Discrete choice and social networks: an analysis of extreme weather uncertainty in travel behaviour

Abstract

In recent years in the UK, and elsewhere, a number of natural events have caused considerable disruptions to travel plans. As extreme weather events are forecast to become more frequent as a consequence of climate change, this paper reports on the methodology behind, and the preliminary results, of an experimental survey on the way long-distance travellers react to weather-related uncertainty. This survey has been developed in line with recent advances in both economic choice theory and travel behaviour research, which hypothesise that travellers, in particular under uncertain conditions, take a number of decisions not in total independence but as a member of a social network. An internet-based survey was conducted with over 2,000 respondents in the two UK cities of London and Glasgow. This paper concentrates on a stated preference experiment that examined travel mode choice between London and Glasgow under different extreme weather conditions. The survey explored the impact of different forms of social dimensions on individual choice in uncertain conditions by asking respondents to state: whether they considered what people in their social circle, or others similar to them (in terms of income, age and neighbourhood), would have done in a similar situation; to guess what those people would have done in a similar situation; and to indicate whether reflecting about the preference of their network, and seeing the percentage of people in their neighbourhood selecting one or another option, would have made them change their mind. Preliminary results from the above questions are discussed and details are given on the following steps of this research.

Full text

1
Discrete choice and social networks: an analysis of extreme weather
uncertainty in travel behaviour
Paper submitted for presentation at the Envecon 2013 Conference, The Royal
Society, London 15 March 2013
Alberto M Zanni* and Tim J Ryley
Transport Studies Group, School of Civil and Building Engineering
Loughborough University, Loughborough LE11, 3TU - UK
This paper represents work in progress, please do not cite without permission
Abstract
In recent years in the UK, and elsewhere, a number of natural events have caused
considerable disruptions to travel plans. As extreme weather events are forecast to become
more frequent as a consequence of climate change, this paper reports on the methodology
behind, and the preliminary results, of an experimental survey on the way long-distance
travellers react to weather-related uncertainty. This survey has been developed in line with
recent advances in both economic choice theory and travel behaviour research, which
hypothesise that travellers, in particular under uncertain conditions, take a number of
decisions not in total independence but as a member of a social network. An internet-based
survey was conducted with over 2,000 respondents in the two UK cities of London and
Glasgow. This paper concentrates on a stated preference experiment that examined travel
mode choice between London and Glasgow under different extreme weather conditions. The
survey explored the impact of different forms of social dimensions on individual choice in
uncertain conditions by asking respondents to state: whether they considered what people in
their social circle, or others similar to them (in terms of income, age and neighbourhood),
would have done in a similar situation; to guess what those people would have done in a
similar situation; and to indicate whether reflecting about the preference of their network, and
seeing the percentage of people in their neighbourhood selecting one or another option,
would have made them change their mind. Preliminary results from the above questions are
discussed and details are given on the following steps of this research.
Keywords: discrete choice, social networks, extreme weather, transport
* corresponding author: a.m.zanni@lboro.ac.uk

2
1. Introduction
The traditional economic approach to analyse choice situations considers individual decision
makers facing a set of alternatives, about which they possess perfect information, who, in
total isolation, only considering their personal preferences, select the alternative bringing
them the highest utility. Over the years, however, theoretical and applied economic studies,
especially when integrating elements of psychology and sociology, have considerably
discussed the validity of this axiom. First, it has been demonstrated that in real life settings
choices are often carried out collectively rather than individually. This, for example,
particularly applies to a range of decisions that are taken by households collectively rather
than by their single members. Second, it has been argued that individuals, when facing a
decision, are affected by the choices and behaviour of others, either the entire society or
particular others, and this has an important impact on their decision making process. All of
these deviations from the traditional theory of choice have a number of practical applications
when analysing choices in the context of transportation.
The aim of this paper is to present work in progress on an application of the second deviation
from the traditional theory of choice mentioned above, and discuss various survey methods
and econometric techniques for the consideration of social dimensions in individual choice.
In particular in this application, whose main purpose was to understand the way long distance
travellers’ behaviour is affected by extreme weather events, an attempt was made to integrate,
to various extents, discrete choice Stated Preference (SP) and Social Networks Analysis
(SNA). Various theoretical and empirical issues are discussed in this paper, and preliminary
results from a large primary dataset are also presented.
The remainder of this paper is organised as follows. Section 2 discusses the theoretical
background. Section 3 briefly introduces SNA, while Section 4 reviews the existing
literature. Section 5 introduces our application to uncertainty in long distance travel, while
preliminary results are presented in Section 6. Section 7 discusses and concludes.
2. Background: why social dimensions in individual choice?
As noted in the introduction, economic theory has traditionally looked at the choice
mechanism as a simple process, albeit well structured, during which perfect information,
consistent preferences, self-interest and expectations play an important role in determining
choice behaviours and outcomes (McFadden, 2001). Both mainstream economic choice
theory and its applications to various contexts have however recognised that such model of
choice may give a limited picture of the array of determinants likely to have a role in
affecting individuals’ choices. This has materialised in the development of different types of
models capable of allowing for various interdependencies within the complex society in
which individuals operate to be rigorously incorporated into economic theory (Zanella, 2007)
and therefore provide a better explanation to complex aggregate social phenomena (Cont and
Lowe, 2010, Soetevent, 2006).
In particular, the concept of social interaction has been introduced by Brock and Durlauf in
their seminal paper (2001:121), as “the idea that the utility or payoff an individual receives
from a given action depends directly on the choices of others in the individual’s reference
group, as opposed to the sort of dependence which occurs through the intermediation of
markets”. Interactions have been identified to be based on preferences, when the decisions

3
over a choice set are dependent on the actions of the other agents; on expectations, when a
decision maker considers the likely outcome of her decisions before making a choice and, for
example, assess the outcome of similar people (peers) facing similar situations; and on
constraints, when different users share a common resource, like in the case of road congestion
(Manski, 2000, Soetevent, 2006). Another important classification discussed in the economic
literature considers endogenous interactions or endogenous social effects, when the
behaviour of an agent is shaped by the way the others behave, exogenous contextual
interactions, when the behaviour of an agent depends on common exogenous shocks or
characteristics to which all group members are subject to (these also include simultaneity or
non-random group selection and general characteristics like race, ethnicity and income), and
correlated effect, when the members of a group behave similarly because they have similar,
unobserved, characteristics, or face similar institutional environments (Ioannides and Zabel,
2008, Krauth, 2006, Manski, 1993, Manski, 2000, Soetevent, 2006).
In terms of their source, social interactions in decision making have been defined as
dependent on two main dimensions, spatial and/or social (Akerlof, 1997, Hayakawa, 2000).
These two dimensions are not always separable and their influence depends not only on the
nature of the effect but also on the examined economic variable (whether it is consumption,
education, job search, unemployment, welfare decisions, etc). The spatial economics and
econometrics literature has for example recognized the necessity of considering that a large
number of economic decisions, especially those related to transportation and residence
(although certainly not the only ones) display a spatial dependency or autocorrelation (Wang
and Kockelman, 2009).
There a number of reasons why social, as well as spatial, influence should be considered
when analysing the particular case of transport situations. First of all, social reasons often
generate the need of travelling. Activities in which individuals engage, and the consequent
travelling, are in fact a direct consequence of the spatial locations of their social contacts
(Carrasco and Miller, 2009, Farber and Paez, 2009, van der Berg et al., 2011). Second,
exchanging information with other individuals in the social space has been identified as an
important strategic tool for travellers, together with personal experience and information
from transport operators (Denant-Boemon and Petiot, 2003, Avineri and Prashker, 2006),
when facing uncertainty due to day-to-day variability in the performance of the transport
systems (Bonsall, 2004). As argued by Schwanen (2008), travellers often react and cope with
uncertainty not individually but as members of a social network. These networks are
therefore an important source of information and decision support for individuals in the
planning of activities and related trips, as they represent relatively low-cost choice heuristic
solutions. Their support to decision-making can materialise in various way. Travellers may
either simply conform to the behaviour of others (observed or unobserved) or directly ask for
suggestions when choosing a departure time, a route, a mode or a vehicle. Neglecting the
consideration of social interactions in the analysis of the way travellers generally behave, and
perceive and react to uncertainty can therefore leave aside important aspects.
When we consider the environmental impact of transport, and ways to reduce it, it is essential
to identify the existence of a critical mass of citizens in a particular area which could trigger
the use of public transportation (or walking and cycling) over private cars (Goetzke, 2008,
Dugundji and Gulyas, 2008), for example. The same applies to the case of vehicle choice,
especially with respect to alternative technology vehicles, whose novelty is likely to create
obstacles to their diffusion, unless potential buyers can observe a considerable number of

4
individuals using them as well as refer to acquaintances for suggestions and testimonials
(Mau et al., 2008, Axsen et al., 2009, Sartzetakis and Tsigaris, 2005).
We have seen above that the influence may come from the overall society, and therefore
based on simple observation or belief, or from a more restricted group of individuals the
decision makers have contacts with. These restricted groups are generally referred to as
reference groups or social networks. These groups are defined in the economic literature as
the group of individuals with whom decision makers are likely to interact to a greater extent
than others, and are therefore assumed to influence their opinion, behaviour and choices
(Hayakawa, 2000), and have often been defined in the existing studies in accordance with
general information that has rather reflected data limitation than research purposes (Soetevent
and Kooreman, 2007), as a proper identification of the individuals likely to have a
considerable impact on choices can be complex. In many cases, data and resource limitations
have forced researchers to define reference groups using common sample characteristics as a
proxy of reference and therefore limit their analysis to anonymous rather than named
networks. For this reason, in recent years, economic choice theory, and transportation
research, have begun to borrow the sociological concepts and methods of Social Network
Analysis (SNA) (Bramoulle et al., 2009, Calvo-Armengol, 2009, Carrasco et al., 2008,
Ioannides, 2006, Carrasco and Miller, 2009, Sunitiyoso et al., 2011)/
3. An introduction to Social Network Analysis (SNA)
Sociological theory defines social networks as the sum of personal networks, which represent
the group of persons (alters) with whom a given individual (ego) considers having a link of
any nature and has contacts with over a lifespan (Degenne and Lebaux, 2005). Social
networks have two main components: actors (persons, groups, organisations) who interact
within each other, and relationships. The latter can be derived, for example, from control,
dependence, competition, and information exchange (Carrasco and Miller, 2009). The main
objective of Social Network Analysis (SNA) is to explore these links between people and
organisations, their formation and their dynamics (Larsen et al., 2009). The ties forming
social networks appear and disappear and have a considerable variability in their intensity
over a life time, and choices made by their members in different situations have also an
important effect on their structures and dynamics (Bidart and Degenne, 2005, Feld et al.,
2007).
In practical terms, in sociological analysis, various survey techniques have been used to
identify personal and social networks and assess their structure and dynamics. Among these
techniques, the name generator appears to be one of the most popular tools. Other
methodologies involve identifying social contacts by using personal sources (like social
media contacts, email address books) or institutional sources (like memberships to clubs,
mailing lists, etc.). In transport settings in particular, travel diaries have been used to identify
social contacts (Axhausen, 2008). The name generator technique identifies the social network
members through in-depth interviewing techniques with questions such as: Who are the
people with whom you discuss matters important to you? who are the people you really enjoy
socialising with? Who are the people you have the most contacts with? (Carrasco and Miller,
2009, Marin and Hampton, 2007). Interviewees reveal first a set of alter names and then
information about their characteristics in order to assess the nature and magnitude of the
relationship (Carrasco et al., 2008).

5
The identification of members and the consequent assessment of the size of the network and
the nature of the relationships are only the first steps in the definition of the structure of a
network. A more rigorous analysis requires the definition of the number of isolates (the
members who are not connected to everyone but only with particular egos), the density of the
network (which is the ratio between the existing number of ties among the members of the
network and the maximum possible given the size of it) and the network sub-groupings.
When the purpose of SNA is to identify social activities (i.e. travel) that can be performed by
the various members individually or in group it is also necessary to assess the potential
activity level between the alters (Carrasco and Miller, 2009), as the activities the individuals
undertake in both their social and geographical spaces have an important impact on the
probability of meeting another individual. Then, the probability of beginning a social
interaction depends on the size of the agents’ current networks and their need for information.
The agents’ utility depends on the similarities with the other agents and how the interactions
with them satisfy their social and information needs. Trust and credibility play an important
role as well (Arentze and Timmermans, 2008).
The discussion above shows that studies aiming at mapping social networks and assessing
their influences on individual decision makers requires a complex and time consuming data
collection effort. Networks are often considerably large and an assessment of their likely
influence also need information on trust, credibility, and relative power of their members.
Snowball sampling techniques have, for example been used to collect information of social
networks (Kowald et al., 2009). Sociograms, graphically showing the connections of
individuals in the social space have also been used (Hogan et al., 2007, Everett, 2006,
Ioannides, 2006).
In this study, we are interested in social networks when they translate into social influence
and therefore affect choice behaviour. In the following section we review the existing
literature considering social influence in the particular case of discrete choices in
transportation contexts.
4. Social interactions in discrete choice: a brief review of the literature
In discrete choice analysis, the effect of social dimensions was first formalised by Brock and
Durlauf for both the binomial (2001) and multinomial case (2002). In accordance with Brock
and Durlauf (2001), the utility of an alternative j for an agent n is specified as follows:
n
jj
n
j
n
jJphU
ε
++
=
(1)
where
n
j
h
represents the deterministic private component of utility (that, in empirical
application normally depends on both the decision-maker’s and neighbourhood’s
characteristics),
j
Jp
the social utility component, where J represents the strength of social
utility and pj denotes the decision makers’ expectations (represented by a conditional
probability measure) of the percentage of others in their neighbourhood selecting the same
alternative j, and
n
j
ε
the random private utility component. Agents’ expectations over the
behaviour of others are non-random, and the main focus is the identification of an
equilibrium for these expectations (Smirnov, 2010). Taking both private and social utility

6
components into consideration, the logit model equation (McFadden, 1974) can be
transformed into:
∑
∈
+
+
=
Ci i
n
i
j
n
j
n
j
Jph
Jph
P)]''(exp[
)]''(exp[
ββµ
ββµ
(2)
Equation (2) specifies the probability for an individual n to select an alternative j over i given
the private and social utility they may gain from the chosen alternative. Extensions of that
model can be found in the more recent papers by Krauth (2006), Zanella (2007) and
Ioannides and Zabel (2008), for example.
Individual consumers could therefore build expectations on the percentage of users of a
particular transport mode (either private car or public transport) in their neighbourhood. This
is the approached followed, for example, by Dugundji et al. (Dugundji and Gulyas, 2008,
Dugundji and Walker, 2005a) in their studies of revealed discrete modal choice data in
Amsterdam. These authors explore the interactions between an individual decision maker and
the aggregate behaviour of other agents in both the individual’s spatial and social networks,
using SNA. Dugundji and Walker (2005a) estimate social and spatial interactions in a mixed
Generalised Extreme Value (GEV) model structure. They discuss five different empirical
strategies to explore social and spatial network interdependencies in individual discrete
choices. These strategies consider a feedback effect as for Brock and Durlauf (2001);
unobserved group heterogeneity, by considering random error correlation among the
members of a specific network as a panel effect (choices of agents in the same group are
treated like the choices over different years by the same person in a typical panel data
approach); and observed heterogeneity, by using alternative specific variables as well as
alternative-specific variance. Dugundji and Gulyas (2008) concentrate instead on the
simulation of choice dynamics over time using a multi-agent based model. Both papers
(Dugundji and Gulyas, 2008, Dugundji and Walker, 2005a) are based on the same dataset and
consider spatial and social heterogeneity, as individuals, differently from Brock and Durlauf
(2002) are assumed to belong to different reference groups. In details, their network
interdependences are taken into account by considering the share of respondents in the same
district, socio-economic group, and neighbourhood (based on postcode) selecting the same
mode.
In such situations, it is reasonable to assume, when modelling travel choices, that there are
unobservable effects likely to have an impact on the choices of different members of a
particular reference group. This could be the case for example of the availability and
accessibility of a certain transport mode in a particular neighbourhood, which will certainly
have an impact of the mode choices of the people living in that neighbourhood. A typical
case of endogenous effect will therefore occur (Dugundji and Gulyas, 2008). Endogeneity
could in fact arise due to consumers self-selecting their neighbourhood (Zanella, 2007) as
well as if multi-directionality of the influence is taken into consideration.
Dugundji and Walker (2005a) tackle the endogeneity issue by allowing a constant correlation
across alternative within a pre-determined spatial (and social) network, without allowing for
this correlation to spread to alternative outside that geographical unit. Their study can
therefore be classified among those using spatial filtering to analyse spatial (and social)
relationships across decision makers (Wang and Kockelman, 2009). However, while their
approach is considered to be effective in analysing social dependence and group-effects

7
among decision makers, some authors have expressed concerned over the capability of
similar models to productively analyse spatial dependence (Smirnov, 2010, Bhat and Sener,
2009). In a recent article, Walker et al (2011) address the issue of endogeneity in choice
models considering social influence using the Berry, Levinsohn and Parker (BLP) two-stage
method. The method, originally used to address similarity among decision makers in a
specific market, involves the decomposition of the error in two different parts, in order to
identify the portion which causes endogeneity. This enables the analysts to distinguish
between the utility relevant to the individual, and that relevant to the peer group. The latter is
then substituted by constants capturing the average effect of the peer group (defined both
spatially, in accordance to respondents’ post code, and socially, based on income similarities)
which are then estimated using linear regression (with the social effect – the fact that each
decision maker choice depend on the choice of others - as explanatory variable). In a linear
setting, an instrumental variable approach is then used to correct for endogeneity. Their
application on a dataset of mode choice in Amsterdam demonstrates the importance of
correcting for endogeneity when analysing social influence in behavioural choice models.
Goetzke and Rave (2011) also considers heterogeneity in their analysis of bicycle use in
Germany.
When the spatial dimension is the only one which is directly considered (simply
neighbouring decision makers), the percentage of other individuals selecting the same
alternative (in this specific case public transport over private car) can also be taken into
account by inserting in the utility specification as from equation (10) a spatially
autoregressive choice indicator (Goetzke, 2008, Adjemian et al., 2010):
n
jj
n
j
n
jWmVU
ε
++=
(3)
Where W is a spatial weight matrix (of dimensions n x n) and mj is a vector that represents the
revealed choices by all individuals. The term Wmj represents therefore the spatially weighted
average percentage of individuals neighbouring the decision maker n selecting the alternative
j. If it is assumed that there is no spatial correlation in the error term, the model specified by
equation (3) can be defined as a spatially autoregressive discrete choice model. This model is
conditional on the observed choice patterns of the neighbouring decision makers and,
therefore, the spill-over process is modelled exogenously (Goetzke, 2008, Anselin, 1988). In
this way, the potential issue of endogeneity is overcome with relatively simple assumptions.
Endogeneity could in fact arise due to consumers self-selecting their neighbourhood (Zanella,
2007) as well as if multi-directionality of the influence is taken into consideration.
Inspiration from both spatial econometrics, as in the case above, and the theory of externality,
has also produced a slightly different approach. This is the case of Paez et al. (2007, 2008) in
their study of telecommuting practices and residential choices. The authors transformed
equation (1) in order to represent social influences:
( ) ( )
[ ]
ntj
ntj
n
jj
nt
nt
n
X
nAyXUAyXUU
n
j
1,1,1,11,111 ,;,...,,;maxmax −−−−
=
(4)
In equation (4), X describes the characteristics of the alternative as well as the decision maker
n,
ntj
y1, −
equals one if the individual n has chosen the alternative j in the previous period t-1,

8
and zero otherwise;
ntj
A1, −
represents instead the previous actions of other decision makers
when facing the same choice set (which are therefore known to the decision maker). These
approaches, where the probability is dependent on variables specific to the decision maker
and the choices of others in a reference group, rather than utility, are sometimes classified as
variants of the Autologistic model in spatial analysis (Smirnov, 2010). Taking into account
the previous decisions by the same agent also formalises learning effects, which have found a
strong empirical evidence in transportation research (Thogersen, 2006)1. Paez et al. (2008)
also integrate elements of SNA by assigning a weight to the influence of others, that is
assumed to be a proxy of the nature of the relationship between the others and the decision
maker. In their application, the weighting variable specifically indicates whether another
individual is part of the decision maker ‘significant others’ or not. The model of Paez et al.
(2008) also allows for the consideration of potential congestion effects, important in a
transportation context, likely to occur if everyone adopts the same practice or decides to use
the same mode.
A similar approach to Paez et al (Paez and Scott, 2007, Paez et al., 2008) is used by Kuwano
et al. (2007) to analyse social dimensions in choice when looking at car purchase decisions.
As noted in the introduction, considering social influence in the analysis of vehicle choice, in
particular alternative technology vehicles, is certainly useful in order to identify possible
barriers to their diffusion. Kuwano et al (2007) utility framework considers past and future
utilities, the influence of habit persistence or variety seeking, and decision makers’
expectations over the choices of members of their reference groups. In particular, the
influence from three types of social interactions on vehicle choice is explored: nation,
neighbourhood, and homogeneous group (based on a classification of households by income
level). Stated preference experiments are used by Mau et al. (2008), to analyse preferences
for new vehicle technology (hybrid and hydrogen fuel cell powered vehicles) in Canada. The
authors carry out four different choice experiments in which respondents, prior to the
completion of the choice cards, are told the hypothetical market share of the different
alternative vehicles they have to select from. Market share information is employed as a
proxy of ‘neighbourhood effect’ and manipulated by the analysts, as a blocking variable in
the experimental design, to understand their effect on consumer acceptability. In a similar
paper, Axsen et al. (2009) combine revealed and stated preference data to analyse
neighbourhood effect in the purchasing of alternative technology in Canada and California.
Spatial interdependence at the neighbourhood level is also analysed by Adjemian et al.
(2010) in their study of revealed car ownership in the San Francisco Bay. Alternative
technology vehicles are also among the transport variables considered by Gaker et al. (2010).
Before each choice card, students are given information about the number of their peers
(other students participating to the experiments) that have selected the different cars as their
favourite one, as well as the number of people who decided not to buy any.
5. An application to the analysis of uncertainty caused by extreme weather conditions
Above, we have discussed the theoretical reasons behind the necessity to extend the
traditional model of choice within discrete choice decisions in transportation context. We
have briefly reviewed a limited but rapidly growing number of studies that provide empirical
evidence on the fact that transportation (as any other consumption or behavioural) decisions,
1 In their 2002 paper, Brock and Durlauf also included the consideration of past society behaviour as an
extension to their basic model.

9
even when taken individually, are influenced by the choices or behaviours of others, to whom
the decision maker has a direct or indirect contact.
The theoretical and empirical frameworks briefly described above inspired the development
of a study on the way long-distance travellers react to uncertainty caused by extreme weather
conditions. In recent years in the UK, and elsewhere in the world, a number of natural events
have caused considerable disruptions to both short and long distance travel plans. As extreme
weather events are forecast to become more frequent as a consequence of climate change,
there is a clear need of studies looking at how travellers react and adapt to these events.
Understanding the changes in travel behaviour brought by extreme-weather related
uncertainty is of great importance for transport operators and both national and local
authorities when setting up prevention as well and contingency plans in order to mitigate the
effects of the disruptions on travellers.
While there are studies on the effect of weather on travel demand, especially on commuting
patterns (Cools et al., 2010, Saneinejad et al., 2012, de Palma and Rochat, 1999) no studies,
at the best of our knowledge, have done so by considering (long distance) travellers
collectively or as members of social networks, and, in particular, their reaction to extreme
weather events (for reviews of the effect of both weather and climate change on transport see
Jaroszweski et al., 2010, Koetse and Rietveld, 2012). As we noted earlier in this paper,
referring to one’s social network when facing uncertainty while travelling is an important
strategy at the disposal of travellers that is often neglected in the analysis of travel behaviour
under uncertain conditions.
From a methodological point of view, as discussed above, few studies have fully integrated
social dimensions in discrete choice settings, in particular in stated preference choice
experiments. And no studies, to the best of our knowledge, have integrated SNA and these
experiments. Although this integration does appear complex, it has a strong theoretical basis
in the economic theory of choice, as well as associated approaches in psychology and
sociology.
Objectives, methodology and survey development
The conceptual framework of our study, in line with what was discussed above, considers
that an individual decision maker’s social network is a relatively low-cost decision heuristics
in uncertain conditions. In particular, our study seeks to provide an answer to the following
research questions:
• In what way is long-distance travel affected by extreme weather conditions?
• How do travellers normally react to failures in the provision of the transport service?
• Do travellers (both prior and during travelling) refer to their social network when
taking travel decisions in uncertain conditions?
• Can discrete choice models effectively be extended to consider that individuals do not
act in isolation but refer to others when making choices?
In order to attempt to answer these research questions and test our hypotheses, it was
necessary to conduct a series of primary data collection efforts. An internet-based survey
instrument was developed through two workshops (March 2010, January 2011) attended by a
number of experts in both travel behaviour and SNA, and two pilot tests
(November/December 2010 and April 2011) on a combined sample of 170 respondents. The
main survey was distributed between August 2011 and February 2012, to over 2,000

10
respondents, split equally between the United Kingdom cities of London and Glasgow.
Quotas were set for age, gender and socio-economic characteristics of respondents.
Survey respondents provided information on their travelling habits, propensity to risk with
respect to extreme weather events, susceptibility to social influence in transport decisions,
propensity to use mobile technology to obtain real-time information on travel conditions,
previous travel experiences on the London-Glasgow route (which was chosen as the case
study corridor), and previous experience of transport service failure due to natural events. In
order to assess respondents social networks, a simple name generator was used, as survey
participants were asked to provide the list of persons “they have regular contact with, and/or
who are the most important to them, and/or those they would want help to discuss personal
matters, and/or those they can trust, and/or those they really enjoy socialising with”. For each
of the contacts, respondents were then asked to indicate whether the particular person lives
with them, the type and length of relationship, and the type and frequency of contacts (by
various means like face-by-face, phone, SMS, email, chat). Respondents were also asked to
indicate which of their contacts they turn to for advice on travel decisions, and, in particular,
who (and why) they would contact if they were experiencing an uncertain situation prior or
while travelling.
Finally, the survey contained an interactive choice experiment focusing on mode choice
under different extreme weather conditions. This experiment not only analysed mode choice
but also explored respondents’ need for social interactions, either direct communication or
simple imitation, in order to inform their choice patterns in uncertain conditions. We discuss
more in details this section of the questionnaire in what follows.
The choice cards
The choice cards that respondents faced in the stated preference section of the questionnaire
depicted hypothetical travel situations first described in terms of the purpose of the trip, its
importance, the persons (if any) travelling with you, and the weather conditions on the day of
travelling (randomly chosen between ‘general fine weather’ and a number of extreme weather
alerts, which included heavy snow, very low temperatures, high winds, dense fog, heavy rain,
and three extreme heat, with temperature of 30, 33, and 36 degrees celsius). Subsequently,
four different methods of travel were compared based on five attributes. The attributes were
departure time, access time to the transport terminal from home, normal duration of the trip,
cost of the trip, likelihood of arriving early, likelihood of arriving on time, likelihood of two
delays levels (up to 5 hours), and egress time (time to reach the final destination once in
London or Glasgow). This was a similar approach to what was proposed by Li et al. (2010) in
their analysis of drivers choice between two different routes which, although we do refer to
general travel behaviour under uncertainty, actually considers a choice under risk as
respondents were given probabilities levels for a certain number of events. Because the focus
of our study was to understand reactions to travel disrupted by extreme weather events,
attributes levels were set up in order for cards to depict situations of high delays (with the
largest probabilities assigned to larger and low probability assigned to earliness and
punctuality). The choice cards were created in accordance to the D-optimality experimental
design principle (Bliemer and Rose, 2011) using the software N-Gene. Eight different choice
cards were distributed to each respondent. Data collected from these choices cards enable us
to estimate a number of figures which have a particular importance in transportation
economics. These are: the value of travel time savings, the value of reliability, the value of

11
expected schedule delay early, and the value of expected schedule delay (Li et al., 2010,
Fosgerau and Karlstrom, 2010). An example choice card is illustrated below.
Table 1. Example of choice card
Air
Train
Car
Coach
WOULD
NOT
TRAVEL
Departure Time
Morning
(between 6am
and 12pm)
Night (after
9pm)
Afternoon
(between 12pm
and 5pm)
Time taken to reach
airport, railway or
coach station + waiting
time
2 hours and 30
minutes
1 hour and 15
minutes
1 hour
Time taken for the
journey in normal
conditions
1 hour and 30
minutes
3 hours and 30
minutes
6 hours and 30
minutes
9 hours and 30
minutes
Cost – single ticket,
and other costs to
reach your final
destination for
Air/Train/Coach; fuel,
parking and motorway
tolls for cars
£170
£40
£210
£60
And you have:
0% probability
of
arriving 10
minutes
early
10%
probability of
arriving 20
minutes
early
0% probability
of arriving 40
minutes
early
0% probability
of arriving 20
minutes
early
10%
probability of
arriving on
time
0% probability
of arriving on
time
0% probability
of arriving on
time
10%
probability of
arriving on
time
50%
probability of
arriving 45
minutes late
60%
probability of
arriving 30
minutes late
50%
probability of
arriving 1 hour
and 30
minutes late
60%
p
robability of
arriving 1 hour
late
40%
probability of
arriving 5
hours late
30%
probability of
arriving 4
hours and 30
minutes late
50%
probability of
arriving 3
hours late
30%
probability of
arriving 2
hours and 30
minutes late
A number of practical and methodological options to integrate social dimensions in the
analysis of individual choices were explored during the development of the survey. As seen
in the previous sections, it is possible to carry out a traditional stated preference experiment
and then assess the degree of social influence in the econometric analysis phase. Unnamed or
anonymous social or spatial networks are therefore used in this case as respondents are
compared with others similar to them based on a number of proxy social (generally income,
working status) and/or spatial variables (residence in the same postcode area). Additional
information about respondents’ ‘social exposure’, ‘likelihood to be influenced by others in
their choices’ or ‘susceptibility to be influenced by opinion leaders’, for example, can also be
used in order to further profile respondents. On the other hand, it is also possible to directly
ask respondents about the potential social influence they may be subjected to when choosing.
There are a number of drawbacks in doing so, as respondents may not be willing to say
directly that their preferences are influenced by others. Nevertheless, such questions can be
used for validating purposes and can also shed more light on the psychology of the

12
consumers. In the next sub-section we present the questions that followed each choice card in
the questionnaire.
The post-choice cards questions
Table 2. Post-choice cards question 1
Q68. When choosing [AIR,TRAIN,CAR,COACH OR NO TRAVEL] have you
considered what other people within your social circle (those identified previously), or
people similar to you (for example in terms of age, income and neighbourhood) would
do in the same situation?
PLEASE TICK ONE BOX ONLY
1. Yes I have considered what people in my social circle would do in the same situation
and chosen as I think they would have
2. Yes I have considered what people in my social circle would do in the same situation
and I have chosen differently
3. Yes I have considered what people similar to me would do and chosen as I think they
would have
4. Yes I have considered what people similar to me would do and chosen differently
5. No, I have decided on my own without thinking what other people would do
6. I do not know
7. Other (PLEASE SPECIFY)
Table 2 shows the first of the post-choice cards question, where respondents were asked
directly whether they had considered what people generally similar to them, and people in
their social circle, would have done in similar situation.
Table 3. Post-choice cards question 2
Q68a: You have chosen [AIR,TRAIN,CAR,COACH OR NO TRAVEL]. What do you
think the majority of other people similar to you (for example in terms of age, income
and neighbourhood) would choose if facing the same situation?
PLEASE TICK ONE BOX ONLY
1. The same as me
2. They would choose [first option not chosen]
3. They would choose [second option not chosen]
4. They would choose [third option not chosen]
5. They would choose [fourth option not chosen]
6. I do not know
Table 3 shows the third of the question, were respondents were instead asked to report what
they thought person similar to them would have done. Using choice data, we will then be able
to check whether respondents were right or not in this occasion.
The questions reported in Table 4 below directly linked the choice cards with the SNA
carried out earlier in the questionnaire. People were asked to state what they thought specific
persons in their social circles would have chosen if facing the same choice cards. The first 5
persons were indicated there, or less if the respondents had named less than 5 friends.
Respondents were also asked about their confidence in reporting their social circle members’

13
preferences and whether reflecting upon their preference had the potential for a change of
mind in their responses.
Table 4. Post-choice cards question 3
Q68b: Please imagine that members of your social circle have to face the same choice as
you. What do you think they would choose?
AIR
TRAIN
CAR
COACH
NO
TRAVEL
I DONT
KNOW
Person 1
Person 2
Person 3
Person 4
Person 5
Q69. How confident are you in knowing what people in your social circle would choose?
TICK ONE BOX ONLY
1. Very unconfident
2. Unconfident
3. Neither confident nor unconfident
4. Confident
5. Very confident
Q70. Would you consider changing your mind, now that we have asked you about their
likely preference?
1. Yes and I would choose: [AIR,TRAIN,CAR,COACH OR NO TRAVEL]
2. No
3. I do not know
Table 5 below illustrates the last of the choice cards questions where indication of ‘market
share’, or percentage of answers for the different categories were given to respondents. In
particular, respondents were said that the shown percentages were recorded in their
neighbourhood. As it was not possible to record the percentages in real time, respondents
were given made up figures which were constructed by varying (+/- 5%) the current mode
share between London and Glasgow.

14
Table 5. Post-choice cards question 4
Q71a. Please now consider that in your neighbourhood in London/Glasgow the
following choices were recorded when facing the same choice card:
Air
Train
Car
Coach
No
travel
Choice
card n.
60
20
5
3
12
4
Would your choice [AIR,TRAIN,CAR,COACH OR NO TRAVEL] remain the same?
TICK ONE ONLY
1. Yes, I would still choose [AIR,TRAIN,CAR,COACH OR NO TRAVEL]
2. No, I would choose [first option not chosen]
3. No, I would choose [second option not chosen]
4. No, I would choose [third option not chosen]
5. No, I would choose [fourth option not chosen]
6. I do not know
7.
I do not believe these percentages are realistic of what people in my
neighbourhood would do in a similar situation.
At the end of the questionnaire, another question, not directly linked to the choice cards,
explored social dimension. It is illustrated in the following table.
Table 6. Additional question.
In a real life situation, would you have needed some additional help in taking the
decision over the different travel situations?
TICK ALL THAT APPLY
1. Yes – I would have needed to access additional information (e.g. transport operators,
MET office)
2. Yes – I would have needed to ask my family/friends for help in deciding
3. No
Who would you have asked in particular?
TICK ONE BOX
ONLY [here the options were the contacts respondents indicated
previously]
Would you have particularly needed help in deciding…?
TICK ALL THAT APPLY
1. Whether to go or not
2. Method of travel (e.g. by car, train, bus, air)
3. The departure time
5. The route
6. Whether the weather conditions would be seriously disruptive or not
7. Whether the cost would be appropriate for such a trip
8. Other (PLEASE SPECIFY)
Questions reported in Table 6 had two main purposes. The first was to assess whether, rather
than referring to their social contacts, respondents would have needed additional information
from transport operators to perform the choices they were asked to carry out. Secondly, we
wanted to understand whether respondents’ need for additional information, and suggestions,
had a role in other moments of the decision process than the final choice of the mode (or not
to leave). For example, we assumed that respondents may have needed assistance in building
their preference over certain attributes of the alternatives only, or to decide between two most

15
favourite alternatives rather than among the four. We wanted, therefore, to explore whether a
semi-compensatory model of choice (Martinez et al., 2009) would have been a better picture
to analyse this type of decisions in uncertain situations.
6. Preliminary results
Sample characteristics
Before presenting some preliminary results, we briefly introduce the sample of our survey.
This was composed of 2,027 respondents, with 1,037 from London and 990 from Glasgow,
with the following main socio-economic characteristics:
• Gender: 41% males, 59% females,
• Age: The average age of respondents is 43 years old
• Status: 50% of respondents are working full time, 11% part-time, 10% are retired, 6%
self-employed, 6% are in education
In terms of social networks, respondents could name up to 30 members of the social circle;
29 respondents named the full 30 members available, whilst 193 respondents did not list
anyone. This represented a total number of 13,022 alters within the sample. The most
frequent number of contacts listed was five (219 respondents, 10.8%). Around half of
respondents named between three and seven contacts (932 respondents, 46.0%); most
respondents had between one and eleven individuals within their social circle (1,572
respondents, 77.6%). The average number of contacts was 6.4. This is lower than all of the
five SNA datasets (across four countries) in the recent transport related paper by Kowald et
al. (2012), which had a reported average number of contacts between 11.9 and 23.9.
Moreover, while some of those surveys did not consider alters living with the respondents,
ours did. The difference in the average number of named contacts is certainly a function of
the internet-based data collection method used here as well as the length of the survey and
nationality of the respondents.
Choice cards data
Table 7 below presents the results of a simple Multinomial Logit Model (MNL) run on the
choice cards data (16,216 observations). In this stage of the analysis, only choice attributes
were included in the utility function and, to facilitate value of time calculations in this initial
stage of the analysis, the different time attributes (access, trip and egress time) were summed
to create a unique total time attribute. The attributes describing the probability of different
travel times entered the utility function as the expected value of schedule delay (ESDL) or
schedule early (ESDE), obtained by multiplying the different value of delay or earliness for
the respective probabilities. All attributes were made alternative-specific in order to detect
their effect on the different modes.

16
Table 7. Multinomial Logit (MNL) – N=16,216
Variable
Coefficient
T-value
AIR
***1.563
3.835
AIRDEPMO
-0.044
-0.819
AIRDEPAF
-0.020
-0.349
AIRDEPEV
-0.023
-0.369
AIRCOST
***-0.006
-18.121
TTIMEAIR
**-0.001
-3.52
AIRESDE
**0.029
2.549
AIRESDL
***-0.002
-5.74
TRAIN
***2.080
5.197
TRDEPMOR
***0.209
3.707
TRDEPAFT
0.023
0.448
TRDEPEVE
*0.100
1.905
TRCOST
***-0.007
-21.223
TTIMETR
***-0.003
-6.451
TRESDE
***-0.047
-3.539
TRESDL
***-0.001
-2.801
CARCOST
***-0.005
-9.633
TTIMECAR
-0.001
-1.009
CARESDE
0.001
0.086
CARESDL
***-0.001
-3.37
COACH
-0.068
-0.111
CODEPMOR
**0.216
2.142
CODEPAFT
0.182
1.913
CODEPEVE
0.095
0.965
COCOST
***-0.010
-8.926
TTIMECO
-0.001
-1.406
COESDE
**-0.056
-2.341
COESDL
-0.001
-1.578
NOTRAV
**-0.758
-2.091
Adjusted R2
0.031
Log-likelihood
-22747.24
Table 7 shows a number of significant coefficient, in particular the cost coefficients for all
modes. Departure time was generally not significant for air travellers. Alternative specific
constants for Air and Train are significant and positive, indicating preference for these modes
over car, while the Coach alternative specific constant is not significant. The significant and
negative sign associated with the coefficient of the alternative specific constant for No travel
means that respondents generally preferred the option of travelling. Additional analysis
considering both weather and context variables, as well as socioeconomic characteristic will
enable us to provide more details on this aspect.
Value of time figures are reported in the table below. This were calculated in accordance with
the scheduling model (Brownstone and Small, 2005, Li et al., 2010) using the coefficients in
Table 7.

17
Table 8. Value of time
Value of time savings
(Btotal time / Bcost)
£ per hour
Air
13.50
Train
25.07
Car
10.56
Coach
5.86
Table 8 shows that train travellers would be willing to pay £25 in excess of their fare in order
to save 1 hour over the whole of their trip (from home to destination).
The effect of weather on mode choice
Table 9 contains a simple comparison of mode choice (taken from the choice cards
responses) under fine weather and three different extreme weather conditions. This
information enables us to make (very) preliminary conclusions on the effect of weather on
mode choice under different weather conditions.
Table 9. Mode choice under different weather conditions
Fine
weather
Heavy
snow icy
road
Diff
Heavy
Rain
Diff
Heat
Diff
Air
37.493
27.466
-10.027
32.058
-5.435
39.032
1.539
Train
33.095
29.765
-3.33
38.27
5.175
32.072
-1.023
Car
13.304
5.661
-7.643
9.54
-3.764
11.429
-1.875
Coach
10.225
5.886
-4.339
8.486
-1.739
7.81
-2.415
No
travel
5.882
31.222
25.34
11.647
5.765
9.657
3.775
Total
100
100
100
100
Table 9 shows that air is the preferred method of travel from London to Glasgow or viceversa
if the weather conditions are fine. The same applies to the case of extreme heat (>30 degrees
centigrade). Train is instead the favourite method of travel in case of heavy snow and icy
road and heavy rain. In the former weather conditions, in 31% of choice cards respondents
would prefer not to travel (up from the about 6% that would not travel between the two cities
even in case of nice weather).
The post-choice cards questions
We briefly discuss now the way respondents replied to the post-choice cards questions we
discussed in the previous section.
In post-choice card question 1, respondents were asked whether they had considered what
other people similar to them (both within and outside their social circle) would have done in
similar situation before expressing their preferences for the different modes. 27% of
respondents said that they did consider the preferences of their social circle, while 9%

18
consider what other general people similar to them would have done. 48% stated they
decided without considering others.
Question 2 asked respondents to guess what people similar to them would have done if facing
the same choice. In 31% of the relevant choice cards respondents did not know. In 54% of
them, respondents said that those similar other would have chosen exactly like them. In the
case of Question 3, where respondents were asked about the preferences of the members of
their social circle, 15% said they were very confident about knowing their peers preference,
48% confident, 28% were neither confident nor unconfident. When asked whether they
would consider changing their mind, 84% said No, 14% I do not know, 2% said yes.
In Question 4 respondents were shown percentages of responses for the different modes (and
not travel option) in their neighbourhood. In the 79% of choice cards where this question was
appended, respondents said that they would have made no changes to their choice. 9% did not
know, and in 4% of cases respondents said they did not believe the shown percentages were
realistic of what people in their neighbourhood would do.
7. Discussions and conclusions
This paper has reported on the work in progress on a study of travel behaviour under extreme
weather conditions. The paper has in particular presented the methodology behind this
research, the experimental survey instruments and some preliminary results. We begin this
section by discussing a number of theoretical and practical issue, then we discuss our
preliminary results. Finally, we give more details on the next steps of our analysis, in
particular the econometric approach.
The first important theoretical issue worth discussing within our attempt to partially integrate
SNA and discrete choice experiments above lays in the fact that with the post-choice cards
questions we are asking respondents’ to report other people preferences. This seems to be an
important alteration to the traditional model of choice, and although this has already been
undertaken when analysts have employed the unitary model of household choice (see
Vermeulen, 2002 for a review of household models in economics), when the preference of
one only member of the household is deemed sufficient to represent the other members
preferences in terms of consumption and labour supply, it is certainly a challenging issue
both from the theoretical and empirical point of view. But another important question arises:
can we trust respondents when they report the preference of members of their networks? The
fact that these others are being indicated as part of their social networks should be enough for
that. Although there are cases in the literature where consumers did not seem to be able to
report the right preferences of other consumers, even when they were very close to (like
between husband and wives) (Beck et al., 2009, Beharry-Borg et al., 2009). The best
approach would have been to interview the members’ of the respondents’ social network,
with a snowball sampling techniques, in order to provide them with real information over
their peers preference facing the same choice. This was however not possible due to financial
constraints.
A look at our preliminary results reveals that in 27% of the choice cards respondents did
admit considering the preference of their social circles before choosing. Much lower was the
percentage of those who said they consider what general people similar to them, and the same
applied when respondents were shown the market share in their neighbourhoods, but this was

19
expectable. These results do however demonstrate some grounds for our hypotheses which
we will be able to further, and much more rigorously, confirm when carrying out deeper
analysis of the choice data.
It is anticipated that our econometric approach will develop as follows. Social dimensions
will then be explored using the methods used in the literature detailed above, for example by
Dugundji et al (Dugundji and Walker, 2005b, Dugundji and Gulyas, 2008) who consider both
spatial and social elements, and Goetzke (2008) who instead mainly considered spatial
elements. We also believe that there may be a potential in employing Latent Class in order to
explore social dimensions, especially if it is considered that the usual causality (I choose
therefore I belong) could be reversed (I belong therefore I choose) in a manner similar to the
application of Morey et al (Morey et al., 2006). However, the approaches mentioned above
have been mainly used in a revealed preference setting. There may be therefore issues in
applying them in a stated preference framework. Importantly, the analysis approach discussed
above does not consider the post-choice cards questions in the analysis. These questions
could be analysed independently, or could be part of a multiple discrete choices approach. At
the moment we are considering the issues involved in this.
The main econometric issue will certainly be the treatment of endogeneity. We have briefly
seen above the way this issue has been treated in the literature. While endogeneity is well
discussed in terms of continuous type of choice (Soetevent, 2006), this issues is less
discussed in discrete choice applications, and the recent paper by Walker et al (2011) in
particular, it is certainly a welcome addition. Whether the approaches used in the literature so
far will be applicable to our dataset remains, however, an unanswered question at his stage.
Finally, we note that our preliminary results do seem to suggest that there is an effect of
weather on mode choice, and further analysis will make our analysis able to detect different
types of social dimensions within travellers’ response to uncertainty caused by extreme
weather conditions to the possible impact of social influence on their choices.
Acknowledgements
The authors wish to thanks the participants to the Social Network Analysis and Transport
Workshops, held in Nottingham, 2-3 March 2010 and Manchester 11-12 January 2011, for
the useful discussions that have inspired this paper. We also thank the Engineering and
Physical Sciences Research Council (EPSRC) and the Economic and Social Research
Council (ESRC) for their financial support of the ‘FUTURENET’ project.

20
Reference list
ADJEMIAN, M., LIN, C. Y. C. & WILLIAMS, J. 2010. Estimating spatial interdependence
in automobile type choice with survey data. Transportation Research Part A, 44, 661-
675.
AKERLOF, G. A. 1997. Social distance and social decisions. Econometrica, 65, 1005-1027.
ANSELIN, L. 1988. Spatial Econometrics: Method and Models, Utrecht, Kluwer.
ARENTZE, T. & TIMMERMANS, H. 2008. Social networks, social interactions and
activity-travel behaviour. Environment and Planning B, 35, 1012-1027.
AVINERI, E. & PRASHKER, J. N. 2006. The impact of travel time information on travellers
learning under uncertainty. Transportation, 33, 393-408.
AXHAUSEN, K. W. 2008. Social networks, mobility biographies and travel: The survey
challenges. Environment and Planning B, 35, 981-996.
AXSEN, J., MOUNTAIN, D. C. & JACCARD, M. 2009. Combining stated and revealed
choice research to simulate the neighbor effect: the case of hybrid-electric vehicles.
Resource and Energy Economics, 31, 221-238.
BECK, M. J., ROSE, J. M. & HENSHER, D. A. The accuracy of proxy responses in a stated
choice setting. 12th International Conference on Travel Behaviour Research, 2009
Jaipur, India, 13-18 December.
BEHARRY-BORG, N., HENSHER, D. A. & SCARPA, R. 2009. An analytical framework
for joint vs separate decisions by couples in choice experiments: the case of coastal
water quality in Tobago. Environmental and Resource Economics, 43, 95-117.
BHAT, C. R. & SENER, I. N. 2009. A copula-based closed-form binary logit choice model
for accommodating spatial correlations across observational units. Journal of
Geographical Systems, 11, 243-272.
BIDART, C. & DEGENNE, A. 2005. Introduction: the dynamics of personal networks.
Social Networks, 27, 283-287.
BLIEMER, M. C. J. & ROSE, J. M. 2011. Experimental design influences on stated choice
outputs: An empirical study in air travel choice. Transportation Research Part A:
Policy and Practice, 45, 63-79.
BONSALL, P. 2004. Traveller behaviour: decision-making in an unpredictable world.
Intelligent Transportation Systems, 8, 45-60.
BRAMOULLE, Y., DJEBBARI, H. & FORTIN, B. 2009. Identification of peer effects
through social networks. Journal of Econometrics, 150, 41-55.
BROCK, W. A. & DURLAUF, S. N. 2001. Discrete choice with social interactions. Review
of Economic Studies, 68, 235-260.
BROCK, W. A. & DURLAUF, S. N. 2002. A multinomial-choice model of neighborhood
effects. American Economic Review - Papers and Proceedings, 92, 298-303.
BROWNSTONE, D. & SMALL, K. A. 2005. Valuing time and reliability: assessing the
evidence from road pricing demonstrations. Transportation Research Part A, 39, 279-
293.
CALVO-ARMENGOL, A. 2009. Peer effects and social networks in education. Review of
Economic Studies, 76, 1239-1267.
CARRASCO, J. A., HOGAN, B., WELLMAN, B. & MILLER, E. J. 2008. Collecting social
network data to study social activity-travel behaviour: an egocentric approach.
Environment and Planning B, 35, 961-980.
CARRASCO, J. A. & MILLER, E. J. 2009. The social dimension in action: a multilevel,
personal networks model of social activity frequency between individual.
Transportation Research Part A, 43, 90-104.

21
CONT, R. & LOWE, M. 2010. Social distance, heterogeneity and social interactions. Journal
of Mathematical Economics, 46, 572-590.
COOLS, M., MOONS, E., CREEMERS, L. & WETS, G. 2010. Changes in travel behavior in
response to weather conditions: whether type of weather and trip purpose matter.
Transportation Research Record: Journal of the Transportation Research Board,
2157, 22-28.
DE PALMA, A. & ROCHAT, D. 1999. Understanding individual travel decisions: results
from a commuters survey in Geneva. Transportation, 26, 263-281.
DEGENNE, A. & LEBAUX, M. O. 2005. The dynamics of personal networks at the time of
entry into adult life. Social Networks, 27, 337-358.
DENANT-BOEMON, L. & PETIOT, R. 2003. Information value and sequential decision-
making in a transport setting: an experimental study. Transportation Research Part B,
37, 365-368.
DUGUNDJI, E. & WALKER, J. Discrete choice with social and spatial network
interdependencies: an empirical examples using mixed GEV models with field and
'panel' effects. 84th Annual Meeting of the Transportation Research Board, 2005a
Washington, D.C.
DUGUNDJI, E. R. & GULYAS, L. 2008. Sociodynamic discrete choice on networks in
space: impacts of agent heterogeneity on emergent outcomes. Environment and
Planning B, 35, 1028-1054.
DUGUNDJI, E. R. & WALKER, J. L. 2005b. Discrete choice with social and spatial network
interdependencies: an empirical example using mixed generalized extreme value
model with field and panel effects. Transportation Research Record: Journal of the
Transportation Research Board, 1921, 70-78.
EVERETT, M. 2006. A graph-theoretic framework for classifying centrality measures. Social
Networks, 28, 466-484.
FARBER, S. & PAEZ, A. 2009. My car, my friends, and me: a preliminary analysis of
automobility and social activity participation. Journal of Transport Geography, 17,
216-225.
FELD, S. L., SUITOR, J. & GARTNER HOEGH, G. 2007. Describing changes in personal
networks over time. Field Methods, 19, 218-234.
FOSGERAU, M. & KARLSTROM, A. 2010. The value of reliability. Transportation
Research Part B, 44, 39-49.
GAKER, D., ZHENG, Y. & WALKER, J. 2010. Experimental economics in transportation.
Focus on social influences and provision of information. Transportation Research
Record: Journal of the Transportation Research Board, 2156, 47-55.
GOETZKE, F. 2008. Network effects in public transit use: evidence from a spatially
autoregressive mode choice for New York. Urban Studies, 45, 407-417.
GOETZKE, F. & RAVE, T. 2011. Bycicle use in Germany: explaining differences between
municipalities with social network effects. Urban Studies, 48, 427-437.
HAYAKAWA, H. 2000. Bounded rationality, social and cultural norms, and interdependence
via reference groups. Journal of Economic Behaviour & Organization, 43, 1-34.
HOGAN, B., CARRASCO, J. A. & WELLMAN, B. 2007. Visualising personal networks:
working with participant-aided sociograms. Field Methods, 19, 116-144.
IOANNIDES, Y. A. & ZABEL, J. E. 2008. Interactions, neighborhood selection and housing
demand. Journal of Urban Economics, 63, 229-252.
IOANNIDES, Y. M. 2006. Topologies of social interactions. Economic Theory, 28, 559-584.
JAROSZWESKI, D., CHAPMAN, L. & PETTS, J. 2010. Assessing the potential impact of
climate change on transportation: the need for an interdisciplinary approach. Journal
of Transport Geography, 18, 331-335.

22
KOETSE, M. J. & RIETVELD, P. 2012. Adaptation to Climate Change in the Transport
Sector. Transport Reviews, 32, 267-286.
KOWALD, M., FREI, A., HACKNEY, J., ILLENBERGER, J. & AXHAUSEN, K. W. The
influence of social contacts on leisure travel: a snowball sample of personal networks.
12th International Conference on Travel Behaviour Research, 2009 Jaipur, India, 13-
18 December.
KOWALD, M., VAN DEN BERG, P., FREI, A., CARRASCO, J. A., ARENTZE, T.,
AXHAUSEN, K., MOK, D., TIMMERMANS, H. & WELLMAN, B. 2012. The
Spatiality of Personal Networks in Four Countries: A Comparative Study. 13th
International Conference on Travel Behaviour Research. Toronto, 15-20 July 2012.
KRAUTH, B. 2006. Social interactions in small groups. Canadian Journal of Economics, 39,
414-433.
KUWANO, M., FUJIWARA, A. & ZHANG, J. A discrete choice model with dynamic social
interactions: framework and application to household choice of vehicle type. 11th
World Conference on Transport Research, 2007 University of California, Berkeley.
LARSEN, J., URRY, J. & AXHAUSEN, K. W. 2009. Social networks and future mobilities.
Report to the Department of Transport, London: University of Lancaster and ETH
Zurich.
LI, Z., HENSHER, D. A. & ROSE, J. M. 2010. Willingness to pay for travel time reliability
in passenger transport: A review and some new empirical evidence. Transportation
Research Part E: Logistics and Transportation Review, 46, 384-403.
MANSKI, C. F. 1993. Identification of endogenous social effects: the reflection problem.
Review of Economic Studies, 60, 531-542.
MANSKI, C. F. 2000. Economic analysis of social interactions. Journal of Economic
Perspectives, 14, 115-136.
MARIN, A. & HAMPTON, K. N. 2007. Simplifying the personal network name generator:
alternatives to traditional multiple and single name generators. Field Methods, 19,
163-193.
MARTINEZ, F., AGUILA, F. & HURTUBIA, R. 2009. The constrained multinomial logit: a
semi-compensatory choice model. Transportation Research Part B, 43, 365-377.
MAU, P., EYZAGUIRRE, J., JACCARD, M., COLLINS-DODD, C. & TIEDEMANN, K.
2008. The 'neighbor effect': Simulating dynamics in consumer preferences for new
vehicle technologies. Ecological Economics, 68, 504-516.
MCFADDEN, D. 1974. Conditional logit analysis of qualitative choice behaviour. In:
ZAREMBKA, P. (ed.) Frontiers of Econometrics. New York: Academic Press.
MCFADDEN, D. 2001. Economic choices. The American Economic Review, 91, 351-378.
MOREY, E., THACHER, J. & BREFFLE, W. 2006. Using angler characteristics and
attitudinal data to identify environmental preference classes: a Latent-Class model.
Environmental and Resource Economics, 34, 91-115.
PAEZ, A. & SCOTT, D. M. 2007. Social influence on travel behaviour: a simulation example
of the decision to telecommute. Environment and Planning B, 39, 647-665.
PAEZ, A., SCOTT, D. M. & VOLZ, E. 2008. A discrete choice approach to modelling social
influence on individual decision making. Environment and Planning B, 35, 1055-
1069.
SANEINEJAD, S., ROORDA, M. J. & KENNEDY, C. 2012. Modelling the impact of
weather conditions on active transportation travel behaviour. Transportation Research
Part D, 17, 129-137.
SARTZETAKIS, E. S. & TSIGARIS, P. 2005. Environmental externalities in the presence of
network effects: adoption of low emission technologies in the automobile market.
Journal of Regulatory Economics, 28, 309-326.

23
SCHWANEN, T. 2008. Managing uncertain arrival times through sociomaterial associations.
Environment and Planning B, 35, 997-1011.
SMIRNOV, O. A. 2010. Modelling spatial discrete choice. Regional Science and Urban
Economics, 40, 292-298.
SOETEVENT, A. R. 2006. Empirics of the identification of social interactions; an evaluation
of the approaches and their results. Journal of Economic Surveys, 20, 193-228.
SOETEVENT, A. R. & KOOREMAN, P. 2007. A discrete-choice model with social
interactions: with an application to high school teen behaviour. Journal of Applied
Econometrics, 22, 599-624.
SUNITIYOSO, Y., AVINERI, E. & CHATTERJEE, K. 2011. The effects of social
interactions on travel behaviour: an exploratory study using a laboratory experiment.
Transportation Research Part A, 45, 332-344.
THOGERSEN, J. 2006. Understanding repetitive travel mode choices in a stable context: a
panel study approach. Transportation Research Part A, 40, 621-638.
VAN DER BERG, P., ARENTZE, T. & TIMMERMANS, H. 2011. A multilevel path of
contact frequency between social network members. Journal of Geographical
Systems, DOI 10.1007/s10109-010-0138-0.
VERMEULEN, F. 2002. Collective household models: principles and main results. Journal
of Economic Surveys, 16, 533-564.
WALKER, J. L., EHLERS, E., BANERJEE, I. & DUGUNDJI, E. R. 2011. Correcting for
endogeneity in behavioral choice models with social influence variables.
Transportation Research Part A, 45, 362-374.
WANG, X. & KOCKELMAN, K. M. 2009. Bayesian inference for ordered response data
with a dynamic spatial-ordered probit model. Journal of Regional Science, 49, 877-
913.
ZANELLA, G. 2007. Discrete choice with social interactions and endogenous memberships.
Journal of the European Economic Association, 5, 122-153.