Travelers’ segmentation based on multimodality behaviors and attitudes
Marco Diana*
Dipartimento di Idraulica, Trasporti e Infrastrutture Civili
Politecnico di Torino
Corso Duca degli Abruzzi, 24 10129 Torino – ITALY
Tel. + 39 011 564 5638 – Fax. + 39 011 564 5699
marco. diana@ polito. it
Patricia L. Mokhtarian
Department of Civil and Environmental Engineering, Institute of Transportation Studies
University of California, Davis
One Shields Avenue Davis, CA 95616 – USA
Tel. + 1 530 752 7062 – Fax. + 1 530 752 7872
plmokhtarian@ ucdavis. edu
June 2008
Abstract
Using data collected from 164 French employees of a transportation institute and 1904 residents
of the U. S. San Francisco Bay Area, we operationalize a segmentation of mobility patterns based
on objective, subjective, and desired amounts of mobility by various modes and overall. We
define a multimodality index from basic concepts of information theory, and we especially focus
on the degree of multimodality in an individual’s current modal mix and desired changes to that
mix. The clusters that result showed some similarities and some differences across countries,
where the latter are likely due to disparities in the sampling strategies and in the land
use/ transportation/ cultural milieux. In both cases, however, the clusters have useful policy
implications, enabling us, for example, to distinguish car users who might be inclined to reduce
car use and increase transit use from those who are largely content with their current modal
baskets.
Keywords
cluster analysis, desired mobility, market segmentation, multimodality, public transport
* Corresponding author
2
1 Introduction
Transport systems are more and more confronted with the need to rationalize the use of
resources and minimize environmental impacts, while maintaining or even improving current
levels of service to support economic development. Encouraging a more consistent use of public
transport is among the favorite strategies to achieve this goal, but results seem to lag behind both
research and policy efforts.
One of the central problems seems to be the possibility of actually substituting trips using a
motorized individual transport mode with trips using a transit service. Several studies have
explored this mechanism with different methodologies and in different disciplinary contexts.
From the point of view of individual differences, one of the most popular techniques consists of
identifying groups of people who have different attitudes and/ or behaviors regarding travel, in
order to define a set of policy actions ( rules enforcement, economic incentives, awareness
campaigns etc.) that is targeted for each group. This user segmentation technique is
commonplace in the marketing literature and several examples can be found in the transport
sector, where segments are defined on the basis of various socioeconomic or personal
characteristics ( TRB, 1977; Dobson and Tischer, 1978; Tardiff, 1979; Gensch and Torres, 1980;
Jensen, 1999; Outwater, Castleberry, Shiftan, Ben- Akiva, Zhou, and Kuppam, 2003; Elgar and
Bekhor, 2004; Anable, 2005; Beirão and Cabral, 2007). An example of how to customize
marketing in the bus industry is given by Beale and Bonsall ( 2007).
Beyond individual characteristics, the possibility of diverting trips is also influenced by
physical constraints, pertaining for example to the pattern of activities across different locations
and the related structure of the transport network. Hence an alternative methodology is to
perform an analysis where the actual trips, more than the persons, are grouped on the basis of
3
their suitability for modal transfer. Beyond the pioneering study of Nash and Hille ( 1968), this
kind of analysis, which perhaps lies more in the purview of the transport engineering discipline,
is proposed for example by Bovy ( 1999), Hensher and Reyes ( 2000) and Massot, Armoogum,
Bonnel and Caubel ( 2006).
The present research offers a contribution that is complementary to the above mentioned
segmentation analyses, by focusing on those aspects that are more linked to the individual degree
of acquaintance with different transport modes. In this respect, Anable and Gatersleben ( 2005)
have shown the importance of the “ predictability” of a transport mode, which in their research
was rated only slightly less important than other instrumental factors such as cost, convenience
and flexibility, at least for some kinds of trips. Predictability is surely influenced by the
performance of the mode itself, but also by the familiarity that the users have with it. The
premise of this study is that familiarity with different transport means could be expressed
through measures of the real and the perceived relative intensity of use of different modes. We
hypothesize that this in turn could be related to the desire to change one’s behavior toward a
different balance of levels of mobility across the various transport means ( which we call the
“ modal basket”). Our point of view is hence somewhat related to the study of the relationship
between mode use habits and mode choices, that is the object of intensive research ( see for
example Verplanken, Aarts, van Knippenberg and van Knippenberg, 1994; Verplanken, Aarts
and van Knippenberg, 1997; Aarts and Dijksterhuis, 2000; Gärling, Fujii and Boe, 2001; Garvill,
Marell and Nordlund, 2003; Bamberg, Ajzen and Schmidt, 2003; Thøgersen, 2006).
Several interesting types of individuals can be postulated a priori. For example, among those
currently using a variety of modes, some could desire maintaining a similar balance, some could
desire to intensify their use of environmentally benign modes ( transit, non- motorized) and
diminish their use of the automobile, and some could desire the opposite. Conversely, some
4
monomodalists may wish to maintain the primacy of their dominant mode, while others may
wish to diversify their modal basket. By focusing on the basket of mobility levels rather than on
separate mode- specific ones, we would like to ascertain its usefulness in assessing the desire to
alter one’s modal basket. This would be of help in giving a preliminary indication of the
influence of self- related factors ( specifically, stated desires) on the switching potential of field
interventions, to be followed by a more thorough evaluation through the consideration of other
psychological dimensions such as attitudes, opinions or beliefs. The relevance of the study of
multimodal behaviors from a policy viewpoint is well shown by Nobis ( 2007), who offers also a
socioeconomic and demographic characterisation of those segments of the population that are
more likely to use different transport modes for their trips.
The objective of our work is then the definition of a new market segmentation approach,
complementary to the other above- mentioned transport market studies, which is based on the
joint consideration of one’s actual and perceived mobility levels by the different transport modes,
as well as the desire to change them in order to modify the modal basket. This goal is achieved
through a comparative study of two different travel behavior survey datasets, one involving a
random sample of individuals living in selected neighborhoods of the San Francisco Bay Area,
and the other comprising employees of the French national transportation research institute
INRETS. In both cases, the actual use ( in terms of the distance traveled or trip frequency) of
different transport modes was measured ( via self- reports), as well as the subjective assessment of
the quantity of travel performed and the desire to increase or reduce the use of any transport
means. These field studies are described in section 2, along with the definitions of the variables
that will be used for the market segmentation.
Given the potentially high number of modes to be considered in our basket, it is important to
try to use synthetic indicators in order to have a manageable number of variables for subsequent
5
analysis. With this objective in mind, in section 3 several kinds of “ multimodality indexes” are
proposed, on the basis of measures of objective, subjective and relative desired mobility levels,
and their usefulness is discussed. These indexes are then used in several cluster analyses that
allow for grouping the survey respondents on the basis of their desire to alter their modal basket,
in relation to their real and perceived intensity of utilization of different transport means.
Repeating the same kind of analysis on somewhat comparable datasets collected in widely
different environments allows for greater insights on the matter, helping to understand the
general validity of the analysis outcomes. In section 4, we present the most interesting cluster
analyses that we performed, including an examination of the socioeconomic characteristics of the
U. S. clusters. We draw some conclusions from our exercise in section 5.
2 Experimental settings and variable definition
The present research makes use of two datasets that are the outcome of two studies,
independently conducted in radically different contexts. The first study was conducted in 1998
with a postal survey sent to 8000 residents in three different communities in the San Francisco
Bay Area, which resulted in 1904 useful observations. The central purpose of the study was to
analyze attitudes toward the act of travelling, and the survey obtained data on those attitudes
( both general, and specific to various types of travel), as well as personality traits and lifestyle
orientations, several measures of mobility, and conventional sociodemographic characteristics.
The second study was an Internet survey that was administered to INRETS staff in 2004.
INRETS is the French National Institute for Transport and Safety Research, with two larger
premises in the Paris and Lyon metropolitan areas and two smaller branches near Lille and
Marseille. 164 cases were considered for analysis out of about 550 workers and students that
6
were contacted. The survey goals and contents were somewhat related to those of the U. S.
survey, but the questions were focused on a specific trip that was randomly selected among those
reported in the introductory part of the survey. This approach should facilitate the quantitative
consideration of elements such as attitudes and personality traits within a modelling framework
( Diana, 2008), along with the customary trip attributes.
The reader is referred to Mokhtarian and Salomon ( 2001) for a more detailed presentation of
the U. S. survey context, purposes and contents, and to Diana ( 2005, 2006) for a thorough
description of the French survey. In the following we focus instead on those variables that were
built from the collected data and that we consider in our subsequent analysis. We select from the
two datasets those variables that describe the actual, perceived and desired mobility levels of the
respondents by different transport modes. Following the definitions given respectively by
Mokhtarian, Salomon and Redmond ( 2001), Collantes and Mokhtarian ( 2007) and Choo,
Collantes and Mokhtarian ( 2005), we will name these three entities Objective Mobility ( OM),
Subjective Mobility ( SM) and Relative Desired Mobility ( RDM).
2.1 Variables measuring Objective Mobility levels
The two datasets differ both in the way they measure those mobility levels and in the set of
transport modes they consider. Concerning OM, people were asked in the U. S. survey for their
typical weekly mileage by each of four transport modes, considering only trips up to 100 miles
one way ( long- distance OM was measured differently and is not considered in this study). The
four modes were: ( 1) driver or passenger in any personal vehicle, ( 2) bus, ( 3) rail and ( 4)
walking, jogging, cycling. The French survey asked for the mean frequency of utilisation of 10
ground transport modes ( bicycle, motorbike, car driver, car passenger, bus, tram, metro, taxi,
7
suburban train and long- distance train) over the previous 12 months through 5- point ordinal
variables ( never, sporadically, 1- 3 times a month, 1- 2 times a week, 3 times a week or more).
For our purposes it is important to try to measure OM in such a way that it can reflect the
degree of familiarity of the respondent with the different means. From the point of view of an
individual, his/ her familiarity with a given mode is probably more related to the amount of time
s/ he spends in using it than to the distance travelled. We would say that, for example, a person
using a bike two hours per day is more acquainted with bikes than a car driver using a car two
hours a week is with his car, although the weekly mileage could be comparable given the
different mean speeds of the two modes. Hence we estimate from the U. S. data the number of
weekly hours spent in each of the four considered modes through a best guess of their typical
speeds. Concerning the French data, we assess instead the number of monthly trips in each of the
10 considered modes from the reported frequencies. Those 14 variables are named with the
prefix “ OM_” followed by a label indicating the mode. We list them in the second column of
block “ A” of Tables 1 ( for the French dataset) and 2 ( for the U. S. dataset).
In the following we focus on the modal basket equilibrium between cars and public transport
modes, so that we need to condense information regarding objective mobility levels of different
modes. Keeping in mind that we look for an OM measure that is a proxy for the familiarity with
the corresponding mode, we consider in the French dataset the sum of OM_ CARDR and
OM_ CARPAX as a measure of the mobility level of the respondent with cars, and we name the
new variable OM_ CAR. We similarly consider the sum of the mobility levels by bus, trams and
metro and we name it OM_ PT. For the U. S. dataset we already have a single variable for car use
( OM_ CAR) and we define OM_ PT as the sum of OM_ BUS and OM_ RAIL, keeping in mind
that long- distance trips were not considered. Finally, OM_ GLOB is the sum across all modes of
the number of weekly hours ( for the U. S. case) or of monthly trips ( for the French case) spent in
8
each mode. These derived variables are reported in the second column of block “ B” of Tables 1
and 2.
2.2 Variables measuring Subjective and Relative Desired Mobility levels
Subjective Mobility and Relative Desired Mobility are measured in both datasets with
ordinal variables. Concerning SM, it is important to stress the added value of considering it along
with the OM measures in our cluster definition. These entities are not to be seen as two different
ways to measure the same thing, since they are complementary in uncovering different aspects of
mobility behaviours. Questions related to OM implied a more analytical review of the real
behaviours of the respondents, and elicited a rational process to evaluate it in quantitative terms.
SM questions have instead to do with feelings, so that mobility is measured on the basis of a
personal norm regarding “ ideal” mobility levels through a semantic scale.
SM scales range from “ I feel I do not travel at all” to “ I feel I travel a lot” ( by that particular
mode), whereas RDM scales go from “ I would like to travel much less” to “ I would like to travel
much more”, passing through the neutral point “ I would like to travel the same amount as now”
( with that particular mode). The U. S. scales have 5 points, whereas the French ones have 10
points when SM is measured and 11 points when RDM is measured.
The considered modes are the same as in the OM analysis, so that as for the OM measures
we will have 14 SM variables and 14 RDM variables that are named following the same
conventions. The 20 variables from the French dataset are reported in the third and fourth
columns of block “ A” of Table 1, whereas the 8 variables from the U. S. dataset are listed in the
third and fourth columns of block “ A” of Table 2.
9
Beyond those variables that are directly taken from the datasets, for our purposes new SM
and RDM aggregate measures for car and public transport use need to be defined, corresponding
to OM_ CAR and OM_ PT in the preceding subsection. However in this case we cannot simply
sum SM and RDM values, since they are ordinal measures. Hence we use here a method that has
been proposed by Wittkowski, Lee, Nussbaum, Chamian and Krueger ( 2004). According to this
method, it is possible to combine n ordinal measures by defining a partial order among the
observations. That is, we say that the observation A is ” greater” than observation B if A is
greater in at least one of the n ordinal measures and smaller in none. Then the combined ordinal
measure for the observation A is given by the number of observations that are smaller than A
minus the number of observations that are greater than A. The interested reader is referred to
Wittkowski et al. ( 2004) for more details and for the study of the statistical properties of the
combined measures. In parallel with what has been done for the OM case, we can thus define for
the French dataset SM_ CAR combining SM_ CARDR and SM_ CARPAX and SM_ PT
combining SM_ BUS, SM_ TRAM and SM_ METRO, and similarly for RDM_ CAR and
RDM_ PT. In the U. S. dataset we will define SM_ PT and RDM_ PT just considering the bus and
the rail modes.
Finally, we compute the global SM and RDM mobility levels across all modes. In the case
of the French dataset, SM_ GLOB and RDM_ GLOB are computed assuming that the ordinal
measures pertaining to the 10 modes under consideration can be combined in successive steps,
according to the hierarchy among these modes that is depicted in figure 1. In doing this we apply
an extension of the above methodology that is presented in Diana, Song and Wittkowski ( 2008).
In short, by taking into account the hierarchical structure of the data it is possible to decrease the
number of pairwise orderings that are ambiguous ( in the sense that the above described partial
order cannot be defined), so that the resulting combined measures are much more informative, as
10
shown in that paper. For the U. S. dataset, we combine the measures pertaining to buses and
trains to derive the measures for the composite “ transit” mode, and then we combine “ walk”,
“ car” and “ transit” to obtain SM_ GLOB and RDM_ GLOB ( see figure 2).
Figures 1 and 2
The composite measures for SM and RDM, as well as the global SM and RDM mobility
levels, are reported in the third and fourth columns of block B of Tables 1 and 2 for the French
and the U. S. datasets respectively.
Tables 1 and 2
3 Multimodality indexes
We can see from tables 1 and 2 that the number of considered variables is quite high. As a
preliminary step, we propose then a methodology that tries to express the same information with
some more synthetic indicators.
In order to define such indicators, we must keep in mind that from our perspective it is not
so important to separately analyse mobility levels with respect to single modes, but to consider
them jointly. Hence, beyond the global mobility level of individuals, we would like to define an
index that measures the degree of multimodality in the individual travel behaviors, i. e. the
“ product differentiation” inside the modal baskets.
We accomplished this by borrowing some ideas from information theory, a branch of
probability theory that allows for measuring the amount of information that is contained in a
message, and then adapting those ideas to our framework. According to this discipline, given a
11
probabilistic experiment having n possible results, with the respective probabilities p1, …, pn
such that Σ=
=
n
i
i p
1
1, the amount of information Hn supplied by such an experiment is given by
the amount of uncertainty that is associated with the experiment itself. Hence it can be measured
by the entropy formula ( Shannon, 1948):
Σ Σ
= =
= − =
n
i i
i
n
i
n i i p
H p p p
1 1
log log 1 . ( 1)
The basis of the logarithm can be arbitrary, although it is customary to take it as 2. In this way,
the amount of information in an experiment with two equally possible outcomes is equal to 1.
Since such a case represents the smallest amount of information ( and hence the largest amount of
entropy) contained in an experiment with two outcomes, this measurement unit is then named
“ bit”, and the information contained in any other such experiment can be measured in bits. Rényi
( 1970, p. 579) proposed a useful generalisation of this measure for incomplete probability
distributions, i. e. when Σ=
<
n
i
i p
1
1:
Σ
Σ
=
= = n
i
i
n
i i
i
n
p
p
p
H
1
1
log 1
' , ( 2)
which can be seen as the probability- weighted average of the contribution to entropy of each
included outcome.
Now, let us come back to our problem. From the outside, we can see the mode choice of an
individual for a generic “ average trip” as a probabilistic experiment, each mode having a given
probability of choice. If we suppose that the characteristics of such a trip are unknown, then the
best approximation to quantifying these probabilities is to consider the relative intensities of use
12
of each mode, which in our case can be computed from the OM variables presented in the
previous section. The entropy of such an experiment is a measure of the uncertainty of the
outcome; the more evenly distributed the choice probabilities of the modes are, the greater the
entropy Hn would be, thus indicating that the final choice is less easily predictable because the
individual does not predominantly use any mode. Hence, the Shannon entropy of such an
experiment can be seen as a measure of the multimodal behavior of the individual, irrespective of
his/ her mobility level. Moreover, by using the Rényi generalisation it is possible not to consider
all the modes that the individual actually uses, but to arbitrarily define the modal basket so that it
better fits our research purposes.
Once having defined the set of n modes under investigation and the corresponding
frequencies of utilisation f1, …, fn ( for example, number of trips or number of hours spent
onboard in a week or a year, as in our OM variables defined in section 2), we compute a Pure
multimodality Index OM_ PI from the above entropy formulas ( 1) and ( 2):
Σ
Σ
= Σ
=
= ⎟ ⎟ ⎟ ⎟ ⎟
⎠
⎞
⎜ ⎜ ⎜ ⎜ ⎜
⎝
⎛
⎟ ⎟ ⎟ ⎟ ⎟
⎠
⎞
⎜ ⎜ ⎜ ⎜ ⎜
⎝
⎛
=
n
i i
n
j
j
n n
j
j
i
f
f
f
f
OM PI
1
1
1
_ log . ( 3)
We take the number of considered modes as the base of the logarithm, so that OM_ PI lies
between 0 and 1. If a given fi is 0, then the addend of mode i in the above summation is set to
zero. When OM_ PI = 0 the individual uses only one mode among those being considered ( or
he/ she does not use any mode at all), whereas when OM_ PI = 1 the individual uses all these
modes with the same intensity.
It is interesting to note that OM_ PI is not influenced by the mean mobility level of the
individual, since only relative frequencies enter in ( 3). One might wonder if this is a completely
13
desirable characteristic in our framework, where we would like to use this index as an indicator
of the degree of acquaintance with different transport modes. Thus we introduce a variant of
OM_ PI that is also sensitive to the mean mobility level of individuals. For this, we define M as
the absolute maximum reported frequency of utilisation of any mode across all observations in
the sample, and therefore nM as the potential maximum total frequency across all considered
modes. Then, we could define a Mobility- level- sensitive multimodality Index, OM_ MI, as
follows:
Σ Σ Σ
= = = ⎟ ⎟
⎠
⎞
⎜ ⎜
⎝
⎛
⎟ ⎟ ⎠
⎞
⎜ ⎜
⎝
⎛
⎟ ⎟⎠
⎞
⎜ ⎜⎝
⎛
+ = ⎟ ⎟
⎠
⎞
⎜ ⎜
⎝
⎛
⎟ ⎟⎠
⎞
⎜ ⎜⎝
⎛
+ = ⎟ ⎟
⎠
⎞
⎜ ⎜
⎝
⎛
⎟ ⎟⎠
⎞
⎜ ⎜⎝
⎛
=
n
i i
n
i
n
i
n
i i
n n
i
i
n
i
f
M
nM
f
f
n M
nM
f
f
nM
nM
f
OM MI
1 1 1
_ log log log 1 log .
However, experimentation indicated that taking the natural logarithm seemed to give the best
results in terms of an optimal blend between sensitivity to the mean mobility level and sensitivity
to the differences of mobility levels across modes, so we define:
Σ=
⎟ ⎟ ⎠
⎞
⎜ ⎜
⎝
⎛
⎟ ⎟ ⎠
⎞
⎜ ⎜
⎝
⎛
⎟ ⎟⎠
⎞
⎜ ⎜⎝
⎛
= +
n
i i
i
f
M
nM
OM MI f
1
_ 1 ln . ( 4)
OM_ MI also lies between 0 and 1, with greater values when the travel behavior is more
multimodal. As before, if a given fi is 0, then the addend of mode i in the above summation is set
to zero. Moreover, if M = 0 then OM_ MI = 0 ( a theoretical circumstance, but with the logical
outcome). OM_ MI also increases with the mean mobility level, and OM_ MI = 1 when an
individual uses all the modes in the basket with frequency equal to M.
To have a more intuitive representation of the differences between the two indexes, we plot
in Figures 3 and 4 the functions representing OM_ PI and OM_ MI, respectively, when only two
modes are considered and M is equal to 15. Note that when f1 = f2 ≠ 0, OM_ PI always equals 1
( maximum entropy for equal shares, regardless of level), whereas OM_ MI only increases to 1 as
f1 = f2 approaches M. Note also the comparison between OM_ PI( 0, 1) and OM_ PI( 0, 15) versus
14
OM_ MI( 0, 1) and OM_ MI( 0, 15). OM_ PI( 0, 1) and OM_ PI( 0, 15) are both equal to zero
( complete monomodality = minimum entropy, regardless of level), whereas the two OM_ MI
measures differ. On the one hand, for OM_ MI( 0, 15) the shares are more disparate ( carry more
information, i. e. less entropy) than for OM_ MI( 0, 1), but on the other hand, the mobility level of
( 0, 15) is much greater than that of ( 0, 1). The outcome of balancing those two factors is that
OM_ MI( 0, 15) is greater than OM_ MI( 0, 1).
Figures 3 and 4
Up till now, the frequencies that entered in equations ( 3) and ( 4) are those measuring the
Objective Mobility levels of individuals. The final step is to generalise the concepts of Pure and
Mobility- level- sensitive multimodality indexes, by considering also “ perceived” and “ relative
desired” frequencies, according to the variable definitions given in section 2. These frequencies
are measured through the semantic ordinal scales there introduced, but here we treat those
ordinal measures as ratio- scaled. Doing so is an approximation, but one that is well- attested and
seemingly robust when there are five or more categories to the scale ( Bollen and Barb, 1981).
Multimodality indexes based on SM measures, SM_ PI and SM_ MI, represent the balance in
current usage across the considered modes as it is perceived by the respondent, whereas RDM_ PI
and RDM_ MI are a measure of the degree of multimodality of the desired changes in the levels
of utilisation of different modes. Hence we point out that RDM indexes do not measure the
change in the degree of multimodality in the basket itself, should the desired changes take place,
since RDM measures are relative to the respondents’ perceptions of current mobility levels and
cannot give absolute estimations of multimodality changes if considered alone. To sum up, in the
15
following we consider 6 different multimodality indexes, i. e. Pure and Mobility- level- sensitive
multimodality indexes with OM, SM and RDM measures.
4 Cluster analyses
Several cluster analyses have been performed with the variables defined in sections 2 and 3.
In the following we present and comment on two of the most interesting types of clusters that
were investigated. The type “ A” cluster solution considers 9 variables, namely the objective,
subjective and relative desired mobility levels in terms of mean values for ( 1) all ground
transportation modes, ( 2) car, and ( 3) public transport. Hence it focuses on the general balance
between car and transit within the actual modal basket, as well as respondents’ perceptions and
desired direction and magnitude of change of their baskets. The type “ B” cluster solution
involves 36 variables for the French case and 18 for the U. S. one, i. e. the mean values for every
considered mode ( the corresponding variables are reported in block A of Tables 1 and 2), the
overall means OM_ GLOB, SM_ GLOB and RDM_ GLOB and finally the pure multimodal ( PI)
indexes. This analysis allows for a more complete consideration of the whole range of ground
transport modes.
Conducting a cluster analysis even on 9 variables, let alone 36, yields cluster centroids that
are too complex to be readily cognitively processed. To simplify the interpretation of the results,
clusters are built only considering the three variables related to objective mobility for the A- type
analysis, and the three overall means ( OM, SM, and RDM) and the three PI indexes for the B-type
analysis. Then we look at the means across groups for the remaining variables of interest to
each cluster. The methodology we employed is a nonhierarchical or k- means procedure where
the number of clusters is selected by the researcher. After several trials it emerged that in both
16
the A- and B- type analyses the solutions with four clusters are the most interesting and
informative given our research goals.
4.1 A- type cluster analysis
The A- type cluster analysis aims at clarifying the relationships among the actual ( OM),
subjective ( SM) and desired use ( RDM) of ( 1) transport modes in general, ( 2) cars and ( 3)
transit. It can hence be seen as a bidimensional analysis where different transport mode uses ( 1-
2- 3) and different ways to intend the term “ use” ( OM- SM- RDM) are jointly considered. As such,
our work builds on previous research that has already investigated the relationships within these
two separate dimensions. Considering the first dimension, pertaining to the relationships among
the uses of different transport modes, there is already a vast literature that deals with “ intermodal
effects” or “ modal diversion”, where the variation in use of a given mode is related to the
variation in use of other modes ( see for example Golob, Van Wissen and Meurs, 1986 and Golob
and Meurs, 1987 for longitudinal studies; Ben- Akiva and Morikawa, 1990 and Hirobata and
Kawakami, 1990 for discrete choice models or Hensher, 2001 for a review on cross- elasticities).
As for the second of the above dimensions, which explores the relationships among
objective, subjective and relative desired mobility, previous research results are also available for
the same U. S. data set analyzed here. Collantes and Mokhtarian ( 2007) and Ory, Mokhtarian and
Collantes ( 2007) study the determinants of subjective mobility and its relationship with actual
mobility behavior, whereas Choo et al. ( 2005) analyze models where relative desired mobility is
a function of objective and subjective mobility, among others. More disaggregated results are
presented by Curry ( 2000, p. 53), which shows that the correlation between objective mobility
and relative desired mobility is generally low for motorized transport modes, and is about 0.3
17
when walking and bicycling are considered. Using different data, Stradling, Meadows and Beatty
( 2004) present some interesting crosstabulations of RDM and the expectation of use of a given
means in the future.
The goal of our A- type analysis is to offer additional insights into the mixed use of different
transport modes that are important from a policy viewpoint. It would be, for example, of interest
to understand if people who intensively use cars desire to reduce their use proportionately more
or less than people who usually travel by public transport, or to determine which transport modes
are more likely to be less used by those who are fed up with travelling. For these kinds of
analyses it is useful to jointly consider the above defined dimensions.
4.1.1 Cluster centroids and mobility levels
Tables 3 and 4 present the resulting A- type cluster centroids for the French and the U. S.
datasets respectively. We can label the resulting clusters based on the three objective mobility
variables around which the clusters were formed, whereas the other values in italics are group
means that were computed a posteriori. Turning first to the French dataset, we have a large first
group of strong car users, a second group of strong public transport users, a small third group of
weak users of both modes and a fourth group of strong users of both modes. We recall that for
the French dataset the reported OM measures are the sum of the number of monthly trips taken
by all the considered modes. Hence it is not meaningful to make “ vertical” comparisons between,
say, OM_ CAR and OM_ PT, since the former is the sum of the number of trips taken by two
modes and the latter the sum of three. For the same reason, French and U. S. OM measures are
also not directly comparable, given the different number of considered modes.
OM measures in table 3 seem to be unreasonably low, but those averages are biased
downward by the fact that the underlying variables are composites of ordinal OM measures
18
whose highest grade is “ more than 3 times a week” ( see section 2) so that higher mobility levels
are flattened around that value. However we believe that this is not a big concern, since for our
research purposes OM variables do not measure the exact mobility levels of individuals, but
rather represent the degree of acquaintance with a given mode, as stated in the introduction. In
that sense, we believe that above a certain threshold utilization level for a given mode, the
corresponding familiarity with that mode does not increase proportionally, so that
underestimations of higher mobility levels are not so relevant in our case.
Tables 3 and 4
In tables 3 and 4, subjective mobility and relative desired mobility measures are the cluster
means of the individual SM and RDM scores that were computed according to the procedure that
was presented in subsection 2.2. The individual scores can theoretically range from –( n– 1) to ( n–
1) when the partial order among the n observations is complete, lower scores indicating lower
SM and RDM levels of that individual in comparison with the whole sample. If the order is not
complete, then the range of the scores is narrowed in proportion to the number of ambiguous
pairwise orderings. However, it is important to understand that, for example, an individual RDM
score of zero does not mean that the individual does not wish to alter his/ her mean mobility level
across the considered modes. It simply means that an equal number of cases are greater than this
one ( i. e. have RDM measures at least as high on all categories, and strictly higher on at least one)
as are smaller than this one. Moreover, the SM and RDM means that are shown in the tables are
to be interpreted in relative terms across the different clusters: specifically, comparisons can only
meaningfully be done among numbers in the same row, i. e. horizontally reading the data.
19
With that in mind, it is interesting to compare the following results with those of Diana and
Mokhtarian ( 2007), that report the same clusters but with a different method to combine SM and
RDM measures, which hence leads to a different cluster interpretation. Their method was based
on heuristic approximations that keep the information on a case’s position relative to the neutral
point in a bipolar scale such as the RDM one. In other words, in that paper it was possible to
understand if the respondents within a given cluster tended to actually like to travel more or less
than what they actually do. The cluster interpretations that can be found in that paper hence give
information that is complementary with the one that is reported in the following, since here the
focus of the analysis is on the above mentioned horizontal comparisons.
If we “ horizontally” read the subjective mobility measures, we can see that they are quite
correlated with the corresponding OM measures, confirming the results of Collantes and
Mokhtarian ( 2007). On the other hand, RDM measures show interesting patterns among clusters
in the French dataset. Those who intensively use a given mode ( i. e. car for groups 1 and 4 and
public transport for groups 2 and 4) would like to travel less than the general average across the
four groups by such a mode and ( except for group 4) would like to travel more by the other
mode. Group 3 seldom uses both modes and is thus willing to increase their use more than the
average. The four groups appear to be less distinct in terms of global RDM levels, so that the
desire to travel more ( or less) overall seems less strongly related to the composition of the modal
basket.
Repeating the same kind of analysis on the U. S. dataset gives us partly different results. This
is not surprising, given the nature of the cluster analysis technique and the radically different
experimental settings. Three out of four groups predominantly use cars, comprising 89% of the
sample, compared to the 54% of respondents belonging to group 1 in the French dataset. People
who predominantly use public transit fall from 29% in the French sample ( Groups 2 and 3) to
20
11% in the U. S. sample ( Group 3). The latter group is much less monomodal toward transit than
the French group 2 and has OM patterns rather similar to the French group 3. However a
cautionary note must be considered when comparing OM measures across datasets, since these
measures are different, as explained in subsection 2.1. Another interesting disparity between the
two samples is that the U. S. analysis did not distill a group that intensively uses both means.
Subjective mobility measures have slightly lower correlations with OM measures compared
to the French sample. This could be due to the coarser SM scale used by the U. S. sample, but
behavioral differences probably also have an influence. In particular, the U. S. group 1 travels
slightly more than group 3 but it “ feels” it travels less. However we do not observe this
discrepancy when looking at the French groups 1 and 3. This could be due to different attitudes
regarding cars and perhaps even more so, public transport in the two samples, since the same
differences between the datasets can be detected when comparing OM_ GLOB and RDM_ GLOB
for groups 1 and 3.
As explained above, RDM scores do not allow us to determine if people belonging to a
group desire to travel more or less on average. Hence we had a closer look at the RDM measures
of the two considered transit modes in the U. S. dataset, namely the bus and the train. Only 4%
wanted to use both transit modes less. In fact, however, the majority of the sample ( 55%) wanted
to travel either by bus or by train the same amount as now, and 80% of those were currently
traveling little or not at all by transit. Only 27% of the sample actually wanted more travel by
train, and 11% wanted more by bus. Overall though, latent demand for travel seems greater in
the U. S. dataset, a result that is probably linked to the different socioeconomic composition of
the sample. This could also be due to the lower levels of use of transit, so that the desired modal
baskets of the two samples are probably closer to each other than is the actual use of different
means.
21
4.1.2 Mobility patterns among different clusters
The similarities and differences between the two datasets that we analysed are useful for
drawing some more general conclusions from our exercise. A preliminary consideration is
related to the existence of individual thresholds concerning the traveling activity, that can of
course be interpreted in view of the long- standing debate on the existence of travel time budgets
( Zahavi and Talvitie, 1980; Zahavi and Ryan, 1980; Supernak, 1982). Several studies have been
published on this matter supporting different points of view, but more recent reviews that draw
on a vast number of works ( Mokhtarian and Chen, 2004; Joly, 2005, 2006) tend to conclude that
unobserved travel time “ desired budgets” exist ( in contrast to observed travel time expenditures),
but vary across individuals, being related to several factors such as socioeconomic status, kind
and patterns of activities at different locations and characteristics of residential areas.
Following this line of research, our results can offer preliminary insights on another possible
source of variation of individual budgets, namely the actual use of different modes. In fact, the
values in tables 3 and 4 and the lack of a clear socioeconomic differentiation between clusters, as
we will see in the following subsection 4.1.3, suggest to us that people have different ideal levels
of use of the different modes. We can see this by separately considering the clusters that show
strong monomodal behavior ( groups 1 and 2 of the two datasets) and then some clusters with
comparable global mobility levels ( namely, groups 1 and 3).
As a general rule, strong users of a given mode would like to bring more balance to their
basket by decreasing the use of this mode more than the average, and increasing the use of the
alternative mode. Hence it is important to jointly look at the use of the different modes to
understand people’s desires. For example, French groups 2 and 4 have the same level of use of
22
public transport, and U. S. groups 3 and 4 are the same for cars, but the corresponding RDM
levels are quite different.
However, turning our attention to objective and relative desired global mobility levels, other
interesting patterns emerge. For example, the U. S. group 3 desires to travel dramatically less than
the U. S. group 1, even if the two OM_ GLOB values are comparable. The decisive difference is
that the group 3 uses more transit than cars, so that the ideal level of use of the two modes is
clearly different. We do not observe the same phenomenon when considering the French groups
1 and 3, so that ideal levels are clearly different also for the two samples.
Our methodology does not allow us to quantify the above two gaps in the ideal levels of use
( between different modes and between different datasets); nevertheless we can infer that ( 1) the
ideal levels of use of the different transport means may be linked more with the balance in the
modal basket than with one’s global mobility levels and that ( 2) when the modal basket is
equilibrated, a comparison of the two samples shows differences in the relative attractiveness of
the modes, that are probably due to the different experimental contexts. For example, the French
sample, comprising employees of a transportation research institute, would be more sensitive to
the negative externalities imposed by the automobile, and perhaps ideologically more inclined
toward environmentally- benign travel modes, than would the more general sample of the U. S.
dataset ( although the northern Californians comprising the U. S. sample would in turn be
expected to be more environmentally aware and proactive than the country as a whole).
These findings show the importance of considering multimodal behaviors, beyond the
absolute mobility levels, in order to understand the direction and magnitude of potential modal
shifts. This fact prompted us to more carefully look into the modal basket mechanism, through
the consideration of the multimodality indexes in the B- type analyses presented in section 4.2.
23
4.1.3 Socioeconomic characteristics of the U. S. clusters
A closer look at the socioeconomic characteristics of the different clusters allows us to gain
further insight concerning the above mentioned classification. This kind of analysis is not so
meaningful on the French dataset, given the very particular characteristics of the sample that
does not allow for an easy generalization of the results and the small number of observations for
three out of the four clusters. In contrast, characterizing the clusters of the U. S. dataset is an
effective way to understand how personal traits relate to multimodality attitudes and behaviors.
Concerning the U. S. A- type clusters, it is particularly interesting to study the differences between
the first two clusters on one hand ( heavily versus somewhat car- oriented) and between car-oriented
and transit- oriented clusters on the other. Light travelers would presumably have greater
within- group variability in terms of socioeconomic characteristics than the other clusters but
offer less insight concerning multimodality behaviors.
Since the socioeconomic variables we consider are ordinal rather than continuous, and ( more
importantly) are not always approximately normally- distributed, a standard analysis of variance
( ANOVA) test of whether the means on each variable are equal across clusters is not necessarily
appropriate. Accordingly, we perform the conceptually similar non- parametric Kruskal- Wallis
test to ascertain whether the group differences that we found are simply due to random variation.
The corresponding p- values are always equal to or below .005, so that we can safely reject the
null hypothesis of no difference among the group populations.
We consider at the outset the number of available cars in the household and the percentage
of time in which a vehicle is available to the respondent, that are respectively shown in figures 5
and 6. Both values are not surprisingly lower for transit- oriented than for car- oriented people.
However it is interesting to note that there is almost no difference between the two car- oriented
24
clusters in these two aspects, so that their distinction is not due to the “ car availability”
instrumental factor. A detectable difference is instead related to sex, since 65% of the heavily
car- oriented people are male, whereas for group 2 this figure is lowered to 53%, roughly the
same as for the transit- oriented cluster. Probably the greater symbolic value of cars for male
drivers ( Steg, 2005) can explain part of this difference.
Figures 5 and 6
The share of persons who did not attend graduate school is roughly equal to 35% across the
four clusters, but transit- oriented persons are much more likely to have completed a 4- year
degree ( figure 7). On the other hand, these latter have a lower income than the average ( figure 8);
the well- known positive relationship between income and educational level is thus not reflected
in our multimodality- based clusters. Strong car users are more likely to be full- time workers, but
other socioeconomic characteristics do not display much difference between the two car- oriented
groups, as shown in figures 5 to 8. To sum up, the use of a multimodality- based clustering
technique cannot easily be represented by segmentations based on socioeconomic characteristics,
since patterns of modal usage are generated from both instrumental and affective factors.
Figures 7 and 8
4.2 B- type cluster analysis
The B- type analysis should clarify the added value of considering the proposed
multimodality indexes within our framework. Our expectation in performing this analysis is to
25
have a closer look at the relationship between the fact of being more or less acquainted with a
variety of transport modes and the desire to alter one’s modal basket, while controlling for the
actual, perceived and desired change in the level of use of several different modes. The interest
in separately looking at the mean mobility level and the multimodal behavior pushed us to
consider “*_ GLOB” variables and PI indexes in the analysis. Given the definitions of our
indexes, an analytically similar procedure would be to define clusters only on the basis of MI
variables. A variant of a B- type analysis that uses MI instead of “*_ GLOB” variables and PI has
been performed, and the cluster solutions are almost equivalent.
4.2.1 Cluster centroids and mobility levels
Tables 5 and 6 present the resulting B- type cluster centroids for the French and the U. S.
datasets respectively. Clusters are presented by ascending OM_ GLOB values. The means for the
six variables on which the cluster formation is based are the ones that are not italicized in the
tables. Their values are standardized since the range of the raw PI variables is up to three orders
of magnitude smaller than that of the “*_ GLOB” variables. The numbers in the tables represent
thus the number of standard deviations the original value is from the mean. This must be taken
into account in the interpretation of the results, since for example the 0.1 values for
RDM_ GLOB in the French groups 2 and 3 do not mean that the respondents in these groups
desire to keep their amount of travel nearly constant on average, but that on average they desire
to change this amount only slightly more than the mean of the whole sample.
The tables report also in italics the group means of the other variables defined in section 2,
that have been directly observed. Their analysis can help to define the characteristics of the
distilled groups. SM scales have a rather intuitive meaning, so that the group means of the
questionnaire scales are directly reported. RDM scales are instead somewhat more cumbersome,
26
since RDM mean values of the cluster centroids are often quite close to the neutral point ( that is
“ 3” in the U. S. dataset and “ 6” in the French one). This would hamper an easy appreciation of
the differences among clusters that actually exist. Hence we decided to define a new 7- point
scale using the symbols “– – –”, “– –”, “–” ( representing the desire to decrease the use of a given
mode), “=” ( neutral point), “+”, “+ +” and “+ + +” ( representing the desire to increase the use of
a given mode). Then we relate this scale to the mean values computed from the scales that we
have in our datasets through the definition of thresholds. These thresholds, reported in table 7,
have been judgementally set in such a way that the differences among clusters are highlighted.
Moreover, the relative width of each corresponding point in the two 7- point scales is the same.
For example, the width of extreme point “– – –” is 3 in the French dataset where the original
scale has 11 points, of which 5 are negative. In the U. S. scale we have 5 points of which 2 are
negative, hence the width of that extreme point will be 3* 2/ 5 = 1.2.
Tables 5, 6 and 7
It can be seen that the analysis gave different clusters in the two samples. For France, the
first two clusters have both their mobility level and their OM_ PI below average, but differ in that
the first cluster contains largely monomodal car users, whereas the second cluster is more
multimodal albeit still auto- dominated. The third and fourth clusters are more multimodal,
differing in that the third group has a mixture of mode- specific desires with respect to altering
their travel, whereas the fourth group is surfeited with nearly every mode. Overall, the first three
groups would like both to increase their amount of travel more than the average and to alter their
modal basket composition by diversifying their desired changes in various modes, whereas group
4 wants the opposite. The U. S. dataset produced four clusters where car use always prevails and
27
respondents would even increase it. Group 1 has a low mobility level and uses only cars, but is
moderately open to increase the use of rail. Groups 2 and 4 have higher mobility and
multimodality levels, but the former would like to sharply decrease its transit usage and thus the
quantity of trips, whereas the latter is mildly deprived with respect to all modes except bus, even
though it is the maximum user of all four considered modes ( car, train, bus and non- motorized).
Group 3 is somewhat the opposite, since it travels less than average, feels it travels more than
that but would like to travel even more both by car and by train.
4.2.2 Mobility patterns among different clusters
B- type clusters can be further interpreted by looking at some other properties of the groups,
namely the values of the italicised variables in the tables. Concerning the French dataset, we
preliminarily notice that these groups are ( obviously) different from that of the A- type analysis,
since we have now two groups ( 1 and 2) that almost exclusively use cars and two others with a
more or less marked predominant use of public transport, as we mentioned above. Group 2
shows also an occasional use of public transport, so that its members feel themselves to be less
monomodal and have a smaller desire to diversify their modal basket compared to group 1.
Groups 3 and 4 travel more than the average and would like to reduce trips taken by both private
and public transport modes. The distinction is that group 4’ s desire to reduce travel is nearly
across the board ( with the low diversity in mode- specific RDM scores shown in the low value of
RDM_ PI), whereas group 3’ s desire is more selective ( and hence, the greater diversity showing
as a higher RDM_ PI).
Another interesting and policy- relevant observation is the strong attractiveness of cycling for
all groups, and the desire to travel by car more as a passenger and less as a driver. Thus it seems
surely worthwhile for policy makers to take those actions that can transform such desires into
28
intentions and, hopefully, actual choices. The measures concerning desired use of non- motorised
modes could be affected by a social desirability bias, beyond the often already- large gap between
desires ( which are influenced by factors such as health concerns) and intentions ( which can take
into account more practical aspects such as scheduling problems). On the other hand, our finding
concerning the differential attractiveness of the car- driver and car- passenger modes is less trivial
and hopefully less affected by such problems, and may be able to give an indication of the
effectiveness of initiatives such as car- sharing and car- pooling. In more general terms, the strong
attractiveness of a less intensively used mode such as the bike helps explain how the first three
groups would like to change their modal basket in a more polarised way. The last group, as
indicated above, contains those who are fed up with travelling and would like to drastically
reduce their trips by every motorized transport means, increasing only cycling Thus, their desired
changes favour cycling exclusively, though as mentioned earlier, this may not mean that the
desired modal basket contains only cycling – it could simply involve greater balance among all
modes.
Since the various modes are considered in a more aggregate way we cannot match in the
U. S. dataset some of the findings of the analysis on the French data. It is however evident that
the attractiveness of nonmotorized transport modes is confirmed for all four U. S. groups,
whereas the analysis did not discern a group that is willing to decrease or keep constant the use
of cars. Groups 1 and 2 are similar to the corresponding French ones in that ( 1) they both have an
intensive use of cars and ( 2) the second group has more familiarity with public transport than the
first one. However here we do not see a replication of the above finding of a desire to increase
the use of public transport for the U. S. group 2. This can be explained by looking at the value of
RDM_ GLOB, which is now negative. This strong desire to decrease the global mobility level
comes at the expense of the level of utilisation of transit.
29
On the other hand, two of the four clusters, comprising 58% of the sample, show an interest
in increasing their rail usage slightly ( group 4, 17%) or moderately ( group 3, 41%), while not
decreasing their ( modest) bus usage. Whether this represents a desire to be in a different location,
where rail is a practical alternative, or a desire specifically for the perceived higher levels of
speed and comfort of rail compared to bus, or simply a bias toward rail as a generally more
attractive mode, it also points to the infeasibility of saturating a metropolitan area in the U. S.
( even one as relatively transit- oriented as San Francisco) to the extent of satisfying the latent
demand for that means of travel. And, as already pointed out, this desired increase in rail travel
would apparently not be realized as a substitute for car travel, since respondents in those two
groups tend to want to increase their ( already substantial) levels of car travel as well.
We finally point out that the B- type clusters here presented are qualitatively quite similar to
those described in Diana and Mokhtarian ( 2007), although the computation of SM and RDM
global mobility levels is different. This makes us more confident in the stability of our findings,
given the well- known high sensitivity of the cluster analysis technique to the definition of the
clustering variables.
4.2.3 Socioeconomic characteristics of the U. S. clusters
As previously done for the A- type clusters ( see section 4.1.3), to see how the clusters might
differ on socioeconomic traits, we restrict our analysis to the U. S. dataset. Concerning sex, 60%
of group 4 is male, whereas for groups 2 and 3 this percentage is 44% and 42% respectively.
Characterizing these clusters with the same socioeconomic variables that were analyzed in
section 4.1.3, namely the number of vehicles in the household, the percentage of time a vehicle is
available to the respondent, educational level and income ( figures 9 to 12 respectively), other
interesting patterns emerge. In this case, we see that groups 3 and 4 have roughly the same car
30
availability but rather different multimodal behaviors and desires, as pointed out in the preceding
section. Not surprisingly, group 1 has a higher than average car availability and the opposite is
true for group 2. Group 1 is less educated and group 4 is more educated than average, whereas
groups 2 and 3 are poorer and group 4 is richer.
Figures 9, 10, 11 and 12
Looking at these figures, we can conclude that the generally more affluent respondents of
group 4 are quite happy with their higher than average mobility levels and modal basket
composition, even if they are open to increasing their already relatively high use of train.
Interestingly, group 2 is in the opposite situation, concerning both socioeconomic conditions and
mobility desire. It is discouraging from a policy point of view to see that what they want, to
improve their satisfaction with travel, is to sharply cut their transit use and thereby become much
more monomodal with respect to the car. This contrasts with group 4, which displays the same
levels of use of transit but is happy with that. This may be a case of the less- affluent desiring a
lifestyle it perceives to characterize the upper class, when the upper class in reality is living, by
choice, the lifestyle that the former group aspires to shed. On the other hand, given the
correlations that exist among income, residential location, and transit supply, these results may
well reflect differences in the latter characteristics between the clusters. For example, the transit
use of cluster 4 may more likely take place on comfortable commuter trains and express buses,
while that of cluster 2 may occur on crowded trams, metro rail, and local buses.
31
5 Conclusions
This paper examined various mode- specific and overall measures of mobility – objective,
subjective, and desired – for a sample of transportation institute employees in France, and a
sample of the general population of the San Francisco Bay Area in the U. S. We focused on
analyzing the modal mixes, or “ baskets”, used, perceived, and/ or desired by individuals, with a
view to identifying population segments having similar modal baskets, and better understanding
the role of modal mix in personal travel patterns and desires.
We conducted two cluster analyses, each resulting in four distinct clusters for each sample.
Although the relative cluster sizes are not necessarily representative of the population at large,
and certainly size- based comparisons across the two samples are unjustified, the types of clusters
identified are of general interest, and have useful policy implications. As a general comment, the
results suggest that different people have different ideal targets for the use of each mode; it is
valuable to explore further the types of people wanting to decrease their car use, increase their
transit use, and/ or increase the multimodality of their modal baskets. The second cluster analysis
of the French sample, for instance, produced two clusters of predominantly car users. Group 1 is
largely monomodal and would like to increase the use of transit. Group 2 uses public transport to
a certain degree, wants to use it more and to decrease the use of cars more sharply than the other
group. Then it would presumably be more effective to market transit to the second group of car
users than to the first.
We also identified people wanting to decrease car driving but increase their use of car as a
passenger; they are potential targets of ridesharing or carsharing campaigns. Finally, a desire to
increase the use of non- motorized modes such as cycling and walking appeared across all of the
clusters from both countries. There may be a large measure of “ wishful thinking” in such
32
responses ( knowing it is “ good for you” but not willing to commit the time and effort to change),
and even if change on that dimension is realized, it may take more the form of recreation rather
than replacing very much auto travel. Nevertheless, the presence of such a desire is at least an
important starting point for behavior change that would have positive health effects even if
limited transportation effects.
The preliminary analyses reported here could be extended in several different ways. With
the same datasets, it would be valuable to explore the identified clusters further, using other
variables in each dataset to flesh out the skeletal picture we currently have of the types of people
in each cluster. The usefulness of the multimodality indexes that we defined could further be
assessed by using them also in different kinds of analyses, for example generalizing the paper of
Choo et al. ( 2005) ( which modeled the relative desired mobility for personal vehicles, alone
among the modes surveyed for short- distance travel) to explore more systematically how relative
desired mobility for various modes differs with degree of multimodality ( among other
explanatory variables). For future data collection efforts, it would be desirable to customize the
measurement of mobility- related variables to be optimally- suited for the current purpose. It
would also be highly desirable to conduct identical ( as much as possible) surveys of the general
population in several different countries, to enable a more systematic exploration of cultural
differences relevant to travel behavior. In any case, the concept of multimodality appears to
make a useful contribution to the study of travel patterns, and one which could be profitably
explored further.
33
References
Aarts, H. and Dijksterhuis, A. ( 2000) The automatic activation of goal- directed behaviour: The
case of travel habit, Journal of Environmental Psychology, 20( 1) pp. 75- 82.
Anable, J. ( 2005) ‘ Complacent car addicts’ or ‘ aspiring environmentalists’? Identifying travel
behaviour segments using attitude theory, Transport Policy, 12( 1), pp. 65- 78.
Anable, J. and Gatersleben, B. ( 2005) All work and no play? The role of instrumental and
affective factors in work and leisure journeys by different travel modes, Transportation
Research Part A, 39( 2- 3), pp. 163- 181.
Bamberg, S., Ajzen, I. and Schmidt, P. ( 2003) Choice of travel mode in the theory of planned
behavior: the roles of past behavior, habit, and reasoned action. Basic and Applied Social
Psychology, 25( 3), pp. 175– 187.
Beale, J. R. and Bonsall, P. W. ( 2007) Marketing in the bus industry: A psychological
interpretation of some attitudinal and behavioural outcomes, Transportation Research Part F,
10( 4), pp. 271- 287.
Beirão, G. and Cabral, J. S. ( 2007) A market segmentation analysis using attitudes towards travel
behaviour. 11th World Conference on Transport Research, Berkeley, USA, ( presentation).
Ben- Akiva, M. and Morikawa, T. ( 1990) Estimation of switching models from revealed
preferences and stated intentions, Transportation Research Part A, 24A( 6), pp. 485- 495.
Bollen, K. A. and Barb, K. H. ( 1981) Pearson’s R and coarsely categorized measures, American
Sociological Review 46( 2), pp. 232- 239.
Bovy, P. H. ( 1999) Urban structure and modal distribution, Public Transport International, 48( 1),
pp. 8- 15.
34
Choo, S., Collantes, G. O. and Mokhtarian, P. L. ( 2005) Wanting to travel, more or less: exploring
the determinants of the deficit and surfeit of personal travel, Transportation, 32( 2), pp. 135-
164.
Collantes, G. O. and Mokhtarian, P. L. ( 2007) Subjective assessments of personal mobility: What
makes the difference between a little and a lot?, Transport Policy, 14( 3), pp. 181- 192.
Curry, R. W. ( 2000) Attitudes toward travel: The relationships among perceived mobility, travel
liking, and relative desired mobility, Master’s Thesis, Institute of Transportation Studies,
University of California, Davis, 252pp.
Diana, M. ( 2005) An exploratory web- based attitudinal travel survey administered to INRETS
staff. Les Collections de l’INRETS – “ Outils et méthodes” n° 12, Arcueil, 187pp, ISBN 2-
85782- 623- 0.
Diana, M. ( 2006) Utilité primaire des déplacements et multimodalité : conception et réalisation
d’un outil d’enquête novateur, Recherche Transports Sécurité, 93, pp. 279- 294.
Diana, M. ( 2008) Making the “ primary utility of travel” concept operational: a measurement
model for the assessment of the intrinsic utility of reported trips, Transportation Research
Part A, 42( 3), pp. 455- 474.
Diana, M. and Mokhtarian, P. L. ( 2007) The individual desire to change a “ modal basket” and its
relationship with the actual and perceived use of different transport means, 11th World
Congress on Transport Research, Berkeley, USA, 41 pp.
Diana, M., Song, T. and Wittkowski, K. M. ( 2008) Studying travel- related individual assessments
and desires by combining hierarchically structured ordinal variables, Research paper
available from the Authors.
35
Dobson, R. and Tischer, M. L. ( 1978) Perceptual market segmentation technique for
transportation analysis, Transportation Research Record 673, pp. 145- 152.
Elgar, A. and Bekhor, S. ( 2004) Car- rider segmentation according to riding status and investment
in car mobility, Transportation Research Record 1894, pp. 109- 118.
Gärling, T., Fujii, S. and Boe, O. ( 2001) Empirical tests of a model of determinants of script-based
driving choice, Transportation Research Part F, 4( 2), pp. 89- 102.
Garvill, J., Marell, A. and Nordlund, A. ( 2003) Effects of increased awareness on choice of
travel mode, Transportation, 30( 1), pp. 63- 79.
Gensch, D. H. and Torres, P. T. ( 1980) Perceived- difference segmentation model for mass transit
marketing, Transportation Research Record 765, pp. 16- 22.
Golob, T. F. and Meurs, H. ( 1987) A structural model of temporal change in multi- modal travel
demand, Transportation Research Part A, 21A( 6), pp. 391- 400.
Golob, T. F., Van Wissen, L. and Meurs, H., ( 1986) A dynamic analysis of travel demand,
Transportation Research Part A, 20A( 5), pp. 401- 414.
Hensher, D. A. ( 2001) Modal diversion, in K. J. Button and D. A. Hensher ( eds.), Handbook of
Transport Systems and Traffic Control, Pergamon, Amsterdam, pp. 107- 123.
Hensher, D. A. and Reyes, A. J. ( 2000) Trip chaining as a barrier to the propensity to use public
transport, Transportation, 27( 4), pp. 341- 361.
Hirobata, Y. and Kawakami, S. ( 1990) Modeling disaggregate behavioural modal switching
models based on intention data, Transportation Research Part B, 24B( 1), pp. 15- 25.
36
Jensen, M. ( 1999) Passion and heart in transport – a sociological analysis on transport behavior,
Transport Policy, 6( 1), pp. 19- 33.
Joly, I. ( 2005) L’allocation du temps au transport – De l’observation internationale des budget-temps
de transport aux modèles de durées, Ph. D. Thesis, Université Lumière Lyon 2, Lyon,
VI+ 477pp.
Joly, I. ( 2006) Stability or regularity of the daily travel time in Lyon ? Application of a duration
model, International Journal of Transport Economics, 33( 3), pp. 369- 400.
Massot, M.- H., Armoogum, J., Bonnel, P. and Caubel, D. ( 2006) Potential for car use reduction
through a simulation approach: Paris and Lyon case studies, Transport Reviews, 26( 1), pp. 25-
42.
Mokhtarian, P. L. and Chen, C. ( 2004) TTB or not TTB, that is the question: a review and
analysis of the empirical literature on travel time ( and money) budgets, Transportation
Research Part A, 38( 9- 10), pp. 643- 675.
Mokhtarian, P. L. and Salomon, I. ( 2001) How derived is the demand for travel? Some
conceptual and measurement considerations, Transportation Research Part A, 35( 8), pp. 695-
719.
Mokhtarian, P. L., Salomon, I. and Redmond, L. S. ( 2001) Understanding the demand for travel: It's
not purely ' derived', Innovation: The European Journal of Social Science Research, 14( 4), pp.
355- 380.
Nash, A. N. and Hille, S. J. K. ( 1968) Public attitudes toward transport modes: A summary of two
pilot studies, Highway Research Record 233, pp. 33- 46.
37
Nobis, C. ( 2007) Multimodality: facets and causes of sustainable mobility behavior,
Transportation Research Record 2010, pp. 35- 44.
Ory, D. T., Mokhtarian, P. L., and Collantes, G. O. ( 2007) Exploring the cognitive and affective
mechanisms behind subjective assessments of travel amounts, Environment and Behavior, 39( 4),
pp. 494- 528.
Outwater, M. L., Castleberry, S., Shiftan, Y., Ben- Akiva, M., Zhou, Y. S. and Kuppam, A. ( 2003)
Attitudinal market segmentation approach to mode choice and ridership forecasting –
Structural Equation Modeling, Transportation Research Record 1854, pp. 32- 42.
Rényi, A. ( 1970) Probability theory, North- Holland, Amsterdam, 670pp.
Shannon, C. E. ( 1948) A mathematical theory of communication, The Bell System Technical
Journal, 27, pp. 379- 423; 623- 656.
Steg. L. ( 2005) Car use: lust and must. Instrumental, symbolic and affective motives for car use,
Transportation Research Part A, 39( 2- 3), pp. 147- 162.
Stradling, S., Meadows, M. and Beatty, S. ( 2004) Who will reduce their car use – and who will
not?, in T. Rothengatter and R. D. Huguenin ( Eds.), Traffic and Transport Psychology –
Theory and Application – Proceedings of the ICTTP 2000, Elsevier, Amsterdam, Netherlands,
pp. 459- 467.
Supernak, J. ( 1982) Travel- time budget: a critique, Transportation Research Record 879, pp. 15-
28.
Tardiff, T. J. ( 1979) Attitudinal market segmentation for transit design, marketing, and policy
analysis, Transportation Research Record 735, pp. 1- 7.
38
Thøgersen, J. ( 2006) Understanding repetitive travel mode choices in a stable context: A panel
study approach, Transportation Research Part A, 40( 8), pp. 621- 638.
TRB – Transportation Research Board ( 1977) Preferences, perceptions, and market segments in
travel behavior, Transportation Research Record 649, 48pp.
Verplanken, B., Aarts, H., van Knippenberg, A. and van Knippenberg, C. ( 1994) Attitude versus
general habit: Antecedents of travel mode choice. Journal of Applied Social Psychology,
24( 4), pp. 285– 300.
Verplanken, B., Aarts, H. and van Knippenberg, A. ( 1997) Habit, information acquisition, and
the process of making travel mode choices, European Journal of Social Psychology, 27( 5),
pp. 539- 560.
Wittkowski, K. M., Lee, E., Nussbaum, M., Chamian, F. N. and Krueger, J. G. ( 2004) Combining
several ordinal measures in clinical studies, Statistics in Medicine, 23( 10), pp. 1579- 1592.
Zahavi, Y. and Ryan, J. M. ( 1980) Stability of travel components over time, Transportation
Research Record 750, pp. 19- 26.
Zahavi, Y. and Talvitie, A. ( 1980) Regularities in travel time and money expenditures,
Transportation Research Record 750, pp. 13- 19.
39
List of figures
Figure 1. Hierarchical structure of the modes in the French dataset; grey modes are the observed
ones
Figure 2. Hierarchical structure of the modes in the US dataset; grey modes are the observed
ones
Figure 3. Plot of OM_ PI for two modes and maximum frequency M = 15
Figure 4. Plot of OM_ MI for two modes and maximum frequency M = 15
Figure 5. Number of personal vehicles per household across the A- type clusters
Figure 6. Percentage of time a personal vehicle is available to the respondent across the A- type
clusters
Figure 7. Educational level of the respondent across the A- type clusters
Figure 8. Household income across the A- type clusters
Figure 9. Number of personal vehicles per household across the B- type clusters
Figure 10. Percentage of time a personal vehicle is available to the respondent across the B- type
clusters
Figure 11. Educational level of the respondent across the B- type clusters
Figure 12. Household income across the B- type clusters
40
List of tables
Table 1. List of variables from the French dataset
Table 2. List of variables from the U. S. dataset
Table 3. A- type cluster solution and group means, French dataset
Table 4. A- type cluster solution and group means, U. S. dataset
Table 5. B- type cluster solution and group means, French dataset
Table 6. B- type cluster solution and group means, U. S. dataset
Table 7. Thresholds in defining the 7- point RDM scale
41
Table 1
List of variables from the French dataset
Travel mode Objective evaluation
of the travel amount
Subjective estimation of
the travel amount
Desired change in the travel
amount
A. Variables present in the dataset
Bicycle OM_ BIKE SM_ BIKE RDM_ BIKE
Motorbike OM_ MOTO SM_ MOTO RDM_ MOTO
Car as driver OM_ CARDR SM_ CARDR RDM_ CARDR
Car as passenger OM_ CARPAX SM_ CARPAX RDM_ CARPAX
Bus OM_ BUS SM_ BUS RDM_ BUS
Tramway OM_ TRAM SM_ TRAM RDM_ TRAM
Metro OM_ METRO SM_ METRO RDM_ METRO
Taxi OM_ TAXI SM_ TAXI RDM_ TAXI
Suburban train OM_ SUBURB SM_ SUBURB RDM_ SUBURB
Long- distance train OM_ TRAIN SM_ TRAIN RDM_ TRAIN
B. Variables built from those above
Car OM_ CAR SM_ CAR RDM_ CAR
Public transport OM_ PT SM_ PT RDM_ PT
Composite over all modes OM_ GLOB SM_ GLOB RDM_ GLOB
Measures
Metric variables:
estimated number of
monthly trips
“ I feel I don’t travel” ( 1)
…
“ I feel I travel a lot” ( 10)
“ I’d like to travel much less”
(– – –) … “ I’d like to travel
the same” (=) … “ I’d like to
travel much more” (+++)
Table 2
List of variables from the U. S. dataset
Travel mode Objective evaluation
of the travel amount
Subjective estimation of
the travel amount
Desired change in the travel
amount
A. Variables present in the dataset
Personal vehicle OM_ CAR SM_ CAR RDM_ CAR
Bus OM_ BUS SM_ BUS RDM_ BUS
Train/ Light rail OM_ RAIL SM_ RAIL RDM_ RAIL
Walk/ jogging/ cycling OM_ WALK SM_ WALK RDM_ WALK
B. Variables built from those above
Public transport OM_ PT SM_ PT RDM_ PT
Composite over all modes OM_ GLOB SM_ GLOB RDM_ GLOB
Measures
Metric variables:
estimated number of
weekly hours spent
on that mode
“ I feel I don’t travel” ( 1)
…
“ I feel I travel a lot” ( 5)
“ I’d like to travel much less”
(– – –) … “ I’d like to travel
the same” (=) … “ I’d like to
travel much more” (+++)
42
Table 3
A- type cluster solution and group means, French dataset
Variable
Group 1
Car-oriented
Group 2
Transit-oriented
Group 3
Neither-oriented
Group 4
Both-oriented
OM_ GLOB 25 40 19 55
OM_ CAR 18 4 3 16
OM_ PT 3 30 10 30
SM_ GLOB - 5 10 - 39 24
SM_ CAR 31 - 60 - 82 13
SM_ PT - 44 66 - 6 62
RDM_ GLOB 0 2 17 - 10
RDM_ CAR - 18 22 43 8
RDM_ PT 26 - 30 - 1 - 45
Number of cases ( 164) 88 33 14 29
NB: The first three rows report the group means of the sum of the
number of monthly trips taken with the considered modes; the
remaining rows the group means of the scores obtained
combining SM and RDM ordinal measures.
Table 4
A- type cluster solution and group means, U. S. dataset
Variable
Group 1
Heavily car-oriented
Group 2
Rather car-oriented
Group 3
More transit-oriented
Group 4
Light
travellers
OM_ GLOB 17 9 15 4
OM_ CAR 13 7 3 3
OM_ PT 1 0 9 1
SM_ GLOB 267 - 26 321 - 118
SM_ CAR 615 270 - 589 - 221
SM_ PT - 79 - 328 1242 - 15
RDM_ GLOB - 2 95 - 236 - 21
RDM_ CAR - 289 - 49 28 100
RDM_ PT 188 156 - 530 - 43
Number of cases ( 1904) 185 696 205 818
NB: The first three rows report the group means of the sum of the hours spent
using the considered modes, the remained the group means of the scores
obtained combining SM and RDM ordinal measures.
43
Table 5
B- type cluster solution and group means, French dataset
Group 1
Very light
travelers,
monomodal car
users, deprived
re other modes
Group 2
Light travelers,
car- dominated
but multimodal,
deprived re
other modes
Group 3
Moderate
travelers, highly
multimodal,
mixed
satisfaction
Group 4
Heavy travelers,
highly
multimodal,
generally
surfeited
OM_ GLOB ( std) - 0.9 - 0.6 0.7 1.2
OM_ PI ( std) - 1.5 - 0.4 0.7 0.8
OM_ BIKE 1 2 3 1
OM_ MOTO 0 0 1 1
OM_ CARDR 11 11 7 10
OM_ CARPAX 4 3 4 7
OM_ BUS 1 2 9 11
OM_ TRAM 1 0 3 4
OM_ METRO 2 4 10 11
OM_ TAXI 0 0 1 0
OM_ SUBURB 0 0 1 4
OM_ TRAIN 0 1 2 1
SM_ GLOB ( std) - 1.5 - 0.1 0.6 0.4
SM_ PI ( std) - 1.7 0.0 0.6 0.3
SM_ BIKE 1 3 4 3
SM_ MOTO 1 1 2 2
SM_ CARDR 7 8 6 8
SM_ CARPAX 4 4 5 6
SM_ BUS 2 4 6 7
SM_ TRAM 1 2 3 4
SM_ METRO 2 4 7 7
SM_ TAXI 1 2 2 2
SM_ SUBURB 1 2 3 3
SM_ TRAIN 2 3 5 4
RDM_ GLOB ( std) 0.2 0.1 0.1 - 0.9
RDM_ PI ( std) 0.5 0.2 0.2 - 2.3
RDM_ BIKE ++ ++ ++ ++
RDM_ MOTO + + + =
RDM_ CARDR – – – – – –
RDM_ CARPAX + + = –
RDM_ BUS + + = – – –
RDM_ TRAM + + + – –
RDM_ METRO + + – – – –
RDM_ TAXI = = = – –
RDM_ SUBURB = = = – – –
RDM_ TRAIN + + = –
Number of cases 25 59 63 17
NB: std = standardized
44
Table 6
B- type cluster solution and group means, U. S. dataset
Group 1
Light travelers,
monomodal car
users, generally
deprived
Group 2
Moderate
travelers,
multimodal but
auto- dominated,
transit- surfeited
Group 3
Light travelers,
multimodal but
auto- dominated,
generally
deprived
Group 4
Heavy travelers,
multimodal but
auto- dominated,
generally
deprived
OM_ GLOB ( std) - 0.3 0.3 - 0.4 1.4
OM_ PI ( std) - 0.8 0.3 0.3 0.7
OM_ WALK 1 2 2 4
OM_ CAR 6 5 4 8
OM_ BUS 0 2 1 2
OM_ RAIL 0 1 0 2
SM_ GLOB ( std) - 0.9 0.6 0.3 0.7
SM_ PI ( std) - 1.0 0.7 0.5 0.6
SM_ WALK 2 3 3 3
SM_ CAR 4 4 4 4
SM_ BUS 1 2 2 2
SM_ RAIL 1 2 2 2
RDM_ GLOB ( std) - 0.2 - 0.6 0.3 0.0
RDM_ PI ( std) 0.3 - 2.7 0.2 0.1
RDM_ WALK ++ + ++ ++
RDM_ CAR + + ++ +
RDM_ BUS = – – – = =
RDM_ RAIL + – – ++ +
Number of cases 642 159 771 332
NB: std = standardized
Table 7
Thresholds in defining the 7- point RDM scale
Grade of the scale
Range of RDM means from
the French dataset ( the
original scale has 11 points)
Range of RDM means from
the U. S. dataset ( the
original scale has 5 points)
– – – [ 1.00 ; 4.00[ [ 1.00 ; 2.20[
– – [ 4.00 ; 5.30[ [ 2.20 ; 2.72[
– [ 5.30 ; 5.80[ [ 2.72 ; 2.92[
= [ 5.80 ; 6.20] [ 2.92 ; 3.08]
+ ] 6.20 ; 6.70] ] 3.08 ; 3.28]
+ + ] 6.70 ; 8.00] ] 3.28 ; 3.80]
+ + + ] 8.00 ; 11.00] ] 3.80 ; 5.00]
45
Figure 1. Hierarchical structure of the modes in the French dataset; grey modes are the observed
ones
Figure 2. Hierarchical structure of the modes in the US dataset; grey modes are the observed
ones
GLOBAL
Bicycle
Bus Tram Metro Suburban Intercity
Car driver Car pax. Urban transit
Moped /
motorcycle
Car Transit
Train
Taxi
GLOBAL
Bus
Walk Transit
Train
Car
46
Figure 3. Plot of OM_ PI for two modes and maximum frequency M = 15
Figure 4. Plot of OM_ MI for two modes and maximum frequency M = 15
47
48
Figure 5. Number of personal vehicles per household across the A- type clusters
Figure 6. Percentage of time a personal vehicle is available to the respondent across the A- type
clusters
0% 20% 40% 60% 80% 100%
Overall
4. Light travelers
3. Rather transit-oriented
2. Rather car-oriented
1. Heavily car-oriented
0
1
2
3
4
> 4
0% 20% 40% 60% 80% 100%
Overall
4. Light travelers
3. Rather transit-oriented
2. Rather car-oriented
1. Heavily car-oriented
0
20%
40%
60%
80%
100%
49
Figure 7. Educational level of the respondent across the A- type clusters
Figure 8. Household income across the A- type clusters
0% 20% 40% 60% 80% 100%
Overall
4. Light travelers
3. Rather transit-oriented
2. Rather car-oriented
1. Heavily car-oriented
Up to high sch.
Diploma
College/ tech
4- year degree
Graduate sch.
Graduate
0% 20% 40% 60% 80% 100%
Overall
4. Light travelers
3. Rather transit-oriented
2. Rather car-oriented
1. Heavily car-oriented
< 15k$
15-< 35k$
35-< 55k$
55-< 75k$
75-< 95k$
>= 95k$
50
Figure 9. Number of personal vehicles per household across the B- type clusters
Figure 10. Percentage of time a personal vehicle is available to the respondent across the B- type
clusters
0% 20% 40% 60% 80% 100%
4. Heavy travelers, multimodal but auto-dominated,
generally deprived
3. Light travelers, multimodal but auto-dominated,
generally deprived
2. Moderate travelers, multimodal but auto-dominated,
transit- surfeited
1. Light travelers, monomodal car users,
generally deprived
Overall
0
1
2
3
4
> 4
0% 20% 40% 60% 80% 100%
4. Heavy travelers, multimodal but auto-dominated,
generally deprived
3. Light travelers, multimodal but auto-dominated,
generally deprived
2. Moderate travelers, multimodal but auto-dominated,
transit- surfeited
1. Light travelers, monomodal car users,
generally deprived
Overall
0
20%
40%
60%
80%
100%
51
Figure 11. Educational level of the respondent across the B- type clusters
Figure 12. Household income across the B- type clusters
0% 20% 40% 60% 80% 100%
4. Heavy travelers, multimodal but auto-dominated,
generally deprived
3. Light travelers, multimodal but auto-dominated,
generally deprived
2. Moderate travelers, multimodal but auto-dominated,
transit- surfeited
1. Light travelers, monomodal car users,
generally deprived
Overall
Up to high sch.
Diploma
College/ tech
4- year degree
Graduate sch.
Graduate
0% 20% 40% 60% 80% 100%
4. Heavy travelers, multimodal but auto-dominated,
generally deprived
3. Light travelers, multimodal but auto-dominated,
generally deprived
2. Moderate travelers, multimodal but auto-dominated,
transit- surfeited
1. Light travelers, monomodal car users,
generally deprived
Overall
< 15k$
15-< 35k$
35-< 55k$
55-< 75k$
75-< 95k$
>= 95k$