Transportation Research Part A 102 (2017) 202–211

Contents lists available at ScienceDirect

Transportation Research Part A

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

Encouraging intermodality: A stated preference analysis

of freight mode choice in Rio Grande do Sul
Ana Margarita Larranaga a,⇑, Julian Arellana b, Luiz Afonso Senna a
a
Industrial and Transportation Engineering Department, Federal University of Rio Grande do Sul, Av. Osvaldo Aranha, 99 – 5° andar, Porto Alegre, RS, Brazil
b
Departamento de Ingeniería Civil y Ambiental, Universidad del Norte, Km. 5 Vía a Puerto Colombia, Barranquilla, Colombia

a r t i c l e i n f o a b s t r a c t

Article history: Brazil’s freight modal split is mainly focused on road transport. The imbalance between dif-
Available online 24 November 2016 ferent transport modes suggests a need to promote alternative modalities to strengthen the
competitiveness and provide a more sustainable economic development. The goal of this
Keywords: paper is to identify logistics managers’ preferences for freight transport service attributes
Freight transport for the case of Rio Grande do Sul in Brazil, and discuss which transport policies could
Mode choice encourage multimodality and more sustainable uses of available transport infrastructure.
Stated preference
In this paper we used the stated preference (SP) technique for collecting data on respon-
Discrete choice model
dents’ choices among hypothetical options. The experimental design was structured using
an Efficient design. Discrete choice models were used to identify the preferences of respon-
dents and discuss some possible sustainable policies that could increase the competitive-
ness of the region. The model structures studied were: multinomial logit; mixed logit as a
special case of random coefficients; mixed logit error components- considering panel effect
and mixed logit error components – including possible correlations between attributes of
intermodal alternatives. Parameters estimated from the models were used to compute sub-
jective value of time savings, which was Euro/t.h 0.34 (R$/t. h 1.088) for the selected model.
The direct and cross elasticity values of the probability of choosing a transport mode for the
different attributes studied show that the shippers significantly value the fulfilment of
delivery time and cost, suggesting that those attributes are the most important ones in
the choice of transport mode in this State of Brazil. Simulation results suggest investments
for increasing the reliability of intermodal alternatives are more effective to encourage
intermodality than cost reductions. Policies and investments to encourage multimodality
should prioritize the increase of intermodal alternatives reliability and combined policies
of cost reduction and reliability.
Ó 2016 Elsevier Ltd. All rights reserved.

1. Introduction

Brazil’s freight modal split is mainly focused on road transport. In the state of Rio Grande do Sul, one of the most populous
states of Brazil, 85.3% of the total cargo is transported by road. This value is above the Brazilian mean of 68.6% transported by
road (Secretaria da Coordenação e Planejamento, 2015). The imbalance between different transport modes suggests a need
to promote alternative modalities to strengthen the competitiveness and provide a more sustainable economic development.

⇑ Corresponding author.
E-mail addresses: analarra@producao.ufrgs.br (A.M. Larranaga), jarellana@uninorte.edu.co (J. Arellana), lsenna@producao.ufrgs.br (L.A. Senna).

http://dx.doi.org/10.1016/j.tra.2016.10.028
0965-8564/Ó 2016 Elsevier Ltd. All rights reserved.
 A.M. Larranaga et al. / Transportation Research Part A 102 (2017) 202–211 203

Planning for a more efficient and sustainable transport system within the state is being carried out through studies that seek
to promote the rationalization of transport flows among different modes, encouraging multimodality. The final aim is to
increase competitiveness in logistics of Rio Grande do Sul that allows better access to domestic and international markets.
Increasing the efficiency of transport systems to improve the competitiveness of a region needs the formulation of appro-
priate transport policies. To formulate adequate transport policies is essential to know supply chain actors’ preferences about
different attributes of available transport modalities. In this regard, assessing firms’ value of service for freight transport in
different modes is important for policy makers, public agencies, local governments and researchers (Danielis and Marcucci,
2007). The goal of this paper is to identify logistics managers’ preferences for freight transport service attributes for the case
of Rio Grande do Sul in Brazil, and discuss which transport policies could encourage multimodality and more sustainable
uses of available transport infrastructure. In this paper, stated preference (SP) technique for collecting data on respondents’
choices among hypothetical options is described and standard econometric discrete choice models are estimated to identify
the preferences and discuss some possible sustainable policies that could increase the competitiveness of the region.
The rest of this paper is organized as follows. Section 2 describes the methodology adopted for models’ formulation; Sec-
tion 3 discusses the data collected, and Section 4 presents the estimation results. The paper ends with conclusions and sug-
gestions for future research.

2. Modelling approach

Discrete choice models were estimated to determine the effect of freight service attributes on the logistics managers’
preferences. Most of the discrete choice models used for travel behaviour applications are based on utility theory
(McFadden, 1974), which assumes that the decision-maker’s preference for an alternative can be reduced to a scalar utility
value. Then, the decision-maker selects the alternative in the choice set with the highest utility value.
Several discrete choice model structures were tested, seeking better fit of the model regarding the data collected. The
model structures studied were: (i) multinomial logit; (ii) mixed logit as a special case of random coefficients (ML-RC model);
(iii) mixed logit error components- considering panel effect (ML-EC model); and (iv) mixed logit error components – includ-
ing possible correlations between attributes of the intermodal alternatives (ML-EC2 model). Models’ estimation was per-
formed using the software Biogeme (Bierlaire, 2003).
Initially simpler structures were tested (i.e. MNL models). The multinomial logit (MNL) (McFadden, 1974) is one of the
simplest discrete choice models and also the most used. It is based on the assumption that the random term eiq of the utility
function is identically and independently distributed according to a Gumbel distribution (Extreme Value Type I). This
assumption for the distribution of residuals is rather simplistic, once they depend on the hypothesis of independence and
homoscedasticity of the residues (Ben-Akiva et al., 2003). The MNL yield to the following probability model (Eq. (1)):

ekV iq
Piq ¼ X kV ð1Þ
e iq
8j
AðqÞ

where k is a parameter related to the common standard deviation of the Gumbel distribution. In practise, it cannot be esti-
mated separately from parameters, and is usually normalized to 1. The MNL considers that estimated parameters associated
to various variables included in the utility function specification are the same for all individuals. In addition, its model struc-
ture restricts the substitution patterns among alternatives and does not allow for correlations due to unobserved factors. To
overcome limitations of the MNL model, more flexible model structures has been adopted.
The Mixed Logit (ML) model is a highly flexible model that can approximate any random utility model (McFadden and
Train, 2000). Mixed logit models can be derived under a variety of different behavioural specifications. The model is defined
on the basis of the functional form for its choice probabilities (Train, 2009). Mixed logit probabilities can be expressed as an
integral of standard logit probabilities over a distribution of the parameters (Eq. (2)):
Z
Piq ¼ Liq ðhÞf ðhÞdðhÞ ð2Þ

where Liq(h) is typically an MNL probability evaluated at a set of parameters h and their density function, f(h), is known as
‘mixing distribution’ (Ortúzar and Willumsen, 2011). The ML model has two basic forms: random coefficients (RC) and error
components (EC). The first version of the ML model considers a random coefficients structure, in which the marginal utility
parameters are different for each sampled individual q, but do not vary across choice situations. The second one considers
that the utility function contains at least two error terms. One error term allows the MNL probability to be obtained (and as
such has the usual IID Extreme Value Type I distribution), while the other has a distribution freely chosen by the modeller,
depending on the phenomenon he needs to reproduce (Ortúzar and Willumsen, 2011).
Taste variations have been represented in models by assuming that preferences are randomly distributed in the popula-
tion. Thus, mixed logit models using random coefficient approach (ML-RC) were estimated, considering normal distribution
for the Total transport time coefficient.
A possible correlation between the responses from the same respondent can also be included when estimating mixed
logit but using the error component approach (ML-EC). The specification of these models is easily generalized to allow for
204 A.M. Larranaga et al. / Transportation Research Part A 102 (2017) 202–211

repeated choices by each sampled decision maker. Subsequently, we included possible correlations between unobserved
attributes in intermodal alternatives. Mixed logit models of error components (ML-EC2) were estimated to represent this
correlation among alternatives.
Parameters estimated from discrete choice models were used to compute subjective value of time savings or, equiva-
lently, the willingness to pay to reduce travel time by one unit. Value of time for freight transport is a critical component
of Cost-Benefit Analysis of transport projects and policies.
Finally, direct- and cross-elasticities were computed to analyse the change in the probability of choosing a transport
mode when a given percentage change in the independent variables occur. The first relate to attributes of the transport mode
under consideration and the second to attributes of competing modes. The direct elasticity was computed with the expres-
sion given by Eq. (3) and the cross-elasticity according with the expression given by Eq. (4) (Ortúzar and Willumsen, 2011).
Elasticities were computed for individuals’ choices and aggregated by sample enumeration techniques, to apply these results
to the entire population of the study area.

EPiq;Xikq ¼ hik :X ikq ð1  Piq Þ ð3Þ

EPiq;Xjkq ¼ hjk :X jkq: Pjq ð4Þ

where EPiq,Xikq: is the elasticity of the probability of choosing Ai with respect to a marginal change in a given attribute Xikq;
EPiq,Xjkq is the elasticity of the probability of choosing Ai with respect to a marginal change in the value of the kth attribute of
alternative Aj, for individual q.

3. Data collection and experimental design

Personal interviews were conducted between January and March 2015. A total of 50 shippers and logistics managers from
Rio Grande do Sul were interviewed. Even though the sample size is smaller than sizes used conventionally for passenger
transport demand modelling, the population of interest is shorter than in passenger studies. Data on disaggregate shipping
decisions are scarce because the number of decision makers on interurban freight transport is not that large. The preference
information to appropriate modelling interurban freight transport demand should come from the shipping managers or
someone with enough knowledge about the firms’ logistic decisions. On the other hand, recent research by Bliemer and
Rose (2005, 2009, 2010), and Rose and Bliemer (2005, 2012) exploring the minimum sample size requirements for stated
choice studies, recommend that a minimum of 30 respondents be sampled for any discrete choice study. In addition, the
experimental design was structured using an Efficient design, which aims to reduce the required sample size to produce
a fixed level of reliability in the parameter estimates (Rose and Bliemer, 2013; Choice Metrics, 2013).
The selection of shippers considered the diversity of production chains, type of cargo, transport volumes, economic value
of the cargo and final destination of the cargo (state, national or international). Previous information about the shippers for
selection purposes was obtained from two sources: (i) Analysis System of International Trade Information-ALICE (Ministério
do Desenvolvimento, Indústria e Comercio Exterior, 2014); and (ii) Business Ranking - Indice Empresarial Grandes (Amanhã,
2014). The type of products transported within Rio Grande do Sul and selected to be included in this study were those with
high density of transport because they represent at least 80% of all products generated in the State. Additionally, we included
the most representative products in the commercial balance of the State, and those products with low added value but with a
strategic importance for the state economy. We selected 22 products, being the most important: footwear, soybeans,
tobacco, vehicles, frozen meat, chemicals and leather.
Each respondent faced a set of 18 choice situations. During the stated preference (SP) experiments logistics managers
were asked to point out their favourite transport option between 3 alternatives available: (i) road, (ii) intermodal considering

Table 1
Attributes and levels description.

Attribute Unit No. Mode Levels description

levels
Total transport cost Hundred 3 Road Present level, +5%, 5%
BRLa Intermodal rail Present level, Road level ±% Rail-road present level
Intermodal Present level, Road level ±% Waterway-road present
Waterway level
Total transport time Hours 3 Road Present level, +12%, 12%
Intermodal rail Present level, Road level ±% Rail-road present level
Intermodal Present level, Road level ±% Waterway-road present
waterway level
On-time delivery percentage % 3 75, 85, 90
Percentage of deliveries delayed more than 2 % 3 5, 10, 15
days
a
Brazilian Real.
 A.M. Larranaga et al. / Transportation Research Part A 102 (2017) 202–211 205

rail, (iii) intermodal considering inland waterway transport. Each alternative was described by four attributes: Total transport
cost, Total transport time, On-time delivery percentage and Percentage of deliveries delayed more than 2 days (see Table 1). The
attributes were selected based on the literature review of national and international relevant papers in the area (Araújo and
Martins, 2002; Cullinane and Toy, 2000; Danielis and Marcucci, 2007; Feo et al., 2011; Guy and Urli, 2006; Hoffman, 2000;
Lirn et al., 2004; Malchow and Kanafani, 2001, 2004; Martins et al., 2005; Nir et al., 2003; Novaes and Vieira, 1996; Santos
et al., 2007; Shinghal and Fowkes, 2002; Steven and Corsi, 2012; Tiwari et al., 2003; Tongzon, 1995, 2009; Tongzon and
Sawant, 2007; Vieira et al., 2006; Vieira et al., 2013). Information obtained from an open survey applied to the same shippers,
asking which attributes they take into account in the shipping decision, complemented the literature review in order to
ensure that the right attributes were included.
The attributes selected appear quite frequently in the literature. The last two refers to reliability. As argued by Brooks
et al. (2012), reliability is a difficult construct because it consists of two components. One is the need for a just-in-time buyer
to have his freight delivered within a delivery window that it is acceptable for production process inputs or retail shelf stock-
ing. The second is the reliability required to reduce stock and minimize costs. In order to account for both dimensions, as the
authors above proposed, we have chosen two attributes: On-time delivery percentage and Percentage of deliveries delayed more
than 2 days. The first attribute refers to the delivery occurring within three hours of scheduled time. The second one, accounts
for delays of more than 2 days off-schedule.
The selected attributes try to ensure realistic responses by describing feasible alternatives for the respondents, as recom-
mended for SP designs (Ortúzar and Willumsen, 2011). The attributes present considerable variation between transport
modes. Other attributes, like frequency of service and security for the cargo, that showed to be important attributes in
the literature, not vary significantly between alternatives in the case study. The theft of cargo is the principal problem con-
cerning security, but currently happens in all modes. On the other hand, we avoid including the frequency attribute because
of the possible correlation that could it present with the reliability measures also included in the choice experiment.
Furthermore, as each respondent had to face a large number of choice situations we decided not to include those attri-
butes in the experiment to minimize task complexity and to avoid respondent burden and boredom. A minimum of three
and a maximum of six attributes are desirable, to place the choice exercises in a realistic context and to avoid fatigue effects
(Pearmain et al., 1991; Ortúzar and Willumsen, 2011). Fatigue effects make respondents simplify their choices (Carson et al.,
1994) by focusing on a smaller number of attributes or simply answering at random or in lexicographic fashion
(Sælensminde, 1999).
Table 1 reports attributes and attribute levels used in the SP experiment.
Each attribute was specified with 3 levels. Attribute levels to be presented to each interviewee were customized for some
groups of cargo from origin and destination information available in a previous study (Rumos 2015, 2006) and information
obtained from Brazil’s transport network, through analysis of spatial data and spatial analysis performed in Transcad (Caliper
Corporation, 2012).
From the information contained in the study Rumos 2015 (2006), we identified origins and most representative destina-
tions for each product under study, considering destinations inside and outside Rio Grande do Sul state. These origins and
destinations (OD) were mapped using Transcad software in order to determine spatial data sets, enabling spatial analysis.
Thus, all alternative routes between each pair OD were identified, considering the different available modes and multimodal-
ity. Using the data analysis of road, rail and waterways networks of Brazil, distances and transport time associated with each
route were estimated. The average speeds considered were 60 km/h for road, 40 km/h for rail and 15 km/h for the waterway.
A GIS-based network analysis allows the acquisition of the cost associated to each route in the transport network. Average
cost per tonne (BRL/t.km) for each mode was determined using secondary information obtained from some companies and
from a federal public (Instituto de Pesquisa Econômica Aplicada - IPEA) report (Pompermayer et al., 2014). The cost value
considered a representative shipment of 37 tonnes.
The experimental design was structured using an Efficient design (Rose and Bliemer, 2009) and implemented in NGene
(Choice Metrics, 2013). Efficient design was used to generate parameter estimates with as small as possible standard errors.
This is particularly important when face sampling restrictions, as in the present study. These standard errors can be pre-
dicted by determining the variance-covariance (AVC) matrix based on the underlying experiment and some prior informa-
tion about the parameter estimates. The efficiency measure used was the Bayesian D_error (Db_error), presented in Eq. (1)
which takes the determinant of the AVC matrix, assuming only a single respondent.
Z
Db error ¼ detðX1 Þ1=L fðbÞdb ð5Þ
b

where X1 is the variance-covariance (AVC) matrix for one individual. The objective became to minimize this efficiency error
and obtain a design with the lowest D-error (Rose and Bliemer, 2009). Initial parameter estimates were adopted from
another study we performed in Brazil, in Mato Grosso do Sul state (Agência Estadual de Gestão de Empreendimentos,
2015). In the generation design was adopted a Bayesian approach of the parameters rather than take them fixed, considering
they follow a uniform distribution. This technique considers a distribution of values for each parameter, generating different
designs through random numbers (Halton sequence was used) and calculating the average efficiency value of all designs. 500
extractions were used for each parameter and assessed the efficiency of each design. The Bayesian approach allowed includ-
ing the uncertainty in the initial parameters’ values.
206 A.M. Larranaga et al. / Transportation Research Part A 102 (2017) 202–211

To enforce realism of the experiment, we categorized shippers according to the transport distance (considering the loca-
tion of producers and consumers) and the type of product. Each segment has a specific design, considering different attribute
levels. The segmentation was considered in the design, performing an efficient Fisher design in which each design is
weighted by the number of shippers that belong to each segment. The Db-error value obtained was 0.154 (average consid-
ering all the designs), which is a satisfactory value.
A total of 6 designs were generated: (i) high value products with destination inside Rio Grande do Sul state and distance
less than 500 km; (ii) high value products with destination inside Rio Grande do Sul state and distance more than 500 km;
(iii) high value products with destination outside Rio Grande do Sul state; (iv) medium/low value products with destination
inside Rio Grande do Sul state and distance more than 500 km; (v) medium/low value products with destination inside Rio
Grande do Sul state and distance less than 500 km; (vi) medium/low value products with destination outside Rio Grande do
Sul state. According to the most representative shipment for each shipper, logistic managers were asked to respond only one
design. Fig. 1 shows an example of choice situation for a shipment representative of group (i), high value products with des-
tination inside Rio Grande do Sul state and distance less than 500 km.

4. Results

The estimations results for the discrete choice models are presented in Table 2.
The ML-EC2 model results (mixed logit error components model – including possible correlations between attributes of
the intermodal alternatives) is not presented in Table 2. The correlation between the intermodal alternative was not signif-
icantly different from zero for the adopted confidence level. Thus, the model collapses to MNL, with the same values esti-
mated for the parameters that the MNL, shown in the table.
The estimated models showed satisfactory overall fit and the signs for the parameters are consistent with microeconomic
theory. That is, the coefficients of Total transport cost, Total transport time, and Delay greater than 2 days had a negative sign,
indicating that the utility of modes decreases with increases in time, cost and delays (more than two days). The variable
related with the percentage of on-time deliveries presented positive sign suggesting that increases in shipments percentage
that satisfy the delivery time increases the propensity of using a particular transport mode.
The alternative specific constants for road and intermodal considering rail modes were significantly different from zero at
the 95% confidence level, for MNL and ML-RC models. The constant for the intermodal alternative considering inland water-
way transport was set to zero. Modal constants capture the average of the unobservable error terms (eiq). Error terms are due
to unobserved attributes, measurement errors, differences between individuals, inaccurate perceptions of attributes and ran-
domness inherent in human nature. The positive signs of the constants show that shippers are more likely to choose the road
and intermodal rail modes over the intermodal waterway mode, all else being equal. A comparison of the alternative specific
constant values suggests that the current tendency is to choose the road alternative when the intermodal railway mode is
also available. In the ML-EC model, the alternative specific constants for road and intermodal rail modes were not signifi-
cantly different from zero at the 95% confidence level. It is possible to say that shippers are more likely to choose the road
and intermodal rail modes over the intermodal waterway (positive signs of constants). However, being not significant, it is
not possible to compare the propensity to choose between road, rail and waterway modes.
We used the following two criteria to select the best model: the likelihood ratio test, and the Akaike information criterion
(AIC). The likelihood ratio test allows the comparison between a general and a restricted version of the model. The MNL

Plano de Logísca e Transportes do Rio Grande do Sul PELT - RS

Indicate which alternative would you prefer

Obs: Values considering a representave shipment of 37 tonnes

Intermodal considering inland

Road Intermodal considering rail
waterway transport
Total transport cost BRL 1700 BRL 1000 BRL 700

Total transport time (including waiting and

5 hrs 12 hrs 24 hrs
transshipment time)

On-time delivery percentage 85% 90% 80%

Percentage of deliveries delayed more than 2

15% 5% 10%
days

Choice: A B C

Fig. 1. Example of choice situation.

 A.M. Larranaga et al. / Transportation Research Part A 102 (2017) 202–211 207

Table 2
Models results.

Variable MNL ML-RC ML-EC

Coefficient p-Value Coefficient p-Value Coefficient p-Value
Total transport cost (hundred BRLa) 0.0993 0.00 0.1420 0.00 0.1610 0.00
Total transport time (h) 0.0200 0.13* 0.0466 0.01 0.0648 0.01
On-time delivery percentage (%) 2.9600 0.00 4.2900 0.00 5.6600 0.00
Percentage of deliveries delayed more than 2 days (%) 5.2600 0.00 5.6800 0.00 8.5200 0.00
Road constant 0.7400 0.02 0.9740 0.01 0.6170 0.48
Intermodal considering rail constant 0.4940 0.03 0.8460 0.03 0.6930 0.55
Panel std – – – – 2.5100 0.00
Intermodal std – – 0.1120 0.00
– Draws = 1000 Draws = 1000
No. observations = 1169 No. observations = 1169 No. observations = 1169
No. shippers = 50 No. shippers = 50 No. shippers = 50
Log Log Log likelihood = 782.000
likelihood = 1110.843 likelihood = 1103.591
Pseudo-R2 = 0.115 Pseudo-R2 = 0.120 Pseudo-R2 = 0.375
AIC = 2233.686 AIC = 2221.182 AIC = 1578
*
Parameter significance levels equal to 85%.
a
Brazilian Real.

model is a restricted version of both, the ML-RC and ML-EC models, which are more general. The rejection of the null hypoth-
esis in both cases implies that ML-RC and ML-EC models are better than the MNL model. Comparisons based on the AIC,
between the three models presented in Table 2, suggest that the ML-EC model (AIC = 1578) fit the data better than did
the alternative models, selecting this version among the estimated models.
Parameters estimated from the ML-EC model were used to compute subjective value of time savings or, equivalently, the
willingness to pay to reduce travel time by one unit. Value-of-time (VOT) measures are valuable in a wide range of public
transport policy and planning applications. The VOT measures allow the conversion of a unit transport time (e.g., one hour)
into monetary value. Many cost-benefit analyses report that savings of time represent the largest portion of the benefit asso-
ciated to transport projects. Although there have been a large number of empirical studies on VOT for passenger transport,
relatively little contributions have been made on freight transport (Konishi et al., 2014; U.S. DOT, 2011). In Brazil, the con-
tribution is even smaller, with almost no reference values for freight transport.
The subjective value of time savings for the ML-EC model was Euro/t.h 0.34 (BRL/t. h 1.088). Table 3 presents the values of
time obtained in other researches in terms of Euros per hour and tonne. The results obtained from different studies vary sub-
stantially from one to another. The variation of freight transport value of time is due to its dependence on several factors,
such as the type of freight and the relative transport distance.
Fig. 2 shows the time values of the different studies in ascending order. The value obtained in this study is in line with
previous results from other studies. The value lies within the range of variation of different research in the literature.
We computed aggregated elasticities to analyse the change in the probability of choosing a transport mode when a given
percentage change in the independent variables occur. To obtain the direct- and cross-elasticities, first we used individual
choices and then aggregated them through the sample enumeration technique (see Eqs. (3) and (4) in Section 2, for direct
and cross-elasticities). Table 4 presents elasticity of mode choice probability with respect to Total transport time, Total trans-
port cost, Percentage of deliveries delayed more than 2 days and On-time delivery percentage.
The direct and cross elasticity values in Table 4 suggest that shippers are highly sensitive to the cost and the fulfilment of
delivery time. Those attributes seem to be the most important ones in the choice of transport mode in this State of Brazil. An
increment in the road transport cost by 1% could diminish up to 4.83% of the market share of this mode. On the other hand,
an increment by 1% on the reliability associated to on-time deliveries in the intermodal alternatives could represent a higher
market share of those alternatives, 1.99% and 3.45% on the rail and waterway alternatives respectively.
Improvements in reliability and cost reductions of intermodal alternatives could significantly decrease market shares of
the competing alternatives. Policies and investments to encourage multimodality in the region should then prioritize a reli-
able performance of modalities using rail and waterway links. Substitution patterns among alternatives are greater when
changes in reliability regarding on-time delivery percentage occur. A 1% reduction in on-time deliveries of the road transport
mode should increase 3.76% the demand of intermodal alternatives. Meanwhile, a 1% increase in on-time deliveries of the
rail intermodal alternative or a 1% cost reduction on the same mode should achieve reductions in the range of 2.5–3.0%
in the demand of road and waterway intermodal alternatives.
After running a series of ‘‘what if” scenarios for the cost and the reliability of the different intermodal alternatives we
obtained the market share predictions of the model while keeping constant the remaining attributes with average values.
We use the model to simulate the market shares of the alternatives and predict the possible effects of implementing different
possible transport policies and investments to enhance multimodality in the case study.
208 A.M. Larranaga et al. / Transportation Research Part A 102 (2017) 202–211

Table 3
Value of time estimates for freight transport (in EURO per hour and tonne).

Reference Country VOT

[1] Fowkes et al. (1991) United Kingdom [0.08–1.26]
[2] Abdelwahab and Sargious (1992) United States 0.32
[3] Widlert and Bradley (1992) Sweden 0.03
[4] Fridstrom and Madslien (1995) Norway Edible refrigerated goods [0.41–340.73] with an average of 14.72
Frozen goods < 0
Edible goods: 0
Raw materials: 0
Processed goods: [0.01–19.85] average 0.64
Semi-processed goods: [0.02–8.48] average 1.59 (⁄⁄)
[5] Kurri et al. (2000) Finland All products: 1.43
Willingness to pay, one-hour reduction in transit time: 0.98
Willingness to accept one-hour increase in transit time: 2.24
Forestry industry: 0.28
Metal industry: 2.03
Electronics industry: 3.22
Consumer goods: 1.44
Technical goods: 0.93
[6] de Jong et al. (2000) Netherlands All products: 2.42
Low value raw materials and semi-processed goods: 2.55
High value raw materials and semi-processed goods: 2.81
Final consumer perishables: 2.35
Final consumer non perishables: 2.15
[7] Bolis and Maggi (2003) Alps Full-loaded shipment: 0.81
[8] Vellay and de Jong (2003) France Subcontracted transport: 36.22
[9] Rand Europe (2004) Netherlands 0.05
[10] Novaes et al. (2006) Brazil High value: 0.25 (⁄)
[11] Beuthe and Bouffioux (2008) Belgium 2.88
[12] Feo et al. (2011) Spain [0.43–0.81]
[13] Liu (2013) Sweden [0.3–15.81]

Table adapted from Feo et al. (2011).

Low High

0.03 0.05 0.08 0.25 0.3 0.32 0.43 0.81 1.43 1.59 2.42 2.88 36.22
- 1,26 - 15.81 - 0.81

VOT
Reference [3] [9] [1] [10] [13] [2] [12] [7] [5] [4] [6] [11] [8]
(*) (**)

Fig. 2. Value of time estimates (in EURO per hour and tonne).

Table 4
Elasticity of mode choice probability with respect to changes in the attributes.

Attributes Road Intermodal rail Intermodal Waterway

Total transport time Road 0.72 0.81 0.74
Intermodal rail 0.14 0.58 0.74
Intermodal Waterway 0.14 0.81 1.90
Total transport cost Road 4.83 2.47 1.34
Intermodal rail 0.90 1.79 1.34
Intermodal Waterway 0.90 2.47 3.16
Percentage of deliveries delayed more than 2 days Road 0.69 0.39 0.16
Intermodal rail 0.12 0.35 0.16
Intermodal Waterway 0.12 0.39 0.63
On-time delivery percentage Road 3.85 2.96 1.08
Intermodal rail 3.76 1.99 1.08
Intermodal Waterway 3.76 2.96 3.45
 A.M. Larranaga et al. / Transportation Research Part A 102 (2017) 202–211 209

Road Rail Waterway

80.0%

Market share
60.0%

40.0%

20.0%

0.0%
1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1
Esmated Cost / Current Cost - Intermodal Waterway alternave

Fig. 3. Variation of demand to the cost of the Intermodal Waterway.

Road Rail Waterway

80.0%
Market share

60.0%

40.0%

20.0%

0.0%
0.75 0.8 0.85 0.9 0.95 1
% On-me deliveries - Intermodal Waterway

Fig. 4. Variation of demand to on-time deliveries of the intermodal waterway alternative.

% On-me deliveries - Rail alternave

0.9 0.92 0.94 0.96 0.98 1
80.0%
70.0%
Market share

60.0%
50.0%
40.0%
30.0%
20.0%
10.0%
0.0%
0.75 0.8 0.85 0.9 0.95 1
% On-me deliveries - Waterway alternave
Road Waterway Rail

Fig. 5. Variation of market shares to on-time deliveries of both intermodal alternatives.

Port infrastructure investments along the implementation of some subsidies and tax incentives could diminish the logis-
tic cost of using some intermodal alternatives. Investments in port facilities such as the acquisition of more efficient equip-
ment for cargo handling, the increment in warehousing space, and the implementation of logistics clusters near inland
terminals could also achieve significant cost reductions making intermodal alternatives more attractive. Subsidies and tax
incentives for the use of intermodal alternatives or for the construction of new vessels could also generate cost reductions.
Fig. 3 shows simulated market shares obtained by applying the selected model using different cost reduction scenarios in the
waterway alternative.
Simulation results in Fig. 3 suggest that is difficult to achieve important modality shifts only by reducing the cost of any
intermodal alternative. Subsidies and tax exemptions for intermodal alternatives should be implemented simultaneously
with other policies or investments focusing in improving travel time reliability and on-time deliveries. In the simulation
case, even a reduction of 80% of the cost in the intermodal waterway alternative will not cause a market share of more than
20% for this intermodal alternative.
Figs. 4 and 5 show simulated market shares obtained by applying the selected model but now considering scenarios with
improvements in the reliability of the intermodal alternatives. According to Table 4, direct and cross elasticities of the water-
way alternative are higher than those of the railway intermodal alternative. Then, Fig. 4 only reflects the effects in market
shares of increasing the reliability of the most elastic intermodal alternative regarding to on-time delivery percentage. On
210 A.M. Larranaga et al. / Transportation Research Part A 102 (2017) 202–211

the other hand, Fig. 5 refers to a scenario considering simultaneous improvements of reliability on both intermodal
alternatives.
Simulation results suggest that investments for increasing the reliability of intermodal alternatives are more effective to
encourage intermodality than those efforts towards transport cost reductions. The waterway alternative could improve its
reliability through dredging and maintenance activities, improving navigation marks and signs, implementing vessel traffic
services, and developing maritime and climate forecasting programs. Meanwhile, the use of advanced information manage-
ment systems and monitoring strategies to keep tracks, signals, bridges, tunnels and level crossings in proper operational
conditions could help to achieve a more reliable railway intermodal alternative.
Intermodal alternatives in the case study zone will become more attractive to logistic managers if the waterways and the
railways offer more reliable and stable transport operational conditions. In this sense, the model predicts that the waterway
intermodal alternative will be more attractive than the railway alternative if above 90% of the deliveries, using the waterway
arrive on time and all the other attributes are the same as today (see Fig. 4). According to Fig. 4, if less than two out of ten
deliveries using the waterway intermodal alternative arrive off- time, then the waterway alternative will be the most used in
the area of study.
Achieving levels above 90% of on-time deliveries in both intermodal alternatives should cause that the market share of
road transport in the study zone to be around 56%. Improving reliability of any of the two intermodal alternatives, or even
better in both simultaneously, will encourage an increase of freight multimodal trips in the zone. Finally, considering an
extreme but desired case of 100% on-time deliveries by both multimodal alternatives, and keeping cost and time in the cur-
rent conditions, will cause a 75% of cargo being transported using multimodal services.

5. Conclusions

This study identified logistics managers’ preferences for freight transport service attributes for the case of Rio Grande do Sul
in Brazil. A stated preference survey was conducted contemplating three alternatives of freight transport: (i) Road, (ii) Inter-
modal considering rail, (iii) Intermodal considering inland waterway transport. Each alternative was described by four attri-
butes: Total transport cost, Total transport time, On-time delivery percentage and Percentage of deliveries delayed more than 2 days.
The estimated models showed satisfactory overall fit and the signs for the parameters are consistent with microeconomic
theory. Among the estimated models, the mixed logit error components was selected. Parameters estimated from this model
were used to compute subjective value of time savings. The value computed was Euro/t.h 0.34 (R$/t. h 1.088), which is within
the range of variation of the different studies in the literature. This measure is useful in public policy and planning applica-
tions, particularly in Brazil where almost no reference values for freight transport. The direct and cross elasticity values of
shippers significantly value the fulfilment of delivery time and cost, suggesting that those attributes are the most important
ones in the choice of transport mode in this State of Brazil. An increment by 1% on the reliability associated to on-time deliv-
eries in the intermodal alternatives could represent a higher market share of those alternatives, 1.99% and 3.45% on the rail
and waterway alternatives respectively.
Simulation results suggest that is difficult to achieve important modality shifts only by reducing the cost of any inter-
modal alternative. A reduction of 80% of the cost in the intermodal waterway alternative, for example, will cause a market
share less than 20% for this intermodal alternative. However, investments for increasing the reliability of intermodal alter-
natives are more effective to encourage intermodality. Achieving levels above 90% of on-time deliveries in both intermodal
alternatives should reduce transport use to about 56%. Policies and investments should prioritize the increase of intermodal
alternatives reliability. Combined policies of cost reduction and reliability increase will be more effective to encourage mul-
timodality in this region of Brazil.

Acknowledgements

The authors appreciate the support received from Secretaria de Infraestrutura do RS - SEINFRA in the project ‘‘Plano
Estadual de Logística de Transportes do Rio Grande do Sul – PELT-RS” and the gratitude to Consórcio STE/SD/Dynatest.

References

Abdelwahab, W., Sargious, M., 1992. Modelling the demand for freight transport: a new approach. J. Transp. Econ. Policy 26, 49–70.
Agência Estadual de Gestão de Empreendimentos, 2015. Plano Estadual de Logística e Transportes - PELT/MS. <http://pelt.agesul.ms.gov.br/> (Access: 1/06/
2016).
Amanhã, 2014. Ranking 500 maiores empresas do Sul. Disponível em. <http://www.amanha.com.br/grandes20anos/metodologia.php>.
Araújo, M.P., Martins, R.S., 2002. Logística de transporte de cargas do agronegócio brasileiro: evidências para uma interpretação à luz da teoria dos custos de
transação. In: IV CONGRESSO BRASILEIRO DE ECONOMIA E SOCIOLOGIA RURAL - 2002, Passo Fundo.
Ben-Akiva, M., Bolduc, D., Walker, J., 2003. Specification, Identification, and Estimation of the Logit Kernel (or Continuous Mixed Logit) Model. Working
Paper, 5th Invitational Choice Symposium, Asilomar, California Brownstone, D., K.E.
Beuthe, M., Bouffioux, Ch., 2008. Analysing qualitative attributes of freight transport from stated orders of preference experiment. J. Transp. Econ. Policy 42,
105–128.
Bierlaire, M., 2003. BIOGEME: a free package for the estimation of discrete choice models, Switzerland. In: 3rd Swiss Transportation Research Conference.
Bliemer, M.C.J., Rose, J.M., 2010. Construction of experimental designs for mixed logit models allowing for correlation across choice observations. Transp.
Res. Part B 46 (3), 720–734. special issue.
 A.M. Larranaga et al. / Transportation Research Part A 102 (2017) 202–211 211

Bliemer, M.C.J., Rose, J.M., 2009. Efficiency and sample size requirements for stated choice experiments. In: Transportation Research Board Annual Meeting,
Washington DC January.
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 preferences experiment. Growth Change 34 (4), 490–504.
Brooks, M., Puckett, S., Hensher, D., Sammons, A., 2012. Understanding mode choice decisions: a study of Australian freight shippers. Maritime Econ. Logist.
14, 274–299.
Caliper Corporation, 2012. Transcad-Transportion GIS Software User’s Guide.
Carson, R.T., Louviere, J.J., Anderson, D.A., Arabie, P., Bunch, D.S., Hensher, D.A., Johnson, R.M., Kuhfeld, W.F., Steinberg, D., Swait, J., Timmermans, H., Wiley, J.
B., 1994. Experimental analysis of choice. Market. Lett. 5, 351–368.
Choice Metrics, 2013. Ngene 1.1 User Manual and Reference Guide. Choice Metrics.
Cullinane, K., Toy, N., 2000. Identifying influential attributes in freight route/mode choice decisions: a content analysis. Transp. Res. Part E 36 (1), 41–53.
Danielis, R., Marcucci, E., 2007. Attribute cut-offs in freight service selection. Transport. Res. Part E: Logist. Transport. Rev. 43, 506–515.
de Jong, G., Gommers, M., Klooster, J., 2000. Time valuation in freight transport: methods and results. In: Ortúzar, J.de.D. (Ed.), Stated Preferences Modelling
Techniques. PTRC Perspectives Series, pp. 231–242.
Feo, M., Espino, R., García, L., 2011. An stated preference analysis of Spanish freight forwarders modal choice on the south-west Europe Motorway of the Sea.
Transp. Policy 18, 60–67.
Fowkes, A.S., Nash, C.A., Tweddle, G., 1991. Investigating the market for inter- modal freight technologies. Transp. Res. Part A 25A-4, 161–172.
Fridstrom, L., Madslien, A., 1995. A Stated Preference Analysis of Wholesalers’ Freight Choice, TOI Report 299/1995.
Guy, E., Urli, B., 2006. Port selection and multicriteria analysis: an application to the Montreal-New York alternative. Maritime Econ. Logist. 8 (2), 169–186.
Hoffman, J., 2000. El potencial de puertos pivotes en la costa del Pacífico sudamericano. Revista de la CEPAL, 71.
Konishi, Y., Mun, S., Nishiyama, Y., Sung, J., 2014. Measuring the Value of Time in Freight Transportation, RIETI Discussion Paper, 14-E-004. <http://www.
rieti.go.jp/jp/publications/dp/14e004.pdf> (Access: 1/06/2016).
Kurri, J., Sirkia, A., Mikola, J., 2000. Value of time in freight transport in Finland. Transp. Res. Rec. 1725, 26–30.
Lirn, T.C., Thanopoulou, H.A., Beynon, M.J., Beresford, A.K.C., 2004. An application of AHP on transhipment port selection: a global perspective. Maritime
Econ. Logist. 6 (1), 70–91.
Liu, X., 2013. Estimating Value of Time Savings for Freight Transport: A Simultaneous Decision Model of Transport Mode Choice and Shipment Size. Örebro
University, Orebro University School of Business, Örebro University, Sweden. <http://www.oru.se/PageFiles/13614/Estimating%20Value%20of%20Time%
20Savings%20for%20Freight%20Transport%20%20A%20Simultaneous%20Decision%20Model%20of%20Transport%20Mode%20Choice%20and%
20Shipment%20Size.pdf> (Access: 1/11/2014)
Malchow, M.B., Kanafani, A., 2001. A disaggregate analysis of factors influencing port selection. Maritime Policy Manage. 28 (3), 265–277.
Malchow, M.B., Kanafani, A., 2004. A disaggregate analysis of port selection. Transport. Res. Part E: Logist. Transport. Rev. 40 (4), 317–337.
Martins, R.S., Lobo, D.S., Pereira, S.M., 2005. Atributos Relevantes no Transporte de Granéis Agrícolas: Preferência Declarada pelos Embarcadores. Rev. Econ.
Agrí. Agron. 3 (2), 173–192.
McFadden, D., 1974. The measurement of urban travel demand. J. Public Econ. 3, 303–328.
McFadden, D., Train, K., 2000. Mixed MNL models for discrete response. J. Appl. Econom. 15, 447–470.
Ministério do Desenvolvimento, Indústria e Comercio Exterior, 2014. Aliceweb-Sistema de Análise das Informações de Comércio Exterior. <http://aliceweb.
mdic.gov.br/> (Access: 10/10/2014).
Nir, A.S., Lin, K., Liang, G.S., 2003. Port choice behaviour – from the perspective of the shipper. Maritime Policy Manage. 30 (2), 165–173.
Novaes, A.G.N., Vieira, H.F., 1996. O nível de serviço logístico-portuário sob a ótica dos exportadores. Gest. Prod. 3 (3), 290–306. Florianópolis.
Novaes, A.G., Gonçalves, B., Costa, M.B., dos Santos, S., 2006. Rodoviário, ferroviário ou marítimo de cabotagem? O uso da técnica de preferência declarada
para avaliar a intermodalidade no Brasil. Transportes 14 (2), 11–17. Rio de Janeiro, RJ.
Ortúzar, J.de D., Willumsen, L.G., 2011. Modelling Transport. John Wiley & Sons, England.
Pearmain, D., Swanson, J., Kroes, E., Bradley, M., 1991. Stated Preference Techniques: A Guide to Practice. Steer Davies Gleave and Hague Consulting Group.
Pompermayer, F.M., da Silva Campos Neto, C.A., Pepino de Paula, J.M., 2014. Hidrovias no Brasil: perspectiva histórica, custos e institucionalidade Instituto
de Pesquisa Econômica Aplicada (Ipea). <http://www.ipea.gov.br/portal/index.php?option=com_content&view=article&id=21597%3Atd-1931-
hidrovias-no-brasil-perspectiva-historica-custos-e-institucionalidade&catid=343%3A2014&directory=1&Itemid=1> (Access: 10/10/2014).
RAND Europe, SEO and Veldkamp/NIPO, 2004. Hoofdonderzoek naar de reistijdwaardering in het goederenervoer, Report for AVV, RAND Europe, Leiden.
Rose, J.M., Bliemer, M.C.J., 2009. Constructing efficient stated choice experimental designs. Transport Rev. 29, 587–617.
Rose, J.M., Bliemer, M.C.J., 2013. Sample size requirements for stated choice experiments. Transportation 40 (5), 1021–1041.
Rose, J.M., Bliemer, M.C.J., 2012. Sample optimality in the design of stated choice experiments. In: Pendyala, R., Bhat, C. (Eds.), Travel Behaviour Research in
the Evolving World. IATBR, India, pp. 119–145.
Rose, J.M., Bliemer, M.C.J., 2005. Bliemer Constructing Efficient Choice Experiments. Report ITLS-WP-05-07, Institute of Transport and Logistics Studies,
University of Sydney.
Rumos 2015, 2006. Estudo sobre desenvolvimento regional e logística de transportes no Rio Grande do Sul/SCP-DEPLAN. Porto Alegre: SCP, 2006.5v.: tab.;
graf., mapas.
Santos, C.H.S., Vieira, G.B.B., Pasa, G.S., Bassanesi, M.M.R., 2007. Exportação nos setores metal-mecânico e automotivo gaúchos: pesquisa de mercado para a
seleção de portos. In: Anais do XIV Simpósio de Engenharia de Produção – Simpep. Unesp, Bauru.
Sælensminde, K., 1999. Valuation of Non-Market Goods for Use in Cost-Benefit Analyses: Methodological Issues Ph.D. Thesis. Department of Economics and
Social Sciences, Agricultural University of Norway.
Secretaria da Coordenação e Planejamento, 2015. Atlas Socioeconômico do Rio Grande do Sul. <http://www.scp.rs.gov.br/atlas/conteudo.
asp?cod_menu_filho=797&cod_menu=796&tipo_menu=INFRA&cod_conteudo=1346>.
Shinghal, N., Fowkes, A.S., 2002. Freight mode choice and adaptive stated preferences. Transport. Res. Part E. Logist. Transport. Rev. 38 (5), 367–378.
Steven, A.B., Corsi, T.M., 2012. Choosing a port: an analysis of containerized imports into the US. Transport. Res. Part E: Logist. Transport. Rev. 48 (4), 881–
895.
Tiwari, P., Itoh, H., Doi, M., 2003. Shippers’ port and carrier selection behaviour in China: a discrete choice analysis. Maritime Econ. Logist. 5 (1), 23–39.
Tongzon, J.L., 1995. Determinants of port performance and efficiency. Transport. Res. Part A: Policy Pract. 29 (3), 245–252.
Tongzon, J.L., 2009. Port choice and freight forwarders. Transport. Res. Part E: Logist. Transport. Rev. 45 (1), 186–195.
Tongzon, J.L., Sawant, L., 2007. Port choice in a competitive environment: from the shipping lines’ perspective. Appl. Econ. 39 (4), 477–492.
Train, K., 2009. Discrete Choice Methods with Simulation. Cambridge University Press, New York.
U.S. DOT, 2011. Departmental Guidance for the Valuation of Travel Time in Economic Analysis. Office of the Secretary of Transportation Memorandum.
<https://www.transportation.gov/administrations/office-policy/2015-value-travel-time-guidance> (Access: 1/06/2016).
Vellay, C., de Jong, G., 2003. Analyse Conjointe SP/RP du Choix du Mode de Transport de Marchandises dans la Région Nord-Pas de Calais. Santa Monica,
Calif.: RAND Corporation, MR-1435-MELTT. <http://www.rand.org/pubs/monograph_reports/MR1435> (Access: 7/1/2015).
Vieira, G.B.B., Pasa, G.S., Santos, C.H.S., BassanesI, M.M.R., Machado, J.K., 2006. O nível de serviço do Tecon Rio Grande a partir da ótica dos usuários. In: Anais
do III Simpósio Internacional de Gestão de Negócios em Ambiente Portuário – SINAP, Santos.
Vieira, G.B.B., Kliemann Neto, F.J., Monfort-Mulinas, A., 2013. Gobernanza en cadenas logístico-portuarias de contenedores: proposición de un modelo
conceptual. Espacios 34 (5), 8–18.
Widlert, S., Bradley, M., 1992. Preferences for Freight Services in Sweden, Paper Presented at WCTR, Lyon.