Analyzing discrepancies between willingness to pay and willingness to accept for freight transport attributes

Transportation Research Part E 89 (2016) 151–164

Contents lists available at ScienceDirect

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

Analyzing discrepancies between willingness to pay and willingness to accept for freight transport attributes María Feo-Valero a, Ana Isabel Arencibia b, Concepción Román c,⇑ a

Instituto de Economía Internacional, University Jaume I de Castellón, Spain Institute of Tourism and Sustainable Economic Development, University of Las Palmas de Gran Canaria, Spain c Institute of Tourism and Sustainable Economic Development, University of Las Palmas GC, Facultad de Economía Empresa y Turismo, módulo D, Campus de Tafira, 35017 Las Palmas G.C., Spain b

a r t i c l e

i n f o

Article history: Received 4 June 2015 Received in revised form 18 January 2016 Accepted 4 March 2016 Available online 29 March 2016 Keywords: Freight transport Discrete choice experiments Stated preference Prospect theory Reference alternative Asymmetric preferences Willingness to pay Willingness to accept

a b s t r a c t In this paper we use discrete choice data to analyze asymmetries in the preference for freight transport attributes. A reference dependent utility specification allowed us to test for the existence of substantial asymmetries in perception of the transport cost. Hence, the re-estimation of our models in the WTP/WTA space helped us to quantify significant discrepancies between the WTP and WTA for the attributes included in the choice experiment, namely transit time, service frequency and delays in delivery time. Results are deemed essential to define alternative services to road capable to attract substantial volumes of freight. Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction There are more and more authors that question the validity of the neoclassical theories of consumer behaviour to analyze individuals’ preferences. In this regard, the incorporation of elements of the prospect theory to model situations that involve decision making has contributed to a better understanding of the aspects that govern individuals’ decisions. The prospect theory was developed by Kahneman and Tversky (1979) and establishes that when people have to make a decision under risk, gains and losses are valued in a different way. In choice situations under risk, they demonstrate the existence of loss aversion, implying that decision makers are more concerned about losses than gains. In other words, individuals prefer to avoid a loss rather than obtain a gain of the same magnitude. Kahneman and Tversky (1979) suggest that the expected utility theory is not a good descriptive model when individuals have to make decisions under risk. They support this statement by showing different choice situations in which preferences do not accomplish the postulates of the expected utility theory. In a more recent work, Tversky and Kahneman (2011) presented the reference dependent utility specification to explain consumer choices in riskless situations, demonstrating that decision making depends on a reference level or status quo, which affects preference formation. The idea behind this theory is also that losses have larger effect on preferences than

⇑ Corresponding author. Tel.: +34 928 45 17 96. E-mail addresses: [email protected] (M. Feo-Valero), [email protected] (A.I. Arencibia), [email protected] (C. Román). http://dx.doi.org/10.1016/j.tre.2016.03.004 1366-5545/Ó 2016 Elsevier Ltd. All rights reserved.

152

M. Feo-Valero et al. / Transportation Research Part E 89 (2016) 151–164

gains. In this way, they broaden their former approach to model choices under uncertainty or risk (Kahneman and Tversky, 1979). Therefore, the utility function considered is asymmetric and is based on three fundamental aspects: (i) reference dependence, (ii) loss aversion and (iii) diminishing sensitivity. Thus, Tversky and Kahneman (2011) consider that the significant discrepancy observed between the minimum amount a person is willing to accept for the loss of a good and the maximum amount he is willing to pay for getting it, is due to loss aversion, among other aspects. In fact, the existence of loss aversion in the perception of all the attributes is sufficient to ensure that the willingness to pay (WTP) is less than the willingness to accept (WTA) (Masiero and Maggi, 2010). Both WTP and WTA measures represent an important input in the social appraisal of different projects and policies. The analysis of transport mode choice has been traditionally based on the estimation of models that consider the specification of symmetrical preferences. In the particular case of freight transport we can find significant contributions in the works of de Jong (2013), Brooks et al. (2012), Samimi et al. (2011), Arunotayanun and Polak (2011), Puckett et al. (2011), Cascetta et al. (2009), Train and Wilson (2008), García-Menéndez et al. (2004), Bolis and Maggi (2003), Shinghal and Fowkes (2002), Abdelwahab (1998), and Jeffs and Hills (1990); among others. As a common characteristic, all models presented in these works give an identical treatment to the effect of increases and reductions in the level of service. These effects are represented by the marginal utility of each attribute and in the case of linear utility functions they correspond to the parameter accompanying this attribute. One of the most important implications of this approach is that the WTP to improve the quality of service is identical to the WTA compensation when quality is reduced. The paper by Hess et al. (2006) is the first reference found in the transport literature which incorporates an asymmetric specification for the utility function. In the freight transport context, the work by Kurri et al. (2007) considered a similar specification when obtaining estimates of the value of time in Finland. In a more recent work Masiero and Hensher (2010, 2011), Masiero and Maggi (2010) and Masiero and Rose (2013) use reference dependent utility specifications to test for different aspects of the prospect theory such as, loss aversion and diminishing sensitivity, comparing results with that obtained when the traditional symmetric model is used. A similar approach is used by Hess et al. (2008), De Borger and Fosgerau (2008) and Rose and Masiero (2010) in the field of passenger transport. In all these papers, choice experiments are created in order to allow for the specification of gains and losses as positive and negative variations with respect to the attribute values in the reference alternative; obtaining, in most cases, a better fit when the asymmetric specification is considered. In this regard, an important issue is how to define the attribute levels in the experiment as this influences responses of the interviewees. Also, a shift in the reference point may affect individuals’ preference formation, producing behavioural reactions to gains and losses. Thus, Masiero and Hensher (2010) found increases in loss aversion for cost and time attributes when negative changes in the reference alternative are produced. Li and Hensher (2012) propose the use of an attribute-specific extended Rank-Dependent Utility Theory framework that allows integrating risk attitudes into freight behaviour modelling. Indeed, traditional freight distribution models work under the implicit assumption of risk neutrality although freight transport decisions are made in a context of uncertainty – this is specially the case for travel time – and are therefore subject to risk. The application of their proposed modelling framework to a sample of road route decisions in Australia shows that freight transporters and shippers have risk-taking attitudes with regard to travel time. An important implication resulting from the asymmetric specification is the ease to derive estimates of the WTP and WTA figures that can be obtained from the ratio between the corresponding marginal utilities. This in turn, allows quantifying the gap existing between these two figures. As many authors have concluded (see e.g. Hess et al., 2008; de Borger and Fosgerau, 2008; Grutters et al., 2008; Masiero and Hensher, 2010, 2011; Masiero and Maggi, 2010; Masiero and Rose, 2013), the obvious differences observed between the WTP and WTA measures, show that the monetary value attached to losses (WTA) is higher than that attached to gains (WTP). In this sense, the assumption of symmetric preferences would lead to overestimate the WTP and to underestimate the WTA, with the corresponding implications for social appraisal. Horowitz and McConnell (2003) indicate that the discrepancy observed between WTP and WTA measures can have two possible interpretations. On one hand, they consider that this difference can be seen as a deficiency of survey methods such us contingent valuation (in contrast with Hanemann, 1991), adding that ‘‘a weak version of this interpretation is that willingness to pay questions measure preference but willingness to accept question do not”. On the other hand, they establish that the discrepancy may exist because individuals do not have neoclassical preferences, ‘‘a conclusion which presumes that the experiments do capture ‘‘true” preferences in both WTP and WTA responses”. In this paper we contribute to this body of literature by analyzing the existence of discrepancies between the measures of WTP and WTA for the most relevant attributes that define modal choice for freight transport. The analysis is focused on long distance corridors that involve the competition of road with more sustainable modes of transport, as recommended by the European transport policy. Our models use data obtained from a discrete choice experiment where decision makers were faced to the choice between road (the current mode) and an intermodal alternative which simulates the transport service by rail or a motorway of the sea service in the corridor linking the regions of Madrid with the Netherlands/Belgium/Northern France/West Germany. The specification of an asymmetric utility function with respect to the reference values provided by the current level of service perceived by freight forwarders helped us to test for the existence of substantial asymmetries in perception of the transport cost, but not in the rest of the attributes. This fact prevented us to directly obtain significant discrepancies between the WTP and the WTA, due to the lack of appropriate statistical tests. However, given that cost parameters for gains and looses appear in the

M. Feo-Valero et al. / Transportation Research Part E 89 (2016) 151–164

153

denominator of the WTP/WTA expressions, discrepancies between the WTP and WTA could exist, even when the impact of gains and losses in some attributes (other than transport cost) is perceived as symmetrical. Hence, the re-parametrization and estimation of our models in the WTP/WTA space will help us to quantify the discrepancies between the WTP and WTA for the attributes included in the choice experiment, namely transit time, service frequency and delays in delivery time. To our knowledge, none of the previous research incorporating elements of the prospect theory involves the analysis of the competition between road and intermodal transport at the international level. For this, we consider that results obtained in this research could provide valuable information for the identification of the most effective policy measures and the evaluation of transport projects within the framework of the European transport policy. Indeed, until now most of the reference values obtained in the area of freight transport correspond to a single measure of the subjective value of attributes. However, when differences between the value attached to improvements and deteriorations in the level of service are allowed in the model, we observe great discrepancies. More specifically, the results obtained suggest that traditional specifications have led to an overestimation of these measures when the intervention involves an improvement in the level of service, and to an underestimation when the level of service is reduced. An accurate estimation of these figures is deemed essential both, to carry out the correct cost–benefit analysis in order to allocate the limited resources for transport infrastructures in the most efficient way possible, and to define alternative transport services to road capable to attract substantial volumes of freight. Indeed, one of the main objectives of the European freight transport policy is to shift substantial amounts of cargo from road transport to rail and maritime intermodal alternatives, because, on the one hand, road infrastructures are saturated in many points of the European network, and expanding the road capacity is comparatively much more expensive than in other modes such as maritime transport; and on the other hand, road transport externalities are much higher than those generated by the short sea shipping or rail. Given the current configuration, the European transport market does not emit the appropriate signals; prices do not represent the real cost, since they neither incorporate social and environmental externalities nor the real cost of the use of the infrastructure. Transport-related externalities are numerous – pollution, accidents, congestion, noise, costs for nature and landscape and additional costs in urban areas (Maibach et al., 2008; Korzhenevych et al., 2014) – and certainly constitute a major market failure in the sector. While all transport modes generate negative externalities, the main beneficiary of the non-internalization of external costs is definitely road transport, which, according to available estimates so far (van Essen et al., 2011; Demir et al., 2015) is the mode that generates more externalities. Failure to internalize external costs alters competition and distorts incentives to use the most efficient ways of transport, not only from an economic point of view but also from a social and environmental perspective. Until now, the European transport policy has been focused in three main lines of action: (i) to try to correct these market failures, so that prices reflect all costs associated with the transport (under fair competition between and within modes); (ii) to promote alternative modes by increasing competitiveness; and, (iii) to develop actions on the European transport network in order to correct the bottlenecks. The European transport policy started to be particularly active in pursuing the aim of rebalancing the modal pattern from mid-90. But despite the efforts, the market is not responding as fast as it is required by the environmental objectives of the EU. A key element of this policy is to get the comparative tort between the road and alternative modes, produced by the failure to internalize the external costs, disappears. For this, two main courses of action are distinguished: firstly, road pricing policies that advocate that the road users bear the external costs generated because of its use, by increasing the price of the road; and secondly, policies such as the Ecobonus and Ferrobonus that subsidizes the price of alternative modes in amounts equivalent to the externalities not paid by the road. Therefore, while the first group of policies increases the price of the reference alternative, the second line of action reduces the price of the intermodal option. Thus, the question then arises if these two actions are equivalent from the demand side. Our findings suggest that they are not as the WTA for deterioration in the current level of service in terms of cost is higher than the WTP for an improvement. The decision to use one way or another is not neutral as the option to increase the price of the road could have different impact in terms of modal shift that further improve the price of the intermodal option in equivalent proportions. The rest of the paper is organized as follows. The second section describes the sample, the questionnaire used and the discrete choice experiment conducted. The main aspects of the methodology are summarized in section three. The fourth section presents the results of the estimation and the discussion thereof, and finally the fifth section concludes. 2. Data and choice experiment The population under study focuses on the producers/distributors of manufactured products that, in 2010, carried out unitized shipments in the corridor linking Madrid with the Netherlands, Belgium, Northern France and West Germany. Concerning the selection of the population, while it is true that freight transport decisions are normally not taken by a single decision-maker but shared by the different agents involved in the logistic chain – shipper/receiver, freight forwarder and transport providers – (see Hensher and Puckett, 2008 for an analysis of the interdependent nature of decision-making within supply chains), in the present application we decided to restrict our population to shippers due to budget constraints. In this period, the corridor selected accounted for 4.3% of the existing traffic between Spain and continental Europe. It is also important to highlight that this is one of the few corridors in Spain where there is real competition between the transport modes analyzed: road, rail and maritime.

154

M. Feo-Valero et al. / Transportation Research Part E 89 (2016) 151–164

The fieldwork was divided into two phases (see Table 1). During the first phase a face-to-face interview was carried out with the individual responsible of transport shipments in the corridor under analysis. The initial total sample consisted of 93 companies located in the Autonomous Region of Madrid. Companies were randomly selected from the directory of Spanish Exporting and Importing Companies developed by the Spanish High Council of Chambers of Commerce (http://aduanas.camaras.org). The questionnaire was structured in four parts which allowed us to collect information about: (i) the characteristics of the company and its logistics perspective; (ii) the characteristics of the reference shipment (cost, transit time, delays, etc.); (iii) the importance of the main attributes that define the transport service as well as the perceived level of quality; and (iv) the preferences of the decision maker by completing two orthogonal stated preferences (SP) games. SP experiments have become a very popular technique as they have the ability to imitate real market situations where, sometimes, it is difficult to observe decisions made by individuals. In our survey, the interviewee participated in two choice games comprised of nine choice tasks each. The first game raised the choice between the current option, and a cheaper hypothetical intermodal alternative but with a worse or similar level service in the rest of the attributes, so that there was a real trade-off for the decision maker. By contrast, the second set posed the choice between the current alternative and a hypothetical intermodal alternative more expensive but with better level service for the rest of attributes considered. It is noteworthy that although in both sets alternatives were unlabeled, the first game assumed, implicitly, the choice between road and rail (that represents the intermodal option road–rail–road); and the second one, the choice between road and a motorway of the sea service (road–maritime–road).1 The pooled information about the companies’ preferences, obtained from the two orthogonal SP experiments carried out during the first phase of the fieldwork, allowed us to estimate preliminary models in order to obtain the parameters priors required for the construction of a D-efficient experiment (see e.g. Bliemer and Rose, 2005) that was conducted during the second phase of fieldwork. The information obtained about the reference shipment enabled us to customize this experiment to each respondent experience. Thus, in a second stage, we created a specific efficient design for every individual by pivoting attributes levels around the reference alternative (Rose et al., 2008). This allowed us to gain realism in the creation of the different choice tasks, improving the quality of the experiment and reducing the hypothetical bias. Because 39 of the 93 companies initially included in the sample did not participate in the second phase of the fieldwork, the final sample consisted of 54 companies and 972 observations. A more detailed explanation about the construction of the experimental design as well as the data description can be consulted in Arencibia et al. (2015). Transport cost, transit time, delays in delivery times and service frequency were the attributes included in the two choice experiments. The levels of variation considered for these variables are presented in Tables 2 and 3. It is important to point out that the levels of frequency and delays were re-defined according to the current level of service provided by the interviewee during the first phase of the fieldwork in order to further increase the realism of the experiment and its variability. Given that variations in the level of delays were defined in absolute terms, the definition of fixed levels could have led to unrealistic scenarios. Moreover, the variation of both the levels of the current and intermodal alternatives allowed us to increase the range of variation. Tables 4–6 detail the levels considered for each of the possible current levels of delays and frequency respectively. A screenshot of the efficient SP experiment is provided in Fig. 1.

3. Methodology The derivation of choice probabilities for discrete choice models is based on the utility maximization decision rule. Assuming that individuals are rational and have perfect information (Manski, 1977), they choose the option that report them maximum utility within their set of available alternatives. As the analyst has imperfect information the utility function is expressed as the sum of a representative or observable component and a not observable random term that accounts for the effect of unobserved elements. Thus, following the interpretation introduced by McFadden (1974), the utility of the alternative i for the individual q is represented by the random variable U iq that is expressed as:

U iq ¼ V iq þ eiq

ð1Þ

where V iq is the observable utility that depends on a vector of measurable attributes, assuming linearity therein; and eiq is the random term that encompasses unobservable aspects and measurement errors. Depending on the assumptions made about the distribution of the random component we will obtain different choice models. Thus, when we assume that error terms eiq distribute iid Extreme Value Type I we obtain the popular Multinomial Logit (MNL) model. Models of the family of Mixed Logit (ML), accounting for random taste heterogeneity, substitution patterns and panel correlation effects, are obtained when considering more flexible error structures (Train, 2009). 1 Authors are aware that the definition of the attribute levels in the two games could condition the WTP/WTA evidence, but in our case it has prevailed the realism when creating the choice experiment. For this reason we decided to do two different games where the intermodal alternative tried to simulate the different options in the market, so that the road user consider the proposed scenarios as realistic options. The other alternative would have been to create a unique design where we would had been forced to impose several constraints on certain attribute levels, in order not to be shown together, and this, in our opinion, would have substantially reduced the efficiency of the design.

Sample Phase I  93 companies. 1 reference shipment per company  Randomly selected from the directory of Spanish Exporting and Importing companies (http://aduanas.camaras.org)

Phase II  54 companies. 58% of the initial sample

Type of interviewee  Face-to-face interview  20/30 min

 Web questionnaire

Interviewee

Questionnaire

Observations

 Person responsible for managing the transport of the reference shipment  Depending on the company this could be managerial staff (23), someone from the logistic department (37) or from the export department (33)

 Characteristics of the company  Characteristics of the reference shipment and its current transport alternative  Importance attached to transport attributes and perceived level of quality  Orthogonal stated choice experiment (2 games of 9 scenario each)

 93  18 = 1674

 Same person answering to the questionnaire in phase I

 Efficient stated choice experiment (2 games of 9 scenario each)

 54  18 = 972

M. Feo-Valero et al. / Transportation Research Part E 89 (2016) 151–164

Table 1 Structure of the fieldwork.

155

156

M. Feo-Valero et al. / Transportation Research Part E 89 (2016) 151–164

Table 2 Attributes and levels for game 1. Attributes – levels

Current alternative

Intermodal alternative

Door-to-door transport cost (Euros per shipment)

1 2 3

Current level

25% 15% 10%

Door-to-door transit time (Days)

1 2 3

Current level

+1 day +2 days +3 days

Delay (Days of delay with respect to the foreseen delivery date) Current level 61 day

1 2 3

0.5 day Current level +0.5 day

+0.5 day +1 day +1.5 days

Delay (Days of delay with respect to the foreseen delivery date) Current level >1 day

1 2 3

0.5 day Current level +0.5 day

+0.5 days +1 days +2 days

Service frequency (No. of weekly departures)

1 2 3

Current level

1 weekly departure 2 weekly departures 3 weekly departures

Table 3 Attributes and levels for game 2. Attributes

Current alternative

Intermodal alternative

Door-to-door transport cost (Euros per shipment)

1 2 3

Current level

+20% +10% +5%

Door-to-door transit time (Days)

1 2 3

Current level

Current level 0.5 day 1 day

Delay (Days of delay with respect to the foreseen delivery date) Current level 60.5 day

1 2

Current level +0.5 day

0.5 day Current level

Delay (Days of delay with respect to the foreseen delivery date) Current level = 1 day

1 2 3

Current level +0.5 day +1 day

1 day 0.5 day Current level

Delay (Days of delay with respect to the foreseen delivery date) Current level >1 day

1 2 3

0.5 day Current level +0.5 day

2 days 1.5 days 1 day

Service frequency (No. of weekly departures) Current level 62 dep/week

1 2 3

Current level

2 weekly departures 3 weekly departures 5 weekly departures (Mon to Fri)

Service frequency (No. of weekly departures) Current level = 3 dep/week

1 2 3

Current level

3 weekly departures 5 weekly departures (Mon to Fri) 7 weekly departures (Mon to Sun)

Service frequency (No. of weekly departures) Current level >3 dep/week

1 2 3

Current level

5 weekly departures (Mon to Fri) 6 weekly departures (Mon to Sat) 7 weekly departures (Mon to Sun)

Table 4 Levels of delays for game 1. Threshold delays

Current level

Current alternative

Intermodal alternative

Level 1

Level 2

Level 3

Level 1

Level 2

Level 3

61 day

0.5 1

0 0.5

0.5 1

1 1.5

1 1.5

1.5 2

2 2.5

>1 day

2 3 4

1.5 2.5 3.5

2 3 4

2.5 3.5 4.5

2.5 3.5 4.5

3 4 5

4 5 6

157

M. Feo-Valero et al. / Transportation Research Part E 89 (2016) 151–164 Table 5 Levels of delays for game 2. Threshold delays

60.5 days

Current level

0 0.5

Current alternative

Intermodal alternative

Level 1

Level 2

Level 3

Level 1

Level 2

Level 3

– –

0 0.5

0.5 1

– –

0 0

0 0.5

1 day

1

1

1.5

2

0

0.5

1

>1 day

2 3 4

1.5 2.5 3.5

2 3 4

2.5 3.5 4.5

0 1 2

0.5 1.5 2.5

1 2 3

Table 6 Levels of frequency for game 2. Threshold frequency

Current level

Current alternative

Intermodal alternative Level 1

Level 2

Level 3

62 dep/week

1 2

1 2

2 2

3 3

5 5

3 dep/week

3

3

3

5

7

>3 dep week

4 5 5 7

4 5 6 7

5 5 5 5

6 6 6 6

7 7 7 7

Fig. 1. Screenshot of the efficient SP experiment.

Traditionally the analysis of mode choice have been based on the estimation of discrete choice models where the utility function is expressed as a linear combination of the attributes xik and the unknown parameters hxk as:

Vi ¼

X hxk xik k

ð2Þ

158

M. Feo-Valero et al. / Transportation Research Part E 89 (2016) 151–164

where the sub-index q corresponding to the individual has been dropped out for convenience. Thus, hxk is interpreted as the marginal utility of the attribute xik (i.e. the impact on utility produced by a marginal change in the attribute), yielding the same figure to evaluate the impact of both, increases and decreases of xik on V i . This interpretation is also valid when we analyze changes in the utility in monetary terms, resulting in identical values for the WTP and WTA figures. When considering the asymmetric specification recently applied in Kurri et al. (2007), Hess et al. (2008) and Masiero and Hensher (2010), the expression of the systematic utility takes the following piece-linear form:

Vi ¼

X þ hxk xþik þ hxk xik

ð3Þ

k  being hþ xk the marginal effect on V i of an increased xik regarding the reference value xk for the attribute k; hxk the marginal

 effect in V i of a decreased xik with respect to xk ; xþ ik ¼ maxðxik  xk ; 0Þ and xik ¼ maxðxk  xik ; 0Þ. In the asymmetry model, parameters signs must be interpreted appropriately. Thus when xi is an undesirable attribute i < 0, e.g. cost), the expected signs for the corresponding parameters are (i.e. that with negative marginal utility @V @xi  þ h if x  x > 0 @V i  i x . hþ x < 0 and hx > 0, as @xi ¼ h if xi  x < 0 x i In contrast, for desirable attributes yi (i.e. those with positive marginal utility @V > 0, e.g. service frequency), the expected @yi  þ h if yi  y > 0 @V i  y signs for the corresponding parameters are hþ . y > 0 and hy < 0, as @yi ¼ h if yi  y < 0 y As noted by Masiero and Maggi (2010), the estimation of different parameters for gains and losses with respect to the reference values allows to test for asymmetries in the utility function and eventually to test for the presence of loss aversion.

3.1. WTP and WTA measures One of the most attractive advantages of using discrete choice models to analyze transport demand is their ability to obtain the WTP figures for improvements in the level of service in a very simple way. WTP measures quantify in monetary terms the effect of policies involving changes in the attributes; being the value of travel time savings one of most widely used in the evaluation of transport projects. Once model estimates are obtained, the WTP measures can be derived as the ratio between the marginal utility for the corresponding attribute and the marginal utility of the cost, which according to discrete choice theory is equal to minus the i =@xk marginal utility of income; that is to say @V . Under the symmetric utility framework, positive and negative variations with @V i =@C

respect a reference value have the same impact on the utility and, consequently, the WTP for improving the level of service and WTA a compensation for reducing it are constrained to be identical. When the asymmetric specification is considered, we are able to obtain a different valuation for gains and losses, and consequently the WTP and the WTA measures differ. In this regard, for undesirable attributes, these measures are defined as:

WTP ¼

hx hþc

and WTA ¼

hþx hc

ð4Þ

In contrast, for desirable attributes we obtain:

WTP ¼

hþy hþc

and WTA ¼

hy hc

ð5Þ

where a convenient negative sign has been added in the above expressions in order to obtain positive figures.2 4. Model results Estimation results corresponding to the specifications discussed above are presented in this section. Different asymmetric models are considered, comparing results with the symmetric specification. Also, some models are estimated in both, the preference and the WTP/WTA space. In all cases, maximum likelihood estimates for the unknown set of parameters were obtained with the software Biogeme 2.0 (Bierlaire, 2003). In order to know the distribution of our data in respect of gains and losses, Table 7 shows the number of choice scenarios in which the intermodal alternative is better, identical or worse for the attributes considered in the analysis, considering the two choice games and the pooled data. Thus, in our pooled data set, to estimate the impact of increases and reductions with respect to the reference value we counted on more than 300 observations for each attribute. During the specification search, several preliminary models were estimated in the preference space, considering both, symmetric and asymmetric MNL specifications. Estimation result corresponding to these models are presented in 2 Note that the subjective value for desirable attributes (such as service frequency), obtained as the ratio between the marginal utility of the attribute and the marginal utility of the travel cost yields a negative figure, which must be considered in absolute value when it is interpreted as the willingness to pay for improving or the willingness to accept a compensation for reducing the quality of service.

M. Feo-Valero et al. / Transportation Research Part E 89 (2016) 151–164

159

Table 7 Distribution of choice scenarios regarding gains and losses. Attribute

Number of choice scenarios Xint  Xcur > 0

Xint  Xcur = 0

Xint  Xcur < 0

Game 1 Cost Transit time Frequency Delay

0 486 93 486

0 0 93 0

486 0 300 0

Game 2 Cost Transit time Frequency Delay

486 0 348 0

0 162 138 157

0 324 0 329

Pooled data Cost Transit time Frequency Delay

486 486 441 486

0 162 231 157

486 324 300 329

Table A1 in Appendix A. The first model MNL1 corresponds to a MNL model considering the symmetric utility specification given by expression (2). This model proved to be our best model in a previous research (see Arencibia et al., 2015), and its results will be compared with those obtained in different asymmetric specifications. MNL2 and MNL3 correspond the asymmetric utility (3) for MNL models, with and without the specification of an alternative specific constant (ASC) in the current option, respectively. It is important to highlight the low significance obtained for the parameter corresponding to reductions in service frequency h F , that in the case of MNL2 was even estimated with the opposite sign to that we expected. According to prospect theory, the loss aversion occurs when the absolute value of the coefficient attached to a loss is greater than that attached to a gain. Looking at the magnitude of the parameters’ estimates, the loss aversion hypothesis is only held for the cost in MNL2, and for all attributes, except the frequency, in the model MNL3. This suggests that the effect of loss aversion could be confounded with the effect of the ASC in MNL2. Analogously, in models MNL4 and MNL5, we specify the reductions in service frequency interacting with the dummy Fm, being equal to one when the number of departures per week is lower than 5, and zero otherwise; this turned the variable to be significant and with the correct sign. In this regard, it is important to point out that different threshold values were tested obtaining a significant negative impact in frequency reduction when the current level of service frequency for the road is not perceived as very high (i.e. less than one departure per day during the weekdays).3 Therefore, for the rest of the models presented in this paper, we considered this specification for this variable. In these two models all parameters have been estimated with the expected sign and most of them are significant at the 95% confidence level. The only exceptions are increases in delay (significant at 93%) for MNL4 and increases in frequency (significant at 90%) and reductions in transit time (significant at 84%) for MNL5. According to the size of the parameters, in MNL4 the loss aversion is held for cost and frequency and in MNL5 is held for all attributes except for delay. Although this deserves a more in deep analysis testing for the statistical significance of the asymmetries, as we will see bellow, these results points out to the difficulties of rail to compete with road transport, as this mode always entails higher transit times and lower service frequency. Also when comparing MNL4 and MNL5 using the loglikelihood ratio test (LR = 3.256) the null hypothesis cannot be rejected at the 92.9% confidence level, suggesting that these two models very are similar, being the MNL5 preferred as it offers results more consistent results with the loss aversion postulates of the prospect theory. ML1 and ML2 in Table 8 correspond to ML models estimated in the preference space, with fixed parameters but accounting for the potential panel correlation existing in SP data. In these cases, the high significance obtained for the standard deviation of the error component r confirmed the existence of correlation among responses belonging to the same respondent. The rest of the parameters were estimated with the expected sign and most of them were significant at the 95% significance level. The magnitude of the parameters in ML2 supports the assumption of loss aversion for all the attributes. However, as these figures represent point estimates subject to some error, the statistical significance of the loss aversion hypothesis should be tested. In the last rows of Table 8, the t-ratio values for test under the null hypothesis bþ þ b 6 0 are reported.4 These results indicate that, in our data set, loss aversion is highly significant for the cost in most of models and for the rest of the attributes only in some cases. In particular, for delays and transit time the loss aversion hypothesis is not supported by any of our models. This should be interpreted as if the decision makers are perceiving the same impact for improvements and worsenings of the same magnitude for these attributes. For the frequency in ML2, the loss aversion is significant at the 91.5% confidence level, considering a one-tail test. 3 In our sample the average service frequency for road is 3.2 departures per week and the percentile P61 is 5, meaning that for near 40% of the sample the current level of service frequency is 5 or higher. Thus, we can infer that figures below 5 are considered as a not very good service. pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 4 t ¼ ðhþ þ h Þ= varðhþ Þ þ varðh Þ þ 2covðhþ ; h Þ.

160

M. Feo-Valero et al. / Transportation Research Part E 89 (2016) 151–164

Table 8 Estimation results. Attributes

Estimates (t-test)

ASC

hASC

Cost+

hC þ

Mean Stand. dev.

Cost

h C

Frequency+

hþ F

Frequency ⁄ Fm

h FFm

Delay+

hþ D

Delay

h D

Transit time+

hþ T

Transit time

h T

Sigma

r

WTA frequency

WTAF

WTA delay

WTAD

WTA transit time

WTAT

WTP frequency

WTPF

WTP delay

WTPD

WTP transit time

WTPT

q2

Adjusted q2 l⁄(0) l⁄(C) l⁄(h) Observations Number of draws (type)

Run time t-test for H0: t-test for H0: t-test for H0: t-test for H0:

hþ C hþ F hþ D hþ T

þ h C þ h F þ h D þ h T

60 6 0=WTAF  WTPF P 0 6 0=WTAD  WTPD P 0 6 0=WTAT  WTPT P 0

ML1

ML2

ML3

ML4

0.627 (2.1) 0.00976 (5.19) – – 0.00513 (5.16) 0.237 (2.71) 0.337 (1.46) 0.273 (1.84) 0.535 (3.2) 0.229 (2.16) 0.638 (2.54) 0.744 5.67 – – – – – – – – – – – – 0.189 0.174 673.739 655.833 546.437 972 2000 MLHS 1:56:36 1.93 0.38 0.92 1.30

– – 0.0115 (6.5) – – 0.00429 (4.93) 0.136 (1.87) 0.469 (2.11) 0.381 (2.77) 0.357 (2.49) 0.378 (4.76) 0.343 (1.65) 0.73 (5.59) – – – – – – – – – – – – 0.186 0.172 673.739 655.833 548.698 972 2000 MLHS 1:26:18 3.42 1.37 0.10 0.15

– – 0.0115 (6.5) – – 0.00429 (4.93) – – – – – – – – – – – – 0.73 (5.59) 109.11 (1.94) 88.72 (2.59) 88.12 (4.29) 11.81 (1.91) 31.12 (2.45) 29.93 (1.69) 0.186 0.172 673.739 655.833 548.698 972 2000 MLHS 8:27:49 3.42 1.70 1.36 1.99

– – 0.0232 (3.2) 0.0169 (1.85) 0.00411 (4.69) – – – – – – – – – – – – 0.775 (5.71) 111.54 (1.86) 101.97 (2.68) 94.9 (4.11) 8.38 (2.19) 19.32 (2.44) 20.87 (1.95) 0.191 0.176 673.739 655.833 545.075 972 2000 MLHS 27:16:40 2.58 1.72 1.94 2.70

As pointed out by Masiero and Maggi (2010), the existence of loss aversion is sufficient to guarantee the existence of discrepancies between the WTP and the WTA in the sense that it implies that the condition WTP < WTA is held. However, the existence of strong asymmetries in the cost coefficients, as happens in our case, could suggest the existence of discrepancies between the WTP and the WTA even if the rest of the attributes are perceived as symmetric, as these parameters go in the denominator of the WTP/WTA expressions. When this happens, the re-parameterization of the model to estimate parameters in the WTP/WTA space (Train and Weeks, 2005) could be a convenient solution to test for the statistical discrepancies between these two figures. Finally, regarding the overall goodness of fit, both the rho-square and the log-likelihood at convergence reported in Table 8, indicate that the asymmetric specifications in the preference space outperforms the symmetric one (Table A1) in all models. According to these two figures, ML models proved to be statistically superior. When comparing ML1 and ML2 using the log-likelihood ratio test (LR = 4.776) the null hypothesis is accepted at the 96.7% confidence level, suggesting that these two models are quite similar; thus the model ML2, without specifying the ASC, would be preferred as it provides more consistent results. Notwithstanding, as our data come from unlabeled experiments, the effect of the ASC has not a clear interpretation.

M. Feo-Valero et al. / Transportation Research Part E 89 (2016) 151–164

161

With the aim of analyzing the statistical significance of the potential discrepancies between the WTP and WTA, the model ML2 is re-parameterized in order to obtain direct estimates of the WTP and WTA measures. To estimate the model in the WTP/WTA space, we follow the recommendation of Train and Weeks (2005) and reparameterize Eq. (3) of the observable utility, obtaining in a non-linear-in-the-parameter function. Thus, the utility of alternative i is expressed in terms of the WTP and WTA measures through the following expression:

V i ¼ hþci cþi þ hci ci þ

 X  X þ hxk xþik þ hxk xik þ hþyr yþir þ hyr yir r

k

! ! X hþy hyr  þ þ  þ þ   r ðh Þx þ ðh Þx ðh Þy þ ðh Þy þ c c c c ik ir ir ik hc hc hþc hþc r k X X     WTAxik hc xþik þ WTPxik hþc xik  WTP yir hþc yþir þ WTAyir hc yir V i ¼ hþci cþi þ hci ci 

V i ¼ hþci cþi þ hci ci þ

X

hþxk

hxk

r

k

Thus, WTP and WTA together with

ð6Þ

hþ ci

and

h ci

are parameters to be estimated.

The results of the estimated models in the WTP/WTA space are shown in Table 8. The model ML3 is the counterpart to ML2, whereas in ML4 a random parameter following the Normal distribution is specified for cþ i . Although the selection of the distribution for the random parameters is a controversial issue, we think that the unconstrained normal has certain advantages with respect to other censored or truncated distributions: (i) it allows analyzing the probability of obtaining marginal utilities with the wrong sign. Any distribution imposing constraints on the sign of the coefficients can simply mask the existence of individuals that are not in the line with the microeconomic principles, and (ii) it does not shift the mean of the distribution (Fosgerau, 2006; Fosgerau and Bierlaire, 2007). During the modelling process, many different specifications were tested, being those presented in Table 8 the ones that provided the best and more consistent results. All parameter estimates resulted significant at the 95% or 90% (WTP for frequency and transit time in ML3 and WTA for frequency in ML4) confidence level. According to the magnitude of the parameters, in all models, point estimates for the WTA are notably higher than those obtained for the corresponding WTP for all the attributes included in the analysis. In order to test for the statistical significance of these discrepancies, an asymptotic t-test under the null hypothesis WTAxk  WTPxk P 0 was performed. Test results, presented at the bottom of Table 8, suggest the existence of discrepancies between the WTP and the WTA measures for most freight transport attributes analyzed. In particular, for model ML4, the one with the better fit, all discrepancies are statistically significant at confidence levels higher than 95% considering the unilateral test. Table 9 compares the WTP and WTA figures for the different models. Although results could vary among these models, in all cases significant discrepancies between these two figures can be observed, which contrast with values obtained for the symmetric model MNL1. In all the models, the higher discrepancy is obtained for the frequency, ranging the ratio WTA/WTP from 9.24 to 13.31. For the rest of the attributes, this ratio ranges from 2.85 to 5.28. We have omitted result from model ML1 where the lack of loss aversion for delay and transit time yields a WTP higher than the WTA. A possible explanation for some of these high discrepancies could be found in the high level of satisfaction perceived for the current transport service in this corridor. In fact, for the attributes considered in this analysis, the obtained satisfaction average scores are higher than 4 in a five-point Likert scale, where 1 represents ‘‘very low” and 5 ‘‘very good” (see Arencibia et al., 2015). This means, that decision makers are not willing to pay very much for improvements, as they are very satisfied with the current level of service, but they would claim high compensations in case of reductions, that may eventually put them below their admissible threshold values. Results obtained in this analysis are therefore consistent with postulates of prospect theory regarding loss aversion hypothesis, where losses are much more valued than gains. Thus, in freight transport, the reduction in cost that a company would accept in the presence of a worsening of one of the attributes that define the transport service is greater than the money they would be willing to pay for improving this attribute in the same magnitude. 5. Conclusions In this paper we use reference-dependent utility specifications to quantify discrepancies between the WTP and WTA for important service attributes that affect mode choice for freight transport, namely, transit time, service frequency and delays. Our case study is focused on routes that analyze modal competition at the international level, which represents one of the contributions of this research from an empirical perspective. In particular, our analysis points to a priority objective of the European transport policy: the diversion of freight traffic from road to more sustainable modes of transport, mainly rail and short-sea shipping. As a result, the information obtained about the discrepancies between the WTP and the WTA measures is essential to properly define alternative services to current road transport, which are able to attract demand. In this regard, the WTA will provide insights about the compensation that must be applied in terms of price reduction when the level of service is lower, whilst the WTP will give guidance about the maximum increase in price that must be set for a better service. Our data set is based on individual-specific efficient SCE created for each respondent by pivoting attribute levels around the reference alternative. This allowed us to gain realism in the hypothetical settings created by the choice experiment. In a

162

M. Feo-Valero et al. / Transportation Research Part E 89 (2016) 151–164

Table 9 Comparison of WTP and WTA measures. Attribute MNL1 WTP = WTA

Frequency (€/departure)

Delay (€/day)

Transit time (€/day)

15.79

62.68

51.76

ML2 WTA WTP Ratio WTA/WTP

109.17 11.81 9.24

88.75 31.12 2.85

88.11 29.92 2.94

ML3 WTA WTP Ratio WTA/WTP

109.11 11.81 9.24

88.72 31.12 2.85

88.12 29.93 2.94

ML4 WTA WTP Ratio WTA/WTP

111.54 8.38 13.31

101.970 19.33 5.28

94.90 20.87 4.55

first approach to the problem, our results demonstrate that asymmetries in the perception of freight transport attributes, different than transport costs, did not resulted statistically significant at a reasonable confidence level in the corridor under analysis. This means that positive and negative deviations with respect to the reference value produce identical impact on decision maker’s utility, in absolute terms. Even though for these models, we failed to statistically test for the loss aversion hypothesis, models exhibiting asymmetries provided better fit to our data set than those with symmetrical preferences. However, the significant asymmetries found in the perception of transport cost suggested the existence of discrepancies between the WTP and WTA measures, as the cost parameters are fundamental for the transformation of utility units into monetary terms. Therefore, the modelling strategy was changed towards the re-parameterization and estimation of the models in the WTP/WTA space, where the statistical significance of the potential discrepancies resulted simple to quantify. According to this modelling approach, the largest discrepancy was found for the service frequency, where the ratio WTA/ WTP reached the highest figure. The interpretation of this value indicates that while decision makers are not willing to pay very much for having an additional departure per week, the compensation claimed in case of suffering a reduction in the level of service is between 9 and 13 times higher. This result provides interesting information to analyze the competitiveness of the road with intermodal alternatives in this corridor. In this sense, those policies trying to favor the rail (where the level of service is in general worse) must be accompanied by strong reductions in the transport cost, if this alternative is not able to offer a level of service frequency equivalent to that provided by the road. In contrast, the WTP figures obtained for saving transit and delay times indicate that policies trying to promote the use of short sea shipping services (with better level of service, in general) should not be accompanied by strong increments in the transport costs. Our results offer a new evidence of the existence of cost-asymmetric response to deviations with respect to reference values, highlighting the importance of incorporating the elements of prospect theory to the analysis of consumers’ decisions. In this respect, it is possible to obtain a better understanding of the decision making process as well as to obtain more accurate measures of both, costs and benefits in transport projects and traffic forecasts. To accomplish this, the application of stated choice techniques capable to create choice scenarios tailored for each respondent specific context resulted greatly helpful. Thus, results such as those presented here would be of great interest for freight forwarders and policy makers in order to obtain a more accurate evaluation of transport projects and policies within the European context. In addition to the analysis presented here, there is scope for future research that could help us to better understand the decision making process. In this regard, the analysis of unobserved latent heterogeneity through the estimation of latent class models will help us to identify differentiated segments of consumers regarding modal choice. Also, the estimation of hybrid choice models incorporating the effect of latent variables related to the perceived service quality and the perceived importance of the main attributes that define the transport service would add interesting information to the findings obtained in this research.

Acknowledgements The authors acknowledge the financial support provided by the Project ‘‘Modelización de previsiones de tráfico de mercancías y posibilidades de transporte intermodal con Europa (PREVITRANS)”. Expediente P4/08. Ministerio de Fomento. Convocatoria de ayudas del Programa Nacional de Cooperación público-privada, Subprograma de proyectos relativos a transporte e infraestructuras, en el marco del Plan Nacional de I+D+i, 2008–2011. This work was also supported by the University Jaume I of Castellon under Grant 13I298 of the Research Promotion Plan 2013; and the Valencia Regional Government under Grant GV/2015/096 supporting R&D Projects of Emerging Research Groups 2015.

163

M. Feo-Valero et al. / Transportation Research Part E 89 (2016) 151–164

Appendix A See Table A1.

Table A1 Preliminary models. Attributes

Estimates (t-test)

ASC

hASC

Cost

hC

Cost+

hþ C

Cost

h C

Frequency

hF

Frequency+

hþ F

Frequency

h F

Frequency ⁄ Fm

h FFm

Delay

hD

Delay+

hþ D

Delay

h D

Transit time

hT

Transit time+

hþ T

Transit time

h T

Sigma

r

q2

Adjusted q2 l⁄(0) l⁄(C) l⁄(h) Observations t-test for H0: hþ C t-test for H0: hþ F t-test for H0: hþ D t-test for H0: hþ T

þ h C þ h F þ h D þ h T

60 60 60 60

MNL1

MNL2

MNL3

MNL4

MNL5

0.349 (3.95) 0.006 (8.86) – – – – 0.090 (2.04) – – – – – – 0.356 (5.79) – – – – 0.294 (4.12) – – – – – – 0.147 0.140 673.739 655.833 574.523 972 – – – –

0.635 (2.64) – – 0.009 (6.15) 0.004 (5.77) – – 0.232 (2.96) 0.044 (0.62) – – – – 0.246 (1.79) 0.536 (3.57) – – 0.206 (2.1) 0.631 (2.67) – – 0.161 0.148 673.739 655.833 546.963 972 3.07 2.28 1.12 1.46

– – – – 0.011 (7.15) 0.004 (5.49) – – 0.110 (1.75) 0.025 (0.37) – – – – 0.369 (2.91) 0.328 (2.58) – – 0.382 (5.32) 0.258 (1.36) – – 0.156 0.144 673.739 655.833 568.505 972 4.19 0.88 0.18 0.60

0.420 (1.8) – – 0.010 (6.32) 0.004 (5.63) – – 0.179 (2.36) – – 0.477 (2.34) – – 0.245 (1.79) 0.483 (3.22) – – 0.218 (2.23) 0.511 (2.18) – – 0.166 0.152 673.739 655.833 562.042 972 3.29 1.29 0.92 1.01

– – – – 0.011 (6.42) 0.004 (4.98) – – 0.140 (1.92) – – 0.482 (2.16) – – 0.378 (2.75) 0.355 (2.48) – – 0.375 (4.75) 0.344 (1.65) 0.727 (5.48) 0.163 0.151 673.739 655.833 563.670 972 4.27 2.23 0.03 0.40

References Abdelwahab, W.M., 1998. Elasticities of mode choice probabilities and market elasticities of demand: evidence from a simultaneous mode choice/shipmentsize freight transport model. Transp. Res. Part E: Logist. Transp. Rev. 34E (4), 257–266. Arencibia, A.I., Feo-Valero, M., García-Menéndez, L., Román, C., 2015. Modelling mode choice for freight transport using advanced choice experiments. Transp. Res. Part A: Policy Pract. 75, 252–267. Arunotayanun, K., Polak, J.W., 2011. Taste heterogeneity and market segmentation in freight shippers’ mode choice behaviour. Transp. Res. Part E: Logist. Transp. Rev. 47 (2), 138–148. Bierlaire, M., 2003. BIOGEME. A free package for the estimation of discrete choice models. In: Proceedings of the 3rd Swiss Transport Research Conference, Ascona, Switzerland. Bliemer, M.C.J., Rose, J.M., 2005. Efficiency and Sample Size Requirements for Stated Choice Studies. Report ITLS-WP-05-08. Institute of Transport and Logistics Studies, University of Sydney. Bolis, S., Maggi, R., 2003. Logistics strategy and transport service choices: an adaptive stated preference experiment. Growth Change 34 (4), 490–504. Brooks, M.R., Puckett, S.M., Hensher, D.A., Sammons, A., 2012. Understanding mode choice decisions: a study of Australian freight shippers. Marit. Econ. Logist. 14 (3), 274–299. Cascetta, E., Marzano, V., Papola, A., 2009. Schedule-based passenger and freight mode choice models for ex-urban trips. Oper. Res./Comput. Sci. Interfaces Ser. 46, 241–250. De Borger, B., Fosgerau, M., 2008. The trade-off between money and travel time: a test of the theory of reference-dependent preferences. J. Urban Econ. 64 (1), 101–115. De Jong, G., 2013. Mode Choice Models. Modelling Freight Transport, pp. 117–141.

164

M. Feo-Valero et al. / Transportation Research Part E 89 (2016) 151–164

Demir, E., Huang, Y., Scholts, S., Van Woensel, T., 2015. A selected review on the negative externalities of the freight transportation: modeling and pricing. Transp. Res. Part 77E, 95–114. Fosgerau, M., 2006. Investigating the distribution of the value of travel time savings. Transp. Res. 40B, 688–707. Fosgerau, M., Bierlaire, M., 2007. A practical test for choice of mixing distribution in discrete choice models. Transp. Res. 41B, 784–794. García-Menéndez, L., Martínez-Zarzoso, I., Pinero De Miguel, D., 2004. Determinants of mode choice between road and shipping for freight transport: evidence for four Spanish exporting sectors. J. Transp. Econ. Policy 38 (3), 447–466. Grutters, J.P.C., Kessels, A.G.H., Dirksen, C.D., Van Helvoort-Postulart, D., Anteunis, L.J.C., Joore, M.A., 2008. Willingness to accept versus willingness to pay in a discrete choice experiment. Value Health 11 (7), 1110–1119. Hanemann, W.M., 1991. Willingness to pay and willingness to accept: how much can they differ? Am. Econ. Rev. 81 (3), 635–647. Hensher, D.A., Puckett, S.M., 2008. Power, concession and agreement in freight distribution chains subject to distance-based user charges. Int. J. Logist.: Res. Appl. 11 (2), 81–100. Hess, S., Rose, J.M., Hensher, D.A., 2006. Asymmetric preference formation in willingness to pay estimates in discrete choice models. Institute of Transport and Logistics Studies Working Paper ITLS-WP-06-12. Available at (accessed October 2015). Hess, S., Rose, J.M., Hensher, D.A., 2008. Asymmetric preference formation in willingness to pay estimates in discrete choice models. Transp. Res. Part E: Logist. Transp. Rev. 44 (5), 847–863. Horowitz, J.K., McConnell, K.E., 2003. Willingness to accept, willingness to pay and the income effect. J. Econ. Behav. Organ. 51 (4), 537–545. Jeffs, V.P., Hills, P.J., 1990. Determinants of modal choice in freight transport – a case study. Transportation 17 (1), 29–47. Kahneman, D., Tversky, A., 1979. An analysis of decision under risk. Econometrica 47 (2), 263–292. Korzhenevych, A., Dehnen, N., Bröcker, J., Holtkamp, M., Meier, H., Gibson, G., Varma, A., Cox, V., 2014. Update of the Handbook on External Costs of Transport. Available at . Kurri, J., Sirkiä, A., Mikola, J., 2007. Value of time in freight transport in Finland. Transp. Res. Rec.: J. Transp. Res. Board 1725 (1), 26–30. Li, Z., Hensher, D., 2012. Accommodating risk attitudes in freight transport behavior research. Transp. Rev. 32 (2), 221–239. Maibach, M., Schreyer, C., Sutter, D., van Essen, H.P., Boon, B.H., Smokers, R., Schroten, A., Doll, C., Pawlowska, B., Bak, M., 2008. Handbook on Estimation of External Costs in the Transport Sector. Internalisation Measures and Policies for All External Cost of Transport (IMPACT). Version 1.1. Available at . Manski, C.F., 1977. The structure of random utility models. Theor. Decis. 8 (3), 229–254. Masiero, L., Hensher, D.A., 2010. Analyzing loss aversion and diminishing sensitivity in a freight transport stated choice experiment. Transp. Res. Part A: Policy Pract. 44 (5), 349–358. Masiero, L., Hensher, D.A., 2011. Shift of reference point and implications on behavioral reaction to gains and losses. Transportation 38 (2), 249–271. Masiero, L., Maggi, R., 2010. Accounting for WTP/WTA discrepancy in discrete choice models : discussion of policy implications based on two freight transport stated choice experiments. Working Papers. Masiero, L., Rose, J.M., 2013. The role of the reference alternative in the specification of asymmetric discrete choice models. Transp. Res. Part E: Logist. Transp. Rev. 53 (1), 83–92. McFadden, D., 1974. The measurement of urban travel demand. J. Publ. Econ. 3 (4), 303–328. Puckett, S.M., Hensher, D.A., Brooks, M.R., Trifts, V., 2011. Preferences for alternative short sea shipping opportunities. Transp. Res. Part E: Logist. Transp. Rev. 47 (2), 182–189. Rose, J.M., Bliemer, M.C.J., Hensher, D.A., Collins, A.T., 2008. Designing efficient stated choice experiments in the presence of reference alternatives. Transp. Res. Part B: Method. 42 (4), 395–406. Rose, J.M., Masiero, L., 2010. A comparison of the impacts of aspects of prospect theory on WTP/WTA estimated in preference and WTP/WTA space. Eur. J. Transp. Infrastruct. Res. 10 (4), 330–346. Samimi, A., Kawamura, K., Mohammadian, A., 2011. A behavioral analysis of freight mode choice decisions. Transp. Plan. Technol. 34 (8), 857–869. Shinghal, N., Fowkes, T., 2002. Freight mode choice and adaptive stated preferences. Transp. Res. Part E: Logist. Transp. Rev. 38 (5), 367–378. Train, K., 2009. Discrete Choice Methods with Simulation, second ed. Cambridge University Press, Cambridge. Train, K., Weeks, M., 2005. Discrete choice models in preference space and willingness-to-pay space. In: Scarpa, R., Alberini, A. (Eds.), Applications of Simulation Methods in Environmental and Resource Economics. Springer Publisher, Dordrecht. Train, K., Wilson, W.W., 2008. Estimation on stated-preference experiments constructed from revealed-preference choices. Transp. Res. Part B: Method. 42 (3), 191–203. Tversky, A., Kahneman, D., 2011. Loss aversion in riskless choice: a reference-dependent model. Q. J. Econ. 106 (4), 1039–1061. Van Essen, H., Schroten, A., Otten, M., Sutter, D., Schreyer, C., Zandonella, R., Maibach, M., Doll, C., 2011. External Costs of Transport in Europe. CE Delft, INFRAS and Franhofer, Delft, Netherlands.