Integration of HSR and air transport: Understanding passengers’ preferences

Transportation Research Part E 71 (2014) 129–141

Contents lists available at ScienceDirect

Transportation Research Part E journal homepage: www.elsevier.com/locate/tre

Integration of HSR and air transport: Understanding passengers’ preferences q Concepción Román ⇑, Juan Carlos Martín Institute of Tourism and Sustainable Economic Development (TiDES), University of Las Palmas de Gran Canaria, Faculty of Economics, Campus de Tafira Modulo-D, 35017 Las Palmas de Gran Canaria, Spain

a r t i c l e

i n f o

Article history: Received 13 March 2013 Received in revised form 4 September 2014 Accepted 5 September 2014

Keywords: Transport networks Intermodal transport Modal integration HSR services stated preference (SP) Willingness-to-pay

a b s t r a c t Different solutions for the integration of high-speed rail (HSR) and air transport could be implemented, ranging from very basic integration to more sophisticated systems which include ticket and handling integration. A discrete choice experiment is conducted to better understand passengers’ preferences. We estimate a number of flexible choice models, taking into account the existence of systematic and random taste heterogeneity. We obtain a range of willingness-to-pay values for service quality attributes, finding some important results that can be used to infer policy conclusions about the real attractiveness of the Air–HSR integrated alternative. In this respect, we find that schedule coordination which reduces connecting time will be crucial. Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction Air and rail passenger transportation have always been seen as substitutes rather than complements when a particular corridor or area is analysed (González-Savignat, 2004; de Rus and Roman, 2006; Román and Martín, 2010; Román et al., 2010; Yang and Zang, 2012). Initially, at the beginning of the twentieth century, the railways dominated the passenger transport demand for long-distance trips, but during the last 60 years the airline industry has undergone an expansion unrivalled by any other form of transport, and the situation was reversed. The rate of technological change in air transport has resulted in falling costs and fares, which stimulated a very rapid growth in demand. However, with the introduction of high-speedtrains, railways are again back on the scene. The competition between these two transport modes is occurring in several areas that affect their competitiveness very differently. Some policy measures and transport mode characteristics make the scenario anything but level. These concern, for example, trips between city centres, location of airports, access to transport terminals, airport security controls vs. more relaxed measures in train stations, and, finally, public money invested in railway infrastructure against full cost-recovery for the airlines and airports. However, recently, we observed a veritable change in the competitive relationship that has previously existed between the HSR companies and the airlines. Nowadays, cooperation is becoming a win–win solution for a number of interurban trips, e.g. some Thalys trains complement the Air France network for the flight between Brussels and Paris Charles De Gaulle airport, or some of Germany’s ICE trains carry a fixed number of seats from Lufthansa passengers, under their respective q The authors acknowledge the financial support provided by Project E 20/08 from the public call: Integración del Transporte Aéreo y Alta Velocidad Ferroviaria: Impactos sobre Accesibilidad y Medio Ambiente, Ministerio de Fomento. Proyectos de I+D+I de la acción estratégica en Energía y Cambio Climático, 2009–2010. ⇑ Corresponding author. E-mail addresses: [email protected] (C. Román), [email protected] (J.C. Martín).

http://dx.doi.org/10.1016/j.tre.2014.09.001 1366-5545/Ó 2014 Elsevier Ltd. All rights reserved.

130

C. Román, J.C. Martín / Transportation Research Part E 71 (2014) 129–141

code-share agreements. These two examples in the EU have catalysed a new era of cooperation instead of competition between air transport and HSR. Therefore, for the near future it is possible to envisage how HSR could play a feeder or distributional role in the air transport industry. Unfortunately, as Yang and Zhang (2012) conclude, the literature on air transport-HSR cooperation is still very scarce. Some studies have partly analysed some of these issues (e.g. Grimme, 2007; Chiambaretto and Decker, 2012; Jiang and Zhang, 2014), but more work is needed. Modal integration may be conceptualised as the well-known practice in the air transport industry known as ‘‘code sharing’’, where two airlines cooperate by offering, respectively, one leg of a two-leg trip. This type of agreement has been studied in Oum et al. (1996), Brueckner (2001) and Gayle (2008). Jiang and Zhang (2014) have extended these previous studies to analyse air transport-HSR integration, and they show that social welfare is not affected by whether or not the hub capacity is constrained, as long as the cross-elasticities of both transport modes are not significant in the overlapping markets. If this is not the case, then welfare is very dependent on the constraints of hub capacity to obtain welfare gains or losses. An increase in the degree of cooperation between HSR operators and airlines could certainly enhance the dominant position of some European hubs, at least in the internal market. It is, however, more frequent to see how airlines are adding new nodes to their networks using HSR stations, so nowadays a passenger can buy a ticket from city A to city B, transferring at an airport which is connected to the HSR network and making part of the trip by HSR. The cases of Charles de Gaulle, Schiphol and Frankfurt are previous European experiences that were developed before the construction of the intermodal HST terminal in Madrid Barajas airport. The aim of this paper is to analyse the key drivers on this trend towards cooperation, in order to promote intermodal connectivity between Air and HSR transport networks in Madrid Barajas airport, focusing on the estimation of passengers’ preferences for the main attributes that define modal integration. The interest goes beyond just how terrestrial access to the airport is going to be changed by the creation of a new HSR station. For this reason, it is those air transport trips which involve at least two flight legs – those being the last one, a trip within the Iberian Peninsula, and a transfer at Madrid Barajas airport – which are the object of this study. Therefore, the new alternative will replace the last leg of the trip by an HSR service, and, consequently, the transfer will involve changing transport mode. In order to succeed in the market, such modal integration, known as AEROAVE (the Spanish acronym for the integration of Air and HSR), has to provide a better alternative than that existing on some routes: that is, the integrated alternative will provide a generalised cost reduction for some trips, and some passengers will evaluate this transport mode interchange as positive. We analyse the preferences of travellers for this new alternative with respect to some basic attributes, like ticket integration, ground-handling integration, price, transfer, and travel time. The empirical analysis is based on individual subjective responses to a discrete choice experiment conducted at Gran Canaria airport with some of those passengers who were connecting at Madrid Barajas airport on some route to a Spanish province in the Iberian Peninsula. We obtained the willingness-to-pay for the different attributes that characterise this new alternative, analysing the main key drivers that need to be identified in order to develop a competitive intermodal alternative. We show that one of the main drivers for this new alternative to be successful is ‘‘the connecting time’’, and for this reason we highlight that the schedule coordination between both modes will be crucial in order to promote intermodality at Madrid Barajas airport. Our findings may prove to be useful for planners, policy makers, and air transport and HSR operators who aim to foster complementarity or cooperation between both transport modes. The next section briefly discusses to what extent HSR and air transport have been competing and are nowadays cooperating more than in the past. Section 3 presents the data, and explains how the experiment was conducted. Section 4 is dedicated to the model and the results obtained. Finally, Section 5 concludes giving some perspective of what needs to be done in order to facilitate the competitiveness of this new alternative.

2. The relationship between HSR and air transport Definitions of high-speed trains are usually based on what speed passengers can achieve on their trips. Thus, experts seem to agree within a range between 200 and 300 kph. Givoni (2006) concludes that an HSR service should be defined as ‘‘high capacity and frequency railway services achieving an average speed of over 200 kph’’(p.1). Nash (2009) suggests that a common definition is ‘‘rail systems which are designed for a maximum speed in excess of 250 kph’’ (p.1), while the European Union EC Directive 96/58/EC defines High-Speed rail as a set of three elements with precise criteria: (1) the infrastructure has to be built specifically for high speed travel or must be specially upgraded for high speed travel; (2) the rolling stock must run at a speed of at least 250 km/h on lines specially built for high speed, and at a speed of 200 km/h on existing lines which have been specially upgraded; and (3) the rolling stock must be designed along with its infrastructure for complete compatibility, safety, and quality of service. Traditionally, air and rail have been considered as competitors, and the relationship between HSR companies and airlines has been dominated by great rivalry. Even the literature on these two modes has mainly focused on the analysis of factors driving the competition between them. The range in which these two high-speed transport modes can compete has been widely studied and, although the findings differ across the transport literature, some agreement has been reached. Before the advent of HSR, trains were unable to compete successfully with flights on trips covering distances over 400 km. Nowadays, air passenger transportation and HSR compete strongly on a range between 400 and 600 km., the latter usually being the dominant transport mode. Other studies have mentioned the interval 750–800 km as the maximum distance that

C. Román, J.C. Martín / Transportation Research Part E 71 (2014) 129–141

131

allows HSR to act as a competitor to air (IATA, 2003). On journey time, the tipping point has been found to be around 2.5 or 3 h for an HSR journey (especially for the case of business trips), which would be equivalent to a 1 h flight; apart from the time spent in the plane/train, other parts of total journey time (such as access, waiting, and egress time) have to be taken into account, given their importance due to the nature of air journeys. Intermodal agreements between airlines and rail operators are beginning to be more popular in the transportation arena. These agreements have been promoted by the benefits accrued by the different stakeholders: airlines, rail companies, intermodal airports, consumers, and policy makers. In the EU, there is some empirical evidence that supports these agreements politically because some environmental externalities can be reduced by transferring passengers from air transport to HSR or by alleviating congestion problems at some European hubs that operate under important capacity constraints. Janic (2011) shows that the HSR substitutive capacity does not act as a barrier to developing air/HSR substitutions at the airport. Even a modest substitution may produce substantial savings in airline costs and passenger delays. It is also true that, in spite of such favourable policy makers’ opinion, air-rail products are still in their infancy, being positioned as ‘niche products’ which are offered by only a few operators, subject to bilateral arrangements. Multilateral cooperation between air and rail operators is made difficult not only because these two modes have been competing and still continue to compete on some of the routes that now could be an object of desirable cooperation, but also because both modes present different business models that have been subject to different constraints in the past: for example, in terms of customer relationship management, pricing policies, terminal lounges, loyalty programmes, etc. The integration of these operators’ culture is one of the main organizational barriers that exist in the design of this intermodal product. That said, it is also necessary to cautiously advise that integration needs a real intermodal platform – an HSR station available inside the airport terminal, in order to develop attractive air-rail products with adequate ticket integration, handling integration, etc. However, while these air-rail product characteristics support and increase the number of intermodal passengers, they do not represent a necessary condition, as some passengers prefer to substitute planes by HSR whenever this possibility exists within the airport terminal. Passengers travel by trains and planes in a complementary way, even without intermodal stations or sophisticated air-rail integrated schemes. Air-rail intermodality in Europe has been previously analysed in different studies that have involved the participation of important stakeholders, mainly policy makers, airports, and airline and railway operators (CEC, 1995, 2001; EC:DGT, 1998, 2004, 2011; IARO, 1998; IATA, 2003). The reports vary widely in extension and coverage, but, in all the cases, they provide interesting results about the degree of cooperation between both transport modes; when and how competition and cooperation is affected; the effects of the frequency of the services of HSR in airport terminals; and the efficiency of the HSR-airport terminal design. All these are some of the essential elements that need to be considered in order to foster modal integration. Other basic characteristics of this new alternative to provide seamless journeys to passengers are interchange conditions and the integration of tickets and handling. Givoni and Banister (2006) have analysed the previous literature on HSR and air transport and found that most of the papers are focused on interurban modal competition. However, there are a number of exceptions that can be cited (Givoni, 2005, 2007; Givoni and Banister, 2006, 2007; Givoni and Rietveld, 2008; Chiambaretto and Decker, 2012; Zanin et al., 2012; Jiang and Zhang, 2014). Chiambaretto and Decker (2012) argued that there are three factors which can be associated with the expansion of intermodal agreements in Europe: (1) the ‘rebirth’ of the rail industry; (2) the difficult trading environment for airlines; and (3) the development of airports which can accommodate intermodal forms of transportation. The idea of such development is based on to what extent HSR stations can be used as additional spokes within the air hub-and-spoke network, creating a win–win situation for the main stakeholders involved – airlines, rail operators, airports and taxpayers. For example, airlines could be better off if they changed some slots from domestic routes to more profitable international routes, and HSR could benefit from the additional traffic and more satisfied passengers. Jiang and Zhang (2014) analyse, in the framework of an endogenous decision-making process, the effects of airline–HSR cooperation with respect to hub airport capacity, economies of traffic density, vertical differentiation between modes, differences in price-elasticities of demand, and heterogeneous passenger types. Readers interested in the welfare effects of this type of modal integration are referred to this recent paper. 3. The data To understand passengers’ preferences for an Air–HSR integrated mode of transport, amongst those who use the intermodal facilities of Madrid–Barajas Airport, we focused on routes linking the Island of Gran Canaria with different cities in mainland Spain, through a connection at Madrid–Barajas Airport. The route LPA-MAD is the fourth largest domestic route in Spain. Although it is evident that connecting passengers at Madrid–Barajas have very diverse origins, the ad hoc sample chosen for this study is considered suitable to assess the main attributes that define the characteristics of this new intermodal product for the following reasons: (1) the respondents are all residents in Spain, and have a good route knowledge of HSRs, and therefore are easily placed in the context of the experiment, as they are able to assess with greater awareness the choice situations presented to them; and (2) the route under analysis is similar to many other routes that connect the main European cities with Madrid, so our results could be easily transferable in this case. This is not, however, the case on other routes, where the first segment is a long-haul flight, e.g. to Asia, North America, or South America. These cases would require a specific study of the effect of connecting time when the time components of in-flight travel differ considerably.

132

C. Román, J.C. Martín / Transportation Research Part E 71 (2014) 129–141

For our research, we conducted a stated choice (SC) experiment where travellers were confronted with the choice between the current alternative (Air–Air) and the Air–HSR option. As well as the usual and normally included attributes in every mode choice model, say travel cost and travel time, our experiment also included variables that are relevant in modal integration1: connecting time, fare integration and baggage integration. As one of the most claimed advantages of the HSR versus air transport is the better accessibility of train stations, access time to destination was also included in the experiment. The primary purpose of the choice experiment was to determine the independent effect of the attributes on the observed choices made by the sample respondents who undertook the experiment (Rose and Bliemer, 2009). To achieve realism and accuracy in the outcomes of the experiment, the attribute levels were customised to reflect each respondent’s current experience. Thus, the alternatives presented in the choice sets were different for each respondent and were defined by pivoting attribute level values around the reference alternative (the current option), taking into account plausible percentage deviations (Rose et al., 2008). All the information needed to define the reference alternative was obtained from a group of preliminary questions included in the questionnaire. Table 1 presents the value of the attribute levels used in the discrete choice experiment (DCE). Travel cost, travel time and access time to destination were defined at three levels in the two alternatives. In order to create choice situations close to each of the respondents’ experience, connecting time for the Air–Air option was defined at the current level. It is evident that the new alternative will affect the connecting time for the current Air–Air alternative, and a more realistic set of values could have been set, considering three different levels with a worse connecting time. However, a fixed level at its current value was preferred for making the experiment simpler. Thus, the variation of this attribute among alternatives was achieved by introducing choice tasks such that the value of this attribute for the Air–HSR alternative was lower in six of the nine scenarios considered in the experiment. This was consistent with the low level of service frequency offered by airlines on some domestic routes, which typically involves higher connecting times.2 Fare and baggage integration for the Air–HSR option were defined at respectively, three and two levels, and these were compared with the current level in the Air–Air alternative, i.e. full integration in both attributes. In this regard, it is important to note that the higher level for fare integration tries to mimic the full fare integration (i.e. one single ticket) of air transport. In contrast, the higher level for baggage integration in the Air–HSR alternative represents a lower level of service than that experienced in air transport.3 This will have certain implications for the interpretation of some model parameters that will be discussed in the next section. An efficient SC design, based on the D-error minimisation criteria, was created with the N-gene software (ChoiceMetrics, 2009). As it was not possible to build a specific design for each respondent (the ideal situation),4 nor to know beforehand the final model specification, we decided to generate a design by considering the attribute level values corresponding to a representative trip for the general structure of the multinomial logit model (MNL). Although we are aware that this strategy could result in a sub-optimal design, the gain in realism would offset this problem by reducing the hypothetical bias (Bradley, 1988). Also, increasing the sample size would positively contribute to the quality of the estimation results. Parameter priors were chosen in order to obtain plausible WTP figures. Estimation results from Román et al. (2010) have been consulted in order to get some idea about the magnitude of the time parameters. Regarding qualitative attributes, as we did not have previous information, priors were chosen taking into consideration that the resulting WTP figures would vary within the range €3 and €9. The nine different choice situations generated for the SC experiment, as well as the parameter priors, are presented in Table 2. The selection of the appropriate design priors is not an easy task, and there is some evidence that priors need to be well related to the preferences that define the utility space. In this regard, Burgess et al. (2014), analysing six different designs according to a range of construction techniques, some of which incorporate priors for the parameters, found that specifying the right model to best describe the underlying preferences of the respondents may then become the limiting factor in the estimation of more complex generalised multinomial models, rather than the design per se. It is not usually debatable that experiments constructed with the correct priors usually perform better when researchers simulate the data, but it is not clear that this performance will translate into better predictions in an empirical study. Bliemer and Rose (2011) compared two Bayesian designs with an LMA design for estimating only the main effects, and found that the designs constructed using prior information resulted in parameter estimates with lower standard errors than those obtained from the LMA design. The survey was conducted between November 2010 and January 2011, and the fieldwork was organised in four waves of interviews, in accordance with the procedure used to find respondents in four locations: the check-in and boarding area at Gran Canaria Airport; different departments of the civil administration in the city; and different departments of the University of Las Palmas de Gran Canaria. For the first two groups, the reference trip was the current trip; and for the rest of the interviewees, the last trip made within the last 12 months. In all the cases, the reference trip involved taking two planes connecting at Madrid Barajas airport. Thus, our stated preference exercise was easier to respond to and contextualise, as all the

1 In fact, these variables are also relevant for any trip that involves any transfer or connection, e.g. air–air alternative, but as it can be seen later on, the level of quality provided by one transport mode or operator is usually better if the coordination between transport modes does not exist. 2 These connecting times in the inter-modal station are going to depend on the degree of cooperation of both transport mode operators regarding their schedule coordination, and also on the development of a true HST node in Barajas airport that on some routes permits passengers to go directly to their final destination. 3 The level of service is lower because the transfer is not as seamless in Air–HSR as it is in Air–Air as the passengers are obliged to change transport modes and to carry their own luggage between terminals. 4 For this, data collection must be carried out in a two-stage process: in the first stage, collect the data for the reference alternative; in the second stage, optimise a design for each individual based on their reference levels. This could be done on the fly if the design generator is linked to the questionnaire, but this requires specialised software to conduct the survey; otherwise, two waves of surveys administered to the same set of respondents are needed (Rose et al., 2008).

133

C. Román, J.C. Martín / Transportation Research Part E 71 (2014) 129–141 Table 1 Attributes and levels. Attributes

Level

Travel cost

Travel time (in-vehicle)

Connecting time1

Access time to destination Fare integration

0 1 2 0 1 2 0 1 2 0 1 2 0

Transport mode Air–Air

Air–HSR

P  20% P P + 20% Tv  20% Tv Tv + 20% Tc

P  50% P  25% P T0v  20% T0v  10% T0v Tc  50% Tc  25% Tc + 25% Ta  40% Ta  25% Ta  10% 2 independent tickets (The airline does not assume any risk in case of delay) 2 independent tickets + one travel insurance to compensate passengers if they have to buy a new HSR ticket in case of delay, without paying an extra cost. One single ticket. The two companies assume all the costs incurred in case of delay

Ta  20% Ta Ta + 20%

1

2

Baggage integration

Full fare integration: One single ticket. (The airline assumes all the costs incurred in case of delay)

0

The baggage is picked-up at Terminal T4 and carried by the passenger to the HSR Terminal (200 m. walking distance) The baggage is picked-up at a Special Airport Terminal within a walking distance of 20 m. from the HSR Terminal

1 2

Full baggage integration: The baggage is checked-in at the origin and picked-up at the final destination

Notes: 1 Hsu (2010) shows that connecting time is mainly affected by the capacities and headways of the connecting and feeder services. P = Air ticket price. Tv = Air in-vehicle travel time (the two legs). T0v = Air + HSR in-vehicle travel time (the two legs). Tc = Connecting time at the airport. Ta = Access time to destination.

Table 2 Choice situations. Attributes

Choice sets Priors

1

2

3

4

5

6

7

Air–Air

Travel cost Travel time (in-vehicle) Connecting time Access time to destination Fare integration Baggage integration

0.01 0.0025 0.0034 0.0034 0.03|0.11 0.09

P Tv Tc Ta + 20% 2 1

P  20% Tv + 20% Tc Ta + 20% 2 1

P Tv  20% Tc Ta  20% 2 1

P Tv  20% Tc Ta + 20% 2 1

P + 20% Tv + 20% Tc Ta 2 1

P + 20% Tv + 20% Tc Ta 2 1

P  20% Tv Tc Ta  20% 2 1

8 P + 20% Tv Tc Ta 2 1

9 P  20% Tv  20% Tc Ta  20% 2 1

Air–HSR

Travel cost Travel time (in-vehicle) Connecting time Access time to destination Fare integration Baggage integration

0.01 0.0025 0.0034 0.0034 0.03|0.11 0.09

P T0v  20% Tc + 25% Ta  25% 2 0

P  25% T0v  20% Tc  50% Ta  40% 0 1

P T0v Tc + 25% Ta  40% 1 1

P  25% T0v Tc  50% Ta  40% 2 1

P  50% T0v  10% Tc + 25% Ta  10% 0 1

P  50% T0v  20% Tc  25% Ta  10% 1 0

P T0v  10% Tc  50% Ta  25% 1 0

P  50% T0v  10% Tc  25% Ta  25% 2 0

P  25% T0v Tc  25% Ta  10% 0 0

respondents were facing the alternative Air–Air, which is based on his/her own experience, with the new alternative based on the new intermodal product in which the necessary air trip between Madrid Barajas and other cities located in one Spanish province of the Iberian Peninsula is substituted by HST. In each task, the respondent was asked to provide which alternative of a pair they prefer. Given the nature of the SC experiment, a face-to-face computer-aided personal interview (CAPI) was developed with the aid of the Sawtooth Software package. After thorough data cleaning, a total of 875 valid questionnaires were collected, resulting in 7875 observations for model estimation. The main characteristics describing our sample of respondents can be summarised as follows: (1) 41 years old on average, male (58% of respondents); resident in the Canary Islands (86%); travelling to Galicia (14%), Andalucía (18%) and

134

C. Román, J.C. Martín / Transportation Research Part E 71 (2014) 129–141

Table 3 Sample description. Residence in canary Islands Destination region Trip frequency (Year) Overnights Trip purpose Access mode Egress mode Destination Airline’s F.F. Program Fare type Activities during the flight Pieces of hand baggage Pieces of checked bags Gender Education level Activity Age In-vehicle travel time Connecting time Access time to destination Net household income Travel cost

Yes 86.1% Levante 31.0% Less than 1 17.5% 1–3 21.1% Work 35.9% Car 65.6% Car 50.9% City centre 69.6% Yes 53.9% Business 1.9% Work related 8.9% 0 5.3% 0 24.8% Male 57.5% Graduate 53.6% Employee 80.8% 41 216 min 112 min 38 min 4272 € 114

No 13.9% Andalusia 17.7% 1–3 54.5% 4–6 30.8% Study 2.9% Taxi 22.3% Taxi 23.5% City suburbs 15.5% No 46.06% Tourist 54.7% Lecture 54.6% 1 78.1% 1 70.6% Female 42.5% Phd 20.7% Self-employed 6.4%

Galicia 14.5% 4–6 17.0% 1 week or more 48.0% Leisure 42.2% Public Transport 11.31% Public Transport 23.3% Other municipality 14.8%

Basque Country 11.2% 6–12 6.9%

Other 25.6% More than 12 4.1%

Visit Friends 14.6% Other 0.8% Other 2.3%

Other 4.5%

Economy 42.7% Music/video 10.3% 2 15.4% 2 4.1%

Other 0.6% Sleep 13.9% More than 2 1.3% More than 2 0.5%

Prof. Studies 14.9% Unemployed 3.3%

Secondary or lower 10.9% Other 9.5%

Other 12.2%

Fig. 1. Importance of design attributes.

Levante (31%); (2) travelling between one and three times per year (54%); (3) staying away for more than one week (48%) for leisure purpose (42%) and work (36%); (4) accessing the airport by private vehicle (66%), with connecting time at Barajas Airport of around 112 min.; (5) using private vehicle (51%) and public transport (23%) to go to the final destination that is located in the city centre (70%); (6) egress time at the destination is 38 min.; (7) belonging to the Iberia frequent-flier

C. Román, J.C. Martín / Transportation Research Part E 71 (2014) 129–141

135

Fig. 2. Importance of other attributes.

programme Iberia Plus (54%); (8) travelling in economy (43%) or tourist class (55%), and most of the individuals read (55%), listen to music (10%), sleep (14%) or work during the trip (9%); (9) check-in one piece of baggage (78%); (10) having a secondary school level of education (11%) or higher education (89%); (11) paying on average €114 per ticket, and belonging to a household with a net family income higher than €3000 per month. A more detailed description of our sample is presented in Table 3. Once the respondents had completed all the tasks of the DCE, they were also asked to rate, according to the degree of importance, the six attributes included in the design of the experiment. Fig. 1 depicts the distribution of the responses on a 5-point Likert scale, where 5 represents ‘very important’ and 1 represents ‘not important at all’. Although all attributes have been considered important, travel cost and travel time are among the most relevant when the individual expresses his travel preferences. Fig. 2 analyses the degree of importance of other attributes that have been commonly reported in the literature as the competitive advantages of HSR over air transport. When the two options are available (i.e. when the second leg of the trip can be made by HSR), safety, punctuality, service frequency, and comfort are the most positively rated.

4. Model and results Under the assumption of random utility maximisation (McFadden, 1974), different discrete choice models were estimated using the SC data set, with the purpose of understanding travellers’ preferences for an integrated Air–HSR transport mode. Our modelling approach was primarily focused on the analysis of factors that determine modal choice when integrated transport alternatives compete with traditional modes. On the other hand, the study of systematic and random taste variation, as well as the panel effects inherent to every SC data set, was also analysed. To meet these objectives, we estimated a number of progressively more flexible models, whose specification included random parameters and the interaction of design attributes and socio-economic variables (for a detailed reference on discrete choice models see Train, 2009). All models were calibrated with the software package Biogeme 2.2 (Bierlaire, 2009) using different procedures to generate random draws, depending on the model specification. The estimation results for different Multinomial Logit (MNL) and Mixed Logit (ML) models are presented in Table 4. The first model MNL1 consisted of a simple linear specification of the attributes included in the discrete choice experiment described in Table 1: travel cost (C), in-vehicle travel time (Tv), access time to destination (Ta), connecting time (Tc), the dummy corresponding to level 1 of baggage integration (BI) and the two dummies corresponding to levels 1 and 2 of fare integration (FI1 and FI2, respectively).The utility specification for the two alternatives is represented by:

U 1ðAirAirÞ ¼ hASC þ hC C 1 þ hT v T v 1 þ hTc Tc1 þ hTa Ta1 ; U 2ðAirHSRÞ ¼ hC C 2 þ hT v T v 2 þ hTc Tc2 þ hTa Ta2 þ hFI1 FI1 þ hFI2 FI2 þ hBI BI

ð1Þ

As we have already mentioned above, fare and baggage integration attributes were presented at their current level (i.e. full integration) for the Air–Air alternative in all the choice tasks included in the experiment. Thus, they were not considered in the specification of the utility of this alternative because of the lack of variability. In contrast, their specification was

136

Table 4 Estimation results. Attributes

Estimates (t-test) MNL1

MNL2

MNL3

MNL4

ML1

ML2

ML3 Log h

ASC Air–Air (ASC)

hASC

Travel cost (C)

hC

Mean

Fare integration level 1 (FI1)

hFI1

Fare integration level 2 (FI2)

hFI2

Access time to destination (Ta)

hTa

Mean Stand.Dev.

Connecting time (Tc)

hTc

Connecting time (Tc) * Work (W)

hTc * W

Travel time in-vehicle (Tv)

hTv

Travel time in-vehicle (Tv) * Work (W)

hTv * W

Baggage integration (BI)

hBI

Baggage integration (BI) * Baggage (B)

hBI * B

Baggage integration (BI) * Baggage (B) * Leisure (L)

hBI * B * L

Mean Stand.Dev.

Sigma l * (0) l * (C) l * (h) Observations Number of Draws (Type) Run Time t-test for H0: hFI1 = hFI2 a b

r

0.707 (11.19) 0.0222 (28.00) – – 0.41 (5.55) 0.515 (6.81) 0.378 (2.64) – – 0.431 (10.7) – – 0.627 (21.22) – – – – 0.0405 (0.63) – – – – – – 5458.53 5451.32 4260.71 7875

0.727 (12.06) 0.0221 (28.18) – – 0.425 (5.88) 0.553 (7.72) 0.285 (2.05) – – 0.436 (10.81) – – 0.635 (21.85) – – – – – – 0.213 (2.93) – – – – 5458.53 5451.32 4256.62 7875

0.725 (11.97) 0.0222 (28.22) – – 0.422 (5.83) 0.551 (7.68) 0.288 (2.08) – – 0.386 (8.12) 0.157 (1.97) 0.593 (17.72) 0.12 (2.47) – – – – 0.192 (2.58) – – – – 5458.53 5451.32 4252.71 7875

1.28

1.35

1.65

1.66

1.1 (10.57) 0.0343 (29.59) – – 0.543 (6.08) 0.786 (8.26) 0.389 (1.76) – – 0.581 (9.54) 0.255 (2.39) 0.797 (16.63) 0.318 (4.29) – – – – 0.215 (1.96) – – 1.92 (23.72) 5458.53 5451.32 3684.72 7875 200 Halton 01:01:55 2.38

1.44 (12.00) 0.0472 (22.60) 0.0294 (11.71) 0.696 (6.44) 0.728 (6.75) 0.843 (3.54) 1.48 (4.40) 0.614 (8.89) 0.258 (2.21) 0.843 (15.89) 0.372 (4.73) – – – – 0.141 (1.18) 0.467 (1.81) 2.27 (21.48) 5458.53 5451.32 3587.69 7875 200 Halton 00:37:01 0.27

1.31 (10.95) 1.42 (29.07) 0.817 (8.87) 0.875 (7.34) 0.827 (7.70) 1.42 (1.27) 0.77 (0.81) 0.632 (8.92) 0.205 (1.71) 0.914 (15.32) 0.347 (4.19) – – – – 3.21 (1.94) 1.82 (1.99) 2.1 (21.26) 5458.53 5451.32 3650.51 7875 100 MLHS 01:15:12 0.39

h

5.776a (9.57) 5.628 [6.47]b

0.325 (0.85) 0.292 [0.08]b

0.211 (0.31) 1.087 [0.58]b

In the estimation of ML3, the cost variable was divided by 100 to avoid overflow in the log-normal distribution. This scale change should be taken into account in the computation of WTP figures. t-test for the coefficient of variation (CV) of log-normal parameters.

C. Román, J.C. Martín / Transportation Research Part E 71 (2014) 129–141

Stand.Dev.

0.702 (10.42) 0.0222 (27.69) – – 0.406 (5.33) 0.507 (6.27) 0.395 (2.65) – – 0.43 (10.68) – – 0.625 (20.65) – – 0.0186 (0.28) – – – – – – – – 5458.53 5451.32 4260.87 7875

C. Román, J.C. Martín / Transportation Research Part E 71 (2014) 129–141

137

included in the utility of the Air–HSR alternative according to expression (1). For this reason, the competitive advantage of the Air–Air option regarding these two attributes is captured by the parameter corresponding to the alternative specific constant hASC , which could also be confounded with other unobserved effects. All parameter estimates presented consistent signs, and are significant at the 95% confidence level, the only exception being baggage integration that presented a very low value for the t-test. In order to understand the effect of this attribute, a new model MNL2 was estimated, considering instead the interaction of baggage integration and the variable baggage (B), being equal to 1 for those individuals who checked in at least one piece of baggage, and 0, otherwise. Although, in this case, the significance of this variable increased, its effect is still not significant at a reasonable confidence level. Thus, we considered another model MNL3 adding the interaction of a third variable: namely, leisure (L), which is equal to 1 if the trip purpose is leisure. The results demonstrate that the effect of baggage integration was significant only for those individuals who checked in some baggage and travelled for leisure purposes5 (which represent 42% of the respondents). This result is consistent with the composition of our sample, as 90.5% of the individuals travelling for leisure purposes checked in at least one piece of baggage, and therefore could positively value the availability of integrated baggage services. In contrast, the effect of this attribute is almost non-existent for work trips, since a significant percentage of these trips (46.5%) involve travellers who did not use the baggage handling services. Model MNL4 also includes the interaction of the variable work (W) that is equal to 1 if the trip purpose is work, and zero otherwise, with connecting time and in-vehicle travel time. In this case, all parameter estimates were significant at the 95% confidence level, and presented the expected sign. The specification of these interactions allowed us to conclude that both connecting and in-vehicle travel time are more negatively perceived for those individuals whose trip purpose is work. Different ML models were estimated in order to analyse panel correlation effects, as well as the existence of random taste heterogeneity. The first one, ML1 consisted of a fixed parameter specification that included an error component represented by a random variable normally distributed with zero mean and unknown standard deviation as a way to account for the potential correlation among the choices of the same respondent.6 The high level of significance obtained for the estimated standard deviation (Sigma) of the error component allowed us to confirm the hypothesis of panel correlation. ML2 and ML3 correspond to a random parameter specification for the travel cost, access time to destination and baggage integration * baggage * leisure parameters, where no correlation is allowed among random parameters, but panel correlation effects are accounted for. During the modelling process, other parameters were allowed to be random, but, as their standard deviations were too small, they were estimated as fixed in the final specification. In the case of the ML2 model, the Normal distribution for all random parameters was considered. Even though this model provided consistent estimates, in terms of the parameter signs and their significance level, the size of the population parameters of the Normal distribution produced a relatively high probability of obtaining an inconsistent sign in certain coefficients. Thus, according to this model, the probability of obtaining a positive marginal utility for the access time to destination was 0.28. A similar result was obtained for the parameter of baggage integration * baggage * leisure, yielding a 0.38 probability of being negative. As this may cause certain problems in the phase of model application, e.g. in the simulation of the WTP distribution, the ML3 model was estimated considering log-Normal distributions for the random parameters, forcing the corresponding marginal utilities to present consistent signs. In order to compare coefficients with other ML models, for the ML3 model we reported the parameters and t-ratios for the underlying normal distribution (Log h), as well as their transformed values corresponding to the log-normal distribution (h). The t-ratio of the coefficient of variation of the log-normal parameters is obtained instead of that for the standard deviation. In this case, except for the cost coefficient, the random parameters substantially reduced their significance level. Other specifications were tested by considering a mixture of normal and log-normal parameters but they did not provide consistent results. It is important to point out that if the decision to choose between ML2 and ML3 were based only on the value of the log-likelihood function, ML2 would be significantly better. However, choosing between different models normally entails a compromise between statistical significance and economic consistency. Thus, in our view, ML3 is more appropriate as this model does not exhibit any inconsistency with the microeconomic principles. Regarding the magnitude of the coefficients, in general, their interpretation appears consistent. Travel and connecting time produce more disutility for work trips; and in contrast with other studies, in-vehicle travel is more negatively perceived than connecting time and access time to destination. The high proportion of in-vehicle travel time with respect to total travel time for these trips (approximately, 59%) could be a possible explanation of this result. In fact, according to the average figures reported in Table 3, in our sample, in-vehicle travel time is nearly six and two times higher than, respectively, access and connecting time. Notwithstanding, a more sensible analysis of this effect would require an experimental design capable of accounting for the effect of time-squared terms, in order to obtain the marginal utility of in-vehicle travel time in terms of the duration of the trip within the vehicle. As an example, this figure differs substantially from the one obtained in the Madrid–Barcelona corridor, where the percentage of in-vehicle time over total travel time is only 30 per cent (Román et al., 2010). As for the qualitative variables, the model results indicate that fare integration is perceived as more important than baggage integration, the latter being significant only for individuals who check in their luggage and travel for leisure purposes. In this regard, it is important to note that the two levels of fare integration were perceived as different, at an 5

In fact, the log-likelihood for this model increases by near 4 points. Note that these assumptions imply that the alternative specific constant (ASC) is normally-distributed among the respondents. Also it is important to note that the specification of the error component lead to a re-scaling of the other coefficients, so that any comparison between MNL and ML models should bear this in mind. 6

138

C. Román, J.C. Martín / Transportation Research Part E 71 (2014) 129–141

Table 5 Willingness-to-pay values. Attribute

Willingness to pay MNL4

ML1

ML2

ML3

Mean

Median

SD

Mean

Median

SD

Baggage integration (€) The traveller checks baggage and the trip purpose is leisure (€) Fare integration level 1 (€) Fare integration level 2 (€)

8.65 19.01 24.82

6.27 15.83 22.92

3.33 16.53 17.27

2.64 13.54 14.15

18.35 19.21 19.48

4.54 26.27 25.19

0.94 20.22 19.25

10.60 19.75 19.30

In-vehicle travel time (€/hour) Trip purpose is other (€/hour) Trip purpose is work (€/hour) Access time to destination (€/hour)

26.71 32.12 12.97

23.24 32.51 11.34

19.70 27.09 14.37

16.33 23.28 13.90

20.30 22.18 33.34

27.13 33.60 10.07

21.05 27.60 5.68

20.10 22.10 12.60

Connecting time (€/hour) Trip purpose is other (€/hour) Trip purpose is work (€/hour)

17.39 24.46

16.94 24.37

14.61 20.33

11.96 16.88

18.19 20.36

20.12 25.42

14.93 19.47

16.52 19.40

acceptable confidence level (higher than 80%), for the first five models. In contrast, for ML2 and ML3 the null hypothesis could only be rejected at very low confidence levels (see Table 4). In all models, the estimated ASC for the Air–Air option had a positive sign. Aside from the competitive advantage of this alternative regarding full fare and baggage integration, this result also suggests the existence of inconveniencies associated with shifting transport modes even with an adequate HST station integrated in the airport. It can be concluded that those preconceived ideas about the comparative advantages of the HSR, in terms of comfort, safety and reliability are not enough to compensate travellers for the necessity of modal interchange. Finally, concerning the overall fit, the model with the highest log-likelihood is ML2, but if we take into consideration the consistency of the marginal utilities, ML3 provides a better performance regarding its adjustment to economic theoretical principles. 4.1. Willingness-to-pay for modal integration Willingness-to-pay (WTP) for the attributes that define the modal integration can be obtained from the estimation of discrete choice models as the ratio between the marginal utility of the attribute and the marginal utility of travel cost, which in turn is defined as minus the marginal utility of income (Jara-Díaz, 2000). In the case of the qualitative variables, the WTP can also be obtained as the ratio between the increment in utility produced by an improvement in the attribute and the marginal utility of income. When a fixed-parameters specification for the utility is used, a point estimate for the WTP, represented by a fixed value, is obtained. By contrast, when random parameters are considered, the corresponding WTP measure is also a random variable, and simulation techniques are usually needed in order to know their probability density function. WTP values corresponding to MNL4 and all ML models are presented in Table 5. For ML2 and ML3, the mean and the median of the simulated WTP distribution are give. The highest WTP is found for saving in-vehicle travel time, with figures ranging from 27.09 to 33.60 €/hour when the trip purpose is work and between 19.70 and 26.71 €/hour when the trip purpose is other, depending on the model. In general, we observe that the value of in-vehicle travel time is higher than the value of connecting time, and the latter is higher than the value of access time to destination. With regard to qualitative attributes, individuals also exhibit a significant WTP for fare integration. It is interesting to note that, for certain models, the two levels of this variable are perceived as similar. It is evident that the nature of fare integration is non-linear, and our results indicate that travellers do not prefer ‘‘one-single’’ ticket to two tickets with some kind of compensation or insurance in the case of delays or disruption of the service. Thus, the operators planning this alternative have adequate opportunity to foresee this characteristic for the future multimodal product, taking into account that this is an important feature which is highly valued by travellers (€16–€25). In contrast, the individuals in our sample did not consider it worthwhile to have the availability of an integrated service of baggage handling. In fact, and in order to be realistic, the highest level of this variable was defined according to the future investment plans at Madrid–Barajas Airport, where a full baggage integration system was not projected in the near future.7 Fig. 3 depicts the probability density function of the WTP measures for models ML2 and ML3 after considering 10,000 random draws of the Normal and log-Normal distributions for the corresponding parameters. Looking at these distributions for the model ML2 we observe that, in the case of access time and baggage integration, there is a significant probability of 7 This result is consistent with Chiambaretto and Decker (2012), who show that passengers at the Paris-CDG train station reject baggage integration. They conclude that this can be interpreted in different forms, but the most plausible one is that baggage integration definitively supposes a benefit-risk trade-off (Shaw, 2011). The authors concluded that passengers preferred to carry the luggage and check-in at the airport instead of on the premises created at the HST station, as they would probably perceive that the risk of losing the luggage would be higher.

C. Román, J.C. Martín / Transportation Research Part E 71 (2014) 129–141

139

Fig. 3. Distribution of the WTP figures. ML2 and ML3.

obtaining a negative WTP. However, this problem is not present in model ML3 when log-Normal distributions are considered, which makes this model much more suitable for policy analysis. The policy analysis should be based on establishing some evidence of the effectiveness of this intermodal integration, and analysing the potential attractiveness of these new intermodal transport products. The evidence shown in this paper could

140

C. Román, J.C. Martín / Transportation Research Part E 71 (2014) 129–141

be used in a cost-benefit analysis that could determine changes in social benefits. The cost-benefit analysis components are usually monetised and expressed in present value terms by discounting costs and benefits according to some rule during the time span of the project. It can be seen that the main monetised benefits of the modal integration accrue as a result of some component of the generalised costs (time) savings to passengers with an emphasis on in-vehicle travel time and connecting time. These last benefits should be promoted with an adequate level of coordination between all the transport operators involved in the development of these products. Those operators will have the benefit of increased revenue if they develop an adequate level of fare integration. In this respect, the evidence found is not conclusive regarding what type of integration (full or with insurance) should be promoted. However, when the analysis is done at the level of baggage integration, the evidence shows that not all the passengers are willing to pay for this service, and the figures obtained for those who will pay do not guarantee that the additional costs of providing these facilities will be covered. It is beyond the scope of this paper to analyse whether integration promotes mode shift from relatively environmentally-unfriendly modes (such as air transport) to more environmentally-friendly modes (such as HSR), and/or other effects of this integration. 5. Conclusions Intermodality between HSR and air transport is gaining momentum worldwide, as a way to rationalise more efficiently both transport modes. Thus, a better knowledge of the importance of the key drivers is necessary in order to promote this new alternative. In this paper, the analysis is based on an (SC) experiment which confronts travellers with the choice between the current Air–Air alternative and the Air–HSR alternative. Our results show that there is certainly some disutility associated not only with the connection in Madrid Barajas airport, but also with the change of transport modes. For this reason, it is highly relevant to have good estimates for the different service attributes that help policy makers and transport managers to develop attractive intermodal alternatives. Preferably, this compensation has to be provided in terms of in-vehicle, connecting, and access time, rather than baggage integration. Fare integration is also very valued. Our results show that passengers value in a similar way the complete fare integration or the soft-integration whereby an insurance company could reduce the uncertainty of the monetary losses for potential delays. We also find a different pattern regarding the preferences between mandatory and leisure trips. In this respect, baggage integration is only valued for leisure travel, and when at least one piece of luggage is checked-in. The disutility associated with travel time is lower for leisure than for mandatory trips. The new intermodal alternative is positively valued in terms of both punctuality and safety associated with HSR. Our results could be used in a cost-benefit analysis that contemplated not only the potential benefits but also the costs of developing such a type of infrastructure. It is very important to bear in mind that investments in some sophisticated facilities that facilitate the baggage integration between these two transport modes might not be socially beneficial as passengers’ WTP does not permit firms to recover this type of expenditure. The costs for the extension of the baggage handling systems at HSR stations and the terminal at the airport are not the only obstacles for this feature, as the wagons of HST are not fitted and prepared to handle any luggage. References Bierlaire, M., 2009. An introduction to BIOGEME Version 1.8, biogeme.epfl.ch. Bliemer, M.C., Rose, J.M., 2011. Experimental design influences on stated choice outputs: an empirical study in air travel choice. Transport. Res. Part A Policy Pract. 45 (1), 63–79. Bradley, M., 1988. Realism and adaptation in designing hypothetical travel choice concepts. J. Transport Econ. Policy XXII, 121–137. Brueckner, J.K., 2001. The economics of international codesharing: an analysis of airline alliances. Int. J. Ind. Organ. 19 (10), 1475–1498. Burgess, L., Knox, S.A., Street, D.J., Norman, R., 2014. Comparing designs constructed with and without priors for choice experiments: a case study. J. Stat. Theory Pract.. http://dx.doi.org/10.1080/15598608.2014.905223 (accepted for publication). Chiambaretto, P., Decker, C., 2012. Air–rail intermodal agreements: balancing the competition and environmental effects. J. Air Transport Manag. 23, 36–40. ChoiceMetrics, 2009, Ngene 1.0. User manual & reference guide. The cutting edge in experimental design. www.choice-metrics.com. Commission of the European Communities, 2001. White paper-european transport policy for 2010: time to decide. Brussels: COM(2001) 370. Commission of the European Communities. D.G. VII, 1995. Optimising Rail/Air Intermodality in Europe. Colin Buchanan and Partners, London. de Rus, G., Román, C., 2006. Economic analysis of the Madrid–Barcelona high speed rail line. Rev. Econ. Apl. 14 (42), 35–79. European commission, 2004. Towards Passenger Intermodality in the EU: Analysis of the Key Issues of Passenger Intermodality. European Commission DG Energy and Transport, Dortmund. European Commission, 2011. White paper. roadmap to a single european transport area – Towards a competitive and resource efficient transport system. Brussels: COM(2011) 144 final. European Commission: Directorate General of Transport, 1998. COST318 Interaction between high speed rail and air passenger transport. Gayle, P., 2008. An empirical analysis of the competitive effects of the Delta/Continental/Northwest code-share alliance. J. Law Econ. 51 (4), 744–766. Givoni, M., 2005. Aircraft and High Speed Train Substitution: the Case for Airline and Railway Integration Unpublished (Ph.D. thesis). University College, London. Givoni, M., 2006. Development and impact of the modern high-speed train: a review. Transport Rev. 26 (5), 593–611. Givoni, M., 2007. Environmental benefits from mode substitution: comparison of the environmental impact from aircraft and high-speed train operations. Int. J. Sustainable Transport. 1 (4), 209–230. Givoni, M., Banister, D., 2006. Airline and railway integration. Transport Policy 13, 386–397. Givoni, M., Banister, D., 2007. Role of the railways in the future of air transport. Transport. Plann. Technol. 30 (1), 95–112. Givoni, M., Rietveld, P., 2008. Rail infrastructure at major European hub airports: the role of institutional settings. In: Priemus, H., van Wee, B., Flyvbjerg, B. (Eds.), Decision-Making on Mega-Projects: Cost-benefit Analysis, Planning and Innovation. Edward Elgar, Cheltenham, Glos, pp. 281–303. González-Savignat, M., 2004. Competition in air transport: the case of the high speed train. J. Transport Econ. Policy 38, 77–107. Grimme, W., 2007. Experiences with advanced air–rail passenger intermodality – the case of Germany. Air Transport Research Society World Conference, Berkeley.

C. Román, J.C. Martín / Transportation Research Part E 71 (2014) 129–141

141

Hsu, S.C., 2010. Determinants of passenger transfer waiting time at multi-modal connecting stations. Transport. Res. 46E (3), 404–413. IARO, 1998. Air rail links – Guide to best practice. International Air-Rail Organisation, Air Transport Action Group. Airports Council International, Brussels. IATA, 2003. Air/Rail Intermodality Study. Final Report. Geneva, International Air Transport Association. Janic, M., 2011. Assessing some social and environmental effects of transforming an airport into a real multimodal transport node. Transport. Res. Part D Transport Environ. 16 (2), 137–149. Jara-Díaz, S.R., 2000. Allocation and valuation of travel time savings. In: Hensher, D., Button, K. (Eds.), Handbooks in Transport, Vol. 1: Transport Modelling. Pergamon Press, Oxford, pp. 303–318. Jiang, C., Zhang, A., 2014. Effects of high-speed rail and airline cooperation under hub airport capacity constraint. Transport. Res. Part B Methodological 60, 33–49. McFadden, D., 1974. The measurement of urban travel demand. J. Public Econ. 3, 303–328. Nash, C. 2009. When to invest in high-speed rail links and networks? (No. 2009–16). OECD/ITF Joint Transport Research Centre Discussion Paper. Oum, T.H., Park, J., Zhang, A., 1996. Effects of airline code-sharing agreements on firm conduct and fares. J. Transport Econ. Policy 30 (2), 187–202. Román, C., Espino, R., Martín, J.C., 2010. Analyzing competition between the high speed train and alternative modes. the case of the Madrid-ZaragozaBarcelona corridor. J. Choice Model. 3 (1), 84–108. Román, C., Martín, J.C., 2010. Potential demand for new high speed rail services in high dense air transport corridors. Int. J. Sustainable Dev. Plann. 5 (2), 114–129. Rose, J.M., Bliemer, M.C.J., 2009. Constructing efficient stated choice experimental designs. Transport Rev. 29 (5), 587–617. Rose, J.M., Bliemer, M.C.J., Hensher, D.A., Collins, A.C., 2008. Designing efficient stated choice experiments involving respondent based reference alternatives. Transport. Res. 42B, 395–406. Shaw, S., 2011. Airline Marketing and Management. Ashgate, Farnham. Train, K., 2009. Discrete Choice Methods with Simulation, second edition. Cambridge University Press, Cambridge. Yang, H., Zhang, A., 2012. Effects of high-speed rail and air transport competition on prices, profits and welfare. Transport. Res. Part B Methodological 46 (10), 1322–1333. Zanin, M., Herranz, R., Ladousse, S., 2012. Environmental benefits of air–rail intermodality: the example of Madrid Barajas. Transport. Res. 48E (5), 1056– 1063.