utility for MaaS-BS
ASC_MaaS BS	2.978***	0.582	0.000
Individual characteristics
Age 18-35	-0.193	0.623	0.757
Age 35-55	-0.151	0.611	0.805
Age 55-65	1.154	0.784	0.141
Older than 65	-0.810
Female	0.076	0.293	0.795
Education Middle	0.773	0.604	0.200
Education High	0.696	0.628	0.268
Education Basic	-1.469
Not empl/Unempl/search	-0.729	0.560	0.193
Student	0.505	0.817	0.536
Employed (Full/part time)	0.210	0.621	0.736
Retired	0.298	1.370	0.828
Working from home (Yes)	-0.284
Income 0-1250€/mo	0.036	0.521	0.946
Income 1250-2500€/mo	-0.282	0.398	0.479
Income >2500€/mo	-0.668	0.551	0.225
“I don’t want to say”	0.915
Flexible Working Hours	0.019	0.292	0.949
Trip related & Decision context
High Interest in MaaS (Yes)	-0.227	0.291	0.436
Work-related purpose (Yes)	-0.133	0.283	0.639
Weekday (Yes)	0.220	0.264	0.405
Time pressure (Yes)	-0.027	0.287	0.925
Travel Time	-0.034***	0.002	0.000
Travel Cost	-0.300**	0.129	0.020
Access Time 0-3 min	0.108	0.083	0.195
Access Time 3-6 min	0.103	0.084	0.222
Access Time 6-9 min	-0.065	0.082	0.429
Access Time 9-12 min	-0.146
MaaS Bundle A	-0.423*	0.250	0.091
MaaS Bundle B	0.358**	0.165	0.030
MaaS Bundle C	0.665***	0.166	0.000
MaaS Bundle D	-0.599
Interactions
Female X MaaS A	0.040	0.164	0.806
X MaaS B	0.286*	0.156	0.066
X MaaS C	-0.301**	0.148	0.043
X Travel Time	-0.005**	0.002	0.024
Utility for MaaS-Taxi
Trip related & Decision context
Travel Time	-0.023	0.020	0.248
Travel Cost	-0.005	0.008	0.534
MaaS Bundle A	-0.147	0.411	0.721
MaaS Bundle B	0.155	0.459	0.735
MaaS Bundle C	0.227	0.509	0.656
MaaS Bundle D	-0.235
Log Likelihood (estimated) Log Likelihood (base) McFadden Adjusted Rho-sq AIC AIC/N BIC	-7320.143 -10391.366 0.434 14998.285 1.856 15339.712

***, **, * Significance at 1%, 5%, 10% level.

Lastly, the probability $P$ for individual $n$ to select transportation mode $j$ in choice situation $t$ among the set of alternatives $q$ ($q$= 1, … j …,Q), is as shown in Equation 3(3) $P_{jnT}=\frac{\text{exp}\left(AS{C}_j+{\displaystyle {\sum }_k{\beta }_k}\left(x_{nkj}\right)+{\displaystyle {\sum }_z{\beta }_z}\left(x_{nz}\right)+{\displaystyle {\sum }_h{\beta }_h}\left(x_{nh}\right)+{\displaystyle {\sum }_{k^{\prime }\subset k}{\displaystyle {\sum }_{z^{\prime }\subset z}{x}_{n{k}^{\prime }}}}*x_{n{z}^{\prime }}+{\displaystyle {\sum }_p{I}_{jp}}{\sigma }_{jp}{\upsilon }_{np}\right)}{{\displaystyle {\sum }_{q=1}^Q\text{exp}}\left(AS{C}_q+{\displaystyle {\sum }_k{\beta }_k}\left(x_{nkq}\right)+{\displaystyle {\sum }_z{\beta }_z}\left(x_{nz}\right)+{\displaystyle {\sum }_h{\beta }_h}\left(x_{nh}\right)+{\displaystyle {\sum }_{k^{\prime }\subset k}{\displaystyle {\sum }_{z^{\prime }\subset z}{x}_{n{k}^{\prime }}}}*x_{n{z}^{\prime }}+{\displaystyle {\sum }_p{I}_{jp}}{\sigma }_{jp}{\upsilon }_{np}\right)}$(3) . The following equation expresses the likelihood function $L_n$ (Equation 4(4) $L_n={\int }_{{\eta }_j}^{}{\prod }_{t\in T}{\prod }_{j\in J}\left[{\left(P_{jnT}\right)}^{y_{jnT}}\frac{1}{{\sigma }_{jp}}\phi \left({\eta }_j\right)\right]$(4) ) which was estimated by simulation.

(1) $\begin{array}{rl}& U_{n\left(S{Q}_{1-m}\right)t}=AS{C}_{SQ}+{\beta }_{SQ}\ast I_{S{Q}_{1-m}}+{\sum }_k{\beta }_k\left(x_{nkSQ}\right)+{\sum }_z{\beta }_z\left(x_{nz}\right)\\ & +{\sum }_h{\beta }_h\left(x_{nh}\right)+{\sum }_{k^{\prime }\subset k}{I}_{S{Q}_{1-m}}\ast x_{n{k}^{\prime }}+{\sum }_{z^{\prime }\subset z}{I}_{S{Q}_{1-m}}\ast x_{n{z}^{\prime }}\\ & +{\sum }_{h^{\prime }\subset h}{I}_{S{Q}_{1-m}}\ast x_{n{h}^{\prime }}+{\sum }_{k^{\prime }\subset k}{\sum }_{z^{\prime }\subset z}{x}_{n{k}^{\prime }}\ast x_{n{z}^{\prime }}+{\epsilon }_{njt}\end{array}$(1)

$\begin{array}{rl}& U_{n\left(j\right)t}=AS{C}_j+{\sum }_k{\beta }_k\left(x_{nkj}\right)+{\sum }_z{\beta }_z\left(x_{nz}\right)+{\sum }_h{\beta }_h\left(x_{nh}\right)\\ & +{\sum }_{k^{\prime }\subset k}{\sum }_{z^{\prime }\subset z}{x}_{n{k}^{\prime }}\ast x_{n{z}^{\prime }}+{\sum }_p{I}_{jp}{\sigma }_{jp}{\upsilon }_{np}+{\epsilon }_{njt}\end{array}$

(2) ${\eta }_j={\sigma }_{jp}{\upsilon }_{np}\sim N\left(0,{\sigma }_{Jp}^2\right)$(2)

(3) $P_{jnT}=\frac{\text{exp}\left(AS{C}_j+{\displaystyle {\sum }_k{\beta }_k}\left(x_{nkj}\right)+{\displaystyle {\sum }_z{\beta }_z}\left(x_{nz}\right)+{\displaystyle {\sum }_h{\beta }_h}\left(x_{nh}\right)+{\displaystyle {\sum }_{k^{\prime }\subset k}{\displaystyle {\sum }_{z^{\prime }\subset z}{x}_{n{k}^{\prime }}}}*x_{n{z}^{\prime }}+{\displaystyle {\sum }_p{I}_{jp}}{\sigma }_{jp}{\upsilon }_{np}\right)}{{\displaystyle {\sum }_{q=1}^Q\text{exp}}\left(AS{C}_q+{\displaystyle {\sum }_k{\beta }_k}\left(x_{nkq}\right)+{\displaystyle {\sum }_z{\beta }_z}\left(x_{nz}\right)+{\displaystyle {\sum }_h{\beta }_h}\left(x_{nh}\right)+{\displaystyle {\sum }_{k^{\prime }\subset k}{\displaystyle {\sum }_{z^{\prime }\subset z}{x}_{n{k}^{\prime }}}}*x_{n{z}^{\prime }}+{\displaystyle {\sum }_p{I}_{jp}}{\sigma }_{jp}{\upsilon }_{np}\right)}$(3)

(4) $L_n={\int }_{{\eta }_j}^{}{\prod }_{t\in T}{\prod }_{j\in J}\left[{\left(P_{jnT}\right)}^{y_{jnT}}\frac{1}{{\sigma }_{jp}}\phi \left({\eta }_j\right)\right]$(4)

$y_{jnt}$= 1 if individual $n$ chooses alternative $j$ in choice situation $t$, 0 otherwise.

Results

Table 1 reports the results of the model. The goodness of fit of the model as indicated by McFadden’s adjusted Rho-Squared is 0.434.

Inertia effects

The respondents’ choice sets include 4 or 5 alternatives depending on their reported transportation mode. More precisely, since all four MaaS bundles include E-Bike Sharing (E-BS), Public transportation (PT), Car sharing (CS), and Taxi, if respondents reported one of these modes as their current mode for a certain trip, their choice set includes four alternatives (which exist within their MaaS bundle). Alternatively, when respondents previously reported a trip based on walking, E-/biking, car as a Driver or Passenger, they have 5 alternatives in their choice set. Respondents could either remain with their current transportation mode (Status Quo-SQ) or choose one of the alternatives available in their MaaS bundle. The model allowed for covariances among the four MaaS modes. Results revealed the highest covariance between MaaS_CS and MaaS_BS (1.916). The smallest covariance exists between MaaS_PT and MaaS_Taxi (0.456), which is expected given the very different nature of these two transportation modes.

The sign of the SQ constant is positive (5.446) and significantly higher than the other 4 constants, indicating that respondents tend to stick with their current transportation mode. About the effects of socio-demographics on inertia, we notice a negative significant coefficient for young millennials (−0.975), indicating that younger age groups are more eager to choose a MaaS mode compared to older age groups. People are sensitive to trip duration (−0.047) and cost (−1.525), which are seen as important factors to choose a MaaS alternative. This sensitivity to price is also confirmed by the positive coefficient (0.299) for free parking and the negative parameter for higher parking cost (−0.171). Regarding the effect of occupational status, we notice that those looking for a job are less eager to stick with their status Quo mode (−0.867). It is also evident that people are sensitive to waiting and parking time, yet the estimated coefficients are not statistically significant.

The tendency to continue using the current transportation mode varies between modes. The estimated inertia effects for car drivers (−1.315) and car passengers (−1.383) signal a lower tendency for these groups to stick with the reported transportation mode, while bikers are mostly inclined to keep using their bike for the reported trip (0.526).

Moreover, we included interaction effects between inertia, and socio-demographic and contextual variables. Results indicate that inertia effects are larger for female Bikers (0.279), Drivers (0.169), and Passengers (0.324). The effect of the variable measuring interest in MaaS was estimated in combination with the type of current transportation mode. The negative signs of the coefficients for Walkers (−0.345), Bikers (−0.529), Drivers (−0.640), and Passengers (−0.382) suggest that in general people with a higher interest in MaaS show a significantly lower tendency to continue using their current mode than average, but this tendency differs between current modes. Moreover, it seems that in case of commuting trips, those who previously walked (−0.642) or traveled as Passengers (−0.269) have a higher tendency to choose one of the MaaS modes. Although users, on average, are sensitive to higher travel expenses and travel time, the interaction effect between travel time and current transportation mode suggests that this sensitivity is lower for car drivers and passengers, followed by walkers (evidenced by the estimated interaction effect of 0.045, 0.031, and 0.013, respectively).

Choice of transportation modes within MaaS bundle

The estimated constants for PT (3.719), CS (1.621), and BS (2.978) as MaaS options indicate that respondents have a higher tendency to choose public transportation, followed by bike sharing, from the inventory of transportation modes in their MaaS bundle. The unemployment and people in search of job are less inclined to choose MaaS public transportation (−0.874). As expected, if their reported travel does not need any transfer (using PT), they are more likely to choose PT (0.162). As for the influence of type of bundle, the results show that subscribing to Bundle A where travelers have to pay per ride leads to a lower tendency of choosing MaaS-PT (−0.224), while Bundle C with 40% discount for public transportation positively impacts the tendency of choosing this mode (0.261). The effect of Bundle D (which offers free usage of PT) is smaller than for Bundle C. An explanation could be that since users pay a relatively high monthly fee (€399) for Bundle D, some of them will be inclined to use other fancier transportation modes even if the use of PT is free. Sensitivity to cost and trip duration can also be examined by the marginal effects. Results show that the marginal effect of cost and travel time is respectively 0.006 and 0.018.

Regarding choosing Car-sharing, results reveal that travel expenses negatively affect the choice of MaaS-CS (−0.024). Coefficients associated with waiting times show that respondents prefer to have a 5 to 10 minutes time lag between ordering and arrival of the vehicle (0.152). Moreover, the pricing scheme of Bundle C (20% discount for CS, 40% discount for PT, and 20% discount per Taxi ride) appears to be the most attractive to choose car sharing (0.357). Although Bundle D gives higher discount on Car-sharing (30%), this bundle shows a smaller effect, perhaps due to a very interesting pricing scheme (free usage) of PT for this bundle. The reported marginal effects (Table 2) of travel cost and time for MaaS-CS are 0.016 and 0.008, respectively.

Table 2. Marginal Effects

	Coeff.	Std.Er	p-val.
Marginal Effect Travel Cost
MaaS-PT	−0.006***	0.9760D-04	0.000
MaaS-CS	−0.016***	0.00021	0.000
MaaS-BS	−0.006***	0.00018	0.000
MaaS-TAXI	−0.003***	0.4923D-04	0.000
Marginal Effect Travel Time
MaaS-PT	−0.018***	0.00015	0.000
MaaS-CS	−0.008***	0.8710D-04	0.000
MaaS-BS	−0.077***	0.00067	0.000
MaaS-TAXI	−0.007***	0.8631D-04	0.000

*** Significance at 1% level.

Regarding the fourth alternative MaaS-BS, its utility decreases with increasing trip duration (−0.034) and travel cost (−0.300). As expected, using bike or ebike as part of MaaS increases as these options are offered for free, which is the case for bundles B (0.358) and C (0.665). Although Bundle D offers free bike as well, respondents are less inclined to choose this transportation mode compared to Bundle B and C, possibly due to the high monthly price of the package, leading them to use otherwise more costly modes. The interaction of gender and bundle shows that BS becomes more attractive for women for Bundle B (0.286) and less desirable for women for bundle C (−0.301). Table 1 also reveals that females are more sensitive to travel time and that an increase in the trip duration de-motivates them to choose this option (−0.005). Table 2 shows the marginal effect of cost is 0.006 and of duration is 0.077 with respect to MaaS-BS.

Finally, not many parameters could be estimated for the MaaS-TAXI mode since this mode was chosen only a few times. The signs of the estimated parameters are as expected, yet not significant. Table 2 shows that the marginal effects for MaaS-Taxi are 0.003 and 0.007 for, respectively, travel cost and time.

Discussion/conclusion

MaaS is attracting a lot of attention from policymakers, industry, and academia with its potential to draw travelers away from using private cars and, therefore, offering a new sustainable transportation mode. The aim of this study was to shed light on the influence of a set of factors on the usage of MaaS transportation modes. The design of the study reflects this focus. The paper discusses an approach to the analysis of mode choice assuming that people have subscribed to this new service. Important to highlight the key role of the consideration of various mobility bundles for what concerns the design of this survey and our model estimation.

Results show that it is not price per se but the combination of monthly fees and the discounts for various transportation modes within a specific bundle that primarily influence the increase or decrease of the use of a specific mode included in the MaaS bundle. Another important and assuring observation is that travelers who currently are drivers or passengers of privately owned cars are less willing to continue using their current mode compared with walkers or bikers. That might be in contradiction with the expectations assuming that getting people out of their private car is more difficult than convincing people who are already multi-modal to use MaaS. Our results, however, should be interpreted keeping in mind that the sample who were recruited were the ones who indicated relatively high interest in MaaS. Preserving such condition was necessary for our study as the respondents would have to assume that they had already subscribed to MaaS.

While the outcome of this work can be used to understand the degree of switching to various transportation modes once MaaS is adopted, the total changes in demand can only be accurately estimated combining these research results with those associated with the prediction of the demand for MaaS subscription. (Caiati, Rasouli, and Timmermans 2019). The prediction of market shares under certain scenarios requires representative activity-travel patterns of a synthetic population to which the probability of choosing a particular bundle (including the ‘none’ option) (e.g. Caiati, Rasouli, and Timmermans 2019) and the estimated coefficients in this study are applied.

Acknowledgments

The authors are grateful to the Netherlands Organization for Scientific Research (NWO) for funding this research. This work is part of the project “SCRIPTS”, which is part of the program “Smart Urban Region of the Future” (SURF).

Disclosure statement

No potential conflict of interest was reported by the authors.