Taste heterogeneity and latent preferences in the choice behaviour of freight transport operators

Transport Policy 30 (2013) 77–91

Contents lists available at ScienceDirect

Transport Policy journal homepage: www.elsevier.com/locate/tranpol

Taste heterogeneity and latent preferences in the choice behaviour of freight transport operators Angela S. Bergantino a,n, Michel Bierlaire b,1, Mario Catalano c,2, Marco Migliore c,3, Salvatore Amoroso c,4 a

Department of Economics, Management and Business Law, University of Bari, Via C. Rosalba 53, 70124 Bari, Italy École Polytechnique Fédérale de Lausanne (EPFL), School of Architecture, Civil and Environmental Engineering (ENAC), Transport and Mobility Laboratory (TRANSP-OR), CH-1015 Lausanne, Switzerland c Energy Department – Transport Section, University of Palermo, Viale delle Scienze, Ed. 9, 90128 Palermo, Italy b

art ic l e i nf o

a b s t r a c t

Available online 19 September 2013

In this paper we show that individual attitudes of road carriers and their latent preferences toward specific freight service attributes do play a role in determining their mode choices. Specifically, we contribute to the empirical literature on freight agents' mode choice by exploring the role of the “perceived importance” of the most relevant service dimensions in determining the attractiveness of two alternatives to “all-road” transport: logistics terminals and road–sea intermodal services. This is carried out through a revealed/stated preference experiment and a mixture of logit framework. Our results support the hypothesis that operators' attitudes towards time, punctuality and risk of loss/damage can significantly enhance the explanatory power of the choice model, thus providing useful information for policy-makers to improve the regional freight mobility system. The “all road” option is preferred by hauliers concerned with the risk of loss/damage but it is, instead, disregarded by those assigning great relevance to punctuality. We also found substantial heterogeneity among respondents: larger firms tend to assign a lower value to time but a higher importance to the risk of loss/damage, especially if shipments are not frequent. In addition, the relevance of service reliability is higher the reliability greater the load size. Finally, we find that the nature of the transported goods significantly influences the choices of operators: when consigning perishables, hauliers tend to prefer the flexibility of a road-related mode. Any policy aiming at fostering the growth of intermodal transport and logistics and to remove obstacles to implementing rationalisation policies in the field of freight transport should take account of these elements. & 2013 Elsevier Ltd. All rights reserved.

Keywords: Stated Preference Intermodal freight transport Attitudes Latent variable Heterogeneity Mixture Logit

1. Introduction The choice of transportation solutions for freight transport is a complex task that involves a great number of agents, with different tastes, perceptions and criteria for the selection of transport alternatives (Woxenius and Bärthel, 2008). Understanding the factors behind these choices is pivotal for a number of

n

Corresponding author. Tel.: þ 39 080 5049038; fax: þ 39 080 5049042. E-mail addresses: [email protected], [email protected] (A.S. Bergantino), michel.bierlaire@epfl.ch (M. Bierlaire), [email protected] (M. Catalano), [email protected] (M. Migliore), [email protected] (S. Amoroso). 1 Tel.: þ41 021 6932537; fax: þ41 021 6932921. 2 Tel.: þ39 091 2386 1909; fax: þ39 091 2384 4425. 3 Tel.: þ39 091 2384 2416; fax: þ 39 091 2384 4425. 4 Tel.: þ39 091 2384 2404; fax: þ 39 091 2384 4425. 0967-070X/$ - see front matter & 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.tranpol.2013.08.002

reasons. On the one hand, from a business perspective, the information can be used in setting-up and marketing new transport and logistic services. On the other hand, from a public policy perspective, the information can be used for the development and use of forecasting models (e.g. for infrastructure investments) or for designing international, national or regional transportation related-measures (e.g. promotion or support for certain transport modes). A typical example of this is the current focus on promoting intermodal transport as a means of reducing congestion and environmental impact from transport related emissions. The growth in freight transport demand registered over the last decades has led, in fact, in many areas, to saturated transportation systems (road networks, in particular), with the consequence of low efficiency and highly negative impacts on the environment. In order to obtain savings in energy consumption, economies of scale and a more sustainable development, many countries have increasingly considered intermodal freight transport. In particular, the Ro/Ro (roll on–roll off) road–sea transport mode

78

A.S. Bergantino et al. / Transport Policy 30 (2013) 77–91

(Stopford, 1999) has been gaining consensus, at least at political level5. This type of intermodal transport uses ships – ferries – designed to carry rolling-stock cargoes for the longest segment of the trip. These ships do not require cranes to load or unload cargo, but trailers are driven on and off the ship's decks6. Many EU initiatives (for instance, Marco Polo I and II) and country specific incentives grant growing support for this transport solution. Insular and less accessible regions should be particularly prone to benefit from the Ro/Ro alternative. However, in many cases, such as for instance Sicily, the largest Island in the Mediterranean and the largest among the Italian regions, transport policies have not been very effective in developing Ro/Ro services nor services functional to the rationalisation of freight mobility. One of the reasons is that it is difficult to design and to evaluate a policy aiming at achieving a modal shift without a sound understanding of users' preferences and, until now, to our knowledge, no specific study has been carried out in this area7. In this paper, we focus on a case study carried out in Sicily where, until now, the development of sea-road intermodal transport services is limited: road captures about 30% of non oil inbound and outbound cargoes albeit Sicily is an island and the land connections from Calabria, the neighbouring region, are qualitatively scarce (Tagliacarne-Unioncamere, 2012). The widespread use of “allroad” transport penalises both the regional economy and the local and national environment. Considering the above, the research aims at identifying the factors which mostly affect road freight agents' choice behaviour, focusing the analysis on the possibility of using road–sea combined transport and logistic terminals. While the first option would generate both internal and external cost savings through to the substitution of road transit with the maritime connection, the latter would reach this result through load factor optimisation. Such terminals are, in fact, break-bulk facilities which can promote a more efficient and sustainable organisation of transport, functional to the development of intermodal transport. They contribute to the optimisation of load factors through truck-to-truck transhipment and consolidation of freights (Daganzo, 1987). This is particularly useful for small scale operators, which cannot otherwise benefit of scale and scope economies and, thus, for Sicilian road haulage industry which is severely fragmented, as it is in the rest of Southern Italy, the so-called Mezzogiorno8. These terminals

5 Particularly useful is the review of the Institute for Global Maritime Studies (2008) in depicting how coastal shipping could reduce traffic congestion. 6 As it is well known, the Ro/Ro combined transport presents several advantages: it does not need complex handling and loading equipments; it imposes minimal investments and maintenance costs (since transport operations are carried out by sea); it reduces the number of trucks on the road network thereby mitigating the environmental impact of freight mobility; it makes the journey of truck drivers far less tiring and risky with respect to the “all road” alternatives; if the driver follows the cargo to the destination, the time spent on board is not counted for driving time limit, eliminating the need of a costly two driver-solution; finally, it allows to exploit economies of scale by assembling hundreds of trucks on the same vessel. 7 Freight operators’ preferences for maritime services in Italian mainland regions has been explored, comparing attitudes towards Ro/Ro short sea shipping, by Bergantino and Bolis (2005 and 2008) and, on more recent data, by Bergantino (2009). The outcomes seem to support the hypothesis that preferences are differently distributed according to firms’ location and territorial infrastructure endowment. It would be useful, thus, to investigate operators preferences in peripherical areas. The Ro/Ro alternative has been the focus of interest of a number of studies carried out in Mediterranean regions, see, among others, the interesting analysis of: García-Menéndez et al. (2004), Blayac (2007) and Feo et al. (2011). Particularly useful in the definition of maritime service alternatives to land transport is the work of Puckett et al. (2011) and Brooks et al. (2012) and the work of Train and Wilson (2008), which focuses on the alternative between rail and barge services in the USA. 8 The reader is referred to the work of Dallari and Brenda (2009) and of MIT (2011).

also improve demand-supply coordination and provide several value added services on-site, such as packaging, quality controls, info-mobility facilities, etc. These two solutions, thus, represent, for road operators, the opportunity of better organising their business so as to increase efficiency and, for society, a promising strategy for a more sustainable freight mobility system. The final outcome of the research are useful insights for the policy makers on the potential benefits of such transhipment centres and on the strategies for more competitive road–sea intermodal services in the freight industry. We use a database collected through a combined Revealed Preferences (RP) – Stated Preferences (SP) survey on ninety road freight firms located in Sicily. They have agreed to participate to the natural experiment for eliciting their preferences for transport attributes and for specific characteristics of the two options. Recent advances in discrete choice modelling, have promoted the treatment of attitudes and perceptions affecting decisionmaking to get a more realistic representation of the choice behaviour (see for instance Ben Akiva et al., 1999, 2002; Morikawa et al., 2002; Walker, 2001; Walker and Ben Akiva, 2002). Rating data, such as responses to attitudinal and perceptual survey questions, are used as indicators of important causal variables that are not directly observable. Following this literature, mostly focusing on passenger travel behaviour, we developed a discrete choice model with attitudinal variables for the freight market. So, we could test the impact of the perceived importance of some relevant transport attributes on road freight operators, when faced with intermodal and logistic services. In detail, we integrated a discrete choice model with latent variables extracted from the responses to survey questions on attitudes, to produce a framework in which the system of equations is estimated simultaneously (hybrid model). In order to allow for heterogeneous responses of carriers to travel time variations and correlations among alternatives, the latent factors were incorporated into a mixture of logit framework (Bolduc et al., 2005). Hence, in line with only a small subset of other studies in the area, our empirical work shows the applicability of the latent variable approach to real world freight transport modelling and allows for agents' heterogeneity. Furthermore, as few other freight transport studies, this research uses disaggregated data and stated preference techniques. In particular, the estimates show that attitudes towards time, punctuality and risk of loss/damage can significantly enhance the explanatory power of the choice model, thus providing useful information for Sicilian policy-makers to improve the regional freight mobility system. We also found that hauliers' different attitudes toward service attributes do play a role: smaller firms tend to assign a higher value to time, while the risk of loss/damage is more important for greater operators, especially if shipments are not frequent. In addition, relevance of service reliability is higher the greater the load size. When consigning perishables, hauliers tend to prefer the flexibility of a road-related mode, as they do when assigning greater importance to the risk of loss/damage; instead, the “all-road” option is less competitive for punctuality-orientated carriers. Any policy aiming at fostering the growth of intermodal transport and logistics should take account of these elements. Section 2 illustrates the scientific background of the research, followed by a description, in Section 3, of the mixed RP–SP survey carried out to produce the choice observations for model estimation. Section 4 details the specification and the estimation of the mixture of logit model used to simulate carrier behaviour. In Section 5 the preliminary analysis for incorporating latent attitudes into the choice model is reported, while Section 6 presents the estimated mixture of logit model integrated with latent factors. Section 7 highlights the implications of the estimation results for the regional policy and section 8 contains some concluding remarks.

A.S. Bergantino et al. / Transport Policy 30 (2013) 77–91

2. Scientific background From the 1980s onwards researchers increasingly employed random utility models to analyse freight agents' decisions on mode of transport. These are econometric tools for the estimation of the demand function in a context of discrete choices (Ben Akiva and Lerman, 1985; Domencich and McFadden, 1975). The modelling techniques have since developed rapidly, especially with reference to passenger travelling choices. In the latter years, a number of studies have underlined the importance of considering taste heterogeneity as an important aspect of mode choice behaviour also in the freight industry (Aruotayanum and Polak, 2011; Feo et al., 2011; Beuthe and Bouffioux, 2008; Bolis and Maggi, 2002; Gopinath, 1995; Kang-Soo, 2002; Maier and Bergman, 2002). Researchers have increasingly recognised that decision makers differ significantly from one another, and the treatment of differences in sensitivities (and hence choices) across individual decision makers is, currently, one of the main areas of interest in choice modelling. Different approaches have been explored in the related literature to accommodate heterogeneity in preferences across individuals, particularly with reference to passenger transport. The basic method is to partition the dataset into mutually exclusive segments, on the basis of socio-economic characteristics relating directly to the decision-makers (Ben Akiva and Lerman, 1985). Notwithstanding the relative simplicity of such a technique, it may not always be possible to “capture” all the taste heterogeneity only through the a priori knowledge of the analyst, since identifying all the relevant discriminators of behaviour is often very difficult. An alternative to exogenous segmentation consists in using random parameters in discrete and continuous mixtures of models (Greene and Hensher, 2003; Hensher and Greene, 2003; Hess et al., 2006). Also attitudes and perceptions may be relevant predictors for differences in choice behaviour. However, the measurability of these elements is rather complicated as they are “latent” and, thus, cannot be observed directly. They can be inferred from other variables or “indicators” such as responses to survey questions about attitudes, perceptions or decision making protocols (Daly et al., 2012; Migliore et al., 2012; Prato et al., 2005, 2012; Choo and Mokhtarian, 2004; Golob, 2001; Walker, 2001). To capture the impact of subjective factors over the decision process, during the last decade a new breed of “hybrid choice” models have been applied. In relation to passenger travel behaviour the existing literature shows that choice models with latent variables are superior to even highly flexible traditional models (Johansson et al., 2006; Yanez et al., 2010). In relation to freight transport, notwithstanding the availability of modelling tools, research taking into account taste heterogeneity among agents is, with a few exceptions, still limited. A pioneering study on the application of mixture models to the freight industry considering latent factors is the work of Gopinath (1995), who, using data from a mixed RP/SP survey on shippers of manufacturing companies in USA, developed mode choice models integrated with attitudinal factors (perceived importance of time and cost), to perform market segmentation. To link the attitude formation sub-model to the choice one, the author exploited both the mixture of logit framework, incorporating the latent attitudes into random taste parameters, and the latent class approach, employing the latent attitudes as a criterion to determine the class membership. He found that in general, the dimension of the company, in terms of employees is relevant and reduces the time sensitiveness, which, however, is also significantly reduced for those operators which have wider acceptable time windows. Furthermore, he found that shippers with higher acceptable delays are more cost sensitive. Those who have a larger turnover, in terms of tonnes moved, instead, are less cost sensitive. Gopinath (1995)

79

finds that the fits of the two latent attitude model compared to the one latent attitude model “improved quite considerably indicating that the observed variations in the importance ratings is better explained” (p. 270). He supports the inclusion of attitudinal elements in defining more refined segments of population, in order to introduce more targeted policy measures. Kang-Soo (2002) estimated two versions of the mixture of logit model, error component and random coefficient logit, with SP data for the mode choice in the freight flow across the Channel Tunnel, showing the superiority of both models over traditional logit and the relevance of taste variations. He too found that the segmenting on the basis of attitudinal elements could improve the explanatory power of the models. Arunotayanum and Polak (2011) dealt with shippers' mode choice behaviour and, through latent class analysis, showed that the conventional practice of using commodity type as the only segmenting variable is not adequate to account for taste heterogeneity. They used stated preference data collected in Indonesia through a hierarchical stated preference experiment. Following this approach, the respondent was presented with descriptions of service quality and flexibility (defined in terms of levels of various service attributes) and then asked to rate each construct based on a semantic scale form 0 (very bad) to 9 (excellent). Secondly, each sampled shipper was asked to select among three alternatives (small truck, large truck and rail) characterised by different timecost combinations, based on scales given to the above constructs and on several contextual variables. They found that, notwithstanding the segmentation by commodity, the accommodation of taste heterogeneity within commodity segments leads to significant improvements in model fit in all segments. It also affects the estimates of the mean effects of cost and time attributes and service attributes, leading to an increase in the estimated parameters. Feo et al. (2011) analysed the viability of a maritime logistics chain in the Motorway of the Sea of South-West Europe and carried out a detailed evaluation of the performance and potential effects of cost-oriented measures – such as eco-bonus and tariff's rebates – in support of traffic reallocation toward the maritime alternative. They conducted a stated preference survey presenting an improved intermodal maritime service in competition with the 100% road mode. Three generic variables were taken into account for both alternatives: transit time, transport cost and delivery time reliability. Furthermore, a frequency variable was also considered for the intermodal maritime option. A mixture of logit model with error components was developed employing the SP variables and additional attributes referring to characteristics of both the forwarder and the shipment. All these works have emphasized the importance of heterogeneity to fully explain freight agents' mode choice behaviour. They also have demonstrated that the patterns of heterogeneity can be systematically related to characteristics of the operators and attributes of the shipments. Based on these findings, we explore the role of taste heterogeneity in the road freight industry focusing directly on operators' attitudes and their underlying factors. To reach this objective, we adopt the generalised methodology for incorporating latent constructs into continuous mixtures of choice models, developed by Walker (2001) and Walker and Ben Akiva (2002). We show that, besides the traditional variables, also the perceived importance of transport and logistical attributes – time, punctuality and risk of loss/damage – influence the attractiveness for road carriers of efficient and sustainable transport modes, such as road–road transhipment and road–sea intermodality. As to heterogeneity, in agreement with the literature, our estimations highlight that attitudes depend on characteristics of both the firm and the shipment. Such outcomes enrich the knowledge on the factors driving hauliers' choice process and

80

A.S. Bergantino et al. / Transport Policy 30 (2013) 77–91

consequently widen the range of transportation planners' actions. In particular, they suggest, along with the estimated elasticities, that, policy measures to develop highly cost-saving and specialized logistic and intermodal services, could promote significant switch from all-road to more environmentally friendly modes.

3. The choice behaviour of road-based carriers in Sicily The input data for the choice model is obtained through RP/SP survey involving ninety Sicilian road freight operators, randomly selected among road-haulage companies, which operated on both short and long-distance segments, active in Sicily in 2008. The registered companies, operating in Sicily, with one or more vehicles, are about 8.419. The sample covers, thus, 1.1% of the total population. Considering, however, only those companies carrying out both short and long distance trips, the population is more than halved (MIT, 2011), and, thus, the sample represents about 3% of the population. Over 80% of the sample has less than twenty vehicles while the rest have more. The sample is skewed towards larger companies as, in the overall population, the incidence of companies with more than 20 vehicles is, for Sicily, about 4%. However, if we consider that, the operators carrying out their services on long haul routes halved in the first three categories but remain stable among the larger operators, the bias is largely reduced. Furthermore, as it is evident from Table 1, the dimensional distribution of road haulage operators is very similar between Sicily and the other Southern regions of the Mezzogiorno (Calabria, Basilicata, Apulia, Molise, Campania, Abruzzo), the outcome of the analysis might be extended, although with the due attention, to larger territories. In particular, we believe that the outcome can be considered valid, in general, for operators working from peripherical regions, such as the Mezzogiorno or other peripherical regions of the Mediterranean (Spain, Greece, ecc.). From Eurostat data, it appears that the in these countries, the structure and size of the operators is comparable with that of the Southern Italian companies (Eurostat, 2010). 3.1. The mixed RP/SP survey Participating road-hauliers were presented with a questionnaire consisting of five parts. In the first section, the respondent is asked to describe the main characteristics of the typical shipment: origin-destination pair, principal commodity type and size in tons, frequency and cargo unit. Furthermore, the respondent is requested to describe his choice set (road versus Ro/Ro road– sea) in terms of transport cost, travel time, percentage of delays and percentage of shipments with damages for goods. The second section of the questionnaire contains eight SP mode choice tasks (Louviere et al., 2000) based on three alternatives: road (Road), road with transhipment at a logistic terminal (LT)9 and Ro/Ro road–sea (Ro/Ro).We considered that decision-makers presented with several attributes might be unable to take all explanatory variables into account and so we built the experimental design only around time and cost. In the third part of the questionnaire, we also introduced questions about the respondent's cut-off for time increase in order to identify the threshold over which the attribute's level is unacceptable for the respondent, regardless of the other variables. Considering individuals' cut-offs can aid in data cleaning as it allows the identification of non-trading choice behaviours. Finally, following Arunotayanum and Polak (2011), we developed the fourth section of the survey asking the respondent to 9 Where loads bound for the same destination can be consolidated. Terminals with this function are on the agenda of the regional government.

Table 1 Distribution rocarriers by dimensional classes. Region

Number of vehicles r3

4–10

10–20

420

Total

Sicily

5,712 67.8%

1,900 22.6%

475 5.6%

332 3.9%

8,419

Mezzogiorno

24,019 65.6%

8,709 23.8%

2,655 7.2%

1401 3.8%

36,628

Italy

90,750 71.7%

23,960 18.9%

4,779 3.8%

4045 3.2%

126,534

rate, through a 1–5 semantic scale (1 ¼ not important, …, 5 ¼ very important), monetary cost, travel time, punctuality, risk of damage for goods and service frequency. In detail, to link the SP experiment with the sampled operators' current experience, the choice scenarios were constructed assuming road as the base option and setting two bi-level attributes for each of the two remaining alternatives: (1) the gap (percent reduction) in terms of monetary cost with respect to road, ΔCLT e ΔCRo/Ro; (2) the gap (percent rise) in terms of travel time with respect to road, ΔTLT e ΔTRo/Ro. The levels of these attributes are based on the outcomes of a nationwide survey about transhipment terminals and on the comparison between road and Ro/Ro road–sea in relation to a sample of o–d links (La Franca et al., 2006). To simplify the choice task, we applied the block experimental design technique. In particular, the full factorial design (16 choice exercises) was subdivided into two blocks of eight choice tasks, to be submitted to two different groups of respondents, by using as block variable the four-way interaction among ΔTLT, ΔTRo/Ro, ΔCLT and ΔCRo/Ro. Table 2 shows the two blocks of choice scenarios representing the full factorial design10. Finally, we collected information about the number and the characteristics of available vehicles and facilities for logistics. Overall the SP experiment yielded 720 choice observations, of which we have been able to use about 88% (632 observations). The remaining 12% contains bad quality data: inconsistent and nontrading choice behaviours. 3.2. Sample characteristics In general, the companies responding to the questionnaire are relatively small and perform few shipments per week: 80% of the respondents own less than twenty vehicles and cargo units, about half of which are very small operators with less than three vehicles and load unit. The remaining 20% operates with more than twenty trucks and cargo units. Most respondents declared to be highly specialised and to operate on a national scale, with origindestination services over 1000 km. About 2/3 of the sample, in fact, operates on distances above 1000 km (50% in the range of 1001–1500 km and 17% above 1500 km). Only about 1/3 of the sample operates below 1000 km distances. Table 3 reports descriptive statistics concerning the RP data collected from the sampled operators considered for model estimation (88% of all interviewed operators). 10 Each block contains an additional choice exercise, which is identical to one of the other eight exercises, so as to identify bad quality information due to scarce understanding of the questionnaire or to low motivation of the respondent. Furthermore, we varied the sequence of choice situations, so as to minimise the errors induced by either learning or fatigue effects.

A.S. Bergantino et al. / Transport Policy 30 (2013) 77–91

Table 2 The two blocks of the SP experiment.

Table 4a Ranking of attributes for large ad haulage operators.

Blocks Scenarios Attributes

ΔTLT (h)

Block I

Block II

Block variable ΔCLT (€)

ΔTRo/Ro (h)

1

þ 25% þ  20% þ þ20% þ

2 3 4 5 6 7 8

þ 25% þ 25% þ 25% þ15% þ15% þ15% þ15%

9

þ 25% þ  20% þ þ20% þ

10 11 12 13 14 15 16

þ 25% þ 25% þ 25% þ15% þ15% þ15% þ15%

þ þ þ    

þ þ þ    

 30%  20%  30%  20%  30%  20%  30%

 30%  20%  30%  20%  30%  20%  30%

81

ΔTLT ΔTRo/Ro ΔCLT ΔCRo/Ro

ΔCRo/Ro (€)

 30% þ þ

 þ20% þ  40%  þ þ þ 10%   40%  þ þ 10%   30% þ þ þ þ20% þ  40%  þ  þ20% þ  30% þ þ þ þ 10%   30% þ þ  þ 10%   40%  þ

Attributes

Monetary cost (%) Travel time (%) Punctuality (%) Risk of damage for goods (%) Service frequency (%)

Perceived level of importance 5 (very imp.)

4

3

2

1 (not imp.)

83.3 11.1 33.3 16.7 22.2

11.1 27.8 33.3 33.3 38.9

0.0 50.0 27.8 16.7 11.1

5.6 11.1 5.6 11.1 22.2

0.0 0.0 0.0 22.2 5.6

Table 4b Ranking of attributes for small operators.

 40%  

Attributes

 þ20% þ  30% þ  þ þ 10%   30% þ   þ 10%   40%   þ þ20% þ  30% þ   þ20% þ  40%   þ þ 10%   40%    þ 10%   30% þ 

Monetary cost (%) Travel time (%) Punctuality (%) Risk of damage for goods (%) Service frequency (%)

Perceived level of importance 5 (very imp.)

4

3

2

1 (not imp.)

83.3 9.7 29.2 16.7 13.9

11.1 48.6 38.9 15.3 19.4

5.6 37.5 22.2 15.3 37.5

0.0 2.8 4.2 31.9 20.8

0.0 1.4 5.6 20.8 8.3

þ : for an attribute, it refers to the high level.  : for an attribute, it refers to the low level. Table 4c Ranking of attributes for operators transporting perishables.

Table 3 Descriptive statistics. Name Number of vehicles and cargo units Number of shipments per week Typical shipment by Road: Time (h) Cost (€) O-D distance (km) Typical shipment by Ro/Ro road–sea: Time (h) Cost (€) O–D distance (km)

Mean 68 2

Min 1 0.25

Max 2970 21

22.2 1062.5 1022.5

10 500 600

96 2000 2380

31.7 1306.5 1364.7

10 220 670

120 5000 3260

Attributes

Monetary cost (%) Travel time (%) Punctuality Risk of damage for goods (%) Service frequency (%)

Perceived level of importance 5 (very imp.)

4

3

2

1 (not imp.)

76.9 19.2 34.6 11.5 19.2

15.4 26.9 34.6 26.9 30.8

7.7 46.2 26.9 23.1 15.4

0.0 7.7 3.8 15.4 34.6

0.0 0.0 0.0 23.1 0.0

Table 4d Ranking of attributes for operator transporting non-perishable goods.

The sampled firms carry out, on average, one/two day journeys. The Ro/Ro mode, as expected, is chosen for longer haul O–D links. The average tariffs are quite similar: road transport costs €1.04/km, while services through the Ro/Ro option €0.96/km; less than 8% difference in price. The Ro/Ro services are intended only for accompanied transport, as there are not yet dedicated services for unaccompanied traffic. The driver, thus, must travel with the truck and the greatest part of the savings from the use of Ro/Ro transport stem only from fuel and usage costs. Tables 4a–4d show the ratings assigned by the interviewed companies to a selection of mode characteristics. The data are segmented by carrier size and type of commodity. One can observe that operator size particularly affects the perception of time, risk of damage and service frequency. In more detail, travel time is considered important or very important (levels 4–5) by 58.3% of small carriers, while only 38.9% of large firms assign importance to this attribute11. Moreover, big hauliers are more sensitive to risk of damage and service frequency than small ones. In fact, only about 30% of these deem as important or very important the two above attributes, against 50–60% of large firms.

11 In this study, a carrier is considered large, if the number of vehicles and cargo units is greater than 20.

Attributes

Monetary cost (%) Travel time (%) Punctuality (%) Risk of damage for goods (%) Service frequency (%)

Perceived level of importance 5 (very imp.)

4

3

2

1 (not imp.)

85.9 6.3 28.1 18.8 14.1

9.4 51.6 39.1 15.6 20.3

3.1 37.5 21.9 12.5 39.1

1.6 3.1 4.7 32.8 15.6

0.0 1.6 6.3 20.3 10.9

In the end, the type of commodity influences particularly the sensitiveness to service frequency, which is defined as important by 50% of carriers transporting perishables and by 34.4% of hauliers dealing with other goods.

4. Specification and estimation In this section we specify the models used to analyse the data and report the results. We start the model development process by discussing the so called “base model” – a simple logit framework – and proceed to compare more articulated models, in order to define the best performing specification, which can be used to infer the role of observed and latent factors. The “base model” is used as a benchmark and compared with more complex specifications

82

A.S. Bergantino et al. / Transport Policy 30 (2013) 77–91

by applying likelihood ratio tests. In the next paragraphs of this section, in fact, we investigate the impact of various assumptions on the modelling strategy. We present the estimation results of two nested choice models, of an error component model and, finally, of a mixture of logit model integrated with a random term to capture the correlation between the road alternatives, along with the panel effect from the SP experiment. All models are coded and estimated in BIOGEME econometric software (Bierlaire (2003); Bierlaire and Fetiarison, 2009; Bierlaire, 2009) and are estimated using all the 632 SP choice observations.

O

Ro/Ro

Logistic Terminal

Road

4.1. Choice model specification and estimation

O The choice set contains three alternatives: (1) road, (2) road with transhipment at a logistic terminal (LT) and (3) Ro/Ro road–sea (Ro/Ro). The first specification, called the base model, includes in the systematic part of the utility function time (percent increase in hours), cost (percent reduction in Euros) and a dummy variable (Freight) assuming the value of 1 in the case of perishable goods and 0 otherwise. In particular, in order to build the SP exercises around the respondents' experience, time and cost are expressed as percent variations with respect to the current time and cost levels of the road mode (thus these attributes equal 0 for road). To explore the crossalternative correlation, two nested logit models are employed (Fig. 1), structured as in the base specification: one model assumes correlation between road and LT (in both cases, trucks are used to convey goods), while the other embodies correlation between logistic terminal and Ro/Ro (both are characterised by transhipment operations to reduce the overall cost of transport to destination). Further, in order to take into account the potential intrinsic correlation among the choices of the same decision-maker, a pure error component model is estimated which includes an individual specific error term with standard normal distribution (ELand) in the utility functions of the land alternatives of the base model. Finally, based on our a priori expectation of heterogeneous sensitivity to travel time12, we also test a mixture of logit model with the time marginal utility consisting of two elements: a standard normally distributed random part (ETime), which varies across individuals, and a deterministic component, which interacts with a dummy variable (FirmSize) measuring operator size (0 if no of vehiclesþ þno of cargo units r20, 1 otherwise). In addition, the mixture of logit model is integrated with an error component (ELand) to capture the correlation between the road alternatives, along with the panel effect from the SP experiment. Appendix A reports the specification tables of all the models presented in this section (Tables A.1–A.5). The estimation results of the different models are reported in Table 5. The coefficients of all models are characterised by expected signs and good statistical significance13. In particular, the alternative-specific values of the coefficient of attribute Freight implies that carriers tend to prefer, ceteris paribus, a road-related mode if they transport perishable goods. It might be considered as a confirmation of the high degree of flexibility associated with road-based transport. The travel time marginal utility is normally distributed, thus, βTime þ βTime-FirmSize  FirmSize represents the marginal utility mean, while sTime is the standard deviation. The estimated value of the 12 According to our data set, the valuation of travel time seems to be more heterogeneous with respect to the monetary cost perceived level of importance. The latter, in fact, shows a low degree of variability among respondents (Table 3). 13 For the mixture of logit models, we simulate maximum likelihood by drawing random realisations from the underlying error process. In particular, the coefficients reported in Tables 4a–d are obtained with 5000 random draws. Reestimating these models with 5000 and 10,000 random draws showed that the parameter estimates are stable (Walker, 2001).

Land

Land-Sea

Ro/Ro

μLand

Logistic Terminal

Road

O

No Transhipment

Transhipment

μTranshipment

Ro/Ro

Logistic Terminal

Road

Fig. 1. Comparison between the base model and the two nested logit models.

latter is statistically significant even at the 99% confidence level, which strongly confirms the hypothesis of taste heterogeneity in relation to time. From the estimated parameters, it is evident that as firm size increases, the sensitivity to travel time variations decreases, ranging on average from  0.243 for smaller firms to  0.1787 for larger firms. It can be explained considering that for small operators – for which few resources in terms of both vehicles and drivers are available – a rise in travel time may be disruptive of service, resulting in loosing the opportunity of new orders. Larger companies can manage time variations more easily adjusting the deployment of fleet and drivers. In all cases, as expected, the values are negative. The value of time (VOT), which was computed for the mixture of logit models through numerical integration14, does not differ a great deal across the estimated models: in all cases, it is about 29 €/h for a truck load. The outcome is in line with the findings of a seminal contribution by Bolis and Maggi (2002) on transalpine

14 If CondP is the choice probability conditional to random parameter density f(w), the value of time is computed as follows:

CondVOT ¼ ðδ CondP=δ TimeÞ=δ CondP=δ CostÞ Z VOT ¼ CondV OTf ðwÞdw:

ω with

A.S. Bergantino et al. / Transport Policy 30 (2013) 77–91

83

Table 5 Estimated discrete choice models (t ratios in parentheses). Coefficients

Models Logit

Nested LogitLand

Nested LogitTranshipment

Error component

Error component and random time parameter

βTime-FirmSize

 0.159 (  8.57) –

 0.153 (  8.33) –

 0.150 (  6.34) –

 0.205 (  8.71) –

sTime

–

–

–

–

βCost

 0.233 (  10.42) 1.280

 0.221 (  9.79) 1.190

 0.221 (  7.02) 1.230

 0.298 (  10.19) 1.580

 0.243 (  7.74) 0.0643 (1.56) 0.104 (5.42)  0.345 (  9.55) 2.120

(4.31) 0.959

(4.36) 0.980

(4.08) 0.874

(3.09) 1.260

(3.13) 1.420

(3.06) –

(3.49) –

(2.68) –

3.220 (4.77)  0.425 (  2.21)

3.200 (5.02)  0.293 (  1.65)

2.950 (3.64)  0.359 (  1.71)

(2.42) 1.510 (5.87) 4.280 (5.23)  0.391 (  1.32)

(2.43) 1.630 (5.47) 4.660 (4.94)  0.420 (  1.31)

μLand a

–

–

–

–

μTranshipmenta

–

1.500 (1.54)b –

–

–

Final log likelihood Rho-squared (no coefficients)c Adjusted Rho-squaredd Value of time (€/h)

 337.666 0.514 0.505 28.93

 336.256 0.516 0.506 29.35

1.100 (0.51)b  337.521 0.514 0.504 28.78

 318.451 0.541 0.531 29.17

 301.541 0.566 0.553 28.24

Attribute coefficients:

βTime

βRoad Freight βLT Freight sLand

βRoad βLT Inclusive value parameters:

a

Parameter determining the error term variance within the nest (Land/Transhipment). t-test against 1. 1Lðβ^ Þ=Lð0Þ, where Lðβ^ Þ is the log likelihood of the available sample and Lð0Þ is the log likelihood of a null coefficients model. d 1Lðβ^ ÞK=Lð0Þ, where K denotes the number of unknown parameters in the model. b c

freight transport corridor (VOT ¼24 Euros/hour for the road transport of a 27 ton load) and, more generally, with the ample review of European, US and Australasian studies carried out by Miao et al. (2011), who determine that, on average, the freight transport VOT is contained between 18 and 28 Euros per hour for a truck load15 (values are expressed in 2008 Euros). 4.2. Specification tests As stated previously, the model specifications reported above are tested through the likelihood ratio (LR) approach in order to verify performance. As the first step of specification testing, the logit model is compared against the nested logit formulations. Only correlation between road and LT is revealed. In fact, if the nested logit model assuming correlation between the land modes is considered (see Fig. 1 and Table 5), the LR statistic equals 2.82, that is greater than the critical value at the 90% confidence level for one degree of freedom (2.71), therefore, the null hypothesis of no correlation can be rejected. On the contrary, the nested logit model embodying correlation between LT and Ro/Ro (see Fig. 1 and Table 5) has a LR statistic of 0.29, that is far lower than the critical value at the 90% confidence level for one degree of freedom (2.71); hence, the null hypothesis (the logit form holds) cannot be rejected. Moreover, this outcome finds further confirmation with the transhipment nest scale factor, μTranshipment, which is not significantly different from 1 (t-test equal to 0.51). 15 For an overview of time value estimates see also Zamparini and Reggiani (2007).

The error component logit model, with a normally distributed term capturing the correlation within the Land set along with the panel effect from the SP experiment, shows an adjusted ρ2 of 0.531, higher than the one of the nested logit form, equal to 0.506, implying the superiority in fit of the former model. Also the assumption of the time marginal utility normality is tested, for the mixture of logit model, using the semi-nonparametric approach described in Fosgerau and Bierlaire (2007). This method tests a distribution postulated a priori for a certain parameter of a discrete choice model against an alternative one (expressed in terms of the former distribution), which is made as general as possible by approximating its density with Legendre polynomials. The test is carried out through the LR statistic, to check the significance of the Legendre polynomial terms. The mixture of logit model (normally distributed time coefficient) and its expanded versions incorporating Legendre polynomial or semi-nonparametric (SNP) terms are compared in Table 6. To estimate these models, we simulate maximum likelihood by drawing 2500 random realisations from the error process. The LR test of the extended models, including one16 or two17 SNP terms (δ1 or δ1 and δ2) against the base mixture of logit model imply that the H0 hypothesis that the time marginal utility is normally distributed cannot be rejected. Therefore, the mixture of

16 Since the second (restricted) is a special case of the first (unrestricted), where all coefficients of the polynomial approximation are set to 0, the LR test is  2 (  301.541þ 301.427)¼0.23, which is definitely lower than the χ2 critical value at the 95% confidence level for one degree of freedom (3.84). 17 In this case, the LR statistic equals  2 (  301.541þ 301.326) ¼ 0.43, which is far below the χ2 critical value at the 95% confidence level for two degrees of freedom (5.99).

84

A.S. Bergantino et al. / Transport Policy 30 (2013) 77–91

Table 6 Testing the normal distribution of time coefficient (t ratios in brackets). Coefficients

Mixture of Logit

Attribute coefficients: βTime  0.243 (  7.74) βTime-FirmSize 0.0643 (1.56) sTime 0.104 (5.42) βCost  0.345 (  9.55) 2.120 βRoad

Mixture of Logit with Mixture of Logit with 1 SNP term 2 SNP terms

 0.266 (  4.80) 0.0637 (1.53) 0.106 (5.06)  0.345 (  9.54) 2.130

 0.244 (  5.38) 0.0642 (1.56) 0.209 (4.25)  0.345 (  9.55) 2.150

(3.13) 1.420

(3.18) 1.400

(3.17) 1.490

(2.43) 1.630 (5.47) 4.660 (4.94)  0.420 (  1.31)

(2.40) 1.620 (5.47) 4.670 (4.94)  0.419 (  1.31)

(2.56) 1.650 (5.50) 4.640 (4.94)  0.438 (  1.36)

SNP terms: δ1

–

δ2

–

 0.115 (  0.50) –

Final log likelihood Rho-squared (no coefficients) Adjusted Rhosquared

 301.541

 301.427

0.00167 (0.01)  0.535 (  2.84)  301.326

0.566

0.566

0.566

0.553

0.551

0.550

Freight

βLT Freight sLand βRoad βLT

logit model in which the time parameter follows a normal distribution is the one adopted in the remainder of the paper.

5. Integrating the choice model with latent preferences In order to improve the model explanatory power, we incorporate attitudinal factors into the specification emerged as more performing in the previous section. In particular, we exploit the data from the questionnaire's fourth section, which reported the respondents' rating, on a 1–5 semantic scale (1¼ not important, …, 5¼ very important), for the following factors: monetary cost, travel time, punctuality, risk of loss/damage for goods and service frequency. 5.1. Preliminary analysis Preliminarily, model estimations are performed to identify the indicators that could impact significantly on the respondents' utility functions. We find that the choice behaviour of Sicilian carriers is affected by the perceived importance of time, punctuality and risk of loss/damage for goods, while the other elements do not play a significant role. The specification of the base logit model integrated with attitudinal indicators is reported in Appendix A (Table A.6). In the estimated model, the time marginal utility depends upon the indicator of the perceived importance of time (ImpTime) and the utility functions of LT and Ro/Ro contain the indicator of the perceived importance of punctuality (ImpPunc). Furthermore, the attitude towards risk of damage (ImpRisk) influences the utilities of the two road modes. The estimated parameters are reported, instead, in Table 7. Firstly, it emerges that the time marginal utility increases, if (as expected) the perceived importance of time rises. Secondly, an increase in the perceived importance of time

Table 7 Estimation of the base logit model integrated with attitudinal indicators. Coefficients

Estimate

Standard Error

t-test

p-value

βTime βTime-ImpTime βCost

 0.163  0.0398  0.239 1.45

0.0189 0.0113 0.0228 0.308

 8.59  3.53  10.46 4.71

0.00 0.00 0.00 0.00

βRoad Freight βLT Freight

0.907

0.316

2.87

0.00

βLT ImpPunc

0.591

0.169

3.49

0.00

βRo=Ro ImpPunc

0.322

0.128

2.52

0.01

βRoad ImpRisk

0.150

0.101

1.48

0.14

βLT ImpRisk

0.0714

βRoad βLT

3.20  0.436

0.103

0.69

0.49

0.685 0.196

4.67  2.22

0.00 0.03

Sample size ¼ 632 observations. Final log likelihood ¼-324.326. Rho-squared (no coefficients) ¼ 0.533. Adjusted Rho-squared ¼0.517.

Table 8 Socioeconomic variables influencing the attitudinal indicators (t ratios in parentheses). Indicators Socioec. Var. Firm size (1/0) Shipment size (tons)

N. shipments/ week

Intercept

ImpTime

 0.378 (  1.47)

–

–

3.676 (12.52)

ImpPunc

–

0.0340 (3.06)

–

2.890 (7.25)

ImpRisk

0.951 (2.00)

–

 0.109 (  1.95)

2.744 (5.07)

reliability curtails road competitiveness. This might be explained by the fact that – given the especially poor state of the road network connecting the Island to the continent, the road mode is considered more subject to traffic congestion effects. Thirdly, the greater the risk of damage indicator the lower the attractiveness of Ro/Ro. Ro/Ro, in fact, implies the inclusion of a third party in the transportation process, augmenting the risk of uncontrolled events. Moreover, if freight is endangered by an unexpected event, while travelling by sea, it is not possible to react immediately. This leads operators to consider more risky and less controllable the maritime alternative. The significance of the parameter, however, is not as high as for the other parameters. Finally, the estimated values of variable Freight confirm that carriers prefer, ceteris paribus, a road alternative to transport perishable goods. We study also the regression relationships between the aforesaid indicators and a set of socioeconomic attributes. Table 8 reports the relevant parameters. In line with the outcome of the choice model estimations, we confirm the fact that the importance of time increases as the firm size reduces (see results reported in Section 4.1). The regression output also shows that the perceived importance of time reliability depends on the shipment size, that is the weight of load in tons. This can be explained by the high fragmentation of Sicilian producers: generally, a high load factor can be ascribed to the consolidation of various shipments. Hence, delaying a single consignment can damage multiple customers, due to a “domino effect”. Moreover, we find out that bigger hauliers, usually operating long-term contracts, attach higher rating to the risk of loss/damage, especially if shipments are not frequent. In fact, if a damage/loss should occur and it were not possible to include the substitutive goods in another large

A.S. Bergantino et al. / Transport Policy 30 (2013) 77–91

Table 9 Correlation across attitudinal indicators. ImpTime ImpTime ImpPunc ImpRisk

1 0.0490 0.234

85

Table 10 The mixture of logit model incorporating the perceived importance of punctuality and risk of loss/damage. ImpPunc

1  0.0746

ImpRisk

1

consignment promptly, the carrier might be obliged to perform a costly urgent delivery not to lose a regular flow of cargo. As a final step of the preliminary analysis, we compute correlation coefficients across the indicators. These are low enough to exclude the possibility of shared latent factors underlying the set of indicators (Table 9). Therefore, every latent attitude is measured by only one indicator.

Coefficients

Estimate

Standard Error

t-test

p-value

βTime βTime-FirmSize sTime βCost βRoad Freight

 0.265 0.0693 0.09  0.364 2.52

0.0353 0.0499 0.0306 0.0412 0.9

 7.5 1.39 2.94  8.85 2.8

0.00 0.16 0.00 0.00 0.01

βLT Freight

1.17

0.493

2.38

0.02

0.804 3.9  0.646

0.666 1.07 0.372

1.21 3.66  1.74

0.23 0.00 0.08

sLand βRoad βLT

ImpPunc coefficients: βLT ImpPunc

1.17

0.493

2.38

0.02

Ro=Ro βImpPunc

0.721

0.427

1.69

0.09

5.2. The mixture of logit model with the perceived importance of punctuality and risk of damage

ImpPunc error standard deviations: 1.3 0.529 sLT ImpPunc

Based on the above findings and given that the random parameter logit model described in Section 4 performs better than the other models, we integrate this specification with the indicators measuring the perceived importance of punctuality and risk of damage in order to obtain a more structured model. We introduce the first indicator (ImpPunc) in the utility functions of LT and Ro/Ro alternatives and the second (ImpRisk) in the utilities of Road and LT. In accordance with the previous estimation (see Table 7), we expect positive sign of their coefficients. We also include, for each attitudinal coefficient, an error component with zero mean and specific standard deviation, to take into account the framework of structural and measurement equations that can be derived when every latent factor is manifested by only one indicator (the equations reported in Appendix B). The specification details of this final model are reported in Appendix A (Table A.7), while the estimation outcome is contained in Table 10. The integrated model parameters are characterised by reasonable signs and, in general, good statistical significance. In particular, the estimated values of the indicator coefficients confirm that, as the importance of time reliability rises, the “all-road” alternative becomes less attractive, because of the higher risk of delay, while high ratings for the risk of damage work against the Ro/Ro option, which is less flexible (even though the statistical significance is not high). As to the travel time marginal utility, since the normal function is characterised by an infinite range of variation, the normally distributed random part of time coefficient can be positive with probability equal to 0.002, which is not high enough to warrant concern for a lack of realism. Compared to the mixture of logit model without latent factors, the adjusted Rho-squared improves, passing from 0.553 to 0.562. Given the estimation robustness, the reader is referred to Section 4.1 for comments on the other estimated values. The signs, in fact, are all confirmed and the dimension of the coefficients show very slight variations. To test the specification robustness, we perform crossvalidation following the approach described in Robin et al. (2009). We implement three experiments, in which a changing subset of the data base (20% of the observations) is saved for validation. The model is re-estimated on the rest and applied to the subset itself to compare each unconditional choice probability assigned by the model to the chosen alternative with the hazard value 1/3 (3 is the number of alternatives) representing the prediction of a purely random model with equal probabilities. We calculate the proportion of observations below this threshold, considered as outliers, for every experiment and for the complete data set. Table 11 reports the outcome of the cross-validation

ImpRisk coefficients:

2.46

0.01

0.722

0.658

1.1

0.27

βRoad ImpRisk

0.306

0.299

1.02

0.31

βLT ImpRisk

0.138

0.2

0.69

0.49

Ro=Ro

sImpPunc

ImpRisk error standard deviations: 2.54 0.591 sRoad

4.3

0.00

sLT ImpRisk

0.54

0.59

ImpRisk

0.361

0.674

Sample size ¼ 632 observations. Final log likelihood ¼  287.124. Rho-squared (no coefficients) ¼0.586. Adjusted Rho-squared¼0.562. Value of time (€/h) ¼29.21

Table 11 Outcome of model cross-validation.

Proportion of outliers (%)

Full data set

Exp. 1

Exp. 2

Exp. 3

16.93

17.86

16.96

15.18

procedure. The proportion of outliers across the three experiments is consistent with that obtained with the full sample (16.93%), which demonstrates the robustness of the specification. In the following section, based on the modelling outcomes, we draw some guidelines for supporting Sicilian policy-makers to take informed decisions with respect to intermodal freight transport and logistics. In particular, we concentrate on the role of transhipment terminals in optimising hauliers' load factors through consolidation of freights, and on the issue of improving Ro/Ro service quality.

6. Implications for the regional policy in favour of intermodal freight transport and logistics 6.1. Estimated elasticities The sensibility of operators to time savings, together with the natural attention to cost savings, also revealed by the values of the elasticities18, which are 0.5 and 2, respectively, should be taken into consideration by policy makers when defining transport and investment policies in these contexts. Cost related considerations 18 The point elasticities are calculated as: ðδ ChoiceProbability=δ Time ðCostÞÞ U ðTime ðCostÞ=Choice ProbabilityÞ.

86

A.S. Bergantino et al. / Transport Policy 30 (2013) 77–91

are more relevant in determining carriers behaviour than time related ones: the effect of the former is four times greater than the effect of the latter. This would suggest that, in general, it would be more appropriate to concentrate transhipment services at few big terminals, where it would be possible to exploit fully economies of scale and scope, fostering load factor optimisation. Although a network of smaller geographically distributed centres on the territory would reduce the average distance to be covered by users, and, thus, the time required to reach them, it would not be able to generate comparable cost savings and it would remain much less attractive to potential users. Policy measures should, hence, concentrate in a few, large infrastructures, localised in strategic areas. Their presence, and the related effect on efficiency, would induce freight carriers to turn to the Ro–Ro alternative, in order to improve the deployment of their fleet. This solution seems reasonable in all those territorial contexts – such as the European-Mediterranean countries – where carriers have limited size and fleet deployment and costs are a major concern. The effect of time, however, it is not homogeneous. As it appears from the estimates, smaller operators have a higher time marginal utility. These firms, in fact, base their activity on a high turnover of a few trucks and drivers. Policy measures aiming at inducing, directly or indirectly, savings in travel time and, thus, fleet utilisation rate would be particularly beneficial to this portion of operators. 6.2. Simulations

The total cost at the terminal would be calculated by summing the value of the time spent for transhipment, 116.84€, to the cost relating to the handling operations which is calculated by multiplying per ton costs of transhipment (2€) by the average current load factor (12.5 tons) and by the average “consolidated” load factor (20 tons). The sum of this to the in-vehicle costs gives an approximation of the generalised cost of the rationalised truck load operation. The results show that logistic terminals can be competitive, in terms of generalised cost, for the outbound freight traffic involving Sicily. In fact, on distances less or equal to 300 km (about the maximum distance from the western outermost side of Sicily to the eastern one, or viceversa), the difference in the generalised cost between road and transhipment is slightly negative (Table 12). On the contrary, for longer connections, transhipment produces significant savings for road operators (Tables 13 and 14). Table 12 Comparison between road and LT for a shipment within Sicily (300 Km). Load factor Value of (tons/truck) Ttranshipmenta (€) All road Logistic terminal

12.5 17.5

– 116.84

Ctransportb (€)

Ctranshipmentc (€)

Ctotd (€)

473.70 338.36

– 25

473.70 480.20  6.50

road C terminal C all tot tot a

Among the possible policy recommendations for inducing a significant modal switch, from all-road to more sustainable transport, both via LT and Ro/Ro a preminent role should be that of the logistic terminals. These infrastructures could support small and medium-sized enterprises to overcome the problem of limited resources: freight carriers could establish cooperative contracts and operate jointly, within the logistic platform; sharing facilities and coordinating orders, so as to reduce their costs and provide better services. Public intervention could stimulate the promotion of automation of handling facilities and the implementation of intelligent transport systems, such as real-time vehicle-location to advise customers on the current position of their goods and updated arrival times, advanced in-vehicle technology to monitor truck performance and loads, travel time information to avoid congested road links and port terminals or accesses, etc. Sharing these facilities within the logistic terminal would generate economies of scale which might determine the feasibility of the investment and which would abate costs. Using the estimated VOT, equal to 29.21€, it is possible to compare, in terms of costs, the use of a logistic terminal with respect to all-road. From a recent investigation carried out by Catalano and Migliore (2005), it is possible to identify the average load factor of road freight transport within and outside the regional boundaries. This is equal to 12.5 tons per truck and 20 tons per truck, respectively. Furthermore, the study identifies that a large consolidation terminal (sized on 500,000 tons handling per annum) produces up to a 40% raise in truck load factor with a 4 h time-investment and average cost of 2 €/ton for the transhipment operations. Taking this value of transhipment cost along with the cost parameter for road transport (MIT, 2011), equal to 1.579 €/km19, and calculating the gains in terms of load factor potentially generated through consolidation, we can estimate the transport costs associated with the use of a logistic terminal and with the “all-road” option in relation to short, middle and long run shipments. The results of simulations are reported in Tables 12–14.

Value of Ttranshipment ¼VOT (29.21 €/h)  Time spent at the terminal (4 h). Ctransport ¼(carrier cost per km (1.579 €/km)  distance)  (Initial load factor/ Improved load factor). c Ctranshipment ¼ average transhipment cost (2 €/ton) Initial load factor. d Ctot ¼ Value of Ttranshipment þ Ctransport þ Ctranshipment. b

Table 13 Comparison between road and LT for a Sicily-Southern Italy shipment (600 Km). Load factor Value of (tons/truck) Ttranshipmenta (€) All road Logistic terminal

– 116.84

Ctranshipmentc (€)

Ctotd (€)

947.40 701.78

– 40

947.40 858.62 88.78

road C all C terminal tot tot a

Value of Ttranshipment ¼VOT (29.21 €/h)  Time spent at the terminal (4 h). Ctransport ¼(carrier cost per km (1.579 €/km)  distance)  (Initial load factor/ Improved load factor). c Ctranshipment ¼ average transhipment cost (2 €/ton)  Initial load factor. d Ctot ¼ Value of Ttranshipment þ Ctransport þ Ctranshipment. b

Table 14 Comparison between road and LT for a Sicily–Northern Italy shipment (1300 Km).

All road Logistic terminal road C terminal C all tot tot a

Load factor (tons/ truck)

Value of Ttranshipmenta (€)

Ctransportb (€)

Ctranshipmentc (€)

Ctotd (€)

20 27

– 116.84

2052.70 1520.52

– 40

2052.70 1677.36 375.34

Value of Ttranshipment ¼VOT (29.21 €/h)  Time spent at the terminal (4 h). Ctransport ¼(carrier cost per km (1.579 €/km)  distance)  (Initial load factor/ Improved load factor). c Ctranshipment ¼ average transhipment cost (2 €/ton)  Initial load factor. d Ctot ¼ Value of Ttranshipment þ Ctransport þ Ctranshipment. b

19 It includes the following costs: vehicle purchase, insurance, taxes, tyre and fuel consumption, maintenance and repair, toll, driver.

20 27

Ctransportb (€)

A.S. Bergantino et al. / Transport Policy 30 (2013) 77–91

To assess the potential market for logistic terminals, we provide, in Table 15, an origin-destination matrix reporting trade flows (in terms of tons) involving Sicily, travelling by road (National Institute of Statistics, data of 2010). About 83% of the total is internal traffic while 17% is directed towards external markets (about 4 million tons per annum), of which about 7% concerns very long haul (north Italian regions and foreign countries). About 12% of the internal traffic and 45% of the outbound movements are commodities for which load factor optimisation would be important: agricultural goods, food-stuffs, machinery, vehicles, leather wares, furniture and other manufactured items, electric and electronic products and clothes (National Institute of Statistics, data of 2010). These products would be the potential market for both logistic terminals and Ro/Ro services. They represent, in terms of weight, a total of 4.11 million tons. To give an idea of the sensitivity of mode choice behaviour to commodity type, we simulate the effect on choice probabilities of a change in the dummy variable Freight, relating to perishable commodities. Recall that this variable has positive and significant coefficients for the Road and LT alternatives: on average, operators tend to prefer a road-related mode when transporting perishables. Passing from one extreme – zero perishables – to the other – 100% perishables – causes a significant rise in average choice probability for the land modes, especially with respect to road (Table 16). Under the perishable freight scenario, the two road alternatives get a global choice share of nearly 35%, although the sampled carriers, which mostly operate on long-haul connections (see Section 3.2) could obtain significant cost-savings through the use of Ro/Ro transport (especially with regard to fuel consumption, maintenance and second driver). The strength of the characteristics of commodity in influencing the choice between road and maritime alternative is probably due to the scarce level of specialization of Ro/Ro services in Sicily. These are not dedicated to freight traffic (the ferries are mainly structured for passenger transport), but also are not organised to meet adequately the requirements of the logistic chain of perishables. Frequency and timing of ship departures are crucial to make the use of Ro/Ro combined transport possible especially for highly perishable commodities (e.g. fresh fruit and vegetables that represent an important driver of the Sicilian exports). These, often, have to be delivered within 24 h. A policy intervention which might overcome this limitation and expand the market for Ro/Ro services would lead to significant advantages for both carriers and

Table 15 Third party road freight transport involving Sicily in 2010. Within Sicily

Tons Shares (%)

From Siciliy to Southern Italy

19,162,602 2,320,439 82.6 10.0

From Siciliy to Central-Northern Italy

From Siciliy to Foreign Countries

1,592,192 6.9

114,316 0.5

Table 16 The effect of a change in dummy variable Freight on choice probabilities. Alternative

Road LT Ro/Ro

Scenario 1: Zero perishables

Scenario 2: 100% perishables

Average choice probability (%)

Average choice probability (%)

8.65 8.98 82.37

19.99 14.49 65.51

87

Table 17 The effect of a change in variable ImpPunc on choice probabilities. Alternative Scenario 1: Punctuality not important

Road LT Ro/Ro

Scenario 2: Punctuality very important

Average choice probability (%)

Average choice probability (%)

24.65 3.19 72.17

7.55 14.33 78.12

Table 18 The effect of a change in variable ImpRisk on choice probabilities. Alternative Scenario 1: Risk of damage not important

Road LT Ro/Ro

Scenario 2: Risk of damage very important

Average choice probability (%)

Average choice probability (%)

9.42 10.92 79.66

15.09 9.65 75.27

the environment. In particular, promoting the creation of pools of road carriers of perishable commodities might grant freight flows large enough to induce shipping companies to set up dedicated services, in line with hauliers' needs. Supporting innovative forms of contracting with shipping companies might provide further stimulus. 6.3. Impact of attitudinal indicators We also evaluate the impact on choice probabilities of changing the perceived importance indicators of time reliability and risk of damage. Recall that the first indicator (ImpPunc) is introduced in the utility functions of LT and Ro/Ro and the second (ImpRisk) in the utilities of Road and LT, with specific coefficients. The change in the punctuality attitudinal indicator from the “not important” level to the “very important” one, for all observations of the estimation dataset, remarkably lowers the average choice probability of all-road, mainly in favour of transhipment (Table 17). The operators for which punctuality, and thus reliability of service, is important, are those which would more easily switch transport mode in favour of intermodal transport, including Ro/Ro services. This implies that policy design should take into account the perceived role of the logistic terminal in rationalising services and providing infomobility facilities to support routing procedures and to improve time reliability. Changes in the risk of damage attitude, on the other hand, do not induce significant changes in mode choice. Risk of damage-averse carriers are not so reactive. As it is reported in Table 18, there is only a slight increase in favour of all road transport. Corrective measures should be introduced in order to lower the risk of damage when Ro/Ro services are adopted. Among these, visual control of the vehicles during navigation, ad hoc insurance products easy to adapt to the specific instances; transparency in damage record of the ships in order to reduce informational asymmetries.

7. Conclusions In this paper, we have analysed the factors determining the choice behaviour of Sicilian road carriers when faced with transhipment-related modal alternatives, so as to draw guidelines to develop intermodal freight transport and logistics. To achieve

88

A.S. Bergantino et al. / Transport Policy 30 (2013) 77–91

these objectives, a data set of 632 choice observations resulting from a mixed RP/SP survey was used. It registered the choice behaviour of Sicilian hauliers in hypothetical scenarios with respect to three alternatives: road transport, road transport with transhipment at a logistic terminal, Ro/Ro road–sea transport. In particular, to simplify the choice task within the SP experiment, choice tasks based on two explanatory variables, travel time and carrier monetary cost, were built. Moreover, the survey asked the respondent to describe the typical shipment, to provide information about the characteristics of the fleet and to declare the perceived importance of relevant mode attributes. Although the dataset is restricted to Sicilian operators, as we have highlighted in the paper, the structure of the road haulage industry in the rest of the Mezzogiorno is comparable with that of the Island. Furthermore, these might be comparable to those of other countries located in the Mediterranean Basin (Eurostat, 2010). Certainly, with the due attention, the results can be extended to other regions with have similar accessibility conditions and road-haulage industry structure. As pointed out in Bergantino (2007), the evaluation of transport service attributes among operators is, in fact, related to the overall accessibility conditions. In some of the regions of countries such as Spain and Greece the contextual elements are very similar to the ones identified in the area interested by this study. Feo et al. (2011), for instance, have found evidence comparable to that described in this work, for those operators located inland, in less accessible areas of Spain (Murcia). The authors use the evidence they obtained to evaluate and support cost-oriented measures – such as eco-bonus and tariff's rebates – as instruments to induce greater modal reallocation towards maritime alternatives. We simulated carrier behaviour referring to random utility theory. In particular, in order to allow for the changing operators' sensitivity to travel time and correlations among alternatives, the empirical analysis was carried out through the mixture of logit framework. Model estimations show that the travel time marginal utility follows a normal distribution and shows heterogeneity around the mean, interacting with firm size (no. of vehiclesþ no. of cargo units). We found that the greater the firm size the lower the marginal utility. The VOT is lower for larger companies, implying a greater organisational capacity. Besides, as expected, hauliers tend to prefer a road-related mode in the case of perishables. While the latter confirms the traditional commodity-type based segmentation, the paper shows that, besides, in line with the results of Arunotayanum and Polak (2011), behaviourally homogeneous segments are defined also by a combination of other attributes related to the attitudes of the decision-makers. To integrate the mixture of logit model with operators' attitudes, in an attempt to improve its explanatory power, we analysed the responses to attitudinal survey questions and found that the perceived importance indicators of travel time, punctuality and risk of damage for goods impact on choices. These indicators are influenced by socioeconomic variables according to the following relationships: the greater the firm size the lower the sensitivity to time; the greater the load factor the higher the importance of punctuality; larger operators assign higher rating to the risk of loss/damage, especially if they do not ship frequently. Given the above results, we incorporated, through the use of attitudinal data, the perceived importance of punctuality and risk of damage into the mixture of logit model with a normally distributed time coefficient. The integrated model points out that, as the importance of punctuality increases, carriers tend to reject the “all-road” mode, which is more likely to produce delays because of congestion. In addition, high ratings for the risk of damage work against the Ro/Ro option, which is perceived to limit the possibility of the carrier to “control” the cargo. Our study supports the assumption that attitudes are important in freight agents' mode choice and shows that a better understanding of

this relationship might yield useful information for planners when designing sustainable transport policies. Although attitudes cannot be easily forecasted, we show that socio-economic variables can be employed to act as proxies and might, thus, solve this problem. The outcome of this paper is in line with previous results: in particular, as Gopinath (1995) pointed out, latent attitude models allow to interpret data in greater details, highlighting the role, for instance, of delay aversion on transportation rate. The most important result, especially for further academic and policy-oriented research of freight choice behaviour, is that policies should be aimed not only towards the realisation of physical infrastructures. A specific and significant role in mode choice is played by perceptions and the beliefs of operators. Policy intervention should aim to reducing the elements of the network which affect the risk aversion of the carriers towards specific attributes (risk of delay; risk of damage; etc.). The abatement of the perceived levels of risks could contribute significantly to promote modal switch toward intermodal and Ro/Ro services. Moreover, governance bodies should pay more attention to the potential of logistic terminals in inducing a significant modal switch, from all-road to more sustainable transport, via Ro/Ro. These infrastructures could support small and medium-sized enterprises to overcome the problem of limited resources pooling together infrastructures and services to reduce their costs through organisational improvements. The terminal would also allow them to provide improved services by sharing their expertise and costs. Indirectly, they could contribute to influence the perceptions of operators with respect to specific risks. Public intervention could stimulate the promotion of automation of handling facilities and the implementation of intelligent transport systems, such as real-time vehicle-location to advise customers on the current position of their goods and updated arrival times, advanced in-vehicle technology to monitor truck performance and loads, travel time information to avoid congested road links and port terminals or accesses, etc. Sharing these facilities within the logistic terminal would generate economies of scale and an abatement of costs. Furthermore, it would allow companies to overcome the size gap and, thus, increase the probability they would choose the intermodal transport option. Considering the sensibility of operators to cost and time savings (elasticity values of 2 and 0.5, respectively), transport service and investment policies should concentrate on establishing large terminals fostering load factor optimisation. Finally, in order to overcome the aversion of carriers for Ro/Ro transport, due to the perceived risk of damage, a number of corrective measures should be introduced, targeted for the risk averse operators, so to induce a greater proportion of the road-haulage carriers to shift towards maritime, eco-friendly, services. These policies might integrate policies oriented mainly to direct cost-abatement, typical of ecobonus or rebates, which, have been in the past few years, the main policy measure used to support modal shift from road to sea. Future work should focus on larger scope applications involving a larger sample of carriers located in and operating from comparable EU regions, in order to understand the role of latent preferences and attitudes at European level. Without an in-depth knowledge of these elements, it would be difficult to define and evaluate policies to support modal rebalance in favour of alternatives to the all-road transport, especially at European level, for which, a strong integration is needed.

Appendix A. Model specifications Please see Table A.1, Table A.2, Table A.3, Table A.4, Table A.5, Table A.6, Table A.7.

A.S. Bergantino et al. / Transport Policy 30 (2013) 77–91

Table A.1 Specification of the base logit model. Coefficients

Table A.5 Specification of the mixture of logit model.

Attributes

Coefficients Attributes Road

Logistic Terminal

Ro/Ro road–sea

βRoad Freight

0 0 0 0 Freightn (1/0)

Time (%) Time (%)∙FirmSizen (1/0) Time (%)∙EnTime ( N[0,1]) Cost (%) 0

Time (%) Time (%)∙FirmSizen (1/0) Time (%)∙EnTime ( N[0,1]) Cost (%) 0

βLT Freight

0

Freightn (1/0)

0

θLand

EnLand

EnLand ( N[0,1])

0

βRoad βLT

[0,1]) 1 0

0 1

0 0

Road

Logistic Terminal

Ro/Ro road–sea

βRoad Freight

0 0 Freightn (1/0)

Time (%) Cost (%) 0

Time (%) Cost (%) 0

βLT Freight

0

Freightn (1/0)

0

βTime βTime-FirmSize sTime βCost

βRoad βLT

1 0

0 1

0 0

βTime βCost

n: sampled operator.

Table A.2 Specification of the nested logit model assuming correlation between Road and Logistic Terminal (the scale parameter at the upper level is normalized to 1). Coefficients

Attributes Logistic Terminal

Ro/Ro road–sea

μLand βRoad Freight

0 0 0 0 Freightn (1/0)

Time (%) 0 Cost (%) 0 0

0 Time (%) 0 Cost (%) 0

μLand βLT Freight

0

Freightn (1/0)

0

μLand βRoad μLand βLT

1 0

0 1

0 0

Table A.3 Specification of the nested logit model assuming correlation between Logistic Terminal and Ro/Ro road–sea (the scale parameter at the upper level is normalized to 1). Coefficients

Logistic Terminal

Ro/Ro road–sea

βRoad Freight

0 0 Freightn (1/0)

Time (%) Cost (%) 0

Time (%) Cost (%) 0

μTranshipment βLT Freight

0

Freightn (1/0)

0

βRoad μTranshipment βLT

1 0

0 1

0 0

μTranshipment βTime μTranshipment βCost

Table A.4 Specification of the error component logit model. Attributes Road βTime βCost βRoad Freight βLT Freight θLand βRoad βLT

Table A.6 Specification of the base logit model integrated with attitudinal indicators.

Road

Logistic Terminal

Ro/Ro road–sea

0 βTime βTime-ImpTime 0 0 βCost Freightn βRoad Freight (1/0) 0 βLT Freight

Time (%) Time (%) Time (%)∙ImpTimena Cost (%) Time (%)∙ImpTimena Cost (%)

βLT ImpPunc Ro=Ro βImpPunc βRoad ImpRisk βLT ImpRisk

0

ImpPuncna

0

0

0

ImpPuncna

0

ImpRiskna

0

βRoad βLT

1 0

0 1

0 0

a

0

0

Freightn (1/0)

0

ImpRiskna 0

0

The indicator value for operator n – the sample mean of the indicator.

Attributes Road

Coefficients

( N

Coefficients Attributes

Road μLand βTime βTime μLand βCost βCost

89

Logistic Terminal

Ro/Ro road–sea

0 0 Freightn (1/0)

Time (%) Cost (%) 0

Time (%) Cost (%) 0

0

Freightn (1/0)

0

EnLand (  N[0,1]) 1 0

EnLand (  N[0,1]) 0 1

0 0 0

Appendix B. Structural and measurement equations when a latent factor is manifested by only one indicator

Table A.7 Specification of the mixture of logit model integrated with attitudinal indicators. Coefficients

Attributes Road

Logistic Terminal

Ro/Ro road–sea

βTime βTime-FirmSize

0 0

Time (%) Time (%)∙FirmSizen (1/0)

sTime

0

Time (%)∙EnTime (  N[0,1])

βCost βRoad Freight

0 Freightn (1/0)

Cost (%) 0

Time (%) Time (%)∙FirmSizen (1/0) Time (%)∙EnTime (  N[0,1]) Cost (%) 0

βLT Freight

0

Freightn (1/0)

0

θLand βRoad βLT βLT ImpPunc

EnLand (  N[0,1]) 1 0 0

EnLand (  N[0,1]) 0 1 ImpPuncna

0 0 0 0

sLT ImpPunc

0

EnImpPunc (  N[0,1])

0

βImpPunc

Ro=Ro

0

0

ImpPuncna

Ro=Ro

0

0

EnImpPunc (  N[0,1])

sImpPunc βRoad ImpRisk

na

ImpRisk

0

0

sRoad ImpRisk

EnImpRisk (  N[0,1])

0

0

βLT ImpRisk

0

ImpRiskna

0

sLT ImpRisk

0

EnImpRisk (  N[0,1])

0

a

The indicator value for operator n – the sample mean of the indicator.

utility U is If the structural equation linking a latent factor (e.g. perceived importance of punctuality/risk of damage) to decision-maker's

U ¼ β  Latent Factor þ random error

ðB:1Þ

90

A.S. Bergantino et al. / Transport Policy 30 (2013) 77–91

and the measurement equation linking the above latent factor to the corresponding indicator (e.g. responses to questions regarding the importance of punctuality/risk of damage) is Indicator ¼ γ  Latent Factor þ η  Nð0; s2 Þ

ðB:2Þ

One can write: Latent Factor ¼ ðIndicator–ηÞ=γ

ðB:3Þ

so that U ¼ ðβ =γ Þ UIndicator–ðβ=γ Þ  η þ random error

ðB:4Þ

References Arunotayanum, K., Polak, J.W., 2011. Taste heterogeneity and market segmentation in freight shipper's mode choice behaviour. Transportation Research Part E-Logistics and Transportation Review 47, 138–148. Ben Akiva, M.E., Lerman, S.R., 1985. Discrete Choice Analysis: Theory and Application to Travel Demand. The MIT Press, Cambridge, Mass. Ben Akiva, M., Walker, J., Bernardino, A.T., Gopinath, D.A., Morikawa, T. and Polydoropoulou, A. (1999) Integration of Choice and Latent Variable Models, Massachussets Institute of Technology, Cambridge. Ben-Akiva, M., McFadden, D., Train, K., Walker, J., Bhat, C., Bierlaire, M., Bolduc, D., Boersch-Supan, A., Brownstone, D., Bunch, D., Daly, A., de Palma, A., Gopinath, D., Karlstrom, A., Munizaga, M.A., 2002. Hybrid choice models: progress and challenges. Marketing Letters 13 (3), 163–175. Bergantino, A.S., 2009. How relevant is the localisation of operators in the evaluation of freight transport services? A comparison North-South Quanto conta la localizzazione dell'operatore nella valutazione dei servizi di trasporto merci? Un confronto Nord-Sud. In: Borruso, G., Forte, E., Musso, E. (Eds.), Economia dei Trasporti e Logistica Economica, Ricerca per l'Innovazione e Politiche di Governance. Giordano Editore, Napoli, pp. 57–72. Bergantino, A.S., Bolis, S., 2008. Monetary values of transport service attributes: land versus maritime ro-ro transport. An application using adaptive stated preferences. Maritime Policy and Management 35 (2), 159–174. Bergantino, A.S., Bolis, S., 2005. An adaptive conjoint analysis of freight service alternatives: evaluating the maritime option. In: Reggiani, A., Schintler, L. (Eds.), Methods and Models in Transport and Telecommunications: Cross-Atlantic Perspectives. Springer-Verlag, Berlin, pp. 181–198. Beuthe, M., Bouffioux, Ch., 2008. Analysing qualitative attributes of freight transport from stated orders of preference experiment. Journal of Transport Economics and Policy 42 (1), 105–128. Blayac, T., 2007. Modelling and estimating a value of travel time savings for sea transport models. An empirical study in stated preferences for the regular lines between the French mediterranean sea-shore and Corsica. International Journal of Transport Economics XXXIV (1), 79–104. Bierlaire, M.,2009. Estimation of Discrete Choice Models with BIOGEME 1.8, 〈http:// biogeme.epfl.ch〉. Bierlaire, M., Fetiarison, M., 2009. Estimation of discrete choice models: extending BIOGEME. In: Proceedings of the 9th Swiss Transport Research Conference (STRC). Bierlaire, M.,2003. BIOGEME: a free package for the estimation of discrete choice models. In: Proceedings of the Swiss Transport Research Conference (STRC). Bolduc, D., Ben Akiva, M., Walker, J., Michaud, A., 2005. Hybrid choice models with logit kernel: applicability to large scale models. In: Lee-Gosselin, M., Doherty, S. (Eds.), Integrated Land-Use and Transportation Models: Behavioural Foundations. Elsevier, Oxford, pp. 275–302. Bolis, S., Maggi, R., 2002. Stated preferences—evidence on shippers’ transport and logistics choice. In: Danielis, R. (Ed.), Freight Transport Demand and Stated Preference Experiments. Franco Angeli, Milano, pp. 203–222. Brooks, M.R., Puckett, S.M., Hensher, D.A., Sammons, A., 2012. Understanding mode choice decisions. A study of Australian freight shippers. Maritime Economics and Logistics 14 (3), 274–299. Catalano, M., Migliore, M., 2005. Logistics terminals planning for an optimal freight mobility system: an application on a regional scale. In: Jaszkiewicz, A., Kaczmarek, M., Zak, J., Kubiak, M. (Eds.), Advanced OR and AI Methods in Transportation. Publishing House of Poznan University of Technology, pp. 160–165. Choo, S., Mokhtarian, P.L., 2004. What type of vehicle do people drive? The role of attitude and lifestyle in influencing vehicle type choice. Transportation Research: Part A 38, 201–222. Daganzo, C.F., 1987. The break bulk role of terminals in many to many logistics networks. Operations Research 35, 543–545. Dallari, F., Brenda, N., 2009. Autotrasporto italiano—Il quadro competitivo, Logistica – giugno 2009, pp. 18–25. Daly, A., Hess, S., Patruni, B., Potoglou, D., Rohr, C., 2012. Using ordered attitudinal indicators in latent variable choice model: a study of the impact of security on rail travel behaviour. Transportation 39, 267–297. Domencich, T.A., McFadden, D., 1975. Urban Travel Demand: A Behavioural Analysis. American Elsevier, New York. Eurostat,2010. Competitiveness in EU road Freight Transport, Paris, July 2010.

Feo, M., Espino, R., Garcìa, L., 2011. An stated preference analysis of Spanish freight forwarders modal choice on the south-west Europe Motorway of the Sea. Transport Policy 18, 60–67. Fosgerau, M., Bierlaire, M., 2007. A practical test for the choice of mixing distribution in discrete choice models. Transportation Research: Part B 41, 784–794. García-Menéndez, L., Martinez-Zarzoso, I., Pinero De Miguel, D., 2004. Determinants of mode choice between road and shipping for freight transport: evidence for four spanish exporting sectors. Journal of Transport Economics and Policy 38 (3), 447–466. Golob, T., 2001. Joint models of attitudes and behaviour in evaluation of the San Diego I-5 congestion pricing project. Transportation Research: Part A 35, 495–514. Gopinath, D., 1995. Modelling Heterogeneity in Discrete Choice Processes: Application to Travel Demand. MIT, Cambridge, MA. (Ph.D. Thesis). Greene, W., Hensher, D.A., 2003. A latent class model for discrete choice analysis: contrasts with mixed logit. Transportation Research: Part B 37 (8), 681–698. Hensher, D.A., Greene, W.H., 2003. The mixed logit model: the state of practice and warnings for the unwary. Transportation 30 (2), 133–176. Hess, S., Bierlaire, M., Polak, J., 2006. ‘Discrete mixtures models’. Paper presented at the 6th Swiss Transportation Research Conference, Monte Verità, Ascona, Switzerland (〈www.strc.ch〉). Institute for Global Maritime Studies, 2008. ‘America's deep blue highway: how coastal shipping could reduce traffic congestion, lower pollution, and bolster national security’, Tufts University, September. Johansson, V.M., Heldt, T., Johansson, P., 2006. The effects of attitudes and personality traits on mode choice. Transportation Research: Part A 40, 507–525. Kang-Soo, K., 2002. Inherent random heterogeneity logit model for stated preference freight mode choice. Journal of Korean Society of Transportation 20 (3), 1–10. La Franca, L., Migliore, M., Catalano, M., 2006. Short sea shipping in Sicily: a future scenario based on infomobility services. In: Puig, J.O., Barbè, R.M., Carcellè, V.G. (Eds.), Maritime Transport III. Barcelona Digital, Barcelona, pp. 597–606. Louviere, J., Hensher, D., Swait, J., 2000. Stated Choice Methods: Analysis and Application. Cambridge University Press, Cambridge. Maier, G., Bergman, E.M., 2002. Conjoint analysis of transport options in austrian regions and industrial clusters. In: Danielis, R. (Ed.), Freight Transport Demand and Stated Preference Experiments. Franco Angeli, Milano, pp. 252–273. Miao, Q., Wang, B.X., Adams, T.M., 2011. Assessing the Value of Delay to Truckers and Carriers, University Transportation Center for Mobility (UTCM) of the Texas Transportation Institute and The National Center for Freight and Infrastructure Research and Education (CFIRE) of the University of Wisconsin Madison. Study commissioned by the U.S. Department of Transportation Research and Innovative Technology Administration, Washington, DC (DOT Grant Nos. DTRT06-G-0020 (CFIRE) and DTRT06-G0044 (UTCM)). Migliore, M., Catalano, M., Lo Burgio, A.e., Maritano, L., 2012. The analysis of urban travellers’ latent preferences to explain their mode choice behaviour. Environmental Impact, Wit Transactions on Ecology and the Environment Series 162, 193–203. Ministero delle Infrastrutture e dei Trasporti (Italian Ministry of Infrastructures and Transport), Comitato Centrale per l'Albo Nazionale delle Persone Fisiche e Giuridiche che Esercitano l'Autotrasporto di Cose per Conto di Terzi (2011), Indagine Relativa al Monitoraggio del Posizionamento delle Aziende di Autotrasporto, 〈www.alboautotrasporto.it〉. Morikawa, T., Ben-Akiva, M., McFadden, D., 2002. Discrete choice models incorporating revealed preferences and psychometric data. Econometric Models in Marketing 16, 27–53. Prato, C.G., Bekhor, S., Pronello, C., 2005. Methodology for exploratory analysis of latent factors influencing drivers’ behaviour. Transport Research Record 1926, 115–125. Prato, C.G., Bekhor, S., Pronello, C., 2012. Latent variables and route choice behaviour. Transportation 39, 299–319. Puckett, S.M., Hensher, D.A., Brooks, M.R., Trifts, V., 2011. Preferences for alternative short sea shipping opportunities: valuation of travel time, reliability and frequency. Transportation Research: Part E 47 (2), 182–189. Robin, Th., Antonini, G.b., Bierlaire, M., Cruz, J., 2009. Specification, estimation and validation of a pedestrian walking behavior model. Transportation Research Part B 43, 36–56. Stopford, M., 1999. Maritime Economics. Routledge Publisher, London. Tagliacarne-Unioncamere, 2012. Atlante delle infrastrutture (Infrastructure Atlas), 〈www.tagliacarne.it〉. Train, K., Wilson, W., 2008. Estimation on stated-preference experiments constructed from revealed-preference choices. Transportation research: Part B 42, 191–203. Walker, J.L., 2001. Extended Discrete Choice Models: Integrated Framework, Flexible Error Structures, and Latent Variables. Massachusetts Institute of Technology. (PhD Thesis). Walker, J.L., Ben Akiva, M.E., 2002. Generalized random utility model. Mathematical Social Sciences 43 (3), 303–343. Woxenius, J., Bärthel, F., 2008. Intermodal road-rail transport in the European Union. In: Konings, R., Priemus, H., Nijkamp, P. (Eds.), The Future of Intermodal

A.S. Bergantino et al. / Transport Policy 30 (2013) 77–91

Freight Transport, Concepts, Design and Implementation. Edward Elgar Publishing, Cheltenham, UK. Yanez, M.F., Raveau, S., Ortuzar, J de, D., 2010. Inclusion of latent variables in mixed logit models: modelling an forecasting. Transportation Research Part A 44, 744–753.

91

Zamparini, L., Reggiani, A., 2007. Freight transport and the value of travel time savings: a meta-analysis of empirical studies. Transport Reviews 27 (5), 621–636.