You are on page 1of 11

Travel Behaviour and Society 31 (2023) 399–409

Contents lists available at ScienceDirect

Travel Behaviour and Society


journal homepage: www.elsevier.com/locate/tbs

Understanding the relationship between mode and destination choices for


personal out-of-town travel
Lisa Aultman-Hall a, *, 1, Anuarbek Onayev b, Jonathan Dowds c, 2, Jeffrey J. LaMondia d, 3
a
Department of Systems Design Engineering, University of Waterloo, Waterloo, Ontario, Canada
b
Department of Civil Engineering, University of British Columbia, Vancouver, BC, Canada
c
University of Vermont - Transportation Research Center, 35 Colchester Avenue, Burlington, VT 05405, United States
d
Department of Civil Engineering, Auburn University, Auburn, AL 36849, United States

A R T I C L E I N F O A B S T R A C T

Keywords: In post-COVID era, more agencies will seek to incorporate long-distance or overnight travel into their travel
Long-distance travel forecasting. This paper leverages a unique survey dataset (n = 440) that queried participants’ propensity to make
Mode choice simultaneous or bundled decisions on mode and destination location for their most recent overnight out-of-town
Intercity travel
personal trip. The random representative sample of Vermont residents indicated they mostly make bundled
Air travel
decisions of destination and mode choice. Binary logistic regressions were estimated to determine (a) socio­
demographic factors (e.g. age, income) and trip travel distance influence the likelihood an individual will bundle
their mode and destination choice and (b) mode-specific travel times, distances traveled and bundled decisions
influence mode choices. Results validate that out-of-town travel destination and mode decisions are sometimes
integrated, and different populations treat this decision process differently. Moreover, choice models of this
behavior require more complex predictors in addition to distance, including whether it is part of a bundled
decision.

1. Introduction predictors will be increasingly important in the post-COVID, the objec­


tive of this paper is to understand factors affecting mode choice espe­
The travel demand models managed by transportation agencies have cially whether participants’ decisions of mode and destination location
traditionally treated intercity and tourism travel in an overly simplistic are made sequential or as simultaneous, bundled decisions. Insights on
manner, focusing instead on more routine, daily, home-based travel to intercity mode and destination choice decision-making are important for
destinations within one’s home community or region. Demand models, social and economic revitalization after the global pandemic as well as
which in the United States (US) usually represent metropolitan regions incorporation of vehicle automation and global carbon emissions across
and in some cases entire states, relegate such intercity trips to external modes. This paper aims to advance our understanding of how in­
status since they typically originate, or are destined for, external spatial dividuals choose modes and destinations to more accurately reflect
zones that are modeled with less detail. Few existing demand models travel behavior in forecasting models. This gap in knowledge is
capture travel demand at the continental and global scales, which we addressed through exploring the personal intercity travel decisions
know many people now routinely travel. This exclusion is due primarily made by a random sample of Vermonters based on a 2017 telephone
to lack of data and has resulted in limited explorations of appropriate survey, the Vermonter Poll (both landline and cell phones were used). In
modeling specifications of mode and destination choice as well as addition to advancing our understanding of intercity mode choice, this
assessment of appropriate predictors, the foci here. paper also measures the relative impacts of (a) mode-specific air travel
Because modeling personal intercity travel with policy-sensitive times between individual trip OD pairs and (b) decision-making

* Corresponding author.
E-mail addresses: lisa.aultman-hall@uwaterloo.ca (L. Aultman-Hall), aonayev@mail.ubc.ca (A. Onayev), jdowds@uvm.edu (J. Dowds), jlamondia@auburn.edu
(J.J. LaMondia).
1
ORCID 0000-0002-8190-9640.
2
ORCID 0000-0003-3420-7790.
3
ORCID 0000-0002-9592-3320.

https://doi.org/10.1016/j.tbs.2023.02.004
Received 5 December 2021; Received in revised form 28 August 2022; Accepted 4 February 2023
Available online 13 February 2023
2214-367X/© 2023 The Authors. Published by Elsevier Ltd on behalf of Hong Kong Society for Transportation Studies. This is an open access article under the CC
BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
L. Aultman-Hall et al. Travel Behaviour and Society 31 (2023) 399–409

processes on out-of-town mode choice. Table 1


We, like others, have anecdotal evidence from prior research, where Structure of existing long distance mode and destination choice models.
we collected trip plans and changes in plans via a web-based longitu­ Model Mode/Destination Model Structure
dinal survey, that overnight trip tour planning is complicated and plans Type
Joint Sequential Both
often change before the trip is undertaken (e.g. the dates of travel,
cancelation, composition of the travel party and even purpose) (Sullivan Research Richards and Ben- Timmermans (1996)
Akiva (1974)
et al., 2016). However, we did not query participants in this prior Adler and Ben- Thrane (2015)Toole
2013–2014 longitudinal survey about the order of their decisions Akiva (1976) et al. (2015)
regarding out-of-town purpose, destination and mode choices. As a LaMondia et al. Moeckel et al. (2015a,b)
result of topics raised by participants in post-survey focus groups, we (2010)
Eugenio-Martin Cho (2013)
designed questions about decision making for the 2017 survey from
(2003)
which data is drawn for this paper. Following a review of prior litera­ Newman and Wu et al. (2012)
ture, this paper contains an explicit framing of potential mode/desti­ Bernardin (2010)
nation decision-making processes; a data description which includes Boyce et al. (1983)
generation of OD-specific total air travel time; modeling results; and Schmid et al.
(2019)
discussion/conclusion sections. Juschten and
Hossinger (2021)
2. Background literature
Regional Fox and Patruni Moeckel et al. (2015a,b)
2.1. Data and modeling (2021) Jonnalagadda et al.
(2001)
The execution of the American Travel Survey in 1995, as well as the
start of several surveys of national travel in the late 1990s in Europe, State FL Dept of Trans. (2018) Cambridge
coincided with increasing interest in making intercity travel models Atkins (2013) Systematics (2014)
Alliance Transportation
more disaggregate. Only limited long-distance travel data have been
Group (2015)
collected since with only our 2013 effort also including an annual report Donnelly (2017)
by each participant (Aultman-Hall et al., 2015). The most comprehen­
sive long-distance survey in the United States was conducted as part of National Outwater et al. Bradley et al. (2016)
the 2001 National Household Transportation Survey (NHTS). In this (2015) Van Bostrand (2011)
case, a dedicated long-distance survey where information about each Janzen and Sivaraman (2015)
trip including mode, purpose, and travel party size was gathered from a Axhausen 2018
sub-sample for trips in the prior 4 weeks. Neither the ATS nor the 2001
NHTS collected information about decision-making processes because
sequential process of destination choice followed by mode choice. Each
the labor of collecting long-distance travel logs was already cumber­
research paper justifies their different technique as attempting to mimic
some. Relatively few intercity mode choice models exist in the literature,
true travel behavior, but no previous research has considered the actual
and even fewer states incorporate mode choice into their statewide
decision-making processes. The Vermonter Poll, used here, is the first
models. In most cases this is likely due to the lack of mode-specific
survey-based effort to our knowledge to tackle the issue of decision-
alternative data. In fact, much research into intercity mode choice is
making processes for intercity travel.
based on stated choice and thought experiments, where alternative
There are few applications of this joint modeling in practice; for
specific variable values are available because they are generated for the
example: Fox and Patruni (2021) use a joint model of seven models and
study (for example: Timmermans, 1996).
over 1700 destination zones in the London regional travel demand
In terms of structuring of mode choice models, researchers have
model. This application requires a complex set of mode-specific vari­
developed sequential structures (where destinations are modeled first
ables and a clearly defined set of destinations to operate. These include
and modes are then modeled separately, for the given destination), joint
costs, in vehicle travel times and other level of service variables for each
structures (where destination choices are correlated with mode choices
mode being considered.
in addition to the typical alternative specific variables), and linked
As long-distance travel volumes increase, many states are improving
structures (where probabilities (often logsums) are used from the
their statewide travel demand models and discussing whether joint
destination model in the mode choice models). Examples of each
destination-mode models are necessary as well as what data are required
modeling structure can be seen in Table 1. Recognizing the need to
to characterize these choices. Table 1 illustrates that most states utilize a
jointly model destinations and mode choices is not new (Adler and Ben-
sequential model structure, presumably due to the lack of data charac­
Akiva (1976) and LaMondia et al. (2010) both documented behavioral
terizing mode choice alternatives. FHWA’s national model originally
motivations for why individuals may link these two decisions), but no
incorporated a joint destination-mode structure, Outwater et al.’s
work has previously documented the process. While not as thoroughly
(2015), but ultimately was presented with a sequential model (Bradley
researched as urban travel decisions, Schmid et al. (2019), Elmorssy and
et al., 2016). The advanced activity-based form included synthetic
Onur (2020), and Juschten and Hossinger (2021) most recently esti­
households simulated in terms of automobile ownership, number of
mated long-distance mode/destination choice models using various
tours, duration, travel parties, destination and mode. But estimation was
techniques. Schmid et al. (2019) estimated joint nested models of
limited by available data. The 1995 ATS and 2001 NHTS long-distance
modes, routes and destinations choices, with an emphasis on mode-
data were augmented with state data from Ohio (2003), Colorado
specific cost variables. Elmorssy and Onur (2020) combine departure
(2010) and California (2012). The effort included 4486 nation-wide
time, destination and mode choices in a nested but sequential process,
zones consisting of counties in rural areas and smaller Census-based
where the choice of departure time was used to determine destination,
zones in urban areas. The joint model was successfully validated
which was then used to determine travel mode. Juschten and Hossinger
against FHWA’s 2008 Traffic Analysis Framework (TAF) county-to-
(2021) estimated a single logistic regression on a joint mode-destination
county origin–destination data by mode (created from the 1995 ATS)
choice, using mode-specific variables (train/bus vs personal vehicle) and
as well as by consideration of sensitivity to certain socioeconomic pa­
generic destination variables not specific to any single zone. Wu et al.
rameters. A similar annual activity-based model is under development at
(2012) also estimated a model of Japanese tourist travel using a

400
L. Aultman-Hall et al. Travel Behaviour and Society 31 (2023) 399–409

ETH Zurich (Janzen and Axhausen, 2018). This model is calibrated trips over 2,000 miles were made by automobile and a few surface trips
based on French mobile phone data and includes a joint model of over 3,000 miles were undertaken. RSG Inc. et al. (2019) and Kim and
destination and mode. In models of regional or local travel the issue of Ryerson (2018) illustrate that air and surface are not two distinct sys­
decision order, nesting and jointness is still debated with some arguing tems but rather an integrated system with complex interactions.
that it is context based (Newman and Bernardin, 2010). There is still a In order to accurately model the overall passenger transportation
gap in terms of data on measuring aspects of decision-making which this system to optimize policies or investments for environmental impacts,
paper seek to help fill. costs or congestion it is necessary to start accurately modeling intercity
One might reasonably assume that models of mode choice between mode choice and this requires better travel data, understanding the
surface and air travel would be important and common. Within the decision-making process and estimation of total air travel times between
northeast corridor of the US, rail dominates for trips under 300 miles origins and destinations. The possibility of automated vehicles being
and air dominates over 300 miles (Chester and Ryerson, 2014). There is used for intercity travel, the increasing focus on aviation greenhouse gas
a distance range where air competes with surface even when the only emissions, and the behavior changes expected due to COVID-19 also
surface mode is driving. Early aggregate corridor level models in the motivates the need to accurately assess travel time trade-offs with
1960s and 1970s were focused on mode (Watson, 1974; Leake and meaningful long-distance travel model parameterization.
Underwood, 1978; Meyburg, 1972). Yet recent long-distance travel
mode choice models are extremely limited in number. Although, con­ 2.2. Framing decisioning making and research questions
ducting descriptive tabulation of revealed mode use is straightforward, a
choice model ideally requires delineation and attribution of both chosen We hypothesize that travelers can engage in one of five decision-
and unchosen alternatives that an individual faces. Long-distance travel making approaches regarding trip purpose, destination and mode
surveys simply document modes chosen and trips taken. However, when making a personal out-of-town trip (Fig. 1). Two of these processes
unchosen modes are not queried and thus must be assumed with (Options 1 and 2 in Fig. 1) are strictly sequential and three of these
modeled attributes. Even if one can assume that both air and surface process (Options 3–5) involve some degree of bundled decision-making
modes were considered for a particular origin and destination pair, air- where two or more trip characteristics are determined simultaneously.
based accessibility is hard to measure and includes many variables. The decision process may unfold in the sequences assumed in traditional
Price, an obvious motivator of travel decisions, is complicated to esti­ 4-step demand models (option 1 in Fig. 1) with purpose being decided
mate for any mode, but especially for air. One way to simplify first, followed by destination then mode. Alternatively, after deciding on
enumeration of mode attributes is to limit the origins and destination. a trip purpose, some people may select a mode, and then choose the
For example, Behrens and Pels (2012) study the implementation of high- destination last (option 2 in Fig. 1). Examples of this decision-making
speed rail between London and Paris to consider its impact on air travel. process would include I need to get away, I should fly somewhere on
They indicate travel time and frequency matter most to business trav­ Spring break or I have to get out-of-town and I think I will drive my car
elers but that price matters most to leisure travelers. In this study, we somewhere this weekend. In other cases, some or all of the decisions are
outline a procedure for estimating surface and air accessibility within bundled into combined decisions. Travelers might decide on all trip
the continental US that we propose will expand the origin destination aspects together as a single bundled decision (option 3 in Fig. 1) as is the
pairs for which intercity mode choice can be studied. case when an individual decides, I want to go on vacation in Iceland and
Cho (2013), in his University of Florida thesis, attempted to develop it’s only accessible by air travel. Mode and destination might be a
the travel time and other attributes associated with all mode options for bundled decision after the trip purpose. (option 4). This would include
trips for Florida as frecorded in the 2009 NHTS. Monzón and Rodrı́guez- the example that I need to get away somewhere, I should fly to Orlando,
Dapena (2006) recommend non-random sampling only for specific OD Florida on Spring break. Finally, purpose and destination might be
pairs to overcome choice modeling challenges. Rohr et al. (2010) esti­ decided first as a bundle followed by a mode choice decision (option 5).
mated a nested choice model for the UK. Ashiabor et al. (2007) esti­ The decision to make a trip to visit one’s aging mother to monitor her
mated a US-based county-to-county mode choice model using a measure medical care, where trip purpose and destination are intrinsically
of airport access based on the 1995 ATS data. Moeckel et al. (2015a,b) linked, followed by a decision to drive, fly or take the train or bus is an
developed a nested Logit choice model for North Carolina using the example of this decision-making sequence.
2001 NHTS. Significant efforts have been invested in modeling mode This paper seeks to answer the following research questions utilizing
choice related to high-speed rail (HSR) in California (Outwater et al., the data from the 2017 Vermonter Poll as well as imputed distance and
2010), although the data have not always been ideal (Koppelman et al., travel times by air and personal vehicle for reported trips:
2011). Models addressing HSR and air substitution in Spain have not
shown that HSR is necessarily a substitute for air travel (Castillo-Man­ 1. Is type of trip decision-making related to sociodemographic charac­
zano et al., 2015). There are enough reports of long-distance mode teristics of the traveler?
choice models to affirm the feasibility of their estimation, but the 2. What sociodemographic factors influence an individual’s likelihood
availability of data for their creation is limiting and part of the context to bundle mode and destination decisions?
we hope to contribute to in this paper. 3. What is the relative importance of (a) this bundling decision and (b)
In general, travel behavior experts assume and agree that long- characteristics of travel to reach the destination on mode choices?
distance mode choice depends on number of people in the travel
party, locale, distance, and rural versus urban locations (Larsen et al., The approaches in Fig. 1 demonstrate the full set of approaches in­
2006). A common mode choice oversimplification is often made in the dividuals can make for planning out-of-town travel destination and
US context that at some trip length threshold most shorter trips are made mode choices. The work presented in this paper presents a scientific
by automobile and longer trips by air. We argue that the association approach to use data to (a) determine which of these processes are more
between mode and distance is not this clear-cut and requires more appropriate to consider in this (and future) study, (b) analyze the mixed
refined assessment of air accessibility between origin and destination. interaction between demographics, trip characteristics, and decision-
Data from our 1-year Longitudinal Survey of Overnight Travel (LSOT) making processes, and (c) characterize the relative importance of each
panel illustrate that this is simply not a strong assumption (Aultman-Hall of these factors on mode choices (when each variable is held constant).
et al., 2016). In this limited convenience sample of only 628 individuals
(229 in Vermont, the location for the data used in this paper), numerous 3. Data
trips to destinations less than 500 miles were made by air and some
under 250 miles were also made by air. Similarly, a notable number of This project utilized Vermonter Poll survey data described below as

401
L. Aultman-Hall et al. Travel Behaviour and Society 31 (2023) 399–409

Fig. 1. Long-distance travel decision-making approaches.

well as attributes of the intercity overnight trips between home origins area therefore includes a range of different levels of airport access with
and destinations estimated using GoogleMaps and FAA’s DB1B data those living in Chittenden County having the best access as measured by
described in the following section. distance from home county to the nearest airport. Airport access is
measured for each respondent based on home spatial location in our
3.1. Data from the Vermonter Poll survey analysis and up to three origin airports and three destination airports are
considered for each person’s most recent trip.
The Vermonter Poll is a telephone-based poll conducted annually by A series of four questions about the respondents’ most recent non-
the Center for Rural Studies at the University of Vermont. The poll work, long-distance trip were added to the 2017 Vermonter Poll as
sample is drawn randomly from Vermont landline and cellular tele­ part of this project. Respondents were asked to think about their last
phone numbers and uses computer-aided telephone interviewing (CATI) overnight trip out-of-town for personal reasons (like to visit family or
for data collection. Responses to the Vermonter Poll survey are collected friends or take a vacation) and asked for the country, state, and place
to represent the population demographics of Vermont, so no weighting containing their travel destination as well as the primary mode of travel
is necessary. The 2017 Vermonter Poll was conducted between the hours on the trip. Respondents were also asked two questions about their trip
of 9:00 a.m. and 9:00 p.m. on weekdays beginning on February 21, 2017 decision process:
and ending on February 28, 2017. The response rate for the 2017 Ver­ Which of the following was most true about how you traveled?
monter Poll was 20.1 %, producing 613 valid responses state-wide in
Vermont. Vermont is a small (population 624,525 in 2017) northeastern • Travel mode (such as driving or flying) was decided first and desti­
state that is predominantly rural. Burlington in Chittenden County, in nation was selected afterward.
the northwest portion of the state, is the only metropolitan statistical • Destination was selected first and then multiple travel modes were
area in the state with a population of 216,751. The study area is non- considered.
typical compared to most travel surveys in that it is state-wide and • Destination was selected first and only one travel mode was
two-thirds non-urban. As shown in Table 2 (all column), the sample is considered.
well-balanced in terms of demographics (only individuals that provided • Destination was selected first and there was only one travel mode
intercity trip information are included in the table). Other researchers available.
included a question about personal well-being on a scale of 1–10 and we • Don’t Know.
included this variable as a possible predictor. Burlington, VT has solid
air service with 578,014 commercial enplanements in 2017. This is Which of the following statements is most true about the trip’s
defined as a small hub airport by the FAA. Larger airports in Manchester, destination?
NH and Boston, MA are 165 and 216 miles from Burlington. The larger
international hub airport in Montreal, Canada is approximately 100 • My exact destination was known and I did not consider any other
miles from Burlington but was not used in this analysis as the data for destinations.
international flights is not included in the FAA database used. The study • Multiple destinations in the same region were considered.

402
L. Aultman-Hall et al. Travel Behaviour and Society 31 (2023) 399–409

Table 2 northeast of the US. While fewer destinations are in the west or central
Vermonter poll sample description (N = 440). US, there are many in Florida, a frequent winter destination for Ver­
Categorical Category N All monters. Recall although the data were collected in February we asked
Variables (only one category shown for binary about the last overnight personal trip and do not know the date of the
variables) trip. It is reasonable to assume that winter holiday trips in December and
Gender Male 194 44.1 % January are represented. As shown in the middle section of Table 2
Female 229 52.0 % above, while the mean one-way trip distance was 478 miles, the sample
missing 17 3.9 % included high variability, with travel to destinations from 10 to 2600
miles one-way.
Income < $25,000 47 10.7 %
$25,000–$50,000 81 18.4 %
3.2. Generating alternative mode attributes for trips
$50,000–$75,000 78 17.7 %
$75,000–$100,000 61 13.9 %
>$100,000 104 23.6 % The attributes of mode-specific travel times used in the mode choice
missing 69 15.7 % analysis were generated as alternative specific attributes using home
origin location and destination location. Our approach excludes non-
Education Level High School or Some College (no degree) 165 37.6 % personal vehicle surface modes (e.g. bus and rail) and private air
Associates or Bachelor’s degree 161 36.6 % travel options. For surface personal vehicles, uncongested highway
Graduate or Professional Education 98 22.3 %
travel time was used from Google API with a 2019 early Sunday morning
missing 16 3.7 %
time. Uncongested travel times were used because we did not know the
time or day of travel. This may be a more reasonable assumption within
Household Size 1 93 21.1 %
Vermont which lacks a major congested metropolitan area. Beyond
2 150 35.3 %
3 69 16.2 % Vermont, the limitations of this assumption are assumed to affect travel
4 69 16.2 % time more.
5 20 4.5 % For air travel, the origin airport access time, and total air travel time
6 19 4.3 %
(including access, egress, airport time and stops/transfers) were desired.
>6 4 0.9 %
missing 16 3.6 %
Estimating air travel attributes is complicated by the many flight route
options available to travelers depending on their ultimate origin and
ultimate destination so the weighted mean is used (weighted by the
Owns Home (versus Rent or Other Arrangement) 344 78.2 %
missing 17 3.9 % relative number of passengers using each route between ultimate origin
airport and ultimate destinations airport in the Federal Aviation
Children Present (versus No Children Present) 145 33.0 %
Administration (FAA) DB1B data). The logic applied here is attempting
missing 16 3.6 % to measure average air access between ultimate origins and destinations
not predict the actual airport choice or flight routing including stops that
Continuous Variables (All N¼440) Min Max Mean SD
was actually used by any individual. Because the logic involved selecting
Age (years) (18 missing) 19 92 54.3 15.6 the closet 3 airports to each ultimate origin (county centroid in Ver­
Personal Well-being (0–10) (15 missing) 0 10 7.9 1.9
Great Circle Distance to Destination (miles) 9.7 2596.4 477.6 617.9
mont) and destination (city center), it was necessary to select a subset of
Personal Vehicle Uncongested Time (minutes) 19 2795 546 637 all commercial airports in the continental US. For this study, 140 com­
Air Weighted Min Time (minutes) (33 missing) 179 770 451 152 mercial airports were included. Airports were selected using the
following logic and calendar year 2016 FAA airport data:

• Multiple destinations in different states, regions, or countries were


All medium and large hub airports.
considered.
Small hub airports were included if they were more than 65 miles (straight
• Don’t Know.
line distance) from a medium/large hub.
Non-hub primary airports with annual enplanements > 90 k.
Extensive beta-testing of the survey questions was conducted prior to
collection to ensure respondents had the best chance of correctly un­
The 3 closest airports were identified using the straight line great
derstanding what was being asked. Specifically, beta-testers answered
circle distance between the home county centroid or the destination city
versions of the questions and then an interview was conducted to
center and the airport latitude and longitude. For the 9 airport pairs the
determine (a) what behavior actually happened, (b) how the respondent
FAA DB1B was queried for all routes that passengers flew in 2017 be­
understood the questions, and (c) if the responses matched the behavior.
tween those airports. Flying time was estimated based on the miles
Of course, there are many factors related to these decisions, including
flown on each segment and a 450 mph average speed. Each stop or
the interplay between personal and household member decisions, but
transfer was assumed to add 90 min. Taxi time was accounted for only at
this survey only allowed the collection of data on the final behavior. Of
the origin airport based on the FAA on-time performance data for July
the 613 total respondents, 552 provided at least the destination country
2017. Origin airport waiting time was estimated to be 120 min for a
for their most recent trip. The 61 non-respondents to this question may
large hub airport and 60 min for all other airports. Destination airport
be non-travelers or simply been unable to recall the last trip meeting the
exit time was assumed to be 30 min. The sum of these air and airport
criteria in the question. In order to facilitate use of the FAA database
times were weighted based on the actual number of passengers who took
described in the next section we limited the analysis to those re­
each route between each origin and destination airport pair in the DB1B
spondents with trip destinations in the continental US. After excluding
dataset.
55 respondents who traveled to an international destination (including
29 trips to Canada, which is not unexpected as Vermont borders the
3.3. Descriptive results
Canadian province of Quebec near the city of Montreal and six re­
spondents’ whose most recent trip was Hawaii or Alaska, 440 observa­
Fig. 3 illustrates how distance from home to destination varied with
tions were included the final analysis. The destinations and reported
income, one of the most significant factors typically influencing long-
mode choices for the 440 analyzed trips are shown in Fig. 2. As one
distance travel. Solid bars indicate the interquartile range, the hori­
might expect for data collected in Vermont, many destinations are in the
zontal line is the median, and points represent outliers. Interestingly, the

403
L. Aultman-Hall et al. Travel Behaviour and Society 31 (2023) 399–409

Fig. 2. Overnight travel destinations (N = 440).

Fig. 3. Distance by household income. (o indicates observations beyond 1.5 times of the interquartile range and * indicates beyond 3 times the interquartile range).

mean distance does not vary significantly by income (ANOVA p = reinforces prior findings that there is no clear breakpoint where air is the
0.575). Persons of all income levels may need to reach destinations at dominate choice of surface modes. Rail and intercity bus seem to be
similar distances. Although, higher income respondents (27 % of the medium distance modes. Surface driving occurs in a limited number of
sample had household income greater than $100,000) were slightly even the longest trips in the sample. Air trips are on average much
more likely to fly (35 % of the flying trips). longer.
Fig. 4 illustrates how mode relates to distance to destination. This The first three variables in Table 3 (Destination Process, Mode

404
L. Aultman-Hall et al. Travel Behaviour and Society 31 (2023) 399–409

Fig. 4. Distance frequency by mode.

Process and Mode) summarize the outcome variables measured in the members are more likely to bundle decisions, and larger households
survey. In terms of destination decision process, the majority of re­ with 5 or 6 members revert back to sequential trip planning. This is an
spondents reported knowing their exact destination and another 9 % intriguing trend highlighting the complication of travel for households
knew the region. Moreover, the majority of people indicated that they of different sizes, and the results also show that the presence of children
did not make an independent mode choice, either they only considered might lead to a slightly higher chance for bundling decisions.
one mode or they reported there was only one mode available (62.2 % Fig. 5 provides data on whether certain modes are preferred as part
and 17.5 % respectively). Respondents traveled predominately by per­ of different decision-making processes. Interestingly, personal vehicles
sonal vehicle (76.8 %) and air (19.1 %) with less than 3 % using train or are notably more common for trips made with bundled decisions,
bus. The Decision Process entry in Table 3 shows how respondents’ whereas air travel is more commonly selected for trips planned
decision-making was classified according to the scheme that was out­ sequentially (Fig. 5a). Perhaps this shows personal vehicles are a default
lined in Fig. 1. Relatively few people made strictly sequential decisions. mode choice for many trips, which is just assumed to be the mode one
A total of 8 people followed the sequential decision-making process takes on an out-of-town trip. Fig. 5b looks specifically at destination
assumed in demand models (Fig. 1, Process 1) and two additional people choices and shows air travel is twice as common for trips where a
determined purpose, then mode, then destination (Process 2). Most destination was known and no other options were considered, compared
people made bundled decisions about mode and destination decisions to trips where multiple destinations were considered.
(Fig. 1, Processes 3 and 4). A small but meaningful proportion of re­ Fig. 5c also shares interesting trends related to mode choices. When a
spondents made unbundled decisions of mode after destination (Pro­ mode is selected first, it is almost exclusively by personal vehicle. This
cesses 1 and 5). Due to small number of observations in some categories, shows that there are many trips where an individual will decide they
a simplified decision process variable was used. Respondents were would like to take an out-of-town personal trip by personal vehicle and
divided into two categories for analysis: decision-makers that bundled then consider destinations accessible by that mode. Alternatively, when
mode and destination (Fig. 1, Processes 3 and 4) and those who deter­ destination is selected first and multiple modes are considered, air travel
mined destination before selecting a mode (Processes 1 and 5). Re­ is more likely to be selected. The likelihood of choosing air decreases
spondents categorized as following Process 2 were excluded from this when only one mode is considered or available. This demonstrates that
simplified categorization. order of decisions is quite important in mode choice, as is the bundling
There were limited patterns in how decision-making processes re­ or sequential process.
lates to destination distance from home (means were not different).
However, when multiple destinations were considered in one region 3.4. Modeling results
these trips had less variation with fewer longer trips; they were closer to
home and represented regional trips by driving. Multiple modes were Two models were estimated, one for decision processes (bundled
considered through a wide distance range. And people who perceived destination/mode or unbundled destination/mode) and one for mode
that only one mode was available to them traveled a large range of choice between driving and flying. The final dataset of 424 respondents
distances. provided enough information about their trip and decision process
Table 4 presents differences in demographics between those who (mode was flying or driving, decision types, and income was provided
bundled their out-of-town travel decisions and those who made desti­ for models using this predictor variable). Both decision models were
nation/mode choices in sequence. Individuals with household incomes estimated as binary logistic regressions using the mlogit package within
less than $25,000 are much more represented as bundled decision­ R Studio. These models assume that alternatives are mutually exclusive
makers (1.8 % to 13.73 %), whereas individuals with incomes greater and error terms are independent and identically (gumbel) distributed.
than $100,000 are more represented as sequential decision makers. This In the decision-process model (Table 5), the dependent variable
is also mirrored by the education levels. Single adults are more likely to described whether the destination and mode were bundled as opposed to
bundle decisions than make the sequentially, households with 3 or 4 the destination being identified first and the mode was identified

405
L. Aultman-Hall et al. Travel Behaviour and Society 31 (2023) 399–409

Table 3 Table 4
Travel/mode frequency observations (N = 440). Differences between sequential and bundled decision makers (Percentages are of
Variable Category Frequency Percent of
respondents who provided information).
full panel Categorical Category Sequential Bundled
Variables (only one category Decision Decision Makers
Destination 1 = My exact destination was 380 86.4 %
shown for binary Makers (N ¼ 354)
Process known and I did not consider any
variables) (N ¼ 70)
other destinations
2 = Multiple destinations in the 40 9.1 % Gender Male 46.3 % 46.0 %
same region were considered Female 53.7 % 54.0 %
3 = Multiple Destinations in 11 2.5 %
different states/region/countries
were considered Income < $25,000 1.8 % 13.7 %
Missing 9 2.0 % $25,000–$50,000 23.7 % 22.0 %
$50,000–$75,000 21.8 % 21.3 %
$75,000–$100,000 21.8 % 15.1 %
Mode Process 1 = travel mode was decided 20 4.4 % >$100,000 30.9 % 27.8 %
first and destination was selected
afterward
2 = destination was selected first 71 14.5 % Education High School or Some 31.2 % 40.2 %
and then multiple travel modes Level College (no degree)
were considered Associates or 35.9 % 38.9 %
3 = destination was selected first 270 62.2 % Bachelor’s degree
and only one travel mode was Graduate or 32.9 % 20.9 %
considered Professional
4 = destination was selected first 70 17.5 % Education
and there was only one travel
mode available Household Size 1 15.6 % 23.2 %
Missing 9 2.0 % 2 39.0 % 34.5 %
3 10.9 % 16.8 %
Mode Personal Vehicle 338 76.8 % 4 10.9 % 17.4 %
Air 84 19.1 % 5 14.1 % 3.2 %
Train 4 0.9 % 6 9.4 % 3.8 %
Bus 6 1.4 % >6 0.0 % 1.1 %
Missing 8 1.8 %
Home Own 84.4 % 80.3 %
Decision Process 1 (Destination Process 2 or 3; 8 1.8 % Rent or Other 15.6 % 19.7 %
(Fig. 1) Mode Process 2) Arrangement
2 (Destination Process 2 or 3; 2 0.5 %
Mode Process 1) Children Children Present 32.8 % 35.1 %
3 (Destination Process 1; Mode 313 71.1 % No Children Present 67.2 % 64.9 %
Process 1,3 or 4)
4 (Destination Process 2 or 3; 41 9.3 %
Mode Process 3 or 4)
political affiliations being more affiliated with not bundling mode and
5 (Destination Process 1; Mode 62 14.1 %
Process 2) destination decisions, compared to democrat, independent, or other
Missing 14 3.2 % affiliations. Interestingly, these affiliations were not correlated with
urban or rural areas, and variables describing urban level were not
Simplified Bundled Destination/Mode (3–4 354 80.4 % significant in the model. Also interestingly, the further away a destina­
Decision above) tion is for a trip (described with GCD straight line distance), the more
Process Non-bundled Destination/Mode 70 15.9 % likely an individual is to consider multiple modes after the destination is
(1 and 5 above)
determined. This is likely related to the regional nature of most trips in
Missing 16 3.7 %
the survey.
Next, a binary mode choice model was estimated to determine (a) the
second, or vice versa. Many independent variables were considered in relative importance of this bundling decision on mode choices and (b)
the estimation process, including demographics, political affiliation, how the characteristics of travel to the destination are connected to the
quality of life, household size, access to data-enabled cell phones, and mode choice, regardless of the decision-making process. The final model
trip characteristics. The best-fit model is shown in Table 5. It is impor­ (seen in Table 6) includes a dummy variable indicating whether the
tant to recognize that while the variables presented in this model are individual reported making a bundled decision of mode and destination.
statistically significant at an 85 % confidence level, this model is not This variable is statistically significant, and the coefficient indicates that
statistically different than a constants-only model, based on an F test individuals who make mode and destination decisions jointly (inde­
with 5 degrees of freedom. This emphasizes the fact that the majority of pendent of travel times and distance traveled) have an inherent pref­
Vermont travelers (84.6 %) automatically linked their destination and erence for, physical need or emotional circumstance necessitating
mode choices, and while there are some statistically significant varia­ driving a personal vehicle in place of flying. It is notable that this de­
tions in the decision process based on the variables presented in Table 3, cision process variable is the only demographic variable to be significant
they do not influence the decision process in any meaningful way. For in the model, despite the testing of multiple combinations of age,
the sake of discussion, households with the lowest incomes and older gender, income, political affiliation, quality of life, household size, ac­
travelers both were slightly more likely to associate a single mode choice cess to data-enabled cell phones, and others. In this dataset, de­
with the destination choice than their higher income and younger mographics did not play a role in mode choices. Note that in Table 2 the
counterparts. Households with the lowest income group were 6.4 times sample is diverse, and the sample design is intended to be representative
more likely to automatically link a mode, likely driving, to a trip of Vermonters.
destination than households with higher incomes. Political affiliation
was also a significant predictor of bundling decisions, with republican

406
L. Aultman-Hall et al. Travel Behaviour and Society 31 (2023) 399–409

Fig. 5. Mode choices by decision making processes.

407
L. Aultman-Hall et al. Travel Behaviour and Society 31 (2023) 399–409

Table 5 intercity usage. The finding that demographics are not influential in the
Logit model of bundle decision maker. type of decision-making or intercity mode choice is not consistent with
Parameter Outcome: Bundled Mode/ past research. It is possible the choices represented in this dataset are
Destination Decision impacted by the spatial structure and mode/destination combinations in
coeff. pvalue odds- New England around Vermont. Or it could be that the trips in prior
ratio surveys are dominated by more frequent longer distance travelers that
Republican Political Affiliation − 0.678 0.071 0.518
are not representative of the general public. Sociodemographic variables
Income <$25 k 1.851 0.073 6.367 require further consideration.
Age 0.035 <0.001 1.036 There are implications for intercity data collection. The location and
GCD Straight Line Distance (1000 s of miles) − 0.346 0.124 1.000 travel-access characteristics of both origins and destinations are essen­
Total N ¼ 341
tial for meaningful analysis and must be added to travel analysis.
log-likelihood of ¡141.26
Additionally, rather than ask for distance in time or miles (as often asked
in surveys), destination and mode considerations should be explicitly
asked. Travel distance and time can be imputed. Survey questions
Table 6
querying if air was considered, and for the origin airport used, could
Alternative intercity mode choice models.
enhance our ability to understand intercity travel decision making
Predictors Personal Vehicle Air Travel McFadden processes.
(base) R2
There are other travel characteristics not captured in this dataset that
coeff. t-stat coeff. t- may influence whether an individual reports a bundled decision sug­
stat gesting a need for joint choice models. Trip duration, structure and ac­
Constant 6.117 <0.001 – – 0.618 tivities at the trip destination(s) may influence an individuals’ need to
Ratio (Personal Vehicle/ − 3.405 <0.001 – – have access to a flexible mode (e.g. personal vehicle). Prior research has
Air Travel Time)
indicated that travel party size, including the number of children, also
GCD Straight Line Distance 1.516 0.040 – –
(1000 s of miles) influences perceptions of cost and flexibility that also may influence an
Used a Bundled Decision 1.174 0.016 – – individuals’ mode choices. We did not have the space to add these
Process questions to the Vermonter Poll. Also recall trip purpose (especially
differentiating between leisure and work) has been shown to influence
travel choices and this survey focused on personal trips only.
4. Discussion

5. Conclusions
These findings provided new relatively robust data that validate the
long-standing assumption that destination and mode are perceived of as
The paper set out to validate an important topic related to intercity
a bundle or package decision. It seems that even when multiple modes
travel: understanding bundled decision making processes linking mode
were available, individuals reported only having one feasible (or
and destination choice. Participants in a statewide telephone survey
appropriate) option. Moreover, this feasible option was not singularly
were asked about decisions for their most recent out-of-town trip for
tied to distance to destination. The context of this scenario can be
personal reasons. They provided the city destination and main mode
evaluated by considering a specific example of intra-regional travel from
used. A large majority (86 %) of respondents reported a bundled or
Vermont. Hartford, Connecticut is a relatively large city with a good
linked decision-making process for destination and mode for their most
airport. Hartford is a four-hour drive from Burlington, Vermont where
recent personal out-of-town trip. This finding demands joint models of
airport access is also good. However, there is no flight from Burlington to
destination and mode be pursued for intercity travel analysis. Most
Hartford and moreover no flight from Hartford to New York City (the
existing models, whether for regional or intercity travel are sequential.
main hubs flown to from Burlington). We used this case with several
When controlling for air and driving times, a dummy variable for type of
Vermonters by asking if they would ever consider flying to Connecticut
decision-making process was significant and improved models of mode
instead of driving. They said no they did not consider flying because it
choice. Our modeling revealed few meaningful relationships between
was not an option. This reveals that the person has indeed considered
decision-making process type and sociodemographic variables or trip
flying and rejected it since possessing the knowledge that flying is not a
length.
practical option compared to driving is inherent knowledge regarding
The analyses also indicated that bundled destination-mode choice
air service levels. Mode and destination are bundled but the extent to
models would benefit from mode-specific travel time variables derived
which people believe they have choices may be more about the service
from specific origins and destinations. Air accessibility between true
levels and availability between their origins and destinations. The case
origins and destinations requires estimation of airport access travel,
of intercontinental travel further complicates our ability to untangle
egress travel, stop/transfer times, and flying times. This approach was
mode and destination.
feasible with publicly available data and is recommended in order to
The modeling results here suggest that traditional joint or sequential
improve intercity mode choice models. Policy questions related to
destination and mode choice models (where distance is the only generic
infrastructure, congestion and environment will soon require more
predictor variable considered and no mode-specific variables are
detailed understanding on long-distance travel. These questions are
considered) are too simplistic. The coefficient values may change after
better answered by more behaviorally-representative models particu­
COVID but the need to include factors beyond distance may be more
larly as we move forward in a world where intercity travel choices are
important in that for a given destination, the relationship between
dictated by post-COVID attitudes and forward still into a world where
driving and flying times seems to be a key factor influencing mode
automated vehicle choices, both personal and shared, offer a unique new
choices. This is not unexpected, but it is usually unmeasured and the
modal option. New automated modes may offer travel times and service
method used here to approximate flying accessibility was successful.
types which vary significantly for those available now.
Generation of air impedance is recommended and methods to do the
There are many opportunities for future research into the decision-
same for rail and intercity bus should be pursued. There is an implication
making process. This study focused on generalized personal travel,
for policy and planning related to new modes such as automated vehi­
and it would be useful to study how decision-making may differ by
cles as a better understanding of the interaction of air and surface travel
purpose, accommodation type, tour structure, travel party, and time of
time may provide the measure of trade offs needed to estimate AV
year. This study also focused on travel from one state, so it would be

408
L. Aultman-Hall et al. Travel Behaviour and Society 31 (2023) 399–409

beneficial to study if these trends are consistent in other states with Cho, H.D., 2013. The Factors That Affect Long-Distance Travel Mode Choice Decisions
and Their Implications For Transportation Policy. Thesis. University of Florida.
similar or different urban development, highway connectivity to
Donnelly, R., Ed., SWIM v2.5 Model Development Report. Oregon Department of
metropolitan centers, and access to air and rail modes. Additional work Transportation. (2017) URL https://github.com/tlumip/model-dev-report/swim
could use the results of this analysis to collect data on alternative des­ 25mdr.pdf.
tinations considered and mode choices to further explore the factors Elmorssy, M., Onur T. 2020. Modelling departure time, destination and travel mode
choices by using generalized Nesterd Logit Model: Discretionary Trips. Int. J. Eng..
involved in this complex relationship. 33(2): 186-197.
Eugenio-Martin, J.L., 2003. Modelling determinants of tourism demand as a five-stage
Declaration of Competing Interest process: A discrete choice methodological approach. Tour. Hosp. Res. 4 (4),
341–354.
Florida Department of Transportation. Use of the Florida Statewide Model. 2018.
The authors declare that they have no known competing financial Fox, J., Patruni, B., 2021. Model of Travel in London Phase 3: Mode and destination
interests or personal relationships that could have appeared to influence choice model estimation. RAND Corporation Final Report.
Janzen, M., Axhausen, K.W., 2018. Decision making in an agent-based simulation of
the work reported in this paper. long-distance travel demand. Procedia Comput. Sci. 130, 830–835.
Jonnalagadda, N., et al., 2001. Development of microsimulation activity-based model for
Acknowledgements San Francisco: Destination and mode choice models. Transp. Res. Rec. 1777 (1),
25–35.
Juschten, M., Hossinger, R., 2021. Out of the City – but how and where? A mode-
We acknowledge funding for this project from the USDOT through destination choice model for urban-rural tourism trips in Austria. Curr. Issue Tour.
the NCST UTC program led by the University of California Davis. We 24 (10), 1465–1481.
Koppelman F., Axhausen K., Charlton B., Miller E., Small K, 2011. Independent peer
appreciate the UVM Center for Rural Studies and their continuing Ver­
review of the California high-speed rail ridership and revenue forecasting process.
monter Poll data collection. Ms. Madeline Suender conducted initial LaMondia, J., Snell, T., Bhat, C.R., 2010. Traveler behavior and values analysis in the
analysis for this project. Ms. Hannah Ullman assisted with graphics and context of vacation destination and travel mode choices: European Union case study.
review. Transp. Res. Rec. 2156 (1), 140–149.
Larsen, J., Urry, J., Axhausen, K., 2006. Mobilities, Networks, Geographies. Routledge.
Leake, G. R., Underwood, J. R. 1978. Comparison of inter-city bi-modal split models.
Contributions Transp. Plann. Technol. 5:1 55–69, https://doi.org/10.1080/03081067808717148.
Meyburg, Arnim H. An Analysis of the Relationships between Intercity Passenger
Transportation and the Socio-Economic Characteristics of Metropolitan Areas,
All authors contributed to the data analysis, study design and Proceedings of the Transport Research Forum. (1972).
interpretation of results. Study design and survey design was led by Lisa Moeckel, R., et al., 2015a. Modeling complex megaregion systems: horizontal and
Aultman-Hall. Material preparation, data collection and analysis were vertical integration for A megaregion model. Int. J. Transp. 3 (1), 69–90.
Moeckel, Rolf, Rhett Fussell, Rick Donnelly. 2015. Mode choice modeling for long-
performed by Jonathan Dowds and Anuarbek Onayev. Model estimation distance travel. Transp. Lett. 7:1 35–46. https://doi.org/10.1179/
and additional variable tabulation were performed by Jeffrey LaMondia. 1942787514Y.0000000031.
The first complete draft of the manuscript was written by Lisa Aultman- Monzón, A., Rodrı́guez-Dapena, A. 2006. Choice of mode of transport for long-distance
trips: solving the problem of sparse data. Transp. Res. Part A: Policy Pract. 40 (7),
Hall but significant portions of the manuscript and graphics were 587–601.
created by Jonathan Dowds, Anuarbek Onayev and Jeffrey LaMondia. Newman, J., Bernardin, V., 2010. Hierarchical ordering of nests in a joint mode &
All authors commented on previous versions of the manuscript. All au­ destination choice model. Transportation 37 (4), 677–688.
Outwater, M.L., et al., 2015. Tour-Based National Model System to Forecast Long-
thors read and approved the final manuscript.
Distance Passenger Travel in The United States. Presented at the Annual Meeting of
the Transportation Research Board, Washington DC. Paper No. 15-4322.
References Outwater, M., Tierney, K., Bradley, M., Sall, E., Kuppam, A., Modugula, V., 2010.
California statewide model for high-speed rail. J. Choice Modell. 3 (1), 58–83.
Adler, T.J., Ben-Akiva, M., 1976. Joint-choice model for frequency, destination, and Richards, M.G., Ben-Akiva, M., 1974. A simultaneous destination and mode choice model
travel mode for shopping trips. Transp. Res. Rec. 569. for shopping trips. Transportation 3 (4), 343–356.
Alliance Transportation Group, Arkansas Statewide Travel Demand Model: Passenger Rohr, Charlene, James Fox, Andrew Daly, Bhanu Patruni, Sunil Patil, and Flavia Tsang.
Model, Austin, Tx., (2015). Modelling Long-Distance Travel in the UK. In The 38th European Transport
Ashiabor, S., Baik, H., Trani, A., 2007. Logit models for forecasting nationwide intercity Conference, Glasgow. Citeseer, (2010) http://citeseerx.ist.psu.edu/viewdoc/
travel demand in the United States. Transp. Res. Rec.: J. Transp. Res. Board 1–12. download?doi=10.1.1.682.1367&rep=rep1&type=pdf.
https://doi.org/10.3141/2007-01 (2007). RSG Inc., Matthew Coogan, Mark Hansen, Richard Marchi, Nancy McGuckin, Megan
Atkins, 2013. Development of Georgia Statewide Model: Development of Statewide Ryerson, and Mike Welch (2019). Air Demand in a Dynamic Competitive Context with
Model Report. Georgia Department of Transportation. the Automobile. Airport Cooperative Research Report 204.
Aultman-Hall, L., Harvey, C., LaMondia, J., Ritter, C., 2015. Design and response quality Schmid, B., Jokubauskaite, S., Aschauer, F., Peer, S., Hossinger, R., Gerike, R., Jara-
in a one-year longitudinal survey of overnight and long-distance travel. Transp. Res. Diaz, S., Axhausen, K., 2019. A pooled RP/SP model, route and destination choice
Procedia 11, 136–153. https://doi.org/10.1016/j.trpro.2015.12.012. model to investigate mode and user-type effects in the value of travel time savings.
Aultman-Hall, L., Harvey, C., Sullivan, J., LaMondia, J., 2016. The impact of long- Transp. Res. A 123, 262–294.
distance tour generation and travel attributes on data collection in the United States. Sivaraman, V., 2015. A Theoretical and Methodological Framework to Analyze Long-
Transportation 43, 6. https://doi.org/10.1007/s11116-016-9754-y. distance Pleasure Travel. Graduate Thesis University of South Florida.
Behrens, C., Pels, E., 2012. Intermodal competition in the London-Paris passenger Sullivan, J., LaMondia, J., Harvey, C., Garrison, C., Aultman-Hall, L., 2016. An analysis
market: high-speed rail and air transport. J. Urban Econ. 71 (3), 278–288. https:// of the long-distance and overnight travel tour-planning process. Transp. Res. Rec.
doi.org/10.1016/j.jue.2011.12.005. 2594, 1–10. https://doi.org/10.3141/2594-01.
Boyce, David E., Chon K. S., Lee Y.J., Lin K.T., LeBlanc L.J., 1983. Implementation and Thrane, C., 2015. Examining tourists’ long-distance transportation mode choices using a
computational issues for combined models of location, destination, mode, and route multinomial logit regression model. Tour. Manag. Perspect. 15, 115–121.
choice. Environ. Plann. A 15:9 https://doi.org/10.1068/a151219. Timmermans, H.JP., 1996. A stated choice model of sequential mode and destination
Bradley, M., Outwater, M.L., Ferdous, N., 2016. Implementation of a practical model choice behaviour for shopping trips. Environ Plan A 28 (1), 173–184.
system to predict long-distance travel for the entire U.S. Population. Transp. Res. Toole, J.L., et al., 2015. Traveled: travel demand estimation using big data resources.
Rec.: J. Transp. Res. Board 2563:1. https://doi.org/10.3141/2563-02. Transp. Res. Part C: Emerg. Technol. 58, 162–177.
Cambridge Systematics. California Statewide Travel Demand Model, Version 2.0: Model Van Nostrand, C., 2011. A Discrete-Continuous Modeling Framework for Long-Distance.
Overview. 2014. Leisure Travel Demand Analysis.
Castillo-Manzano, J.I., Pozo-Barajas, R., Trapero, J.R., 2015. Measuring the substitution Watson, P.L., 1974. Comparison of the model structure and predictive power of
effects between high speed rail and air transport in Spain. J. Transp. Geogr. 43, 2. aggregate and disaggregate models of intercity mode choice. Transp. Res. Rec. no.
https://doi.org/10.1016/j.jtrangeo.2015.01.008. 527.
Chester, M.V., Ryerson, M.S., 2014. Grand challenges for high-speed rail environmental Wu, L., Zhang, J., Fujiwara, A., 2012. Dynamic analysis of Japanese Tourists’ three stage
assessment in the United States. Transp. Res. A Policy Pract. 61, 3. https://doi.org/ choices. Transp. Res. Rec. 2322, 91–101.
10.1016/j.tra.2013.12.007.

409

You might also like