Content uploaded by Edoardo Marcucci
Author content
All content in this area was uploaded by Edoardo Marcucci
Content may be subject to copyright.
Content uploaded by Edoardo Marcucci
Author content
All content in this area was uploaded by Edoardo Marcucci
Content may be subject to copyright.
1
A stated ranking experiment to study policy acceptance:
the case of freight operators in Rome’s LTZ
Marcucci
1
E., DIPES, University of Roma Tre, Italy
Stathopoulos
2
A., DISES University of Trieste, Italy
Gatta
3
V., “Sapienza”, University of Rome, Italy
Valeri
4
E., DISES, University of Trieste, Italy
ABSTRACT: City logistics require an understanding of several issues seldom accounted for
in current research. Policies might produce unsatisfactory results because behavioural and
contextual aspects are not considered. Relevant data is crucial to test hypotheses and forecast
agents’ reactions to policy changes. Despite recent methodological advances in modelling
interactive behaviour, the development of apt survey instruments is still lacking to test policy
acceptability. This paper expands and innovates the methodological literature by describing a
stated ranking experiment to study freight agent interactive behaviour and discusses the
experimental design implemented to incorporate agent-specific priors when efficient design
techniques are employed.
1
DIPES/CREI, University of Roma Tre, Via Gabriello Chiabrera, 199 – 00145 Rome
Italy, edoardo.marcucci@uniroma3.it
2
DEAMS/CREI, University of Trieste, Piazzale Europa, 1 - 34100 Trieste, Italy,
amandairini.blombergstathopoulos@phd.units.it.
3
DSPAS, Sapienza University of Rome, Piazzale A. Moro 5 - 00185 Rome, Italy,
valerio.gatta@uniroma1.it
4
DEAMS, University of Trieste, Piazzale Europa, 1 - 34100 Trieste, Italy, eva.valeri
@phd.units.it.
2
Keywords: urban freight distribution, stated ranking experiments., agent-specific interaction.
JEL: R48, C83, C93
Introduction
Cities are characterised by relevant economies of density and proximity, produce
ideas, innovations and generate economic growth that irradiates to other areas. At the
same time, however, they consume more goods than they produce and, consequently,
need to be supplied from outside. They are characterised both by concentrated
research and service production as well as various negative externalities among which
the most prominent are: congestion, visual intrusion, environmental and acoustic
pollution. Their impact is particularly high in densely populated areas where
economic activities are concentrated and generate a consistent, strong and, usually,
rigid demand for public and freight transportation. Decision makers have adopted
policies with the intent of optimising the movement of both passenger and freight so
to foster a sustainable development via the decoupling of economic growth from
transport demand. The most frequently implemented urban freight policies need to be
analysed and evaluated considering a host of these factors. These include: policy
characteristics, linkages with the problems they should solve, external effects,
distribution of impacts among the different stakeholders, the correct level of analysis
of the phenomenon, the data needed to evaluate policy results, the most likely
reactions deriving from the policies implemented and, last but not least, the models
adopted to forecast policy impacts used to provide policy-makers with the relevant
information needed for taking relevant decisions.
This paper illustrates the potential of using a stated ranking experiment (SRE) to elicit
the relevant data for successfully estimating and quantifying the preferences of
stakeholders within an urban freight transport (UFT) context. In particular the study
3
reports an application of the techniques described to a case study concerning UFT in
Rome’s limited traffic zone (LTZ). We propose an innovative methodology to
investigate both retailer’s and carrier’s sensitivity to changes in policy packages that
are simultaneously considered possible by the local authorities (transport regulators)
and acceptable by the main stakeholders (retailers, own-account
5
and carriers).
The paper describes the definition, development and administration of a SRE in a
real-life context where effective policy interventions (e.g. access charging, time
windows, loading/unloading (l/u) bays) are envisaged and evaluated for
implementation. Ideally, the experiment proposed will enable the researcher to
identify both overall ex-ante policy acceptability as well as policy acceptability by
single stakeholder influenced by the policy mix implemented. Urban freight
distribution is a phenomenon deeply intertwined and influenced by interaction effects
among the actors involved; the approach described in this paper identifies not only
effective and efficient measures but also, among these, the subset that can be
considered acceptable, if not by all, by the greatest number of actors possible.
The innovative features of the methodology proposed relate to the contemporaneous
consideration of both demand and supply operators instead of, as is usually done, just
studying the two facets as separate phenomena. Under this respect our approach
proves complementary to the widely used Freight Quality Partnership (FQP) that,
however, adopts a more descriptive and qualitative stance.
The paper is structured as follows. Section 2 reviews the literature on both agent
interaction analysis in the freight sector as well as that of stated preference and
experimental design. The description of the study context is reported in section 3
5
By own-account we intend a specific group of retailers that, predominantly, auto-
produce their own freight transportation services.
4
while section 4 describes the development of the survey instrument. Section 5
illustrates the deployment of the survey and section 6 concludes.
1 Literature review
1.1 Freight and agent interaction: an overview
Freight modelling is to date typically performed by means of aggregate models that
provide no satisfactory account of the critical role individual actors play in the
decision making process. This represents a substantial limitation especially for policy
interventions aimed at changing the reference scenario and altering agents’ relative
convenience of past actions. This paragraph illustrates some recent findings of a
behavioural approach to freight modelling, in general and to UFT in particular. This
innovative method accounts for the most relevant complexities deriving from modern
logistic supply chain activities. Hensher and Figliozzi (2007) argue that standard
approaches do not fully account for the complexity of freight movements at different
geographical scales. What is more, new delivery methods (e.g. JIT) and customer
driven freight services (e.g. electronic commerce) have made UFT more complex thus
paving the way to highly specialised third-party logistic providers. Within the group
of disaggregate models (e.g. inventory models and logistic optimisation) behavioural
models explicitly consider stakeholders’ utility maximization efforts. When dealing
with behavioural models, one has to clearly and unequivocally identify the key
decision makers to develop a modelling framework adopting an actor-based micro-
simulation approach capable of describing and forecasting the behaviour of the
specific actors involved (Liedtke and Schepperle, 2004). Various authors (Gray, 1982;
5
Southworth 2003; Wisetjindawat et al., 2005; de Jong and Ben-Akiva, 2007; Hensher
and Figliozzi, 2007; Samimi et al., 2009; Yang et al., 2009; Roorda et al., 2010)
consider UFT the most appropriate field of application for developing actor-based
micro models. Freight movements are as relevant as the underlying motivations
determining the relative convenience of each stakeholder in taking a specific action or
making a given choice. Structural behavioural analysis represents a substantial
improvement with respect to standard modelling techniques. The specific advantages
of explicitly allowing for behavioural considerations when modelling freight
movements become evident when considering network and micro-simulation
modelling, land use/transport network with feedback effects, the relevance of physical
characteristics of logistics networks. Previous modelling approaches mostly
abstracted from these aspects. These innovations have introduced greater realism in
the analysis by explicitly accounting for the behavioural aspects influencing and
motivating freight stakeholders when: 1) choosing among different strategies, 2)
dealing with specific constraints, 3) accounting for incentives, 4) interacting with
others. These facets are for UFT policy analysis, acceptability and impact assessment.
In fact, interactions between existing and prospective constraints posed by new
policies, motivations to choose a particular strategy or a set of constraints may change
when the state of the world is altered. For example, policy changes influencing fuel
prices, land use patterns and pricing strategies modify the constraints and alter the
relative convenience of each option. Puckett and Greaves (2009) argue that it is
important to jointly consider both the instruments available to policy makers and the
set of drivers influencing freight travel behaviour to gain a better understanding of the
potential impacts the policies implemented might have on market outcomes. This is
exactly what policy makers would like to know ex-ante before actually implementing
6
a given policy. It is not only important to identify a type of incentive/disincentive with
a relevant impact but also be able to understand and quantify its impact given the
reference context. To do so one has to understand which type of decision makers are
involved, how they interact, under which constraints they operate, on which specific
freight service attribute they negotiate and what sort of interaction is actually going on
among them.
Some new approaches have been recently developed to tackle the issues raised in this
section. The most prominent promoters of interactive choice experiments (IACE) for
analysing urban freight transport are Brewer and Hensher (2000), Puckett and
Hensher (2006; 2008). Usually both financial and sample size issues render this
approach difficult to implement for real-life applications. Only a limited number of
buyers of road freight transport services or transport providers are willing to
participate in a study and hence it is difficult to guarantee a sufficient participation in
order to obtain reliable parameter estimates
6
.
1.2 Experimental design: an overview
Stated choice (SC) experiments have a long-standing tradition dating back to the early
eighties. In fact, we can trace the first contributions in this field back to the works of
Louviere and Woodworth (1983) and Louviere and Hensher (1983). Choice
experiments have progressively been employed in a variety of research fields among
which the most prominent applications have been in transportation, marketing,
environmental evaluation and economics. While transportation has witnessed path-
6
Hensher and Puckett (2008) have provided a solution to this issue by developing
minimum information group inference (MIGI) a less data demanding methodology
even if equally capable of producing relevant results. Their illustration indicates the
critical areas where specific efforts are needed to gain a better understanding of UFT
related decision making.
7
breaking contributions in discrete choice modelling, historically, the most relevant
advances in choice experiment design have emerged in marketing and economics
7
.
A choice experiment aims at acquiring high quality data to generate reliable and
useful estimates of the parameters of interest. Depending on the research question
considered, one may adopt a different response format among: choice, ranking or
rating which plays a relevant role since it is linked to the way data can be analyzed
once acquired (Johnson and Desvousges, 1997; Ortúzar and Garrido, 1994; Crask and
Fox, 1987; Louviere, 1992, 1988; Aaker and Day, 1990) and to the reliability of the
responses obtained.
Estimation of statistically significant parameters, especially when small samples are
used (as is usually the case in empirical research), may be aided (impaired) by a good
(poor) experimental design. Thus, the choice of a specific experimental design is not
irrelevant with respect to the research conclusions reached.
An experimental design is, de facto, a matrix of values containing the levels of the
attributes that will constitute the SC survey. The analyst has to optimize the allocation
of the attribute levels to the design matrix given his research goals. Historically, the
most common strategy adopted has been to ensure attribute levels that are
uncorrelated or orthogonal (Louviere et al., 2000). However, more recently, efficient
design, an alternative and innovative approach, has been developed by numerous
researchers (Huber and Zwerina, 1996; Kanninen, 2002; Kessels et al., 2006; Sándor
and Wedel, 2001, 2002, 2005; Ferrini and Scarpa, 2007). The logic underlying
efficient design hinges upon the consideration that orthogonality is not related to
relevant and desirable properties of the discrete choice models employed to analyze
7
Qualified systematizations of both advanced and introductory scientific knowledge
for discrete choice modelling include, among others, the following: Ben-Akiva, and
Lerman (1985), Hensher et al. (2005), Louviere et al. (2000), Train (2003), Marcucci
(2005).
8
SC data. Discrete choice models, such as multinomial logit or probit, commonly used
for estimation, using SC data, are not linear and do not require zero correlation
between the attributes of the design
8
. Almost twenty years ago Hensher and Barnard
(1990) clarified the distinction between design orthogonality and estimation-data
orthogonality evidencing that this property is not always preserved in model
estimation. This last characteristic would only be guaranteed if the differences in
attribute levels were orthogonal rather than the levels themselves. In other words, the
attribute correlation structure should not be utilized as the sole or main design criteria
and, indeed, a more important element is the correlation of the differences in the
attributes.
Huber and Zwerina (1996) made the first attempt to link the statistical SC properties
to the econometric models used to treat such data. The authors showed that by
relaxing orthogonality conditions the asymptotic standard errors of the parameter
estimates (e.g the square roots of the diagonal elements of the asymptotic variance-
covariance matrix –AVC–) can be reduced. Researchers have, in many cases, used
Monte Carlo simulations to calculate the AVC even if it can be determined
analytically by taking the second derivatives of the log-likelihood function (Rose and
Bliemer, 2005). When constructing an efficient design it is easier to define, evaluate
and consider a single value instead of assessing the whole AVC. Various analysts
have proposed different efficiency measures (e.g. d-efficiency, a-efficiency) to
measure the desirability of the design obtained.
8
This would be important to detect independent effects when employing linear
models.
9
2 The study context: the roman freight limited traffic zone
This section describes the roman LTZ context where the SRE has been implemented.
The institution of a LTZ in Rome’s historical centre can be traced back to the late
eighties when a 5 km
2
area was restricted to non-resident vehicles. The bans on traffic
apply to passenger and freight vehicles alike. Access and circulation in the larger peri-
central area termed “LTZ Anello Ferroviario” (LTZ– Railway Ring) is prohibited to
pre-Euro-1 and Euro-1 light and heavy vehicles. The central area, focus of this study,
has a more detailed legislation in place. It corresponds to a 4 km
2
area in the historical
centre. Least polluting vehicles (Euro 1 and later) alone are allowed to enter the LTZ
with access permission awarded for free only to residents while other agents (e.g.
retailers and freight carriers) must pay an access fee. The scheme operates during
daytime hours (passenger cars: 06.30–18.00 Monday to Friday and 14.00–18.00 on
Saturday). The passenger and freight LTZ largely overlaps where the latter is aimed at
goods vehicles and operates between 10.00–14.00 and 16.00–20.00. The yearly
permit costs 565€ per number plate. Initially, the local police enforced the scheme
manually and this resulted in many vehicles entering the zone illicitly. The system has
subsequently been automatized using cameras and optical character recognition
software. Specific time windows apply for access and parking of freight vehicles.
Nonetheless, a wide range of freight operators is exempted from payments. A
synthetic summary of the regulatory regime presently in place is shown in Table 1.
10
Table 1 – Main regulatory characteristics of Rome freight LTZ
Indeed, the regulation is essentially designed to incentivate the use of third account
operators while discouraging lengthy parking of own account vehicles, given the
shortage of on-street parking in the area. Time windows are currently not
systematically enforced. The scheme, due to the many exceptions, can hardly be
considered as a congestion reducing policy nor can it be classified as a pure
environmental low emission zone (LEZ) since vehicle emissions standards are not
currently part of the scheme. However, the exclusion of Euro-1 and below and the fee
General regulation
Laden weight < 35 q Laden weight > 35 q
Transit and parking allowed
from 20.00 to 10.00 and 14.00
to 16.00 and prohibited
otherwise
Transit and stopovers permitted from 20.00 to 7.00
and prohibited otherwise
Exceptions from time window (around the clock transit and parking)
Laden weight < 35 q Laden weight > 35 q
1. Transport of perishable
foods, pharmaceuticals,
newspapers and precious
goods
1. Trucks with justified request detailing time,
place and route (for instance house moving)
2. All courier and transport
companies
operating as third account (if
enrolled in the “National
registry of auto transporters”)
3. Trucks involved in cleaning
and maintenance services on
account of the municipality or
ATAC
Fee reductions 50% reductions offered for electric cars and 25% reduction for
CH4, GPL and hybrid motor/fuel
11
reduction for alternative fuels suggest that environmental objectives prevail over
efficiency goals.
3 Development of the survey instrument
This section describes the rationale behind the use of separate designs by agent-type
and illustrates the different components of the questionnaire administered.
Fundamentally, when studying urban supply chains one has to figure out what are the
main driving forces at the base of supply agents behaviour. Beyond mapping the main
problems and policy solutions surrounding urban freight distribution, policy
administrators need to understand the perspectives and roles of different stakeholders
in the logistic chain. Receivers, carriers and forwarders are traditionally considered as
essential stakeholders in urban freight logistic system analysis (Ogden, 1992). The
current SRE concentrates on representing three main supply chain agents: carriers,
retailers and own-account operators. The first two, transport operators and retailers
that receive the goods, are well identified in the literature. Stakeholder consultations,
specific studies of the roman context and an analysis of the current regulation, all
suggest it is essential to include own-account operators as well. This separate
treatment and differentiation of the design according to agent-type represents an
important advancement to adequately describe the heterogeneity in needs and
problem perception among agents. Indeed, the insight gained from the meetings with
the stakeholders regarding which agents to represent and the issues potentially
generating more tension among them, proved an important source of guidance in this
process.
The first issue to be dealt with is the definition, selection and development of the
attributes to include in the SRE for each agent. In particular, we illustrate in detail
12
how we moved from the stakeholder consultation stage to the attribute definition. In
doing so we highlight and motivate which specific attributes were included in the
final questionnaire design. Indeed, the level of joint policy acceptance was the main
criteria for attribute inclusion. Following the justification for inclusion we report how
each attribute was defined and structured in levels and ranges. An important point to
keep in mind is the progressive differentiation of the attributes modelled that were
progressively differentiated by agent-type. This procedure was adopted to account for
real-world agent-type constraints and preferences. The choice of attributes was, to a
large extent, based on the results from the stakeholder surveys. The following sections
overview the attributes included, describe their characterization and illustrate the
reasoning behind our choices.
3.1 Attributes included in SRE
Each alternative in the SRE is described by a set of attributes that can take several
levels to describe ranges of variation when the alternatives are presented to the
respondents. For example, when choosing between alternative city logistics policies
one usually, among key attributes, encounters entrance fees and loading-unloading
regulations. Respondents are asked to rank alternative versions of the policy differing
in attribute quality and quantity.
In order to acquire the necessary data to assess the ex-ante acceptability of city
logistic policies in Rome’s LTZ we defined the attributes used in the experiment by
drawing on three main sources, namely: a) literature survey; b) previous quantitative
studies on city freight distribution in Rome; c) series of focus group meetings with
relevant expert stakeholders.
13
We performed an extensive review of the current city logistics literature with an
agent-based perspective that indicated a set of potentially conflicting policy
components when regarded and evaluated from each different agent-type perspective
in the chain. For instance night-time deliveries were considered efficiency enhancing
by carriers but considered only to contribute towards operational cost increases by
retailers. However, before pondering any differentiated agent-specific design it was
necessary to select the attributes to include in the SRE.
Reviewing the quantitative studies on city logistics previously carried out in Rome
(Filippi and Campagna, 2008; STA, 1999) and considering the series of stakeholder
surveys organized provided the greatest contribution to the definition of the attributes
to be included in the SRE. An important phase of the expert surveys
9
was the
questionnaire asking the respondents to indicate the policies reputed most appropriate
to mitigate the identified city logistics problems (Stathopoulos et al. 2010).
In a following stage the results were evaluated according to several criteria to ensure
an appropriate attribute selection for the SRE. The criteria applied were the following:
saliency, shared support and plausibility with respect to changes of the current
scenario. The attributes selected by the stakeholders in the consultation stage could
not automatically be used in the SRE but were revised according to the criteria above.
Volvo REPORT (2010) provides a detailed overview of the link between the
stakeholder survey results and the attributes used in the SRE. Among the policies that
gathered the largest support in the stakeholder survey five macro-policy categories are
9
The results presented here are part of a greater study (Volvo Research and
Educational Foundation, project SP-2007-50 - Innovative solutions to freight
distribution in the complex large urban area of Rome) where a great deal of attention
was paid to the attribute definition phase. A group of experts was interviewed and
long-lasting discussions ascertained which where the most relevant and informative
attributes to include in the study so to correctly characterize the policy intervention
measures to be tested.
14
represented: vehicle, information, loading/unloading, distribution and access policies.
Not surprisingly, among the top rated policies we encounter those inducing least costs
to users (e.g. incentives and an information provision services) in line with the well-
known equity-efficiency trade-off. (Stathopoulos et al 2011) .
Out of the twelve policies, based on the criteria of relevance and acceptability, six
attributes were selected to undergo pilot testing with real operators, namely:
1. number of l/u bays;
2. probability to find l/u bays free;
3. time windows;
4. exemption from time windows;
5. entrance fees;
6. exemptions from entrance fees.
Each of these six attributes have been on the political agenda for a long period and all
were perceived as realistic measures to be included in future policy mixes.
3.2 Agent specific SRE
Following the pilot of the SRE with operators and in line with the differentiation
required by the efficient design, some respondent-type differentiation of the choice
tasks was necessary. In Table 2Table 2 an overview of the content of the SRE for
each agent-type is reported.
The presence of the time window attribute only for own-account operators represents
the main difference among agent-types. This is due to an anchoring affect around the
status quo condition. Indeed, only own-account operators are currently de facto facing
time window restrictions, whereas carriers operating as third account can access the
LTZ at all times, along with a series of other exemptions awarded according to
Formattato:
Inglese
15
specific goods categories. As described for the exemptions, thus, the introduction of
restrictions for operators that have none in real life is very penalizing. In line with
these observations and the results from the pilot study, the attribute was included only
for own-account agents.
Table 2 - Content of SRE per agent-type
Own-account Retailer Carrier
N. of
exercises
10 ranking exercises 10 ranking exercises 10 ranking exercises
Attribute
considere
d in SRE
•
number of l/u bays
• prob. l/u bays available
• time window
• LTZ access fee
• number of l/u bays
• prob. l/u bays available
• carrier LTZ access fee
• number of l/u bays
• prob. l/u bays available
• LTZ access fee
Response
format
ranking: own-account and
potential partner
ranking: retailer and
partner
ranking: carrier and
partner
Regarding the response format, the SRE took shape as a ranking among three policy
options, where one was the status quo LTZ regulation. The agents were asked to rank
policy bundles according to their preferences. They were also solicited to indicate
whether a policy was considered unacceptable and thus not part of their policy-
ordering. For each choice task the respondent is also asked to perform the same
ranking procedure for their typical commercial partner. This means requiring
respondents to state, to the best of their knowledge, the ranking their freight partners
would provide among the available options and whether any of the alternatives would
be considered unacceptable by their partners. In Table 3 an example of a SRE task is
reported.
Table 3 - Example of a ranking task
16
Policy 1 Policy 2 Status quo
Loading/Unloading bays 400 800 400
Probability to find L/U bays free 20% 10% 10%
Entrance fee 1000 € 200 € 600 €
Policy ranking:
Which ranking of the policies, in your view, would
your partner provide?
After selecting the attributes to include in the SRE, the next important step is to
determine the appropriate levels and ranges for each attribute.
The levels that characterize the attributes should ideally be both plausible and policy
relevant, although a choice experiment may also test currently unavailable but
possible alternatives (e.g. a new mobility control policy). In defining the levels it is
important to consider the number of levels, how they are spaced among them and
their range of variation. The attributes, levels, distribution and range are illustrated in
Table 4.
Table 4 - Attribute levels and ranges used in the SRE
Attribute
Number of
levels
Level and range of attribute
(sq underscored)
Loading/unloading
bays:
3 400, 800, 1200
Probability to find l/u
bays:
3 10%, 20%, 30%
Time windows: 3
OPEN from 18:00 to 08:00 e from 14:00 to 16:00;
OPEN from 20:00 to 10:00 e from 14:00 to 16:00;
OPEN from 04:00 to 20:00
Fees: 5 200€, 400€, 600€, 800€, 1000€
The first issue is to determine the number of levels to include. For instance a two-
level attribute only allows for the estimation of linear effects. Yet, the indirect utility
17
function of an attribute may exhibit non-linear effects and for this reason it is often
more informative to include more than two levels to describe an attribute, when
appropriate, and to allow for the estimation of non-linearities in the utility deriving
from different levels.
A second issue is how to distribute the levels. The literature recommends that levels
be evenly spaced to aid interpretation of the coefficients. What is more, if levels are
also symmetrical with respect to the status quo, this allows for the control of
asymmetrical effects related to gains and losses.
The ranges of the levels are of particular importance. Indeed, a sufficiently wide range
of levels should be used to avoid respondents ignoring the attribute due to a lack of
variations. The level range is particularly important for the price attribute that is used
to calculate implicit prices of other attributes using willingness to pay (WTP)
estimates. Moreover, the payment vehicle should be chosen to match the setting.
As may be observed in Table 4 all attributes are characterized by at least three levels.
This allows for controls for non-linear effects in the attribute levels during estimation.
Joint stakeholder meetings were an important source of information concerning the
attribute distribution and range. On this occasion the six selected attributes were
presented and agents asked to provide indications of ranges. Typical questions posed
were: “What is the minimum increase in the number of loading and unloading bays
you would consider necessary?” for each attribute. Based on the ranges provided by
the stakeholders a maximum increase for each attribute was defined for the two l/u
bays and the fees. For the time windows, instead, the stakeholders were asked to
suggest two alternative scenarios to the current one: the first representing a minimum
increase desirable for operators and the second defining a maximum sustainable
reduction concerning the number of hours. Moreover, a meeting with local policy-
18
makers, responsible for promoting and planning changes to the LTZ regulations was
organized.
In the relevant meetings both the feasibility of fee increases and the likely
construction of l/u bays were discussed. Based on comments from local planning
functionaries these attributes were further redefined to achieve realism and properly
mirror plausible policy changes.
Drawing on these results the minimum and maximum points of the attribute ranges
were defined. For the l/u bay attributes the minimum coincides with the current
situation. Instead the range is extended to reflect the stakeholder opinions and the
three levels are then equally distributed. This implies that the policy scenarios only
proposed an increase in the levels. The time window attribute was reduced from five
to three levels due to its complexity. Great effort was dedicated to define one
improved and one deteriorated level for the time window attribute. Due to the
qualitative nature of the attribute it was not possible to ensure that the levels were
evenly spaced. Lastly, the entrance fee attribute was defined to vary in both directions
with respect to the status quo level of approximately 600€. Since past policy changes
have been quite abrupt, the attribute proposed for the SRE had a wide range of
variation going from 200€ to 1000€. The quantitative nature made it a simple task to
ensure that the levels were both symmetrical and evenly spaced over the five levels.
4 Deployment of the survey
4.1 First contact with potential interviewees
Potential interviewees were contacted by mail before approaching them in person for
face-to-face interviews. In fact, various contact methods were considered in the first
19
instance and one evaluated in practice. Contacting potential interviewees by phone
was tested but, after a pilot attempt (30 phone calls were made) with a low success
rate, we reverted to a more traditional and expensive mail contact.
A standard contact letter was prepared to explain both the motivations and scope of
the research. Each letter was completed with the individual contact information and a
signature of a member of the research team to provide some personalization and an
institutional guarantee for the research project. The letter also provided all the
standard guarantees concerning privacy issues and data treatment and dissemination
10
.
Once the letters were sent and the control letter received
11
, we transferred the
information to the interviewers who could then start contacting the various
interviewees.
4.2 Overview of efficient design in four waves
Efficient design is especially desirable in a context characterised by: 1) established
difficulty to contact freight operators and to gain the necessary information due to
privacy issues, 2) lack of interest among agents, 3) lack of appropriate prior
information needed to map specific logistic chains and, 4) the generally high costs of
face-to-face interviews. Indeed, a more efficient design not only improves data quality
but also leads to cost savings. For instance statistically efficient designs may require
smaller numbers of respondents while allowing researchers to extract richer
preference and choice information. Researchers should always try to use the most
10
The letters were progressively sent out according to interviewing needs. In fact, the
letters were in general mailed around one week ahead of the planned interviews.
Particular attention was paid to both the timing and need for sufficient potential
contacts to perform the forecasted interviews for each wave. The mailing was also
performed according to geographical and density of contact criteria.
11
Within all mailing waves we included a letter addressed to ourselves to ensure that
once we received it the other addressees would, most likely, have also received it.
20
efficient designs available but this is much more so in our specific research context
for the motivations reported above.
In what follows, a brief overview will be given as to the design criteria used in each
of the four waves of the SRE. The current work is based on sequential experimental
design theory. This means there may be an evolution of the design that is upgraded in
several, so-called waves, where each wave represents a change in the structure of the
design incorporating the findings from prior interviews. Ideally the sample should be
distributed in such a way to interview 10% of the sample in each of the first three
waves, whereas the largest portion should be saved for the last wave, roughly
representing 70% of the interviews so to provide confirmative results.
4.2.1 First wave
The novelty of the attributes and the lack of any prior studies to rely on in the
definition of the sign and dimension of the coefficients lead the team to test different
approaches. In the course of the work three design strategies were tested. In the first
instance a d-efficient design with very broad priors and the sign of the coefficient of
the attribute was tried. Due to the low precision of the priors used, characterised by
large standard deviation of the coefficients, it was not possible to make the design
converge based on the limited sample size planned for the first wave of interviews. In
the second stage an orthogonal experimental design was tested. This approach implies
that each column containing attributes in the design matrix is perfectly uncorrelated
with every other attribute (Louviere and Woodworth, 1983). It proved impossible to
generate a design with the criteria of orthogonality given the small number of choice
sets defined (9 sets). Due to the inconvenience of working with a design in blocks,
where a segment of the design is given to each respondent, given the small sample-
size foreseen for the first wave a third approach was devised. The third and final
21
design tested was a fractional factorial design. This implies that only a subset of the
possible level combinations appears in the design. Given that six attributes were
present in the initial design, the number of combinations of the design would be equal
to 2
5
× 2
3
× 2
2
= 1,024. Instead, nine choice sets were created with Ngene 1.0
software, which were only a selection of the complete factorial design.
4.2.2 Second wave
For the second wave of the design some important novelties were incorporated. Based
on the estimates from the first wave it was possible to obtain indications of the
magnitude and sign of each coefficient. Based on these results differentiation in the
SRE design properties is introduced. A first aspect of differentiation concerns the
attributes to utilize. As described earlier, several among the attributes originally tested
were eliminated following the pilot survey. However, even for the four attributes
selected, some agent-specific considerations were made. The main difficulty
concerned the time window where econometric estimates were not plausible. Since
attribute improvements proved irrelevant for carriers and retailers, given that neither
operator currently abide by time window restrictions, it was decided that the time
windows be used solely for own-account operators. Moreover, a differentiation in the
design priors was introduced. Given that estimates of attribute coefficients were
available for each agent-type they were incorporated marking the refinement process
needed to implement an efficient design where efficiency refers to the precision with
which coefficients are estimated. Efficient designs produce reliable parameter
estimates for a given sample size or, alternatively, can produce attribute estimates of a
pre-determined level of reliability at a lower cost. In our case, we applied the widely
used d-efficiency criterion along with other criteria used in finalizing the design:
o level balance: each attribute appears equally often and;
22
o utility balance: options in each choice set have similar probabilities of being
chosen.
Since nine choice sets were created for each SRE, level balance could only be ensured
for the three-level attributes.
4.2.3 Third wave
The third wave should ideally confirm and solidify the coefficient estimates derived
from prior waves in view of the final and most comprehensive one. The main novelty
of this wave was the inclusion of non-linearities in attribute level effects. By
estimating effects coding on all attributes, it was possible to control for non-linear
effects
12
. Substantial level-specific effects were found for the fee attribute and in
several cases for the remaining attributes. This lead to the specification of a non-linear
design. At this stage all attributes were defined as agent-specific. It should be
mentioned that when defining the priors for the coefficients, not only a mean prior but
also a prior distribution was proposed. Different distributions can be used and, in our
case, depending on the attribute modelled, normal or uniform forms were used.
4.2.4 Fourth wave
The design of the fourth wave chiefly confirmed the approach previously used. In
conclusion, the criteria used to model the design in the previous waves was
characterised by the following elements:
o agent specific models;
o priors based on estimates of ranking data in previous waves;
o effects coded priors where appropriate;
12
An advantage of effects coding over dummy coding is that it avoids correlation
with the baseline estimate.
23
o unitary or normal distribution of priors according to a priori beliefs;
o use of d-efficiency criterion to select design;
o use of further design criteria such as level balance and utility balance;
o inclusion of a control for ranking consistency.
Since the last wave of interviews involved, by far, the greatest number of interviewees
an additional feature was introduced to ensure the quality of the data gathered. In
previous waves one set of ten identically ordered ranking tasks was administered to
all respondents in a given agent-specific group. However, for the fourth wave, to
avoid acquiring low data quality due to problems deriving from specific task
positioning (e.g. incomplete comprehension of early task or fatigue in the later) we
developed an algorithm for shuffling the tasks so to ensure each task appeared in
different positions within the SRE in the three different versions of the choice task
created for each agent-type.
5 Summary, conclusions and future research
The paper reports a synthetic literature review of both agent interaction in freight and
experimental design followed by a description of the study context and the roman
freight LTZ. We model preferences of three different agent-types and their likely
interactions with their “typical” business partners. The section overviewing the
development of the survey instrument includes a description of the essential activity
of organizing focus group meetings. The stakeholder meetings proved fundamental
for identifying the main freight distribution problems in Rome’s LTZ. This phase
produced a clear view of the perceptions of the main problems and possible solutions
foreseen by the three stakeholder-types involved in this phase: local policy makers
demand (retailers) and supply (transport providers). The main output from this
24
consultation phase was the identification of the attributes considered most critical for
inclusion in potential policy-mixes to be implemented. Several criteria were employed
in selecting the specific attributes used in the SRE. This approach assured two
positive outcomes. On the one hand it provided attributes considered relevant by
interested stakeholders and, on the other, it identified attributes viewed as significant
and important for a balanced group of stakeholders. In fact, policy evaluations ought
to address both relevant and collectively important issues/attributes aimed at
providing policy-makers with indications of potentially effective and acceptable
solutions. Subsequently, the paper describes in detail the various phases of the
development and refinement of a SRE for three different agent-types in Rome’s LTZ.
In fact, a major innovation of the present research is the sub-division of the analysis to
consider three different agent-types: carriers, retailers and own-account. Most of the
recent literature on city logistics acknowledges, in principle, the importance of agent-
specific measures. The present study has acquired the necessary data to formulate
analytically sound and empirically verifiable propositions incorporating knowledge of
agent-specific behaviour. The main problems and potentially feasible solutions
identified in stakeholder surveys were extremely useful in the progressive
specification of the various attributes purposely conceived to map the preferences of
each agent-type. Innovative solutions were also adopted in the questionnaire design
strategy pertaining to a novel use of prior information to seize the trade-offs of
different agent-types. More precisely, the design strategy relied on state-of-the-art
efficient design theory
13
.
The data acquired will allow for the estimation of agent-specific models that are
useful for analyzing the most promising and potentially acceptable policy-mixes. The
13
The questionnaire was implemented thanks to the newly released Ngene 1.0
software by Choicemetrics.
25
results obtained are not only reliable but also relevant under a policy implementation
and evaluation scenario. The research produced is not only innovative under several
aspects but also provides socially relevant results. In brief, the research approach
described in this paper allows for the:
1. identification of the most relevant problems for the LTZ in Rome for the main
significant stakeholders;
2. enumeration of potentially feasible and relevant policies based on
stakeholders’ opinions and preferences;
3. the design of a SRE differentiated using agent-specific attributes and
specification.
The data acquired open the door to several promising future research explorations. A
central extension concerns the estimation of potential shared acceptability of policy
interventions by “couples of agents”, namely retailers and freight carriers. Moreover,
it would be of interest to detect potential distribution channel effects for each category
of goods. Another important extension would be to include and evaluate other
relevant attributes in the policy mix scenarios such as time window exemptions,
entrance fee exemptions, etc. The reactions to such policies are likely to be strongly
differentiated by agent-type and have rarely been explored experimentally in past
research. A further point that would be relevant to investigate relates to reaction to
extended “what if” scenarios. This would allow practitioners to predict the degree of
acceptance and foresee behavioural adjustments as a response to wider contextual
changes, such as fuel-price changes, tax restructurings or changes in related policies
such as parking.
Finally, we would like to stress the great benefits provided by the methodology
proposed in terms of greater accuracy of the estimates obtainable given a specific
26
budget for interview administration or, alternatively, the reduction of the budget
needed to reach a predetermined level of accuracy . This last aspect may be crucial in
different empirical research situations.
Acknowledgement
We gratefully acknowledge funding from Volvo Research and Educational
Foundation, project SP-2007-50 and from the Italian Ministry of University and
Research, Prin 2008 (prot. 2008YEPPM3_005) Methods and models for estimating
the efficacy of urban freight distribution strategies. We would also like to thank the
participants in the Kuhmo Nectar Conference in Valencia 2010 along with the
interviewees participating in the various stakeholder surveys. Lucia Rotaris and Silvia
Ferrini read an earlier version of the manuscript and their comments helped us in
greatly improving the quality of the paper.
6 Bibliographical references
Aaker, D., Day, G., (1990), Marketing Research, Wiley and Sons.
Ben-Akiva, M. and S.R. Lerman (1985) Discrete Choice Analysis: Theory and Application to
Travel Demand, MIT Press, USA.
Brewer, A. and Hensher, D.A. (2000). “Distributed work and travel behaviour: The dynamics
of interactive agency choices between employers and employees”, Transportation,
Vol. 27, pp. 117-148.
Crask, M., Fox, R., (1987), “An exploration of the interval properties of the three commonly
used marketing research scales: a magnitude estimation approach”, Journal of the
Market Research Society, 29 (3), pp. 317-39.
27
de Jong, G., and M. Ben-Akiva (2007) A micro-simulation model of shipment size and
transport chain choice. Transportation Research Part B 41: 950–965.
Ferrini, S., and R., Scarpa (2007) Designs with a-priori information for non-market valuation
with choice-experiments: a Monte Carlo study, Journal of Environmental Economics
and Management, 53:342-363.
Filippi, F. and A. Campagna (2008), “Indagine sulla distribuzione delle merci a Roma
nell’ambito dello studio di settore della mobilità delle merci a Roma”, CTL: Centre
for Studies on Transport and Logistics and ATAC.
Gray, R. (1982) Behavioural Approaches to Freight Transport Modal Choice. Transport
Reviews, 2(2), pp. 161-184.
Hensher, D. A. and Puckett, S. M. (2008) “Power, concession and agreement in freight
distribution chains: subject to distance-based user charges”, International Journal of
Logistics Research and Applications, 11: 2, 81-100
Hensher, D., and Figliozzi, M. A. (2007). "Behavioural insights into the modelling of freight
transportation and distribution systems." Transportation Research Part B:
Methodological, 41(9), 921-923.
Hensher, D.A., Rose, J.M. and Greene, W.H. (2005) Applied Choice Analysis: A Primer,
Cambridge University Press, UK.
Hensher, D.A. and Barnard, P.O. (1990) The Orthogonality Issue in Stated Choice Designs, in
Fischer, M., Nijkamp, P. and Papageorgiou, Y. (Eds.), Spatial Choices and Processes,
North-Holland, Amsterdam; 265-278.
Huber, J. and K. Zwerina (1996) The Importance of Utility Balance in Efficient Choice
Designs. Journal of Marketing Research 33, 307-317.
Johnson, L., Desvouges, W., (1997), Estimating Stated Preferences with Rated-Pair data:
Environmental, Health, and Employment Effects of Energy Programs, Journal of
Environmental Economics and Management, 34, 79-99.
Kanninen, B.J. (2002) Optimal Design for Multinomial Choice Experiments. Journal of
Marketing Research 39, 214-217.
28
Kessels, R., Goos, P. and Vandebroek, M. (2006) A comparison of criteria to design efficient
choice experiments, Journal of Marketing Research, 43(3), 409-419.
Liedtke, G., and H. Schepperle (2004) Segmentation of the Transportation Market with
Regard to Activity–based Freight Transport Modelling, International Journal of
Logistics: Research and Applications, Vol. 7, No. 3, pp. 199-218.
Louviere, J., (1988), Analyzing Decision Making Metric Conjoint Analysis, Quantitative
Applications in the Social Sciences (67), Sage University Paper, SAGE Publication
Inc.
Louviere, J., (1992), Experimental Choice Analysis: Introduction and Overview, Journal of
Business Research, 24, pp. 89-95.
Louviere, J.J. and D.A. Hensher (1983) Using Discrete Choice Models with Experimental
Design Data to Forecast Consumer Demand for a Unique Cultural Event. Journal of
Consumer Research 10, 348-361.
Louviere, J.J., and G. Woodworth (1983) Design and analysis of simulated consumer choice
or allocation experiments: an approach based on aggregated data. Journal of
Marketing Research 20, 350-367.
Louviere, J.J., D.A. Hensher, and J.D. Swait (2000) Stated Choice Methods—Analysis and
Application. Cambridge University Press, UK.
Marcucci E., (Ed), (2005) Introduzione ai modelli a scelta discreta per lo studio dei trasporti,
(An introduction to discrete choice models to study transportation) Carocci Editore,
Roma.
Ngene 1.0, Choicemetrics Pty Ltd, http://www.choice-metrics.com.
Ogden, K. W. (1992). Urban goods movement: a guide to policy and planning, Ashgate,
Aldershot, UK.
Ortuzar, J., Garrido, R. (1994) “A practical assessment of stated preferences methods”,
Transportation, 21, pp. 289-305.
29
Puckett, S. M., and Greaves, S. (2009). "Estimating Policy-Induced Changes in Freight
Vehicle Emissions in Sydney." Institute of Transport and Logistics Studies Working
Paper, The University of Sydney.
Puckett, S.M., Hensher, D.A. (2006) Modelling interdependent behaviour utilising a
sequentially-administered stated choice experiment: analysis of urban road freight
stakeholders, Paper presented at the 2006 Conference of the International Association
of Travel Behaviour Research, Kyoto, Japan.
Puckett, S.M., Hensher, D.A. (2008) The role of attribute processing strategies in estimating
the preferences of road freight stakeholders under variable road user charges.
Transport. Res. E., 44,379-395.
Roorda, M. J., R. Cavalcante, S. McCabe, and H. Kwan (2010) A conceptual framework for
agent-based modelling of logistics services. Transportation Research Part E: Logistics
and Transportation Review, Vol. 46(1), pp. 18-31.
Rose, J.M., and M.C.J. Bliemer (2005) Constructing efficient choice experiments. Report
ITLS-WP-05-07, Institute of Transport and Logistics Studies, University of Sydney.
Samimi, A., A. Mohammadian, and K. Kawamura (2009) Integrating supply chain
management concept in a goods movement microsimulation. IEEE International
Conference on Service Operations, Logistics and Informatics, July 22 - 24, 2009,
Chicago, IL, USA.
Sándor, Z., and M. Wedel (2001) Designing Conjoint Choice Experiments Using Managers’
Prior Beliefs. Journal of Marketing Research 38, 430-444.
Sándor, Z., and M. Wedel (2002) Profile Construction in Experimental Choice Designs for
Mixed Logit Models, Marketing Science 21(4), 455-475.
Sándor, Z., and M. Wedel (2005) Heterogeneous conjoint choice designs. Journal of
Marketing Research 42, 210-218.
Southworth, F. (2003) Freight Transportation Planning: Models and Methods. Transportation
systems planning: methods and applications, by K. G. Goulias.
30
STA. (1999). "Quadro conoscitivo del problema della distribuzione delle merci nel Centro
Storico di Roma." Rome.
Stathopoulos, A., E. Valeri, E. Marcucci, V. Gatta, A. Nuzzolo, A. Comi (2010) “Urban
freight policy innovation for Rome’s LTZ: A stakeholder perspective”, in Macharis,
C., and Melo, S. (Eds.), Multiple views on City Distribution: a state of the art,
Edward Elgar Publishing, London, UK.
Train, K. (2003) Discrete Choice Methods with Simulation. Cambridge University Press, UK.
Volvo Research and Educational Foundation (2010), Innovative solutions to freight
distribution in the complex large urban area of Rome, Final report, project SP-
2007-50.
Wisetjindawat, W., S. Matsumoto, and K. Sano (2005) Supply Chain Simulation for
Modeling the Interactions in Freight Movement. Journal of the Eastern Asia Society
for Transportation Studies, 6: 2991 – 3004.
Yang, C. H., J. Y. J. Chow, and A. C. Regan (2009) State-of-the Art of Freight Forecasting
Modeling: Lessons Learned and the Road Ahead. Proceedings, 88th Annual Meeting
of the Transportation Research Board, Washington DC, 11–15 January 2009.
