Conference PaperPDF Available

PREFERENCES FOR SHARED MODES OF LOCAL PUBLIC TRANSPORT USERS IN THE URBAN LAST-MILE

Abstract and Figures

Shared transport creates an opportunity to facilitate the last mile connections of public transport (PT) trips. Nevertheless, user preferences for using such shared modes as last-mile connection have hardly been studied. To explore such preferences within urban areas we have designed and conducted a stated choice experiment in the province of Utrecht, the Netherlands. In the experiment respondents were able to choose from shared bicycles, e-bikes, e-scooters, and e-mopeds to reach their final destination from a PT stop. A sample of 285 respondents considered their last-mile mode choice of a recent PT trip in light of the new options. Results show that shared (e)-bicycles are generally preferred over e-scooters and e-moped; still a majority of the urban PT travellers prefers not to use shared modes. We also found age and cycling experience to be important determinants for using shared modes. Last-mile travel times have a limited impact on the preferences shared mobility as last mile whilst frequent PT use has a negative relationship with using shared modes.
Content may be subject to copyright.
PREFERENCES FOR SHARED MODES OF LOCAL PUBLIC TRANSPORT USERS IN1
THE URBAN LAST-MILE2
3
4
5
Roy J. van Kuijk, Corresponding Author6
Delft University of Technology7
Faculty of Civil Engineering & Geosciences8
P.O. Box 5049, 2600GA, Delft9
The Netherlands10
Phone: +31 (0)15 278 91 2911
E-mail: r.j.vankuijk@tudelft.nl12
13
Gonçalo Homem de Almeida Correia14
15
Niels van Oort16
17
Bart van Arem18
19
20
Word Count: 6422 words +5 table(s) ×250 =7672 words21
22
23
24
25
26
27
Submission Date: January 20, 202128
Van Kuijk, Correia, Van Oort, Van Arem 2
ABSTRACT1
Shared transport creates an opportunity to facilitate the last mile connections of public transport2
(PT) trips. Nevertheless, user preferences for using such shared modes as last-mile connection3
have hardly been studied. To explore such preferences within urban areas we have designed and4
conducted a stated choice experiment in the province of Utrecht, the Netherlands. In the experi-5
ment respondents were able to choose from shared bicycles, e-bikes, e-scooters, and e-mopeds to6
reach their final destination from a PT stop. A sample of 285 respondents considered their last-mile7
mode choice of a recent PT trip in light of the new options. Results show that shared (e)-bicycles8
are generally preferred over e-scooters and e-moped; still a majority of the urban PT travellers9
prefers not to use shared modes. We also found age and cycling experience to be important deter-10
minants for using shared modes. Last-mile travel times have a limited impact on the preferences11
shared mobility as last mile whilst frequent PT use has a negative relationship with using shared12
modes.13
14
Keywords: Shared Mobility, Mode Choice, Last Mile, Micromobility, Transit15
Van Kuijk, Correia, Van Oort, Van Arem 3
INTRODUCTION1
Public transport (PT) brings travellers from one stop to another. To access PT, a traveller needs2
to bridge the distance between his/her origin and the stop. After alighting at the destination stop,3
again, the travellers needs to bridge the distance to reach the destination. These trip stages are4
commonly known as the first and last-mile of PT trips.5
Satisfaction with PT trips is strongly related to the travel experience in the first and last-6
mile (1). Negative experiences with these trip stages could, therefore, limit the willingness for7
people to use PT. To improve livability and sustainability in cities, there is an on-going challenge8
to improve PT systems and, thus, the first and last-mile of PT trips.9
Considering a trip from home to work, the first-mile is home-bound and the last-mile10
activity-bound. For the return trip back home this is vice versa. The work of Hoogendoorn-Lanser11
et al. (2) shows that mode choice differs between home-bound and activity-bound trip stages. One12
explanation is that the availability of transport alternatives generally differ. For the home-bound13
legs, instead of walking, many people can use privately-owned means such as bicycles or cars.14
These could help smoothen this specific trip component. As a consequence, the activity-bound trip15
part is still problematic as it cannot be easily covered by private vehicles.16
In this paper, we specifically analyse first-mile trips that are home-bound and the last-mile17
to be activity-bound. Therefore, the last-mile throughout the paper will consists of a trip stage that18
takes a person from a stop (of a bus or tram) to the activity at the destination location (or vice19
versa). It is mainly this part of a PT trip where shared modes can potentially fill the gap left by the20
absence of private vehicles; they expand the set of transport means to conduct the last-mile.21
Shaheen and Chan (3) define shared modes as transport means that can be accessed by peo-22
ple to enable transport on a short-term "as-needed basis". Shared micro modes can be considered23
as an important sub-category of shared modes that have been experiencing a significant demand24
growth over the recent years. These micro modes are characterized by their small size and the25
presence of an electric (support) engine. Examples of shared micro modes are: shared bicycles,26
e-bikes, e-scooters, and e-mopeds. They are suitable for large-scale urban use given their small27
size and limited local environmental impacts (i.e. noise and pollutants).28
Nevertheless, there is limited knowledge about which shared modes are preferred by trav-29
ellers in the last-mile. Yan et al. (4) found a correlation between age and acceptance of shared30
bicycles in the last-mile; younger people show a higher propensity to use them. Adnan et al. (5)31
considered shared bicycles for train travellers in Belgium. Yan et al. (6) focused on ride-sourcing32
for urban PT travellers in Michigan, United States. The study of Yap et al. (7) included autonomous33
driving vehicles as a last-mile alternative for train travellers in the Netherlands.34
Apart from these few studies focussing on user preferences for shared modes in the last-35
mile, fortunately, there are more studies having a more general scope. Most of these studies focus36
on the user preferences for shared bicycles and e-scooters. Generally, the following aspects deter-37
mine mode choice: level of service, trip characteristics, individual and household characteristics,38
weather and season, and built environment (e.g. Buehler (8)). We provide an overview on shared39
mobility preference regards only the first three aspects as they affect mode choice on a larger spatial40
and temporal scale.41
With regards to service requirements, shared bicycles users view convenience, availability42
and saving money as important attributes for participating in shared bicycle schemes (9). The43
study of Bachand-Marleau et al. (10) shows that shared bicycles mainly replace walk and cycle44
trips, yet also PT trips. Fitt and Curl (11) report that shared e-scooters replace walk trips (58%)45
Van Kuijk, Correia, Van Oort, Van Arem 4
and only limitedly replace cycle trips (6%). Bai and Jiao (12) produced e-scooter trip statistics1
(average distance, travel time, and speed) from two US cities and found those range between2
typical walking and cycling values.3
Related to trip characteristics, Fishman (13) found shared bicycle use to peak during week-4
day rush hours and around midday in the weekends. Most bike-sharing scheme members use5
shared bicycles for commuting (14)(15). McKenzie (16) does not find a shared e-scooter demand6
pattern with commute peaks and concludes that e-scooters are used for other trip purposes. On7
the contrary, a German field test showed that e-scooters are most likely to be used for commute,8
business and leisure trips (17). Fitt and Curl (11) found more nuanced findings on shared e-scooter9
trip purpose. First time users are most likely using them for fun or out of curiosity; subsequent10
users are more likely to conduct commuting or shopping trips or use e-scooters for visiting family11
and friends.12
With respect to individual characteristics, many researchers found gender to have an effect13
on using shared bicycles. Less than 20 percent of the trips within the London bike-sharing scheme14
are made by women (18). The percentage of women in bike-sharing schemes in Melbourne and15
Brisbane are, respectively, 23 and 40 percent (9). The limited use of shared bicycles by women16
likely relates to a lower inclination for cycling in general. In the UK, the USA and Australia,17
typically low-cycling countries, most cycle trips (65-90%) are conducted by men Buehler and18
Pucher (14). This differs from the situation in the Netherlands where many people cycle; in this19
country, women cycle more than men (19).20
The literature review of Fishman (13) shows that shared bicycle users tend to have a higher21
average income, a higher education status, and a paid occupation. A Chinese survey showed that22
gender, current cycling level, familiarity with shared bicycles and positive attitudes towards cy-23
cling being environmental friendly have the largest impact on using shared bicycles (20). Montreal24
(Canada) bike-sharing data shows that PT use, combined PT-bicycle use and driver’s license pos-25
session are important determinants for using shared bicycles (10) With regards to shared e-scooters,26
Jiao and Bai (21) found that these are mainly used by young, male and high educated people. In27
contrast, they found a negative relationship with household income. Degele et al. (22) studied28
the individual characteristics of shared e-scooter users in Stuttgart, Germany. They found most29
e-scooter users to be male and between the age of 25 and 35.30
Not only shared mode attributes will have an impact on user preferences in the last-mile31
context. Also PT (main stage) attributes impact the last-mile distance accepted by PT travellers.32
Typically, researchers refer to this distance as the catchment area of a PT stop. Daniels and Mulley33
(23) show that the size of the catchment areas is related to the PT mode being egressed; people34
tend to accept longer egress distances for trips made by train over trips made by bus.Brand and35
Hoogendoorn (24) and Rijsman et al. (25) show that an increase of speed and service frequency36
for bus and tram is related to larger catchment areas. They also show that for longer last-mile37
distances, fewer people walk and more people cycle.38
To recap, we found that most studies do not focus on a last-mile context. The studies which39
do only focus on a limited set of modes. Therefore, at this point, it is difficult to explore the trade-40
offs made by PT travellers in the last-mile; this hinders the possibility to understand which ones41
should be provided in each case. Another hiatus in research is that user preference studies tend to42
focus on PT trips done by train; PT trips done using a bus or tram are generally not considered.43
As a consequence, this leaves the exploration of shared modes as a last-mile solution for local PT44
largely untouched.45
Van Kuijk, Correia, Van Oort, Van Arem 5
This paper contributes to these knowledge gaps by addressing the preferences of urban bus1
and tram users towards shared modes in the last-mile. Multiple types of shared modes are included:2
bicycles, e-bikes, e-scooters, and e-mopeds. The key methodology to explore user preferences for3
these modes is by estimating discrete choice models. Input data is collected employing a stated4
choice experiment. Using this method, we aim to identify the shared modes of transport which5
facilitate the last-mile of many urban PT travellers. This could improve their door-to-door PT6
travel experience. The study area covers the Utrecht province, the Netherlands. This is a suitable7
study area as an extensive bus and tram network is available, the use of bicycles is largely accepted,8
and population densities show a good diversity.9
We structured this paper as follows. First, we describe the research methodology. Then,10
we provide more details about the data collection. Next, we provide the results of our study and11
discuss the main findings. Finally, we provide conclusions based on this study and set out recom-12
mendations for future research.13
METHODOLOGY14
This section starts with setting out the survey structure. Next we provide the choice context and15
attribute levels and the design of the stated choice experiment.16
Survey structure17
The Qualtrics tool (26) was used for the development and distribution of the online survey. The18
survey consisted of four separate blocks. In the first block respondents were asked to provide basic19
personal information. In the second block, the respondent had to provide detailed information20
about one recent trip made by local PT. Each trip which included a bus or tram was considered a21
valid trip for this study; this means that some respondents provided a trip which also included the22
use of trains. Table 1 shows the individual and trip characteristics which had been included for the23
initial model estimation.24
TABLE 1 Included individual and trip characteristics
Individual characteristics Trip characteristics
• Age • Travel Date
Gender Departure & arrival time
Household composition Fare type
Gross monthly income Fare reimbursement by employer
Posession of driver’s license Travel purpose
Weekly car use Home-bound direction
Weekly bicycle use Modes used
Weekly public transport use (in chronological order)
Purposes for PT trips Departure and arrival
in the previous 6 months location population density
Through the Google Maps API we were able to obtain the postal code of the activity-25
bound destination. We cross-referenced this information with a database from the Dutch census26
consisting of postal codes and a 5-level urbanization classification (27). In this study, only PT trips27
Van Kuijk, Correia, Van Oort, Van Arem 6
with a class 4 or 5 (urbanized and strongly urbanized) destinations had been considered. PT trips1
to destinations with a lower urbanization classification have been considered to be sub-urban; these2
were out of the scope of this study. In the third block the stated choice experiment was provided.3
The next sections provide detailed information about the experiment.4
FIGURE 1 The shared modes presented to the respondents in the last-mile mode choice ex-
periment: bicycle, e-bike, e-scooter, and e-moped
Choice context5
In the stated choice experiment, participants reconsidered their last-mile mode choice of a recent6
PT trip with a bus or a tram. The shared modes shown to the respondents in this experiment were:7
bicycle, e-bike, e-scooter, and e-moped. We presented these modes both graphically and textually.8
Figure 1 shows how the shared modes were depicted to the respondents.9
Respondents could also choose not to use a shared mode. In this case they could choose to10
use a private mode (including walking) or cancel the trip. The latter could be chosen when none of11
the alternatives, and their attributes, would not satisfy their travel needs.12
We expected that many participants were not acquainted with shared mobility. They might13
also have issues understanding novel modes of transport, in particular e-scooters and e-mopeds.14
For that reason a complete introduction about the modes was provided to the participants. First,15
we provided a graphic overview of the included modes along with some brief characteristics about16
them. We listed the applicable driver requirements, traffic regulations and propulsion system.17
For some modes the driver needs to wear a helmet or possess a driver’s license. For each mode18
the maximum speed and network type (e.g. foot path or bicycle lane) were mentioned. Besides,19
information was provided about the presence of an electric (support) engine and the need for human20
effort.21
We also mentioned the context of using these shared modes with respect to the access/payment22
method, pick-up/return policy, and pricing scheme. We expected these to be most crucial for un-23
derstanding the specific choice context. These aspects have been formalized within our experiment24
and were mentioned to the respondents. Travelers need to use the same PT smart card as in bus25
or tram. The shared vehicle can be (un)locked by tapping this card onto the vehicle’s card respon-26
der. Return policies for using shared modes differ between the urban and sub-urban environment.27
Urban users face a one-way renting policy, allowing them to leave their vehicle at any safe place28
near their destination. The price for using the shared modes were presented as additional to the29
PT fare. All shared modes had fixed prices and there were no limitations in place regarding the30
mileage driven and time spent with these vehicles.31
Before starting the experiment, we listed the attributes which were varied during the exper-32
iment and explained the meaning of each of these attributes. To strengthen the understanding of33
Van Kuijk, Correia, Van Oort, Van Arem 7
the experiment, we repeated the trip details of the reference trip.1
We provided 12 choice situations to the respondent. Figure 2 depicts what was presented2
to respondents online. Alternatives which require a valid driver’s license were not shown to re-3
spondents who did not possess a driver’s license. After the choice experiment itself, we added the4
fourth and final part to our survey. Here, respondents stated to what temporal extent (5-unit Likert5
scale: (almost) never - (almost) always) they perceive difficulties with regards to visual, hearing6
and physical impairment. In the survey, physical impairment referred to the ability of walking and7
climbing stairs. We also surveyed whether the respondent experiences difficulties understanding8
the PT system.9
FIGURE 2 Example of one of the experiment choice tasks
Attribute levels10
We varied the alternative attributes such that they represent different last-mile and PT (main trip11
stage) contexts for the participant’s reference trip. PT attributes were included to explore the12
impact of the main trip part on mode choice behaviour in the last mile of existing trips. Therefore,13
we explicitly mentioned that the choice experiment also requires the participant to consider their14
stop of choice where they alighted the PT vehicle and started their last mile. The alternatives15
Van Kuijk, Correia, Van Oort, Van Arem 8
represent stop locations which can be situated nearer or farther from their destination with respect1
to the current stop location.2
Table 2 provides an overview of the included attributes and its levels. For the last mile we3
varied the costs (pricing) and the travel time of the shared modes. Prices were varied such that they4
represented realistic market pricing. We were also interested in how PT travellers would respond5
to shared modes without any additional charge. We therefore decided to set one of the pricing6
levels for shared bicycles and e-bikes to zero. The travel times were varied such that they translate7
last mile distances between 300-1500 metres. We chose to present travel times to the respondents8
rather than travel distances; This was done as we expected that respondents relate travel impedance9
more to travel time.10
For the main part of the trip, we varied the PT frequency and PT in-vehicle time as binary11
attributes. The PT frequency was either equal or double relative to the reference trip value. The PT12
in-vehicle travel time was either set equal or 25 percent lower value with respect to the reference13
trip.14
TABLE 2 Attribute levels varied in the stated choice experiment
Shared Shared Shared Shared Private
Bicycle E-bike E-scooter E-Moped Mode
Last-mile attributes
Travel time (minute) [2;6;10] [1.5;4.5;7.5] [1.5;4.5;7.5] [1;3;5] [3;9;15]*
Travel costs (euro) [0;0.75;1.5] [0;1;2] [0.5;1;1.5] [1;2;3]
PT (main stage) attributes**
Frequency (%) [100,200] [100;200] [100;200] [100;200] [100;200]
In-vehicle travel time (%) [75;100] [75;100] [75;100] [75;100] [75;100]
* = Travel time for private modes were provided with walking time as a reference
** = Reference trip of respondent is considered base level (100%)
Van Kuijk, Correia, Van Oort, Van Arem 9
Experiment design1
We constructed the experiment design by means of the NGENE software package (28). Given the2
novelty of the research subject, we preferred an orthogonal design which enables estimating the3
effects independently of other included attributes. We used a orthogonal fractional factorial design4
for the choice experiment. This design consisted of 36 choice tasks. Therefore, we separated three5
different blocks out of this design in order to limit the participant’s cognitive load.6
MODEL SPECIFICATION AND ESTIMATION7
We estimated three different types of discrete choice models and compared their performances8
to find the most statistically significant model representation of the respondent’s stated choice9
behaviour. Apart from a multinomial logit (MNL) model, we estimated more complex ones such10
as nested logit (NL) models, panel-effect models and mixed logit (ML) models. Background11
information on these discrete choice model specification can be found in e.g. Train (29) and12
Hensher et al. (30).13
Each of the estimated models is based on the random utility maximization framework. This14
assumes that every respondent nof each choice task twould only choose alternative iwith utility15
Uint, when Uint >Ujnt ,j6=i.16
describes how utility function Uint is defined for the MNL model. The utility function17
Uint consists of the systematic utility Vint and the error term εint . The systematic utility consist of18
β0
int being a vector of attribute-specific parameters and xint being a vector of the attribute values.19
The MNL model assumes that the error terms are independently and identically (IID) Gumbel20
distributed.21
4Uint =Vint +εint =β0
int xint +εint,Cn(1)22
23
NL model specifications can be suitable in case choice probabilities between pairs of alter-24
natives are correlated. We estimated multiple NL models and did not find grounds to continue the25
estimation of NL models; i.e. we were not able to estimate a statistically significant nest parameter.26
Therefore, we decided to focus on alternative model specifications.27
The panel-model mixes the specification of a MNL model with a panel effect error term.28
This panel effect error term accounts for the correlation between error terms over multiple choice29
tasks for the same respondent. This enables the random error term εint to be Gumbel IID for each30
respondent and choice task. In our model specification, we consider the panel effect parameter αin
31
to be N(0,sigma) distributed for each alternative iand respondent n. describes the utility function32
within the panel-effect model specification.33
4Uint =Vint +αin +ε0
int ,Cnt (2)34
35
A ML model allows for the estimation of distributions around parameters which are the36
result of heterogeneity of preferences between respondents. This coefficient ξint represents the37
distribution independently over each respondent and choice task. In our model, we consider ξint to38
be N(0,σint ). describes the utility function in the ML model specification.39
4Uint =Vint +αin +ξint +ε0
int ,Cnt (3)40
41
Van Kuijk, Correia, Van Oort, Van Arem 10
We estimated all models using the Python Biogeme software package (31). We aimed to1
identify the model which provides the most statistically significant representation of the choice2
behaviour. We therefore defined a number of performance measures: the percentage of correctly3
predicted choices, the log-likelihood ratio and the adjusted Rho-squared as performance measures.4
The first is a direct measure of the predictive power, while the latter two are measures for the5
model’s statistical fitness to the data.6
We started with the estimation of a MNL model. During this process, we extended the7
MNL model by adding parameters until the log-likelihood ratio stopped improving significantly.8
This MNL model was considered the best MNL model. Next, we introduced the variables which9
provided significant parameters in the MNL model in the specifications of the panel-model and10
the ML model. The more complex structure of these models causes some parameters to become11
insignificant. We therefore left these parameters out and reintroduced some of them such that we12
could not significantly improve the log-likelihood ratio anymore. In all estimated models, “not13
using shared modes” was defined as the reference alternative with its alternative-specific constant14
fixed to zero.15
DATA COLLECTION16
This section describes the context in which stated choice data was collected. Next, we provide the17
statistics of the people who participated in the survey.18
Case study: province of Utrecht19
We collected our survey data in the province of Utrecht, the Netherlands. This province is densely20
populated (904 inh./km2) and is located in the centre of the Netherlands. Many of the 1,34 million21
people inhabitants live in cities such as Utrecht, Amersfoort or Veenendaal.22
The province of Utrecht possesses a large local public transport network. As such, it com-23
plements the extensive network of train services which provides direct connections to many Dutch24
regions. Within the cities and their direct surroundings, local PT is provided mainly through bus25
services, yet there are also a few tram services available. Most of these are provided on high26
frequency schedules (>6 vehicles/hour).27
We have conducted our survey from January 20 until February 17, 2020 which did not28
include the covid-19 health crisis. Only people who had done a local PT trip in Utrecht over the29
last three months were included in the survey. We used two different methods to approach these30
PT users. One consisted of a direct approach by e-mail. We used the e-mail addresses collected in31
a recent in-vehicle survey (with the purpose to sample the composition of the local PT user group).32
Another consisted of an marketing campaign broadcast on in-vehicle displays to prompt travellers33
to visit the survey website.34
Sample statistics35
Our survey and the included stated choice experiment was answered by a final sample of 28536
urban respondents. We have compared the sample statistics with the composition of the local PT37
travellers in Utrecht. Information about the population was established by means of an in-vehicle38
survey, which was conducted pre-pandemic for 14 consecutive days (January, 20 - February, 2,39
2020) and included all bus and tram service. Its sample of 2363 travellers is considered to provide40
a good presentation of the population of local PT travellers in the province of Utrecht. Table 341
compares our study’s sample statistics and the Utrecht local PT travellers composition.42
Van Kuijk, Correia, Van Oort, Van Arem 11
TABLE 3 Sample statistics of the survey respondents compared to the characteristics of local
PT travellers in the province of Utrecht, the Netherlands
Local PT
travellers Urban
Utrecht sample
Gender
Male 43% 48%
Female 57% 52%
Age
<18 years 17% 4%
18-25 years 42% 27%
26-45 years 23% 31%
46-65 years 14% 29%
>65 years 4% 9%
Gross monthly income
< 2000 euro 64% 47%
2000-4000 euro 10% 33%
> 4000 euro 26% 20%
Household composition
Single-adult, children 8% 4%
Single-adult, no children 29% 34%
Multi-adult, children 33% 21%
Multi-adult, no children 29% 41%
Driver’s license possession
Yes 56% 70%
No 44% 30%
Private car access
(Almost) always 35% 52%
Sometimes 27% 22%
(Almost) never 38% 26%
Trip purpose
Commute 34% 39%
Education 33% 18%
Visit family/friends 11% 11%
Leisure 7% 5%
Shopping 5% 5%
Business 6% 10%
Medical reasons 2% 10%
Other 1% 2%
Van Kuijk, Correia, Van Oort, Van Arem 12
The sample statistics show that survey participants tend to be older than the average pop-1
ulation. Our sample under represents travellers below the age of 25. This is also reflected in2
household composition and income, as many older people have already experienced life events3
such as starting to live together or getting a first job. Moreover, this likely explains the higher num-4
bers in driver’s license possession and private car access. It also causes the trip purposes in our5
sample to be different than in the population; the sample consists of fewer education trips. Instead,6
the travellers in the sample make more commute, business, and medical trips.7
Our sample is also characterized by a more equal gender distribution; this is contrary to8
practice where more women can be found in local PT. The number of travellers with a high income9
is under-represented in our sample. We consider the deviation between population and sample as10
self-selection bias. We argue that travellers who are more interested in local PT (e.g. frequent PT11
users, dependency on local PT) are more likely to participate in our survey.12
RESULTS AND DISCUSSION13
First the performances of the estimated discrete choice models are compared. Thereafter, we14
provide the model estimation results from the best performing model.15
Model performance comparison16
We estimated a MNL model, a Panel-effect model and a ML model. We compared the perfor-17
mance of each of these models in order to find the model specification which describes the choice18
data the best. The model performances were compared by means of the following measures:19
Log-likelihood ratio, Rho-squared, and Adjusted rho-squared. The comparison of the model per-20
formances is set out in Table 4.21
We found that the performance of the panel-effect and ML model are comparable. Both22
models outperform the original MNL model. Because the ML model provides additional informa-23
tion (heterogenity in preferences and a panel effect) we consider the ML model to have the best24
performance.25
TABLE 4 Overview on the performance measures of the considered model specifications
MNL Panel-effect ML
Sample size 3420 3420 3420
Included parameters 67 63 66
Initial log-likelihood -5020.15 -5020.15 -5020.15
Final log-likelihood -3154.301 -3109.152 -3106.211
Likelihood ratio test 3731.427 3821.725 3827.608
Rho-squared 0.372 0.381 0.381
Adjusted Rho-squared 0.358 0.368 0.368
Van Kuijk, Correia, Van Oort, Van Arem 13
Mode characteristics1
Table 5 provides the results of the estimated model. Alternative-specific constants, βConst ant,2
describe the average part of the utility of each alternative which is not explained by any of the3
included variables. We found all constants, except for shared e-scooters, to be negative and sig-4
nificant. This indicates that most shared modes have a certain variance in utility that cannot be5
explained by the explanatory variables. In general, we think that attitudinal beliefs contribute most6
to this unexplained utility. The constant for e-bike is less negative than the constant for the non-7
electric bicycle has. Apparently, it is more difficult to explain all variance in utility for e-bikes than8
for bicycles. It could be that limited experience with e-bikes plays a role here. Van Cauwenberg9
et al. (32) mention that perceptions on safety, battery range, social stigma ("e-bike is cheating")10
and the weight of e-bikes could also be important factors.11
To add, the panel effect σPanel are large for both cycling modes, yet it is only statistically12
significant for shared e-bikes. This shows that the choice for shared e-bikes of an individual re-13
spondent are correlated over the experiment. The σPanel for shared e-bikes is so large that the sum14
of βConstant and σPanel can be larger than zero, implying that the unexplained utility of share e-bikes15
can be larger than for the non-sharing alternative. This shows that for some people there is a larger16
chance they will prefer the use of shared e-bikes over not sharing.17
We found that travel time only significantly affects the use of shared cycling modes and not18
using a shared mode. It could be that the last mile itself causes most disutility, with the distance19
covered therefore being less important. Another explanation might be that the travel time ranges20
of the other modes are smaller, such that people do not perceive a significant difference between21
travel time variations.22
Model results also show that travel costs have a negative impact on the likelihood of using23
a shared mode. We only found a weak statistical significant result for shared bicycles and shared24
e-scooters. It could be that users perceive these modes to add more value and are therefore less25
sensitive for pricing changes. Research on the use of e-scooters suggests that users might not only26
pay for transport, yet also for the fun related to using e-scooters (17)(11).27
Individual characteristics28
The final model results show that age is a strong determinant for choosing a shared mode in the last29
mile. We found that younger people (< 26 years) are more likely to use a shared mode; in contrast30
to older people (>45 years) who are less likely to use a shared mode. The one exception is the use31
of e-mopeds by older people.32
This is in the general line of other studies on shared mobility (4, 21, 22). Attitudes and33
capabilities which are more prevalent among young people could provide an explanation for this34
age effect. Alonso-González et al. (33) studied the attitudes towards other emerging transport35
paradigms: demand-responsive transport and mobility-as-a-service. They found that young peo-36
ple over-represent user groups with positive attitudes towards sharing and multi-modal lifestyles.37
These groups are characterized by their flexibility as they do not feel committed to a single mode38
of transport. In addition, these groups seem to be more tech-savvy as they feel very comfortable in39
using mobile apps for using transport services.40
Furthermore, we found that having a low income (<2000 euro) relates positively with the41
preference for shared cycling modes. On the contrary, our results show a negative correlation42
between lower incomes and the preference for shared e-scooters. This could be explained by43
the travel costs attribute ranges for e-scooter being larger and higher than for the bicycle. In44
Van Kuijk, Correia, Van Oort, Van Arem 14
TABLE 5 Urban results from the ML model specification
Shared Shared Shared Shared Not
Bicycle E-bike E-scooter E-moped Sharing
coef. t-test coef. t-test coef. t-test coef. t-test coef. t-test
Mode
Charact.
βASC -1.29 -2.20** -2.07 -6.04** -0.710 -1.59 -2.89 -2.57**
σPanel -1.14 -0.81 -1.69 -2.29** 0.284 0.30 -0.0703 -0.02
βCosts -0.693 -1.82* -0.995 -1.57 -0.841 -1.72* -1.28 -1.23
σCosts -0.208 0.50 -0.327 -0.49 -1.16 -1.44 -1.00 1.57
βTime -0.0826 -3.55** -0.0528 -1.98** -0.0328 -0.99 -0.116 -1.31 -0.0768 -4.50**
Individual
Charact.
βCarhigh 0.565 1.81* 0.542 2.00** 0.523 1.27
βCyclehigh 0.758 1.83* 1.01 3.97** 0.488 2.29** -0.991 -2.27**
βPT high -0.284 -2.15** -1.15 -4.69**
βIncomelow 0.562 2.56** 0.533 2.84** -1.18 -3.78**
βChildren 0.843 2.37** 0.997 4.73** 0.472 1.92* 0.788 2.80**
βYoung 1.11 3.38** 2.71 3.05**
βOld -1.80 -2.94** -1.59 -4.10** -1.57 -5.07**
βD.Walk -0.331 -0.84
βD.Underst.0.440 1.80*
Trip
Charact.
βCommute -0.338 -1.80* -0.439 -1.97** -0.816 -3.33** 1.26 2.42**
βEd ucation -0.695 -2.45** -0.763 -1.90*
βSubscription 0.512 2.83** 0.459 2.70** 1.37 5.44** 1.04 2.28**
βWeekend -0.884 -2.63** -0.946 -3.52** -2.18 -5.02** -1.18 -1.86*
βTrans f er -0.400 -1.53 -0.670 -1.92*
βEgresstrain 1.10 1.72* 1.68 3.65**
βCycle f irst -0.909 -2.45** -0.593 -1.56
βCyclelast 0.936 1.49 2.38 3.09**
Base = not sharing
** = significant on a 95% confidence level. * = significant on a 90% confidence level
some scenarios, the shared bicycle was available without any additional charge which could attract1
travellers who have less money to spend. Other studies show a negative correlation between income2
and the propensity to use shared e-scooters and e-mopeds (21)(34). It is uncertain whether this3
effect will also hold in the last-mile context. Aguilera-García et al. (34) state that people with4
higher incomes will replace shared e-mopeds by privately-owned vehicles as soon as they are able5
to afford these. This substitution behaviour which will likely not be present in the context of the6
Van Kuijk, Correia, Van Oort, Van Arem 15
last-mile; PT travellers already chose to use public transport in the first place. For that reason, it is1
likely that their last-mile mode choice is determined by the same mode and trip attributes as their2
initial choice for PT was based upon.3
The situation on income and shared bicycles seems to be more nuanced. Fishman (13)4
found a positive relationship between income and the use of shared bicycles. Although, he also5
found the ability to save money to be an important driver to use shared bicycles. Our finding that6
people with lower incomes have a higher preference for shared bicycles than other people could7
be explained by the nature of the Dutch cycling context. Cycling is common for many people and8
is not limited to higher income groups. In addition, shared bicycles enable PT travellers to have a9
cheap alternative in the last-mile.10
We were not able to establish a gender effect based on the collected data. In contrast11
to other studies on shared mode preferences Fishman et al., Buehler and Pucher, Goodman and12
Cheshire (9, 14, 18). Also here, we expect that the Dutch cycling culture plays an important role.13
The findings of Ton et al. (35) seem to underpin this statement. In their elaborate study on Dutch14
active mode choice behaviour also no gender effect was found.15
We also found mobility behaviour to be an important choice determinant. Travellers who16
also cycle regularly (min. 4 days/week) are more likely to choose any type of shared mode. In17
addition, people who regularly use PT are less likely to use a shared bicycle or e-scooter. We expect18
habits and the commitment towards PT to play an important role for this group. Consequently,19
they might have a more reserved attitude towards sharing. We expect habits and the commitment20
towards PT to play an important role for this group. Many studies show that uni-modal trips by21
shared bicycles often are a replacement for PT trips (9)(13)(15)(36). This indicates that many PT22
travellers prefer substituting PT completely by using a bicycle rather than combining these two23
modes.24
Lastly, we found a weak effect for travellers who experience difficulties understanding25
the PT system. They are more likely to conduct the last mile without using a shared mode. We26
expected this result as the use of shared modes adds complexity to the trip. We did not find an27
effect for PT travellers who experience difficulties with walking on (non-)sharing.28
Trip characteristics29
We first found that trip purpose affects the likelihood of using shared modes. People travelling to30
work show a lower preference to use shared cycling modes or e-scooter, yet a higher preference31
to use e-mopeds. The effect found for shared e-scooters seems to be in contrast with previous32
studies; most of them emphasize that many trips by e-scooter are made for commuting purposes33
(11, 15, 17). We think that attitudes, and more specifically with regards to status or image, are34
likely to play a role for commute trips35
Model results also show that having a local PT subscription increases the likelihood of36
using any shared mode. One explanation might be that they are committed to using PT and are37
open to shared modes for improving their trips. As these people only pay out-of-pocket cost for38
the shared modes, they might feel a lower financial burden to include shared modes in their trip.39
Next, we found that weekend travellers have a lower preference for using a shared mode40
than traveller during weekdays. This effect is the strongest for e-scooters. It is likely that weekend41
trips are being less frequent which lowers the need for improving the travel performance.42
Furthermore, our results show that travellers who made a transfer between local PT services43
(bus-to-bus for example) are less likely to use a shared e-scooter of e-moped in the last-mile. It44
Van Kuijk, Correia, Van Oort, Van Arem 16
might be that these travellers do not want to increase their trip complexity by using other modes.1
Moreover, current cycling in the first and last-mile impacts the preference of using a shared2
mode. People who cycled from their origin location to a PT stop are less likely to use a shared3
bicycle in the last mile. Contrastingly, last-mile cyclists are more likely to use an shared e-mopeds.4
We argue that travellers do not want to cycle on both trip ends, as this might require a greater5
amount of physical effort. On average, last-mile cyclists cover a longer distance than last-mile6
walkers (24)(25). For that reason they might feel a higher need to improve the last-mile by using a7
shared mode.8
CONCLUSION9
The aim of this paper was to identify which shared modes of transport are preferred by urban PT10
travellers in the last mile. The integration of these modes with local PT could improve the overall11
door-to-door travel experience. Up to now, trade-offs between multiple shared modes, specifically12
in the context of the last-mile, have only been limitedly considered in scientific literature. We esti-13
mated discrete choice models based on data from an stated choice experiment. In the experiment,14
respondents reconsidered their last-mile mode choice in a recent PT trip made by bus or tram. The15
sample was collected among 285 urban PT travellers in the province of Utrecht, the Netherlands.16
We found that shared bicycles, both conventional and electric bicycles, are generally pre-17
ferred over e-scooters and e-mopeds. A majority of the urban PT travellers, however, prefers not18
to use a shared mode in the last-mile. Those travellers will mainly walk to their destination. These19
results indicate that PT travellers do not necessarily want to avoid physical exercise. This is likely20
also the result of the limited distance which needs to be covered in the urban last-mile. This could21
also explain the low propensity to use shared e-scooters and e-mopeds in the last mile.22
The preference for shared (e-)bicycles could partially be explained by existing high cycling23
levels in the Netherlands. Shared e-bikes are generally not preferred over non-electric bicycles.24
However, current cycling behaviour has a stronger effect on the preference for shared e-bikes than25
for bicycles; this is the case for frequent cyclists and people who already cycle in the last-mile.26
Our study confirms the importance of age on user preferences for shared modes, which27
was also found in other studies. We found a strong relationship between age and the preference28
for shared modes; people up to the age of 26 are more willing to use shared modes in the last-mile29
than older people. In addition, we were able to find a strong heterogeneity in preferences among30
this group of young PT travellers. This shows that shared modes are not necessarily adopted by all31
young people.32
We found that frequent PT users are less likely to use shared modes for their last mile. We33
expect this is caused by their existing travel habits; they might not be open to consider alternatives34
for the current last-mile travel behaviour. Interestingly, we found that having a PT subscription35
increases the probability to use shared modes.36
As being said already, existing cycling behavior is an important determinant for using37
shared modes in the last mile. Regular cyclists are more likely to prefer shared modes (e-mopeds38
excluded). In case PT travellers already cycle in the first mile, they are less likely to use a shared39
bicycle in the last mile. A possible explanation is that a local PT trip with cycling on both sides40
of the trip is less attractive to many people than travelling directly by a bicycle. However, if trav-41
ellers currently use a privately-owned bicycle in the last mile, they are more likely to shift to use42
an e-moped in this trip stage. These modes might better meet the needs of these travellers, while43
leaving out the need for having a privately-owned bicycle available at the egress stop.44
Van Kuijk, Correia, Van Oort, Van Arem 17
In contrast to other studies, we did not find a gender effect. We expect that Dutch cycling1
culture had a significant impact on our experiment; cycling is well-established in the Netherlands2
such that, other than in many countries, women cycle more than men. We found that shared modes3
(except e-mopeds) are less preferred by commuters. Student travellers show a lowered preference4
for shared cycling modes. Although, PT trips made during the week are more likely to be combined5
with a shared mode in the last mile than weekend trips. This suggests that PT trips made on a more6
structural basis are more likely to be combined with a shared mode in the last mile.7
Given our results, public transport planners need to consider providing shared modes at8
local PT stops; these could bring added value to a significant number of PT travellers. We, however,9
do not advise to provide these at all stops, as many vehicles would probably be left unused. Major10
PT stops and transfer hubs seem to be most suitable as demand for shared modes can be centralized11
here more easily. Compared to decentralized distribution, this will induce longer last-mile travel12
times, but our results show that this is not very important for users of shared modes in this trip13
stage.14
We advise to focus on conventional shared bikes as this will likely attract the most users.15
The supply of shared e-bikes would specifically help to replace the use of private bicycles in the16
last mile. This is interesting if there is a need to lower the demand for bicycle parking at PT stops17
and stations, specifically if existing parking capacity is insufficient. Both e-scooters and e-mopeds18
are likely to only serve niche markets. Although, the e-scooter might address the needs of some19
young people. We advise to follow-up developments on the e-scooter markets; it is not unlikely20
that e-scooters will appeal to a larger group and for more structural trip purposes. Based on these21
conclusions we identified a number of important directions for future research. Most importantly,22
more insights are needed with regard to the attitudes which determine the preferences for shared23
modes. This requires a more qualitative research approach and the inclusion of testing attitudinal24
statements. In addition, it is important to address the found heterogeneity in user preferences. This25
will further support the actions to be taken by PT operators and authorities to improve the last-mile26
and the overall door-to-door travel experience. This could for example be done by latent class27
analysis. Further, we would advise exploring shared mode preferences in different contexts. It is28
likely that the built environment, infrastructure and cycling infrastructure had an important role in29
our research findings. Specifically, we suggest studying shared mode preferences outside the urban30
context and in a low-cycling context.31
ACKNOWLEDGEMENTS32
This research is part of a collaboration between the province of Utrecht and the Smart Public33
Transport Lab of Delft University of Technology. The authors would like to thank the province34
of Utrecht for their support to this study, specifically regarding their contribution to the survey.35
We would also like to thank Danique Ton for her help and support in conducting the stated choice36
experiment.37
Van Kuijk, Correia, Van Oort, Van Arem 18
REFERENCES1
1. Susilo, Y. O. and O. Cats, Exploring key determinants of travel satisfaction for multi-modal2
trips by different traveler groups. Transportation Research Part A: Policy and Practice,3
Vol. 67, 2014, pp. 366–380.4
2. Hoogendoorn-Lanser, S., R. Van Nes, S. P. Hoogendoorn, and P. Bovy, Home–Activity5
Approach to Multimodal Travel Choice Modeling. Transportation research record, Vol.6
1985, No. 1, 2006, pp. 180–187.7
3. Shaheen, S. and N. Chan, Mobility and the Sharing Economy: Poten-8
tial to Overcome First-and Last-Mile Public Transit Connections Permalink9
https://escholarship.org/uc/item/8042k3d7. UC Berkeley: Transportation Sustainability10
Research Center, 2016.11
4. Yan, Q., K. Gao, L. Sun, and M. Shao, Spatio-Temporal usage patterns of dockless bike-12
sharing service linking to a metro station: A case study in Shanghai, China. Sustainability13
(Switzerland), Vol. 12, No. 3, 2020.14
5. Adnan, M., S. Altaf, T. Bellemans, A. u. H. Yasar, and E. M. Shakshuki, Last-mile travel15
and bicycle sharing system in small/medium sized cities: user’s preferences investigation16
using hybrid choice model. Journal of Ambient Intelligence and Humanized Computing,17
Vol. 10, No. 12, 2019, pp. 4721–4731.18
6. Yan, X., J. Levine, and X. Zhao, Integrating ridesourcing services with public transit : An19
evaluation of traveler responses combining revealed and stated preference. Transportation20
Research Part C, Vol. 105, No. July 2018, 2019, pp. 683–696.21
7. Yap, M. D., G. Correia, and B. van Arem, Preferences of travellers for using automated22
vehicles as last mile public transport of multimodal train trips. Transportation Research23
Part A: Policy and Practice, Vol. 94, 2016, pp. 1–16.24
8. Buehler, R., Determinants of transport mode choice : a comparison of Germany and the25
USA. Journal of Transport Geography, Vol. 19, No. 4, 2011, pp. 644–657.26
9. Fishman, E., S. Washington, N. Haworth, and A. Watson, Factors influencing bike share27
membership: An analysis of Melbourne and Brisbane. Transportation Research Part A:28
Policy and Practice, Vol. 71, 2015, pp. 17–30.29
10. Bachand-Marleau, J., B. Lee, and A. El-Geneidy, Better understanding of factors influ-30
encing likelihood of using shared bicycle systems and frequency of use. Transportation31
Research Record, , No. 2314, 2012, pp. 66–71.32
11. Fitt, H. and A. Curl, E-scooter use in New Zealand: Insights around some frequently asked33
questions, 2019, p. 21.34
12. Bai, S. and J. Jiao, Dockless E-scooter usage patterns and urban built Environments: A35
comparison study of Austin, TX, and Minneapolis, MN. Travel Behaviour and Society,36
Vol. 20, No. January, 2020, pp. 264–272.37
13. Fishman, E., Bikeshare: A Review of Recent Literature. Transport Reviews, Vol. 36, No. 1,38
2016, pp. 92–113.39
14. Buehler, R. and J. Pucher, International Overview: Cycling Trends in Western Europe,40
North America, and Australia. City cycling, 2012, pp. 9–29.41
15. Transport for London, Barclays Cycle Hire customer satisfaction and usage survey: Mem-42
bers only: Wave 9 (Quarter 3 2014/15), 2015.43
Van Kuijk, Correia, Van Oort, Van Arem 19
16. McKenzie, G., Spatiotemporal comparative analysis of scooter-share and bike-share usage1
patterns in Washington, D.C. Journal of Transport Geography, Vol. 78, No. May, 2019,2
pp. 19–28.3
17. Hardt, C. and K. Bogenberger, Usage of e-Scooters in Urban Environments. Transporta-4
tion Research Procedia, Vol. 37, No. September 2018, 2019, pp. 155–162.5
18. Goodman, A. and J. Cheshire, Inequalities in the London bicycle sharing system revis-6
ited: Impacts of extending the scheme to poorer areas but then doubling prices. Journal of7
Transport Geography, Vol. 41, 2014, pp. 272–279.8
19. Harms, L., L. Bertolini, and M. Brömmelstroet, Spatial and social variations in cycling9
patterns in a mature cycling country exploring differences and trends. Journal of Transport10
& Health, Vol. 1, No. 4, 2014, pp. 232–242.11
20. Guo, Y., J. Zhou, Y. Wu, and Z. Li, Identifying the factors affecting bike-sharing usage and12
degree of satisfaction in Ningbo, China. PLoS ONE, Vol. 12, No. 9, 2017, pp. 1–19.13
21. Jiao, J. and S. Bai, Understanding the shared e-scooter travels in Austin, TX. ISPRS Inter-14
national Journal of Geo-Information, Vol. 9, No. 2, 2020.15
22. Degele, J., A. Gorr, K. Haas, D. Kormann, S. Krauss, P. Lipinski, M. Tenbih, C. Kop-16
penhoefer, J. Fauser, and D. Hertweck, Identifying e-scooter sharing customer segments17
using clustering. In 2018 IEEE International Conference on Engineering, Technology and18
Innovation (ICE/ITMC), IEEE, 2018, pp. 1–8.19
23. Daniels, R. and C. Mulley, Explaining walking distance to public transport: The domi-20
nance of public transport supply. Journal of Transport and Land Use, Vol. 6, No. 2, 2013,21
p. 5.22
24. Brand, J. and S. Hoogendoorn, Modelling Multimodal Transit Networks Integration of bus23
networks with walking and cycling, 2017.24
25. Rijsman, L., S. Hoogendoorn, E. Molin, and T. Teijl, Walking and bicycle catchment areas25
of tram stops : factors and insights, 2019, pp. 0–4.26
26. Qualtrics, Qualtrics. Provo, UT, USA, 2013.27
27. Centraal Bureau voor de Statistiek, Statistische gegevens per vierkant en postcode 2018-28
2017-2016-2015, 2019, pp. 1–32.29
28. Choicemetrics, Ngene 1.1. 1 user manual & reference guide. Sydney, Australia: Choice-30
Metrics, Vol. 19, 2012, p. 20.31
29. Train, K. E., Discrete choice methods with simulation. Cambridge university press, 2009.32
30. Hensher, D. A., J. M. Rose, and W. H. Greene, Applied choice analysis: a primer. Cam-33
bridge university press, 2005.34
31. Bierlaire, M., PythonBiogeme : a short introduction. Transport and Mobility Laboratory,35
Ecole Polytechnique Fédérale de Lausanne., 2016.36
32. Van Cauwenberg, J., I. De Bourdeaudhuij, P. Clarys, B. de Geus, and B. Deforche, E-bikes37
among older adults: benefits, disadvantages, usage and crash characteristics. Transporta-38
tion, Vol. 46, No. 6, 2019, pp. 2151–2172.39
33. Alonso-González, M. J., S. Hoogendoorn-Lanser, N. van Oort, O. Cats, and S. Hoogen-40
doorn, Drivers and barriers in adopting Mobility as a Service (MaaS) – A latent class41
cluster analysis of attitudes. Transportation Research Part A: Policy and Practice, Vol.42
132, No. November 2019, 2020, pp. 378–401.43
34. Aguilera-García, Á., J. Gomez, and N. Sobrino, Exploring the adoption of moped scooter-44
sharing systems in Spanish urban areas. Cities, Vol. 96, No. July 2019, 2020, p. 102424.45
Van Kuijk, Correia, Van Oort, Van Arem 20
35. Ton, D., D. C. Duives, O. Cats, S. Hoogendoorn-Lanser, and S. P. Hoogendoorn, Cycling1
or walking? Determinants of mode choice in the Netherlands. Transportation Research2
Part A: Policy and Practice, Vol. 123, No. August 2018, 2019, pp. 7–23.3
36. ZHU, W., Y. PANG, D. WANG, and X. YU, Travel Behavior Change after the Introduc-4
tion of Public Bicycle Systems: A Case Study of Minhang District, Shanghai. In Urban5
Planning Forum, 2012, 5, p. 18.6
Article
Emerging access/egress transport modes to and from railway stations may play a vital role in the future performance and usage of public transport. To learn about these modes, their acceptability, and attractiveness, we performed a case study at Delft Campus train station in the Netherlands, using a stated preference experiment. We investigated travellers’ preferences towards shared bicycles, shared e-steps, shared e-scooters, automated vehicles (individuals and shuttles), and the importance of time, costs, and availability of these modes to access or egress this small-sized hub. Furthermore, we studied the impacts of two contextual situations: weather conditions and carrying luggage, affecting mode choice. The results indicate that travel costs have much higher importance than travel time for accessing or egressing from a small hub, and Autonomous Vehicles (AVs) usage is positively valued as first/last mile transport modes compared to other alternatives. Increasing 10% time and cost of all modes indicate that the demand for individual and collective AVs falls sharply, whereas the change in demands of shared bikes is negligible. A significant impact of context effect variables in the utility of travellers was also observed when these variables interacted with the leading travel attributes. The potential market share of the modes indicates that shared bike usage will continue to be strong, and automated vehicles will find their position at future stations. These findings could provide critical criteria for designing future small/medium multimodal hubs.
Article
Full-text available
In recent years, the popularization of dockless shared electric scooters (e-scooters) across many American cities has provided a great opportunity to reduce short-distance automobile trips. However, there is not enough research that examines e-scooter usage patterns and their association with the urban environment. Moreover, the question of whether this association would vary across different cities also remains unanswered. To bridge these gaps, this study investigated e-scooter ridership in Austin and Minneapolis using GIS hotspot spatial analysis and negative binomial regression models. The spatial analysis results showed that the densest e-scooter usage happened in downtown areas and university campuses in both cities. However, the temporal characteristics of the two cities' e-scooter usage patterns turned out to be different: in Austin, afternoons, and weekends experienced greater e-scooter traffic, whereas Minneapolis showed larger evening ridership and stable daily vehicle miles traveled throughout the week. Finally, regression results showed that proximity to the city center, better access to transit, and greater land-use diversity positively correlated with higher e-scooter ridership in both cities. Compared to single-family residential areas, office and institutional land use were more likely to relate to higher ridership in both cities. Curiously, the statistically positive relationship with commercial areas and parks only existed in Austin. This study contributes to transportation literature and practice by providing empirical evidence on e-scooter trips in the U.S. cities and by highlighting the importance of local uniqueness by comparing two cities.
Article
Full-text available
This paper investigated the travel patterns of 1.7 million shared E-scooter trips from April 2018 to February 2019 in Austin, TX. There were more than 6000 active E-scooters in operation each month, generating over 150,000 trips and covered approximately 117,000 miles. During this period, the average travel distance and operation time of E-scooter trips were 0.77 miles and 7.55 min, respectively. We further identified two E-scooter usage hotspots in the city (Downtown Austin and the University of Texas campus). The spatial analysis showed that more trips originated from Downtown Austin than were completed, while the opposite was true for the UT campus. We also investigated the relationship between the number of E-scooter trips and the surrounding environments. The results show that areas with higher population density and more residents with higher education were correlated with more E-scooter trips. A shorter distance to the city center, the presence of transit stations, better street connectivity, and more compact land use were also associated with increased E scooter usage in Austin, TX. Surprisingly, the proportion of young residents within a neighborhood was negatively correlated with E-scooter usage.
Article
Full-text available
The dockless bike-sharing (DLBS) system serves as a link between metro stations and travelers’ destinations (or originations). This paper aims to uncover spatio-temporal usage patterns of dockless bike-sharing service linking to metro stations for supporting scientific planning and management of the dockless bike-sharing system. A powerful visualization tool was used to analyze the differences in usage patterns in workdays and weekends. The travel distance distributions of using dockless bike-sharing near metro stations were investigated to shed light on the service area of the dockless bike-sharing system. Agglomerative hierarchical clustering was applied to analyze differences in usage patterns of metro stations located in different areas. The results show that the usage patterns of dockless bike-sharing on weekends are different from those on workdays. The average travel distance using the dockless bike-sharing system at weekends is significantly larger than that of workdays. The travel distance distribution could be nicely fitted by the Fréchet distribution of the Generalized Extreme Value (GEV) distribution family. The usage characteristics of shared bikes are correlated with land use and population density around metro stations. No matter in urban or suburban areas, there is a great demand for bike-sharing in densely populated areas with intensive land development, such as university towns in suburban areas. This study improves the understandings regarding the usage patterns of the DLBS system serving as a link between the final destinations (or originations) and metro stations. The results can be helpful to the operation and demand management of DLBS.
Article
Full-text available
The United States is currently in the midst of a micro-mobility revolution of sorts. Almost overnight, U.S. cities have been inundated with short-term rental scooters owned and operated by start-up companies promising a disruption to the urban transportation status-quo. These scooter-share services are presented as a dockless alternative to traditionally government-funded, docking station-based bike-sharing programs. Given the rapid rise of electric scooter companies, and how little is known about their operations, there is pressing public interest in understanding the impact of these transportation-sharing platforms. By exploring the nuanced spatial and temporal activity patterns of each of these platforms, this research identifies differences and similarities between dockless e-scooters and existing bike-sharing services. The findings from this research contribute to our understanding of urban transportation behavior and differences within mobility platforms.
Conference Paper
Full-text available
Since the introduction of the internal combustion engine (ICE), modern mobility has fundamentally changed. Today, it is impossible to imagine modern transportation without motorized vehicles. Powered Two-Wheelers (PTWs) represent a relevant section of such vehicles. In recent years, electric low-powered PTWs, also called e-scooters, became a notable trend. Without excessive emissions and noise, as well as convenient dimensions, e-scooters can help European cities ease their increasing problems with traffic, emissions, and parking. In this paper, we present the results of a real-life field test, performed in the City of Munich, Germany. By providing six vehicles, 38 subjects were able to incorporate an e-scooter into their everyday lives. Recorded with travel diaries and a pre-post-survey, usage and attitudes were evaluated. Results show that a majority of daily trips is suitable for e-scooters and charging infrastructure is sufficient for this kind of vehicle, while subjective safety, weather conditions, and baggage capacity are restricting attributes.
Article
Full-text available
Inspired by the success of private ridesourcing companies such as Uber and Lyft, transit agencies have started to consider integrating ridesourcing services (i.e. on-demand, app-driven ridesharing services) with public transit. Ridesourcing services may enhance the transit system in two major ways: replacing underutilized routes to improve operational efficiency, and providing last-mile connectivity to extend transit’s catchment area. While an integrated system of ridesourcing services and public transit is conceptually appealing, little is known regarding whether and how consumers might use a system like this and what key service attributes matter the most to them. This article investigates traveler responses to a proposed integrated transit system, named MTransit, at the University of Michigan Ann Arbor campus. We conducted a large-sample survey to collect both revealed preference (RP) and stated preference (SP) data and fit a RP-SP mixed logit model to examine the main determinants of commuting mode choice. The model results show that transfers and additional pickups are major deterrents for MTransit use. We further applied the model outputs to forecast the demand for MTransit under different deployment scenarios. We find that replacing low-ridership bus lines with ridesourcing services could slightly increase transit ridership while reducing operations costs. The service improvements offered by ridesourcing mainly come from reductions in wait time. Though relatively small in our study, another source of improvement is the decrease of in-vehicle travel time. Moreover, we find that when used to provide convenient last-mile connections, ridesourcing could provide a significant boost to transit. This finding verifies a popular notion among transit professionals that ridesourcing services can serve as a complement to public transit by enhancing last-mile transit access.
Article
Full-text available
First and last-mile access to and from public transport stations/stops is a major problem for encouraging public transport use. Bicycle sharing schemes have shown potential to fill this gap. Consequently, railway operators in the Netherland and Germany have started their own bike sharing schemes. Majority of the studies examined the preferences for using bike share schemes for larger cities. This study analyses the collected stated preference survey data for the use of bicycle sharing scheme for last mile travel, which is recently launched in small/medium sized cities of Belgium. Within this scheme a single docking station is available and users need to return bicycle at the same station. The survey also includes questions on respondents attitudes towards friendliness-to-cycling. The hybrid choice modelling framework is used to investigate preferences of users. Usual explanatory variables such as temperature, rain conditions, distance, rental cost, gender and age are found significant, which confirms the findings of earlier studies. Along with these; last-mile (to home) and an interaction term between rental cost and duration to keep bicycle are found significant, which indicate the negative effects of having a single docking station. Availability of escorting facility from parents/colleague/friends (a more common phenomenon in small/medium cities) also has a negative effect on the use of the examined bike share scheme. Paper also discusses promotional campaigns and marketing efforts according to the obtained results for making such schemes more successful.
Article
Within the context of the sharing economy, innovative mobility solutions are emerging in urban areas to enable people short-term access to means of transport on an as-needed basis. Within this new trend, moped scooter-sharing is experiencing a great boom in many cities worldwide, particularly in Europe. The generalization of electric moped scooter-sharing services can bring benefits such as lowering noise and air pollution or reducing road congestion, with important implication for improving the livability and quality of life in urban areas. To date, many contributions have been conducted to characterize innovative mobility options such as car-sharing or bike-sharing, but almost no efforts have been devoted to explore the use of moped scooter-sharing services. Based on the information collected through an online survey disseminated in different Spanish cities, we developed a generalized ordered logit model to identify the key drivers determining the adoption and frequency of use of moped scooter-sharing services in urban areas. The research concludes the main role played by some sociodemographic and travel-related variables, such as age or level of education, while personal opinions and attitudes were not generally found statistically significant. The research found somehow similar profiles for occasional and frequent users of moped scooter-sharing, then public authorities should make further efforts to overcome the barriers related to the first contact with this mobility option.