You are on page 1of 15

Netw Spat Econ

DOI 10.1007/s11067-014-9224-z

Sustainability SI: Exploring Heterogeneity in Cycle


Tourists’ Preferences for an Integrated Bike-Rail
Transport Service

Ching-Fu Chen & Wen-Chieh Cheng

# Springer Science+Business Media New York 2014

Abstract Bicycle tourism has recently become an important niche tourism market, and
one that is growing along with rising awareness of the need for sustainable develop-
ment. However, one major concern is that such tourism might generate additional car
journeys, as cycle tourists often put their bikes in cars and drive to the location where
their cycling will take place. This paper focuses on an integrated bike-rail transport
service, and examines cyclists’ preferences with regard to this service by using a
discrete choice experiment in Taiwan. The preferences of different groups segmented
by cycle tourists’ recreation specialization level and willingness-to-pay (WTP) for
service attributes were examined. Data were collected using an on-site choice experi-
ment survey, with choice sets designed based on service attributes in the integrated
bike-rail transport service context. Mixed logit models were analyzed to explore the
varied preferences of the respondents. The results revealed that cycle tourists in general
are concerned about the service attributes of the integrated bike-rail transport service,
such as price, type of bike storage, bike storage location and service frequency. In
addition, varied preferences and differences in the WTP for the attributes were found
among the different groups.

Keywords Integrated bike-rail transport . Bicycle tourism . Recreation specialization .


Willingness-to-pay . Heterogeneity

1 Introduction

As tourism-related transport is increasingly recognized as a source of environmental


degradation, transport and tourism agencies have begun to pay more attention to
developing non-motorized tourism transport systems which meet the principles of
sustainable development at a destination level (Lumsdon 2000). In this context, cycling

C.<F. Chen (*) : W.<C. Cheng


Department of Transportation & Communication Management Science, National Cheng Kung
University, 1, University Road, Tainan 701, Taiwan
e-mail: cfchen99@mail.ncku.edu.tw
C. Chen, W. Cheng

has drawn considerable attention due to its wide array of societal and environmental
benefits (Lamont 2009; Pucher et al. 2010; Sener et al. 2009). According to Sustrans
(1999), the development of cycle tourism is of importance due to several reasons. For
example, cycle tourism is positive at generating local trade and offers a number of
business opportunities, particularly in rural areas. It is also an environmentally sustain-
able form of tourism, with minimal impact on the environment, and can help reduce
traffic congestion. Bicycles are thus an increasingly important mode of vacation
transport (Ritchie 1998), with cycle tourism being recognized as an important and
growing niche market, which has the potential to provide a range of economic, social
and environmental benefits to tourist areas and the wider community (Lumsdon 2000;
Lumsdon and McGratch 2011; Ritchie 1998). However, one major concern is that such
tourism might generate additional car journeys, as cycle tourists often put their bikes in
cars and drive to the cycling destination. For example, evidence from the UK’s first
long-distance cycling route, obtained by Cope et al. (1998), found that 85 % of
holidaymakers cycling on the Tarka Trail in the South-West of England arrived by
car, despite the network being served by a nearby rail station, which only 23 % of such
visitors used. Cycle tourism might thus lead to an increase in car trips if no convenient
and environment-friendly transport services are provided for the carriage of cycles. In
addition, where an acceptable level of public transport provision is made for the
carriage of cycles, inter-modality then makes a significant contribution to the long-
term environmental gains associated with this form of tourism (Lumsdon 2000), in
particular for those who are with higher level of recreation specialization, an important
measure in the research field of leisure and outdoor recreation. Therefore, designing
and providing an environmentally-friendly public transport service that recreational
cyclists accept is of considerable importance and in accordance with the call for
research on environmental sustainability in transportation networks, although little
research has yet been conducted into this specific topic.
To address this research gap, the present study advances past studies related to
cycling tourism and explores heterogeneity in cycle tourists’ preferences for an inte-
grated bike-rail transport service. The remainder of the paper is structured as follows.
Section 2 describes the concept of recreation specialization, while Section 3 outlines the
econometric model used in this work, i.e. the Mixed Logit (ML) model, and explains
the concept of willingness-to-pay (WTP). Section 4 describes the experimental and
survey design, while Section 5 reports the results of the analysis. Section 6 then
concludes the paper by offering a discussion of the results.

2 Recreation Specialization

The heterogeneity of participants’ preferences complicates product (or service) provi-


sion efforts that seek to both maximize participant satisfaction and provide sustainable
services. In light of this, recreation specialization, first proposed by Bryan (1977) as a
means of exploring within-group variability among recreational participants, is a useful
and effective measure to investigate preference heterogeneity. Recreation specialization
provides a general framework for understanding the attitudinal and behavioral differ-
ences of individuals in an outdoor recreation context. According to Bryan (1977),
recreation specialization refers to “a continuum of behavior from the general to the
Sustainability SI

particular, reflected by equipment and skills used in the sport and activity setting
preferences”. Over time, participants would move along a continuum from low in-
volvement and general recreational interests, to high involvement and more specific
ones (Bryan 2000). Progress from beginner to expert within a specific recreational
activity, defined as the level of recreation specialization, results in an increase in skill,
experience, and commitment to the activity, as well as changes in value orientation in a
person’s attitudes, behaviors and setting preferences (Bryan 1977).
Regarding the measurement of recreation specialization, McIntyre and Pigram
(1992) proposed a multidimensional construct consisting of behavioral (e.g., experi-
ence), cognitive (e.g., level of skill), and affective dimensions (e.g., enduring involve-
ment) which are interrelated and mutually reinforcing. These three dimensions, as
suggested by Scott and Shafer (2001) in their synthesis of specialization research, are
deemed to capture the essence of the construct. The behavioral dimension refers to past
experience of an activity and familiarity with a setting. The cognitive dimension
indicates the knowledge and skills an individual acquires. Finally, the affective dimen-
sion represents an individual’s attachment, which is related to commitment, centrality-
to-lifestyle, or ego involvement.
Recreation specialization frameworks have been proposed for a variety of outdoor
recreation groups, such as birders, anglers, hunters, canoeists, hikers, wildlife viewers
and boaters (Hvenegaard 2002; Jett et al. 2009). Past studies have examined the
relationships between recreation specialization and variables such as leisure value
(Bryan 1977; Oh et al. 2005), place attachment (Bricker and Kerstetter 2000),
conservation attitudes and behaviors (Dyck et al 2003; Miller and Graefe 2000; Oh
and Ditton 2008), management actions (Oh and Ditton 2006), setting preferences
(Hopkin and Moore 1995), and site choice behaviors (McFarlane 2004). In general,
the results of these earlier studies reveal that as individuals progress from a low to
high specialization level, they possess higher levels of environmental concern, more
responsible behaviors and greater preferences for sustainable programs. Different
levels of specialization lead to different preferences and site choices, with more
specialized individuals seeking settings that will test their skills and enable them to
exert greater control (Bryan 1977). Recently, recreation specialization has been
combined with stated preference choice models to investigate the variations in
participants’ preferences. One example is Oh and Ditton’s (2006) study of angler
preferences, which used the stated preference choice approach to elicit participants’
preferences with a set of hypothetical choice sets that included the most important
decision attributes and their related levels of specialization. In their work, the spe-
cialization level was used to segment participants into different groups by a cluster
analysis, followed by estimating conditional logit models for each segment to enhance
the understanding of group differences in preferences. The results revealed that high
specialization individuals are expected to be more likely to report higher net WTP
values for service attributes than low specialization ones.

3 Econometric Model

Stated preference choice models are used in this study to estimate the utility associated
with individuals’ evaluations of a designed set of multi-attribute bike-rail transport
C. Chen, W. Cheng

services based on random utility theory. The ML model is often used to avoid the
limitations of the Multi-nominal Logit (MNL) one, and this extends the standard
MNL model due to its capability to estimate the distribution of preferences for the
attributes examined in an experiment. Another advantage of the ML model is that
it relaxes the IIA assumption, and accounts for the fact that each individual makes
choices among alternatives which are not independent (Amador et al. 2008;
Cherchi and de Dios Ortúzar 2006, 2008; Train 2003). The ML model is in fact
a generalization of the standard MNL model, allowing for random taste variation,
unrestricted substitution and correlation in unobserved factors (McFadden and Train
2000). The ML model has been recently applied to investigate the heterogeneity in
individual taste preference in mode-choice of pro-environmental transport, such as
electric vehicles (Rasouli and Timmermans 2013).

3.1 Mixed Logit (ML) Model

The ML model, also known as the random parameter logit model, considers the
heterogeneity that exists among individuals. Since the parameters in the ML model
are assumed to be random, the utility of alternative j for an individual i is defined by
0
U ij ¼ βi X ij þ εij ð1Þ
β′i
where is a vector of random coefficients of the variables for individual i and varies in
the population with density f(βi/θ), where θ are the fixed parameters of this distribution
(usually the mean and covariance).
Given the value of βi, since the εij is the IID extreme value, the conditional
probability that individual i selects alternative j is a simple logit as

eðβ X ij Þ
0

Lij ðβÞ ¼ X 0 ð2Þ


eðβ X ij Þ
k

If βi is not given, however, the unconditional choice probability (Pij), i.e. the ML
probability, is the integral of Lij(β) over all values of βi as
Z
Pij ¼ Lij ðβÞf ðβ=θÞdβ ð3Þ

Since the integral does not have a closed expression, the ML probabilities
cannot be calculated analytically. In practice, researchers rely on simulation
methods which allow the maximum simulated likelihood estimator of the param-
eters θ that characterize the distribution of βi to be obtained (Train 2003). The
probabilities obtained depend on the distribution assumption. Typically, four main
distributions, including normal, lognormal, uniform and triangular, can be imposed
on random parameters (Train 2003) The uniform distribution is suitable for the
case of dummy variables, while the log-normal form is for the case that the
response parameter needs to be a specific sign (e.g., non-negative). Of these, the
normal distribution is the most well-known and widely applied (Hensher and
Greene 2003), and is thus adopted in the current study.
Sustainability SI

3.2 Willingness-to-Pay (WTP)

The discrete choice estimation of willingness-to-pay (WTP) was developed based on


Lancaster’s economic theory and random utility theory. In a choice experiment, the
participants are asked to show their trade-off between prices and other attributes, and
thus the willingness-to-pay value can be acquired (Train 2003).
Once the logit model has been estimated, the implicit price of each attribute can be
estimated to examine the respondents’ valuations of service attributes based on changes
in attribute levels. More specifically, the marginal WTP for each attribute can be
estimated as the ratio of an attribute coefficient over the price coefficient:
∂V =∂x β
WTP ¼ − ¼− x ð4Þ
∂V =∂p βP

where βx and βP are the coefficients of attribute x and price, respectively.

4 Experiment and Survey Design

A stated preference choice experiment approach is adopted in this study, because it


allows for (a) more variation in price and (b) better control over information provision
and choice set consideration. For the stated preference experimental design, a pilot
survey is conducted to elicit the main service attributes of green rail services that
consumers are concerned about. Ten initial attributes, including price, fixed bike rack,
passengers and bikes on the same coach, service frequency, train service during non-
holidays, need for a reserved seat, number of train stops, provision of on-line reserva-
tions, and package tours are selected based on the features of integrated bike-rail
transport services. A 5-point Likert-type scale is used in the pilot survey to measure
respondents’ perceptions of the importance of the ten attributes. Based on the responses
of 35 respondents with experience of using green rail services, fixed bike rack,
passengers and bikes on the same coach, service frequency, train type (i.e. service
level) and shipping price are the five key attributes. This finding is then used in the
experimental design. The five attributes and the respective levels used in this research
are shown in the Table 1.
A full factorial design for the five attributes with their corresponding levels comprising
the choice experiment provides a total of 11,664 (i.e. 36 ×24) combinations of alternatives.
While a full factorial design permits the main effects of each attribute, as well as the effects
of interactions among all the attributes, to be identified, in practice it is rather difficult for
the participants to answer for all the alternatives. An orthogonal fractional factorial design
that generates only a small subset of all possible combinations while maintaining the
orthogonality of the full factorial designs is therefore employed, and after removing three
dominant choice sets, this produces 24 choice sets. Furthermore, a blocking design is
added to group the choice sets to reduce the burden on respondents. In this, the 24 choice
sets are divided into four blocks of six choice sets. Each choice set is designed under three
choices (i.e. A, B and C). Choices A and B are two hypothetical alternatives with different
levels of the five attributes, whereas Choice C represents the ‘no-choice’ when neither
Choices A or B are chosen. An example choice set is shown in Fig. 1.
C. Chen, W. Cheng

Table 1 Attributes and levels used in the choice experiment

Attributes Description Levels

Service level Service levels of integrated bike-rail 1. Ordinary servicea


service. (NT$ 90/ 100 min)
Different passenger fares and 2. Semi-express service
traveling times (NT$ 90/ 85 min)
are associated with service levels. 3. Express service
(NT$ 140 / 70 min)
Bike rack Ways to store bike during the journey 1. Tied on the handraila
2. Stored in the luggage area
3. Fixed bicycle rack
Locations for passenger and Passengers and bikes are located 1. On same coach
bike on the same coach or not. 2. Not on same coacha
Frequency Frequency of services per day 1. 2 services /day
2. 4 services/ day
Shipping price Shipping price is calculated based on the 1. 50 % of passenger ticket price
percentage of the price of passenger 2. 75 % of passenger ticket price
ticket (varying at different service 3. 100 % of passenger ticket
levels) . price

a
Treated as reference base in mixed logit models

In addition to the choice experiment questions, details of the respondents’ level of


specialization with regard to bicycles were also collected. The specialization level is
measured by three dimensions, i.e. behavior (six items), cognition (five items) and
affection (six items), which are adapted from past studies (Bricker and Kerstetter
2000; Dyck et al. 2003; McFarlane 2004; McIntyre and Pigram 1992). The behavioral
dimension was assessed using open-ended questions, such as number of years as a
cyclist, frequency of cycling, average distance of each ride, maximum distance for a
ride in the past year, expenditure on cycling, and expenditure on cycling apparel and
safety equipment. The measurement scale for each item, as shown in Table 2, is
dependent on the item characteristic. The cognitive dimension was assessed by five
items using a 5-point Likert type scale ranging from 1 (= strongly disagree) to 5 (=
strongly agree). A sample item reads as “My cycling skill is good”. The affective
dimension was measured by six items using a 5-point Likert type scale, as noted
above. A sample item reads as “Cycling is important to me”.
Data were collected by on–site choice experiment survey interviews, which were
conducted at various well-known bike trails in southern Taiwan from December, 2008
to January, 2009. Cycle tourists who gathered at rest areas along the bike trails were
approached by a trained interviewer and briefed on the research purpose. After
ensuring that the cycle tourists had heard about the integrated bike-rail transport
service provided by the TRA, the interviewer asked about their willingness to take
part in the interview. If they responded positively, the respondents were then asked to
complete the questionnaire. Applying the convenient sampling technique due to
limited time and manpower, on-site interviews were conducted with a total of 222
respondents. As each survey consisted of six choice sets, a total of 1,332 (i.e. 222×6)
usable observations were obtained for model estimations. The majority of the sample
Sustainability SI

Alternatives

A B C
Attributes

Service level Express service Semi-express service


(NT. 140 / 70 minutes) (NT$ 90/ 85 minutes)
Bike rack Storage at the luggage area Fixed bicycle rack

None

Locations for on same coach not on same coach


passenger and
bike
Frequency Two services /day Four services /day
Shipping price 50% of the passenger price (i.e. 100% of the passenger price
NT$ 70) (i.e. NT$ 140)
Choice
Fig. 1 An example choice set

Table 2 Summary statistics of recreation specialization

Dimension/Items Mean S.D.

Behavior dimension (six items, α=.70)


B1: number of years as a cyclist (years) 2.17 2.45
B2: frequency of cycling ( times per week) 2.35 1.69
B3: average distance of each ride (km) 24.33 21.63
B4: maximum distance for a ride in the past year (km) 66.46 61.30
B5: amount of money invested in cycling (NT$) 21025.23 28116.48
B6: amount of money invested in cycling apparel and safety equipment (NT$) 3791.89 5160.55
Cognitive dimension (five items, α=.76)
C1: cycling skill is good 3.65 0.76
C2: discuss cycling skills with other people who cycle 3.62 0.81
C3: knowledge of how to maintain bikes 3.27 0.86
C4: awareness of information about cycling activities 3.38 0.82
C5: awareness of information about bikes from books or the internet. 3.23 1.01
Affective dimension (six items, α=.86)
A1: cycling is important to me 3.77 0.82
A2: cycling is enjoyable to me 4.28 0.63
A3: cycling can let me relax 4.31 0.62
A4: cycling is a way of self-expression 3.73 0.79
A5: my life is highly related to cycling 3.28 0.89
A6: cycling is a way to get along with friends 3.63 0.88
C. Chen, W. Cheng

was male (72.1 %), and 52.7 % were married. Around 40 % of the respondents were
aged 21–30, followed by those aged 31–40 (25 %). Around half of the respon-
dents stated they had a monthly salary of between NT$ 20,001 (around US$ 670)
and NT$ 50,000 (around US$1,675), followed by those with a monthly salary of
less than NT$ 20,001 (30 %). The descriptive statistics of the behavioral dimen-
sion for the sample included an average of 2.17 years as a cyclist, riding 2.35
times per week, 24 km per ride, a maximum distance of 66 km in the past year,
NT$ 21,025 invested in cycling, and NT$ 3,791 spent on related apparel and
safety equipment. The average scores of cognitive items ranged from 3.23 to 3.65,
while those of the affective items from 3.28 to 4.31. Table 2 reports the basic
statistics of recreation specialization.

5 Results

This study adopts two approaches to explore heterogeneous preferences. The first
is an ML model with interaction terms of socioeconomic variables interacting with
alternative-specific attributes. The second is a segmentation approach that un-
covers underlying segments. As different segments of recreational specialization
are postulated to have different preference structures, we employ the cluster
analysis approach to classify segments, and then use a stated preference choice
model to explain and predict preferences for each segmented group.

5.1 Segments of Recreation Specialization

Due to different scale measurements being used for the three dimensions of
recreation specialization, we follow the procedure used by Dyck et al. (2003)
and combine measures of multiple specialization dimensions into a single index of
recreation specialization. The standard scores, i.e. Z-scores, for each item score are
first computed, and the respondents’ Z-scores are then averaged across items
within each dimension to produce a single score for each of these. Finally, the
dimension scores are summed to provide an overall index of recreation speciali-
zation for each respondent. Based on the index of recreation specialization, the
sample is divided into equal-sized thirds as low, medium, and high specialization
groups. As shown in Table 3, the respondents’ overall Z-scores ranged from −4.14
to 5.99. More specifically, there were 74 low specialists (−4.14~−054), 74 medium
specialists (−0.53~07.6) and 74 high specialists (0.81–5.99).

Table 3 Groups by recreation specialization

Recreation specialization group Mean Minimum Maximum

Low (n=74) −1.54 −4.14 −0.54


Medium (n=74) 0.05 −0.53 0.76
High (n=74) 2.08 0.81 5.99
Sustainability SI

5.2 Results from the Mixed Logit Models

The ML model allows for variations in preferences across individuals, and adjusts for
error correlation across the choices made by each individual. In order to apply the ML
model, assumptions about the distribution of preferences have to be made. Typically,
four main distributions can be imposed on random parameters when using ML models,
which are normal, lognormal, uniform and triangular (Train 2003). We assume that
preferences relating to four service attributes are heterogeneous and follow a normal
distribution, while preferences towards price are homogenous. Therefore, separate
parameters are estimated for each individual for each of the four service attributes
along with a single parameter for all the price attributes. Fixing the price coefficient has
several advantages (Revelt and Train 1998). First, it ensures the right sign of the price
coefficient, as a normally distributed price coefficient implies that some respondents
may prefer a service with higher cost, which is counter-intuitive. Second, a fixed price
coefficient implies that the distribution of WTP for the remaining attributes follows the
same distribution as the attribute coefficients. Finally, fixing at least one of the
coefficients in the model helps empirical identification. The models are estimated by
maximum simulated likelihood using 1,000 Halton draws (Train 2003).
Table 4 shows the results obtained from three mixed logit models: ML1 for the
whole sample, ML2 for the high recreation specification segment, and ML3 for the low
recreation specialization segment. The overall fit of the model is measured by
McFadden’s Pseudo-R2 (i.e. ρ2). The value of ρ2 should be above 0.1 if the model is
to be accepted, whereas a value between 0.2 and 0.4 is considered as an extremely good
fit (Louviere et al. 2000). The explanatory power of the ML1 model was relatively
high, with a goodness-of-fit measure, i.e. McFadden’s ρ2, of 0.235. All the coefficients
are statistically significant and intuitively correct, except for the service levels. First,
price (β=.01) was found to have the expected negative sign and be statistically
significant at the 5 % level, implying that respondents prefer alternatives that are less
costly. With regard to the service level attributes, both semi-express service and limited
express service were not significant. This may be because the current integrated bike-
rail transport service is only provided within a small region, and thus the difference in
travel times among various service levels are not significant to the respondents. The
positive signs for ‘luggage area’ (β=.66) and ‘fixed bike rack’ (β=1.42) imply that
respondents are more likely to favor better services to store and protect their bikes than
simply ‘tied on the handrail’ during the trip. Additionally, note that the preference for
‘fixed bike rack’ is higher than ‘luggage area’. Regarding the attribute of whether
passengers and bikes are located on the same coach or not, respondents preferred that
they are together (β=.57). For the attribute of frequency, as expected, respondents
favored more frequent train services (β=.35), which were believed to increase flexi-
bility with regard to trip planning.
In addition, significant standard deviations for the three attributes (express service,
fixed bike rack and passengers and bikes on same coach) were found, indicating that
the data support choice specific unconditional unobserved heterogeneity for these
attributes, and that some respondents might prefer lower levels of them.
Turning to the model estimates for both low and high recreation specializa-
tion segments, the overall fits of the ML2 (low recreation specialization seg-
ment) and ML3 (high recreation specialization segment) models also indicate a
C. Chen, W. Cheng

good fit, i.e., ρ 2 = .243 and .277, respectively (See Table 4). Several
observations can be made based on the estimated results.
First, respondents from both groups, as expected, did not prefer high prices, but did
express a preference of a fixed bike rack, consistent with those in the ML1 model. This
implies that cycle tourists favor a lower shipping price and better equipment for storing
their bikes on the coach, regardless of the level of recreation specialization. Second,
respondents in the low specialization group significantly prefer a higher frequency of
train services. This suggests that cycle tourists who belong to this segment treat cycling
as a casual recreational activity, and thus hope to have more train services to choose in
terms of convenience when they decide to ride for leisure. However, the coefficient of
frequency is not statistically significant for the ML2 model, implying that respondents
with high recreation specialization are more likely to be serious cycle tourists who have
planned a timetable for their cycling activities in advance, and thus require less
flexibility. Third, the high recreation specialization group also preferred to have their
bikes stored on the same coach as they are sitting in. This might be explained by the
fact that since they invested more money in their bikes than the general population, they
want to make sure they are secure and remain within view. Last, differences in
preference towards types of facility for storing bikes exist between respondents with
different levels of recreation specialization. More specifically, the high recreation

Table 4 Results of mixed logit models

Attributes variables ML1: all sample ML2: low ML3: high


specialization specialization
Parameters (t- Parameters (t-value) Parameters (t-value)
value)

Price −0.01 (−4.16**) −0.02 (−2.72**) −0.02 (−2.48**)


Semi-express service 0.06 (0.46) 0.08 (0.28) −0.03 (−0.14)
Express service −0.32 (−1.60) −0.39 (−0.89) −0.23 (−0.67)
Luggage area 0.66 (4.67**) 1.10 (3.33**) 0.79 (3.02**)
Fixed bike rack 1.42 (5.98**) 1.90 (3.13**) 2.05 (4.45**)
Passengers and bikes on same coach 0.57 (4.23**) 0.68 (1.55) 0.61 (2.40**)
Frequency 0.35 (2.90**) 0.62 (2.32**) 0.47 (1.90)
Standard error for attributes
Express service 0.57 (2.58**) 1.43 (2.91**) 0.68 (1.67)
Fixed bike rack 0.73 (4.48**) 1.12 (3.23**) 1.07 (3.64**)
Passengers and bikes on same 0.56 (2.57**) 2.37 (4.09**) 0.57 (1.32)
coach
Frequency 0.32 (1.28) 0.57 (0.80) 0.32 (0.65)
Sample size 222 74 74
Number of experiments 1,332 444 444
LL(β) −626.14 −207.70 −198.81
LL(0) −818.61 −274.36 −274.97
ρ2 0.235 0.243 0.277

* and ** denote that the results are statistically significant at the 10 and 5 % levels, respectively
Sustainability SI

specialization segment prefers fixing bikes on fixed bike racks (β=2.05) to storing
them in the luggage area (β=.79). This preference for the low recreation specialization
segment is smaller, i.e. 1.90 for the fixed bike rack and 1.10 for the luggage area.
Overall, these results show that respondents with different levels of recreation special-
ization have different preferences for the integrated bike-rail transport service.

5.3 Results of WTP

To evaluate the impact of service attributes on the respondents’ choices, their WTP for
service provisions or improvements are calculated by using the coefficient estimates of
the ML models. Table 5 reports the mean WTP for attributes such as luggage area,
fixed bike rack, passengers and bikes on same coach, and frequency for three ML
models. Note that since the estimates of service type (i.e. express service and semi-
express service) were not statistically significant, their WTP values are therefore not
calculated. As expected, both recreation specialization segments obtain distinct values
of WTP for service attributes.
For the whole sample model, the mean WTP for a better bike storage service, from
‘tied on the handrail’ to ‘storing in the luggage area’ and to ‘fixed rail rack’, is NT$ 59
and NT$ 102, respectively. Note that the latter is much more than the former. The mean
WTP for a better bike storage location, from ‘not on same coach’ to ‘on same coach’, is
NT $50. Finally, the mean WTP for an additional train service per day, i.e. a better
service frequency, is NT$ 30. These results indicate that respondents are willing to pay
more for better service quality in general.
Comparing the low and high recreation specialization models (see Table 5), a
number of differences are found in the WTP for service attributes. For example, the
low recreation specialization segment (NT$ 46) is willing to pay more for a better bike
storage service, moving from ‘tied on the handrail’ to ‘parking at the luggage area’,
than the high recreation specialization segment (NT$ 67). In contrast, the high recre-
ation specialization segment (NT$ 153) is willing to pay more for a better bike storage
service, from ‘tied on the handrail’ to ‘fixed bike rack’, than the low recreation
specialization one (NT$ 90). In addition, the high recreation specialization segment is
willing to pay NT$ 77 to sit with their bikes in the same coach instead of in a different
one, whereas the low recreation specialization segment is not willing to pay for this
better service. In terms of more frequent services, the low recreation specialization

Table 5 Mean WTP for service attributes (NT$)

Models ML1: all sample ML2: low specialization ML3: high specialization
Variables

Luggage area 59 (36,76) 46 (37,55) 67 (60,73)


Fixed bike rack 102 (95,109) 90 (76,104) 153 (134,172)
Passengers and bikes on same coach 50 (44,56) 44 (17,70) 77 (71,83)
Frequency 30 (20,40) 37 (17,57) –

1$ ≒32 NT$; 95 % confidence intervals in brackets


C. Chen, W. Cheng

segment is willing to pay (NT$ 37), but the high recreation specialization segment is
not.
According to the findings above, implications are made as following:

(1). Improvement in equipment for parking bikes and increase in diversity of services
Based on the results, the cyclists were found to favor the train service attributes
such as facility for settling their bicycles and sharing the same car with their
bikes. However, the preferences for higher service level (i.e. express service in
this study), sharing the same car, and fixed bike rack are heterogeneous, sug-
gesting the service provider can remodel their services and equipment with
consideration of different needs for heterogeneous passengers. For instance, in
order to improve the space utilizing for settling bikes in the train car, it is
suggested to have both luggage area and foldable seats to increase the flexibility
of transportation. It is also possible to add fixed bike racks in the cargo car, and
attached it to the regular passenger train to satisfy passengers’ comfort and heir
bikes’ safety.
(2). Price differentiation
Results from WTP estimation indicated that the bicyclists were willing to pay
more for being able to stay in the same car with their bikes and fixed bike racks.
Since these two attributes were valued the most by subjects in this study, the
service provider may want to set different prices for trains with different quality
levels of equipment to attract more heterogeneous passengers.

6 Discussion and Conclusion

Along with the increasing popularity of cycle tourism, there has been growing interest
in developing an environmental-friendly transport system to help cycle tourists trans-
port their bicycles to their destinations. We demonstrate the advances over past studies
related to cycling tourism and explore heterogeneity in cycle tourists’ preferences for an
integrated bike-rail transport service. Specifically, this paper identifies cycle tourists’
preferences with regard to the integrated bike-rail transport service among different
recreation specialization groups by using the stated preference approach and ML
models. Since recreation specification reflects a continuum behavior from the general
to the particular by equipment and skills used in the sport and activity setting prefer-
ences (Bryan 1977), it provides a suitable measure to cluster cyclists into various
specialization groups that we assume each group members hold similar preferences
in transport-related services such as the integrated bike-rail transport service.
The results of the ML model show evidence that cyclists in general are concerned
about the service attributes of an integrated bike-rail transport service, such as price,
bike storage type, bike storage location and service frequency. Several differences in
preferences towards service attributes, such as bike storage location and service
frequency, are found between the high and low recreation specialization groups. In
this study, we employ the construct of recreation specialization to account for the
heterogeneity in preferences among cyclist tourists. Cycle tourists with a high level of
recreation specialization are in favor of storing their bikes in the same coach they are in,
Sustainability SI

while the preference for this is not significant for the low recreation specialization
group. In contrast, the low recreation specialization group prefers more frequent
services, although this is not true for the high recreation specialization one. Therefore,
this study provides evidence that heterogeneous preferences exist among cycle tourists
segmented by the level of recreation specialization. The level of recreation specializa-
tion reflects on three dimensions: behavior, cognition and affection of the underlying
recreation activity. In the context of cycle activity, a cyclist with a higher level of
recreation specialization tends to cycle frequently, spend more money on cycling-
related activities or items, own (use) a better quality bicycle, and get more involved
with cycling in his or her daily life. Thus, it is reasonable to justify their preferences in
attributes such as bike storage location and service frequency for the integrated bike-rail
transport services of cyclists with a higher recreation specialization when they take rail
transport to a tourism destination for the cycling purpose. In line with past studies
(Dyck et al. 2003; Kuentzel and Heberlein 1992; McFarlane 2004; McFarlane et al.
1998; McIntyre and Pigram 1992), recreation specialization is thus seen as an effective
variable to explore the specific preferences of different cyclist segments, and one that
can provide useful implications for travel and tourism managers.
The WTP results reveal that cycle tourists are willing to pay more for better service
quality. In addition, the WTP for better storage is the greatest among all service
attributes, indicating the high importance of this for cycle tourists when they consider
using an integrated rail-bike transport service. Therefore, providing a better storage
service to take care of tourists’ bicycles should be a priority issue for rail transport
manager or marketer to target cycling tourists using the integrated rail-bike transport
service. Significant differences in the WTP pay for service attributes are also found
between high and low recreation specialization cyclists, with the latter willing to pay
more for a better bike storage in the luggage area than the former. In contrast, high
recreation specialization cycle tourists are willing to pay more for a better bike storage
using a fixed bike rack than low recreation specialization ones. Furthermore, the high
recreation specialization segment is willing to pay for keeping their bikes in the same
coach they are sitting in, whereas the low recreation specialization one is not. In
contrast, the low recreation specialization segment is willing to pay for more frequent
services, while the high recreation specialization segment is not. Integrated rail-bike
transport service providers can thus use the results of this work to develop more
effective marketing strategies to meet the needs of distinct segments of cycle tourists.
Recreation specialization is used as a segmentation variable in this work to explore
the differences in cycle tourists’ preferences. Although this study combined the three
specialization dimensions into a single index to produce a continuum of specialization,
treating the dimensions as a continuum may mask the differential effects of the
dimensions (McFarlane 2004). Techniques utilizing various specialization dimensions
as segmentation variables, and integrating these with choice models to explore varia-
tions in individual tastes or preferences, are thus suggested for future research. Alter-
native models accounting for consumer taste variations, such as a latent class model
(Greene and Hensher 2003) or nested logit one (Train 2003), can be estimated and
compared in the context of integrated bike-rail transport service choice in future studies.
Although this work was able to effectively capture variations in preferences by using
the recreation specialization level, this heterogeneity might also be uncovered in other
factors in relation to cycle tourists’ specific characteristics, such as trip purpose, travel
C. Chen, W. Cheng

party composition and travel party size. The issue of heterogeneity in preferences and
how to find more appropriate sources of heterogeneity in order to design more desirable
services for such travelers is thus one issue that deserves further study.
It should be noted that the current study collected a sample of cycle tourists by a
convenient sampling technique, and only focused on locations in southern Taiwan, due
to limited time and manpower. Although the results provide insights into cycle tourists’
preferences with regard to the integrated bike-rail transport services and the differences
among various specialization segments, they cannot be generalized to a wider popula-
tion without more validation being undertaken. More research evidence from other
regions in Taiwan or other countries is thus suggested.

References

Amador FJ, González RM, de Dios Ortúzar J (2008) On confounding preference heterogeneity and income
effect in discrete choice models. Netw Spat Econ 8:97–108
Bricker KS, Kerstetter DL (2000) Level of specialization and place attachment: an exploratory study of
whitewater recreationists. Leis Sci 22:233–257
Bryan H (1977) Leisure value systems and recreational specialization: the case of trout fishermen. J Leis Res
9:174–187
Bryan H (2000) Recreation specialization revisited. J Leis Res 32:18–21
Cherchi E, de Dios Ortúzar J (2006) Income, time effects and direct preferences in a multimodal choice
context: application of mixed RP/SP models with non-linear utilities. Netw Spat Econ 6:7–23
Cherchi E, de Dios Ortúzar J (2008) Empirical identification in the mixed logit model: analysing the effect of
data richness. Netw Spat Econ 8:109–124
Cope AM, Doxford D, Hill AI (1998) Monitoring tourism on the UK’s first long-distance cycle route. J
Sustain Tour 6:210–223
Dyck C, Schneider I, Thompson M, Virden R (2003) Specialization among mountaineers and its relationship
to environmental attitudes. J Park Recreat Adm 21:44–62
Greene WH, Hensher DA (2003) A latent class model for discrete choice analysis: contrasts with mixed logit.
Transp Res B 37:681–698
Hensher DA, Greene WH (2003) The mixed logit model: the state of practice. Transportation 30:133–176
Hopkin TE, Moore RL (1995) The relationship of recreation specialization to setting preferences of mountain
bicyclists. J Leis Res 14:47–62
Hvenegaard GT (2002) Birder specialization differences in conservation involvement, demographics, and
motivations. Hum Dimens Wildl 7:1–36
Jett JS, Thapa B, Ko YJ (2009) Recreation specialization and boater speed compliance in Manatee zones.
Hum Dimens Wildl 14:278–292
Kuentzel WF, Heberlein GD (1992) Does specialization affect behavioral choices and quality judgments
among hunters? Leis Sci 14:211–226
Lamont M (2009) Reinventing the wheel: a definitional discussion of bicycle tourism. J Sport Tour 14:5–23
Louviere JJ, Hensher DA, Swait JF (2000) Stated choice methods and analysis. Cambridge University Press,
Cambridge
Lumsdon L (2000) Transport and tourism: cycle tourism – a model for sustainable development? J Sustain
Tour 8:361–377
Lumsdon L, McGratch P (2011) Developing a conceptual framework for slow travel: a grounded theory
approach. J Sustain Tour 19:256–279
McFadden D, Train K (2000) Mixed MNL models for discrete response. J Appl Econ 15:447–470
McFarlane BL (2004) Recreation specialization and site choice among vehicle-based campers. Leis Sci 26:
309–322
McFarlane BL, Boxall PC, Watson DO (1998) Past experience and behavioral choice among wilderness users.
J Leis Res 30:195–213
McIntyre N, Pigram JJ (1992) Recreation specialization reexamined: the case of vehicle-based campers. Leis
Sci 14:3–15
Sustainability SI

Miller CA, Graefe AR (2000) Degree and Range of specialization across related hunting activities. Leis Sci
22:195–204
Oh C, Ditton R (2006) Using recreation specialization to understand multi-attribute management preferences.
Leis Sci 28:369–384
Oh C, Ditton R (2008) Using recreation specialization to understand conservative support. J Leis Res 40:556–
573
Oh C, Ditton R, Anderson DK, Scott D, Stoll JR (2005) Understanding differences in nonmarket valuation by
angler specialization level. Leis Sci 27:263–277
Pucher J, Dill J, Handy S (2010) Infrastructure, programs, and policies to increase bicycling: an international
review. Prev Med 50:106–125
Rasouli S, Timmermans H (2013) Influence of social networks on latent choice of electric cars: a mixed logit
specification using experimental design data. Netw Spat Econ. doi:10.1007/s11067-013-9194-6
Revelt D, Train K (1998) Mixed logit with repeated choices: households’ choices of appliance efficiency level.
Rev Econ Stat 80:647–657
Ritchie BW (1998) Bicycle tourism in the South Island of New Zealand: planning and management issue.
Tour Manag 19:567–582
Scott D, Shafer CS (2001) Recreation specialization: a critical look at the construct. J Leis Res 33(3):319–343
Sener IN, Eluru N, Bhat CR (2009) An analysis of bicycle route choice preferences in Texas, US.
Transportation 36:511–539
Sustrans (1999) Cycle tourism TT21. Sustrans, Bristol
Train K (2003) Discrete choice methods with simulation. Cambridge University Press, Cambridge

You might also like