Content uploaded by Francesco Piras
Author content
All content in this area was uploaded by Francesco Piras on Dec 04, 2019
Content may be subject to copyright.
Sottile, Piras, Meloni 1
COULD A NEW MODE ALTERNATIVE MODIFY PSYCHO-ATTITUDINAL FACTORS
AND TRAVEL BEHAVIOUR?
Eleonora Sottile
University of Cagliari
Via San Giorgio 12, Cagliari
Tel: (+39) 070-6756405; Email: esottile@unica.it
Corresponding author:
Francesco Piras
University of Cagliari
Via San Giorgio 12, Cagliari
Tel: (+39) 070-6756405; Email: francesco.piras@unica.it
Italo Meloni
University of Cagliari
Via San Giorgio 12, Cagliari
Tel: (+39) 070-6756405; Email: imeloni@unica.it
Sottile, Piras, Meloni 2
ABSTRACT
There is ample consensus that, besides objective characteristics, psycho-attitudinal factors play a
key role in influencing people’s mode choice. Hybrid choice models use these theoretical
frameworks so as to include latent constructs for capturing the impact of subjective factors on
mode choice. But recent work in transportation research raised the question about the ability of
hybrid choice models to derive policy implications that aim to change travel behaviour, given the
focus on cross-sectional data. To address this problem we designed a survey for collecting
longitudinal data (socioeconomic and psycho-attitudinal) so as to evaluate, on the one hand, the
long term effects on travel mode choice of the implementation of a new light rail line in the
metropolitan area of Cagliari (Italy), on the other to detect any changes in the psycho-attitudinal
factors and/or in socio-economic characteristics after implementation of those measures. In
particular, the objective of the study is to analyse whether these changes in individual
characteristics are able to affect mode choice from a modelling perspective, through the
specification and estimation of hybrid models. Our results show that latent variables were not
significantly different over waves, showing that the impact of the psychological construct
remained stable over time, even after the introduction of the new light rail. Additionally, we
found some evidence that the variables that explain the latent variables could change over time.
Keywords: Latent variables, Discrete choice models, Light rail, Panel data, Behaviour change
Sottile, Piras, Meloni 3
INTRODUCTION
There is ample consensus that, besides objective characteristics, psycho-attitudinal factors play a
key role in influencing people’s mode choice and their travel behaviour (e.g. (1), (2)). Different
theoretical frameworks have been developed over the years for studying the effect of cognitive
factors on behaviour, such as the Theory of Planned Behaviour (3), Theory of Interpersonal
Behaviour (4) and Schwartz’s Norm Activation Model (5).
Only in the last two decades have models of disaggregate decision-making started to use
these theoretical frameworks so as to include latent constructs for capturing the impact of
subjective factors on mode choice. These kinds of models, called hybrid choice models (HCM)
or integrated choice and latent variable (ICLV) models, proposed in the 1980's by McFadden (6)
and Train et al. (7), only became popular in 2002 following the work of Ben-Akiva et al. (8). An
increasing number of researchers have begun to adopt these models in the context of
transportation and logistics. Some examples include the study of travel mode choice (9) (10) (11)
(12), route choice (13) (14), departure time (15), fuel/vehicle type choice (16) (17), freight (18),
etc. Hence, hybrid choice models appear to be powerful and useful methods for improving
existing representations of decision-making (19) and for providing recommendations for travel
demand management policies. However, a recent work in transportation research (20) raised the
question about the ability of hybrid choice models to derive policy implications that aim to
change travel behaviour, pointing out two potential issues. Firstly, latent attitudes and
perceptions are partly endogenous with respect to travel behaviour and the analyst cannot be sure
about the direction of causality in attitudes and perceptions and choices (see for example (21)).
Moreover, latent variables and the observed choice could be influenced by the same underlying,
unmeasured factors. Secondly, the data collected are typically cross-sectional, measured at a
single moment in time. This means only between-person comparisons based on differences in
latent variables can be evaluated, and not within-person comparisons and how latent variables
may change over time due to a variation in socio-economic (SE) characteristics or the
implementation of travel demand management policies. Similar criticism has already been raised
in social psychology (e.g. (22), (23)).
Notwithstanding these requirements, the majority of recent papers infer policy
implications that are not adequately supported by the data used for hybrid choice model
estimation. So far, only a few works have tried to address the above issues, such as (24) and (25).
Jensen et al. (24) estimated a hybrid choice model using data gathered from a two-wave stated
preference experiment, before and after respondents experienced driving an electric vehicle, to
investigate the impact of individual preferences and attitudes on the choice between an electric
and a conventional vehicle. Beck and Hess (25) proposed a new hybrid choice model framework
to examine security preferences and attitudes in international air travel, conducting a repeated
stated preference experiment at two different points in time.
Surprisingly, to the authors’ knowledge no research has been published on the role and
evolution of cognitive factors before and after the implementation of a new travel alternative in
the choice context. While research has been conducted on the influence and casual effect of the
built environment on travel behaviour over time (26) or whether an information measure could
create a shift in psychological variables (27), none have directly examined whether the
introduction of a structural measure is able to change individuals’ cognitive factors.
The aim of the present paper is first of all to design a survey for collecting longitudinal
data (socioeconomic and psycho-attitudinal) so as to evaluate, on the one hand, the long term
effects on travel mode choice of the implementation of a new light rail line in the metropolitan
Sottile, Piras, Meloni 4
area of Cagliari (Italy), on the other to detect any changes in the psycho-attitudinal factors and/or
in socio-economic characteristics after implementation of those measures. In particular, the
objective of the study is to analyse whether these changes in individual characteristics are able to
affect mode choice from a modelling perspective, through the specification and estimation of
hybrid models that use for the same sample the data collected for these two moments in time.
Something we would like to stress is that the design and implementation of the survey for
collecting longitudinal data represents a contribution as well as the main criticism of the paper.
The difficulty in gathering information is well known; in this case the difficulty increased
because of the need to interview the same sample three times in the space of five years.
Using the data collected, this paper attempts to provide a contribution to the research,
analysing, from a statistical and modelling perspective, the evolution of travel behaviour over
time as well as individuals’ intrinsic characteristics (socioeconomic and attitudinal) following a
change in the context characteristics.
The first results of model estimation were not intended to derive policy implications, but
rather to understand whether the criticism raised about these models can actually be supported,
for the case at hand, by a scientific result.
The remainder of the paper is organized as follows: the next section describes the study context
and data collection, and this is followed by an exploratory analysis of the data. The hybrid choice
model set up is presented in the paragraph modelling framework followed by discussion of the
model results. The conclusions and suggestions for further research are outlined in the last
paragraph.
STUDY CONTEXT AND DATA COLLECTION
The transport context chosen for this study is a corridor linking the city centre of Cagliari (Italy)
to a university/hospital complex (Cittadella Universitaria), where in February 2015 a new light
rail route (METROCAGLIARI) went into service. Unexpectedly, in September 2015, the public
transport agency of the metropolitan area of Cagliari introduced a new bus route (“University
Express”), that connects one of Cagliari’s largest residential areas with the complex. Note that
the light rail line serves a fairly short corridor, and for some people is more convenient, in terms
of travel time, to use the bus because of its more extensive network. The Cittadella is a major trip
attractor, thus a large number of people could be intercepted. The number of people potentially
attracted daily to the Cittadella is just over 10,200: 1,784 university and hospital employees
(17.5%), 7,872 students (77.2%) and 580 (5.7%) for hospital admissions, medical examinations,
patients’ visitors, etc.
Well aware of the difficulties that a survey for collecting longitudinal data entails, a
major promotional campaign was conducted via traditional communication channels (postcards,
press conferences, TV and daily newspapers) and social media to intercept the greatest possible
number of respondents. Potential candidates were contacted via mailing lists provided by both
the university and hospital, requesting them to complete the questionnaires (8,847 invitation
mails were sent). The possibility of winning an iPad or a € 100 gift voucher was offered as an
incentive.
The programme, called Cittadella Mobility Style, comprised three surveys (Figure 1),
conducted at three different times:
• first survey wave, conducted one year before implementation of the measures (new light
rail line and new bus route), to capture individual travel patterns, SE characteristics and
psychosocial factors.
Sottile, Piras, Meloni 5
The aim was to intercept as many university students, university and hospital staff and
visitors to the hospital for medical examinations, admissions, etc., as possible. The target
population of potential light rail users was identified from all car drivers living along the
light rail corridor.
The first questionnaire was filled in by 2,886 individuals, 2,163 questionnaires were
complete (74.9%) and 723 partially completed (25.1%).
• Second wave, carried out three and four months respectively after the light rail began
operation, to evaluate any behaviour change.
The only goal of this wave was to monitor the travel behaviour of all respondents to the
first wave, after introduction of the light rail service (we were not aware of the
forthcoming new University Express bus route at that time), collecting the necessary
information for analysing the phenomenon. The second questionnaire was e-mailed to all
2,163 individuals who accurately completed the first one. 740 (34.2%) questionnaires
were completed. The second wave results are beyond the scope of this study, hence they
will not be described here.
• Third wave, conducted two years after the second following introduction of the new bus
service, to assess 1) whether the measures had long term effects and 2) to detect any
change in individuals’ SE and psycho-attitudinal characteristics.
This survey was designed such that the data collected were perfectly comparable with
those gathered in the first wave and included the same questions that appeared in the first
questionnaire.
However, having analysed the responses to the second wave, it was deemed useful to
include some questions about long term travel behaviour. As in the two years that had
passed, some of the participants might no longer travel to the Cittadella (students had
graduated, contracts finished, etc.) the participants were asked whether they used the light
rail to travel to other destinations/for other reasons and the trip characteristics (frequency,
purpose, etc.). The third questionnaire was also e-mailed to all 2,163 individuals who
accurately completed the first one, 522 (24.1%) questionnaires were completed and 464
(88.9%) of respondents travelled to the Cittadella.
A total of 350 people participated in all three surveys. However only 61% (215
individuals) who travelled to the Cittadella were included in the sample analysed for assessing
travel behaviour. The sample analysed for studying psycho-attitudinal factors and for hybrid
model estimation comprised instead 149 individuals, as out of the 215 identified, we eliminated
those who could not use at least two of the alternative travel modes considered (car, bus and
light-rail). Thus, all those included in the final sample had access to a car for the commute to the
Cittadella (Table 1).
The final sample of 149 individuals, just 6.9% of the questionnaires gathered in the first
wave, is the criticism raised above, concerning the collection of longitudinal data. During five
years, for different reasons, we lost 93% of the respondents.
DATA ANALYSIS
Table 2 provides an overview of the socio-economic variables and their descriptive
characteristics in the first (2013) and third wave (2017). Socio-economic characteristics included
gender, age, occupation, educational level, household composition and household income.
Clearly the number of students decreased significantly in the third wave, due to the fact
that the majority of them had graduated (percentage of graduates passed from 26.8% to 40.3%)
Sottile, Piras, Meloni 6
and had started work. Consequently, respondents’ levels of income have also increased, and as a
result we also observe an increase in car ownership in the third wave, in line with other studies
(28) (29). Moreover, some participants no longer live with their parents, proven by a slight
decrease in the average number of household members (Table 2).
Note that 26.1 % of the sample changed trip origins, and the number people who moved
from a location close to the metro corridor to a location distant from it and viceversa is equal to
9.3%.
Analysing the responses to the third questionnaire, it was possible to examine travel
behaviour, and hence modal share, of all the respondents who travelled to the Cittadella
following implementation of the measures. Figure 2 shows the modal share for the first and third
waves. Generally, there has been a slight decrease (-6.0%) in the use of the private car for
travelling to the Cittadella between two waves. Interestingly there is a sharp drop in the number
of people using the bus and an increase in travellers by light rail. This is in line with the findings
of other works (e.g. (30)), which argue that a rail transit service is able to attract significantly
more passengers than an express bus service.
Examining the data in detail, we observe that 16.1% individuals changed from private car
to bus and light rail, but 10.7% individuals changed from bus to private car. This increase in
private car use could be attributed to a change in the SE characteristics of the sample, especially
the average increase in income and a greater level of car ownership (note that in the meantime
many students had received research grants for furthering their education or for medical
specialisation or are now working) which are certainly two determinants of travel mode choice.
Evolution of psycho-attitudinal factors
One important aspect of the study concerned the definition and subsequent evaluation, in
two different moments of time, wave A and wave C, of those psycho-attitudinal factors that are
able to impact on travel behaviour and/or viceversa. To identify those factors to be measured, we
selected two focus groups, in order to understand the peculiarities, benefits/disadvantages,
motivation and limits associated with use of the private car and of public transport in general.
The first group was made up of habitual public transport users, the second of car drivers. Both
groups were invited to discuss identical topics, so as to capture the differences associated with
diverse travel habits. Analysis of the results enabled to identify, under the supervision of a team
of environmental psychologists, 36 items, measured by means of the 5-point Likert scale. Table
3 shows the questions asked, along with summary statistics of the responses in both waves of the
survey. T-tests were used to detect any significant relationship between the scores of the two
waves. Note that no significant differences were detected in attitudes between who dropped out
in wave 3 and who participated in the other waves.
At an aggregate level, some differences were detected in the answers to psycho-
attitudinal questions, with some average responses being significantly different over the two
waves. Examining the psycho-attitudinal factors measured, it emerges that during the time
elapsed between the two surveys:
1. Users reported greater attachment to their cars, consistent with the greater number of
car owners. In particular, there was a significant difference in the indicator A3, suggesting that
the perception of the car as the only means of transport compatible with daily commitments is
greater in spite of the introduction of the light rail and the new bus line. This is due to the limited
coverage of the light rail service in the urban area and to the fact that the new bus route only
serves a residential corridor, where few shops or public offices exist. Moreover, note that in the
Sottile, Piras, Meloni 7
third wave some students had started work, and therefore have less flexible working hours than
students.
2. By contrast, the sample exhibited less aversion to public transport, in keeping with the
findings of a well-controlled field experiment conducted in Sweden by Pedersen et al. (31). It is
interesting to observe a significant difference in the indicators B2 and B7. The first, related to the
regularity of the public transport service, scored lower in wave C, indicating that after the
introduction of the light rail, that runs along a rail corridor and is not slowed by vehicle traffic,
the sample consider the public transport service more reliable in terms of punctuality of the
service and fixed travel times. The second, concerning what kind of people use public transport,
also scored lower in the third wave. This is because the light rail is used not only by people who
cannot afford a car but also by high-income earners, who appreciate the benefits in terms of
travel time and a greater level of comfort than the bus.
3. Individuals were less willing to use the light rail. A possible explanation for this result
is that before it went into service, the expectations surrounding the new alternative stirred
people’s curiosity, mainly because, as found in other works, e.g. (30) (32), the light rail could be
perceived to offer a higher service quality than the bus.
4. For pro-environmental behaviour and environmental awareness statements, on average
we did not find any significant differences. This result shows a weak correlation between these
two constructs and the introduction of a structural measure in the choice context.
Interestingly, there are some differences in psycho-attitudinal factors across population
segments. For instance, taking into account the construct Attachment to the car several
distinctions can be observed in the answers provided by men and women across the two waves,
confirming that the psycho-attitudinal factors can change over time and this variation may
depend on individuals’ characteristics.
A confirmatory factor analysis was performed prior to modelling choices in order to
identify one or more latent dimensions (called factors or components) underpinning a set of
items or variables. Table 4 shows the factors with the linked items.
JOINT HYBRID CHOICE MODEL
The model framework we used is a latent variable model jointly estimated on a two-wave panel
dataset as shown in Figure 3. The discrete choice model (DCM) in our hybrid model is a
multinomial logit that incorporates latent variables to measure individual attitudes.
Because we collected data in two different time waves, before and after implementation
of the new light rail, the hybrid structure for each wave is jointly estimated to control for scale
differences between the two datasets to detect any differences in individual preferences and
attitudes between the two waves.
As in typical discrete choice models, we define
the utility that each individual q
associates with alternative j respectively in the first (w = A) and third wave (w = C). The discrete
part of the joint discrete choice model can be specified as:
Where
is a vector of individual socio-economic characteristics,
is a vector of travel
mode alternative attributes, is a latent variable, , ,
are vectors of coefficients
associated with the variables and are the alternative constants.
are the independently
Sottile, Piras, Meloni 8
and identically distributed Gumbel error terms for each wave and
is the scale
parameter that yields the same variance in both wave utilities.
Following the framework of hybrid choice models, we model each psychological
construct as a latent variable that depends on the socio-economic characteristics of each
individual q:
Where is the intercept, is the vector of the coefficients associated with the socio-economic
characteristics and
is the normally distributed error term, with zero mean and standard
deviation
.
can be different from the socio-economic characteristics included in the
discrete choice model and all coefficients are allowed to vary between waves.
The measurement equation of the discrete choice model is defined by a dummy variable
that takes the value one if the alternative chosen has the highest utility, zero otherwise:
where is the set of alternatives available to individual q in each wave.
The measurement equation of the latent variable is given by a set of R indicators
according to the following expression:
Where
is the intercept,
is the coefficient associated with the latent variable and
is an
error term that can have any distribution Q (assumed to have zero mean and standard deviation
. All the coefficients are also allowed to vary between the first and third wave.
Indicators are expressed in a five-point numerical scale, so the measurement equation of the
indicators is expressed as an ordered logit:
Where are thresholds defined respectively as , , , .
Because we assumed that
is i.i.d. Gumbel across alternatives, in each wave, the
probability that decision-maker q chooses alternative j is given by:
The joint probability for an individual q making the choice j for each period w is the
integral over the distribution :
Where
and
are the distribution of the latent variable and the
indicators, respectively.
The joint hybrid choice model was estimated using the software PythonBiogeme (33).
Sottile, Piras, Meloni 9
MODEL RESULTS
Three modes were considered to be available for the trip to Cittadella: 1) private car, 2) bus, 3)
light rail. The time and cost of travel for each mode were determined for each commuter, based
on the location of the person’s home and work. We simulated, for each individual, the values of
the attributes of the non-chosen available alternatives using Citlabs CUBE software. Travel time
was differentiated as walk time (for private car, bus and light rail) and in-vehicle time (for
private car, bus and light rail). Walking time enters separately for car, public transport and light
rail. In-vehicle time enters separately for car, public transport and light rail. The cost was entered
specific in the utility function of car, bus and light rail. Before estimating the joint hybrid choice
model, we tested various discrete choice models to gain a better knowledge of the phenomenon.
The most relevant estimation results of the best specifications are summarized in Table 5.
The estimation of the discrete part alone is shown in the first column of Table 5. The sign
of all level of service (LOS) coefficients is in line with the microeconomic theory.
The frequency attribute, which measures the number of times/year the trip to the
Cittadella is made, was positive when incorporated into the utility function of the car, indicating
a habit effect in car use.
In terms of socio-economic characteristics, not surprisingly, the level of personal income,
the number of cars per household and the car ownership positively affect the utility of the car
mode.
The scale parameter (θ) that allows for heteroscedasticity between waves was not
significantly different from one, meaning that the two datasets have the same variance.
Several hybrid choice models are estimated for measuring the effect of individual
characteristics on mode choice, accounting for attitudes and perceptions collected at two
different moments in time (first and third wave surveys). Importantly only the latent variable
Attachment to the car was found relevant for the purpose of the study. We also tested a number
of interaction terms between LOS variables (i.e. car travel time and public transport (PT) travel
time) and the latent variable Attachment to the car but none of the results turned out to be
statistically significant.
The last four columns in Table 4 show the results of estimating two hybrid choice models
that also include the latent variable Attachment to The Car. In model AC1 we used the indicators
of the psychological factor collected only in the first wave while model AC2 was estimated using
indicators of the psychosocial factor collected in both the first and third waves. This kind of
specification allowed us to quantify the error committed using the indicators of the psychological
factor collected in the first wave, for both waves.
As can be observed all the coefficients of the discrete part in the two hybrid choice
models are in line with the results of the DCM alone.
The latent variable was positive in both models for each wave, indicating that people who are
more attracted by the car for its high level of comfort, flexibility and shorter travel time have a
smaller level of disutility associated with the private vehicle.
As the coefficients of the latent variable model were introduced specific for each wave,
we were able to take into account whether 1) the Attachment to the car changed over time, even
if the analysis of indicator values (Table 3) did not detect significant changes, and 2) Attachment
to the car affects mode choice in different ways. We found that, for both models, the coefficients
associated with the latent variable were not significantly different over waves, showing that the
impact of the psychological construct remained stable over time, even after the introduction of
the new light rail and bus line. There is an intuitive explanation for this phenomenon: a certain
number of people during the timeframe of the survey started to use the car, developing a stronger
Sottile, Piras, Meloni 10
car dependence, so those who showed less attachment to the car due to the introduction of the
light rail and the new bus route were replaced by these.
The structural model for the first wave indicates that men have a higher level of
attachment to the car. This result is in line with the findings of other works (e.g. (32)), showing
that men, generally, tend to be more car dependent. Also, the number of car per driver in the
household positively impacts car attachment, in keeping with the results of the discrete part of
the model.
Remarkable, the explanatory variables relevant in the structural equation of the first wave
were no longer relevant in the structural equation of the third wave, except for the number of car
per driver in the household. This result shows that the explanatory variables related to the latent
variable can change over time, validating the criticism raised in (20), whereby using cross-
sectional data does not allow for within-person comparisons.
Table 6 shows the probability of choosing car and elasticity of demand for car with
respect to cost by car.
The disaggregate direct elasticity is computed according the following expression:
Where Pq is the choice probability that individual q chooses the car and cjq is the cost,
expressed in euros, associated with the alternative car for individual q. The results are similar in
all models, since the latent variable does not modify the marginal utilities.
However, the latent variable affects overall utility, hence choice probability. The model
AC1 slightly underestimates the choice probability in the third wave, compared to model AC2,
and this effect is more pronounced when the analysis is performed for those categories which we
found relevant for explaining the Attachment to the car. Nevertheless, the difference in choice
probability is quite small, and this might be viewed as an additional proof of the stability in the
overall level of the latent attitude between the two waves.
DISCUSSION AND CONCLUSIONS
This works focuses on the attempt to assess, in quantitative terms, the evolution of psycho-
attitudinal factors over time before and after the implementation of a new light rail and bus
service and their role in the change process.
This study brought to light the importance of, and at the same time the difficulties
encountered in, collecting longitudinal data. Unfortunately, with surveys comprising different
phases spread over time a large number of the initial participants may fall by the wayside, in
spite of a major promotional campaign. In the case at hand however, note that the metropolitan
area of Cagliari is relatively small (population 431,819), and for this reason it is difficult to
intercept a large number of individuals to ensure, after processing the responses, that sufficiently
large samples are obtained. Clearly this problem has repercussions on the results obtained for
behaviour change, as in numerical terms they may appear irrelevant. Moreover, the time gap
between the survey and implementation of the light rail (2 months and 2 years) may influence
attitudes towards a certain means of transport and be a limitation when analysing the results.
The lack of any standard questions on attitudes to be adopted in the transportation field
and the fact that we did not follow a specific psychological theory could pose a potential
limitation in interpreting the results. However, we argue that even the use of standardised
questionnaires may involve potential errors since they might not perfectly fit the specific case at
hand.
Sottile, Piras, Meloni 11
Another important result obtained concerns confirmation of considering data gathered
before and after the implementation of policy measures, also with respect to those psycho-social
attributes that could play a crucial role in change processes and might vary over time. If these are
not constantly and sufficiently measured, they would deprive us of the opportunity both to
measure them in modelling terms and analyse and evaluate whether they influence travel
behaviour or viceversa are affected by behaviour. Even more so when modelling results are
intended to provide indications for intervention policies aimed at incentivising preferably car
drivers to change their travel behaviour.
While this is one of the first attempts at performing an analysis of this kind and further
work is warranted, our findings show, from a modelling perspective, that the explanatory
variables used in the structural equation of the latent variable changed between the two waves,
accounting for within-person comparisons. As discussed by (20) the use of cross-sectional data
would not have allowed us to understand this kind of aspect and would have led us to derive
inappropriate policy implications from the model.
However, we are also aware that the estimation of the hybrid choice model allowed us to
understand that the latent factor Attachment to the car remains unchanged, even after the
introduction of the new bus route and light rail line. But, if these kinds of variables are stable
over time, this means for policy makers that the implementation of a structural measure does not
suffice to impact significantly individuals’ cognitive factors. These findings support the idea of
other studies that only the presence of a strong shock in the choice context (such as the
prohibition of a mode alternative) or the implementation of personalised information campaigns,
which focus on those factors that could diminish this emotional attachment, are able to trigger a
shift in people’s psycho-attitudinal characteristics.
Because of the small sample size, this study did not allow to define a rule about the
implications of Hybrid Choice Models, but it does provide a cue to investigate this issue in
different contexts using a robust sample. If 10, 100, 1,000 studies find the same conclusions,
these could become the correct answer to the criticism surrounding HCMs.
AUTHOR CONTRIBUTION
The authors confirm contribution to the paper as follows: study conception and design: Eleonora
Sottile, Francesco Piras; data collection: Italo Meloni; analysis and interpretation of results:
Eleonora Sottile, Francesco Piras, Italo Meloni; draft manuscript preparation: Eleonora Sottile,
Francesco Piras. All authors reviewed the results and approved the final version of the
manuscript.
Sottile, Piras, Meloni 12
REFERENCES
1. Koppelman, F. S., & Pas, E. I. (1980). Travel-choice behavior: models of perceptions,
feelings, preference, and choice. Transportation Research Record, (765).
2. Gärling, T., Gillholm, R., & Gärling, A. (1998). Reintroducing attitude theory in travel
behavior research: The validity of an interactive interview procedure to predict car use.
Transportation, 25(2), 129-146.
3. Ajzen, I. (1991). The theory of planned behavior. Organizational behavior and human
decision processes, 50(2), 179-211.
4. Triandis, H. C. (1979). Values, attitudes, and interpersonal behavior. In Nebraska
symposium on motivation. University of Nebraska Press.
5. Schwartz, S. H. (1977). Normative influences on altruism1. In Advances in experimental
social psychology (Vol. 10, pp. 221-279). Academic Press.
6. McFadden, D. (1986). The choice theory approach to market research. Marketing
science, 5(4), 275-297.
7. Train, K. E., McFadden, D. L., & Goett, A. A. (1987). Consumer attitudes and voluntary
rate schedules for public utilities. The Review of Economics and Statistics, 383-391.
8. Ben-Akiva, M., Walker, J., Bernardino, A. T., Gopinath, D. A., Morikawa, T., &
Polydoropoulou, A. (2002). Integration of choice and latent variable models. Perpetual
motion: Travel behaviour research opportunities and application challenges, 431-470.
9. Paulssen, M., Temme, D., Vij, A., & Walker, J. L. (2014). Values, attitudes and travel
behavior: a hierarchical latent variable mixed logit model of travel mode choice.
Transportation, 41(4), 873-888.
10. Vij, A., Carrel, A., & Walker, J. L. (2013). Incorporating the influence of latent modal
preferences on travel mode choice behavior. Transportation Research Part A: Policy and
Practice, 54, 164-178.
11. Kamargianni, M., & Polydoropoulou, A. (2013). Hybrid choice model to investigate
effects of teenagers' attitudes toward walking and cycling on mode choice behavior.
Transportation Research Record: Journal of the Transportation Research Board, (2382),
151-161.
12. Abou-Zeid, M., & Ben-Akiva, M. (2011). The effect of social comparisons on commute
well-being. Transportation Research Part A: Policy and Practice, 45(4), 345-361.
13. Bhat, C. R., Dubey, S. K., & Nagel, K. (2015). Introducing non-normality of latent
psychological constructs in choice modeling with an application to bicyclist route choice.
Transportation Research Part B: Methodological, 78, 341-363.
14. Prato, C. G., Bekhor, S., & Pronello, C. (2012). Latent variables and route choice
behavior. Transportation, 39(2), 299-319.
Sottile, Piras, Meloni 13
15. Thorhauge, M., Haustein, S., & Cherchi, E. (2016). Accounting for the theory of planned
behaviour in departure time choice. Transportation Research Part F: Traffic Psychology
and Behaviour, 38, 94-105.
16. Daziano, R. A., & Bolduc, D. (2013). Incorporating pro-environmental preferences
towards green automobile technologies through a Bayesian hybrid choice model.
Transportmetrica A: Transport Science, 9(1), 74-106.
17. Glerum, A., Stankovikj, L., Thémans, M., & Bierlaire, M. (2013). Forecasting the
demand for electric vehicles: accounting for attitudes and perceptions. Transportation
Science, 48(4), 483-499.
18. Bergantino, A. S., Bierlaire, M., Catalano, M., Migliore, M., & Amoroso, S. (2013).
Taste heterogeneity and latent preferences in the choice behaviour of freight transport
operators. Transport Policy, 30, 77-91.
19. Vij, A., & Walker, J. L. (2016). How, when and why integrated choice and latent variable
models are latently useful. Transportation Research Part B: Methodological, 90, 192-217.
20. Chorus, C. G., & Kroesen, M. (2014). On the (im-) possibility of deriving transport
policy implications from hybrid choice models. Transport Policy, 36, 217-222.
21. Kroesen, M., Handy, S., & Chorus, C. (2017). Do attitudes cause behavior or vice versa?
An alternative conceptualization of the attitude-behavior relationship in travel behavior
modeling. Transportation Research Part A: Policy and Practice, 101, 190-202.
22. Borsboom, D., Mellenbergh, G. J., & Van Heerden, J. (2003). The theoretical status of
latent variables. Psychological review, 110(2), 203.
23. Schwartz, S. H. (1978). Temporal instability as a moderator of the attitude–behavior
relationship. Journal of Personality and Social Psychology, 36(7), 715.
24. Jensen, A. F., Cherchi, E., & Mabit, S. L. (2013). On the stability of preferences and
attitudes before and after experiencing an electric vehicle. Transportation Research Part
D: Transport and Environment, 25, 24-32.
25. Beck, M. J., & Hess, S. (2017). On the Stability of Preferences and Attitudes: A Hybrid
Model of Air Travel Preferences at Two Different Points in Time (No. 17-05934).
Presented at 82rd Annual Meeting of the Transportation Research Board, Washington,
D.C., 2017.
26. van de Coevering, P., Maat, K., Kroesen, M., & van Wee, B. (2016). Causal effects of
built environment characteristics on travel behaviour: a longitudinal approach. European
Journal of Transport & Infrastructure Research, 16(4).
27. Jariyasunant, J., Abou-Zeid, M., Carrel, A., Ekambaram, V., Gaker, D., Sengupta, R., &
Walker, J. L. (2015). Quantified traveler: Travel feedback meets the cloud to change
behavior. Journal of Intelligent Transportation Systems, 19(2), 109-124.
28. Dargay, J. (2007). The effect of prices and income on car travel in the UK.
Transportation Research Part A: Policy and Practice, 41(10), 949-960.
Sottile, Piras, Meloni 14
29. Oakil, A. T. M., Manting, D., & Nijland, H. (2016). Determinants of car ownership
among young households in the Netherlands: The role of urbanisation and demographic
and economic characteristics. Journal of transport geography, 51, 229-235.
30. Ben-Akiva, M., & Morikawa, T. (2002). Comparing ridership attraction of rail and bus.
Transport Policy, 9(2), 107-116.
31. Pedersen, T., Friman, M., & Kristensson, P. (2011). Affective Forecasting: Predicting and
Experiencing Satisfaction with Public Transportation 1. Journal of Applied Social
Psychology, 41(8), 1926-1946.
32. Beirão, G., & Cabral, J. S. (2007). Understanding attitudes towards public transport and
private car: A qualitative study. Transport policy, 14(6), 478-489.
33. Bierlaire, M., & Fetiarison, M. (2009, September). Estimation of discrete choice models:
extending BIOGEME. In Swiss Transport Research Conference (STRC).
Sottile, Piras, Meloni 15
LIST OF FIGURES
FIGURE 1. Steps of the programme Cittadella Mobility Styles
FIGURE 2. Modal share between first and third wave
FIGURE 3. Model framework
Sottile, Piras, Meloni 16
LIST OF TABLES
TABLE 1 Data collection
First survey wave
(wave A)
Second survey wave
(wave B)
Third survey wave
(wave C)
People contacted
8,847
2,163
2,163
Questionnaires completed
2,163 (24.4%)
740 (34.2%)
522 (24.1%)
Participants travelling to the Cittadella
2,163 (24.4%)
516 (23.8%)
464 (21.4%)
Individuals who participated in all three waves 350 (16.1%)
Individuals who travelled to Cittadella 215 (9.9%)
Individuals who had available at least two alternatives among car, bus and light rail 149 (6.9%)
TABLE 1 Socio-economic characteristics between first wave and third wave survey
Variables
First wave
(wave A)
Third wave
(wave C)
N.
%
AVG
N.
%
AVG
Tot
149
-
-
149
-
-
Gender (male)
66
44.3%
-
66
44.3%
-
Age
35.1
39.1
Age 18_30
73
49.0%
-
60
40.3%
-
Age 31_40
23
15.4%
-
30
20.1%
-
Age 41_60
51
34.2%
-
48
32.2%
-
Over 60
2
1.3%
-
11
7.4%
-
Employment status
Student
71
47.7%
-
38
25.5%
-
Employed
76
51.0%
-
102
68.5%
-
Unemployed or retired
2
1.3%
-
9
6.0%
-
Level of education
Low (High school and
lower)
81
54.4%
-
49
32.9%
-
Medium (graduates
or master degree)
40
26.8%
-
60
40.3%
-
High (Higher than master
degree)
8
5.4%
-
40
26.8%
-
Number of household members
-
-
2.94
-
-
2.8
Children
35
23.5%
-
33
22.1%
-
Own car
118
79.2%
-
138
92.6%
-
Number of cars per household
-
-
1.9
-
-
1.8
Personal income per month
Income € 0 - 1,000
76
51.0%
-
59
39.6%
-
Income € 1,000 - 2,000
50
33.6%
-
66
44.3%
-
Income € 2,000 - 4,000
18
12.1%
-
21
14.1%
-
Income > € 4,000
5
3.4%
-
3
2.0%
-
Sottile, Piras, Meloni 17
TABLE 3 Psycho-attitudinal factors between first wave and third wave survey
ITEMS
To what extent do you agree with the following statements?
(Assign a score from 1=not at all to 5=very much)
Wave A
Wave C
Diff
T-stat
Mean
St dev
Mean
St dev
ATTACHMENT TO THE CAR
A1. The car is the most convenient means of transport in terms of
trip time
3.37
1.22
3.56
1.11
-0.19
-1.39
A2. The car offers a high level of comfort (comfort, privacy,
flexibility, etc.) that other forms of transport do not provide
3.90
1.09
4.03
0.96
-0.13
-1.07
A3. The car is the only means of transport compatible with daily
commitments (work, school runs, shopping, etc.)
3.05
1.29
3.33
1.26
-0.27
-1.86
A4. Driving is a pleasurable experience
2.86
1.31
2.76
1.31
0.10
0.66
A5. Driving gives a feeling of freedom that other means of
transport cannot provide
2.98
1.28
3.01
1.29
-0.03
-0.18
A6. Car use is a habit: one does not consider available alternatives
every time
3.50
1.24
3.45
1.37
0.05
0.33
A7. Owning a nice car is a sign of prestige and a status symbol
1.83
1.13
1.91
1.18
-0.08
-0.6
A8. The car is a means of self-expression and a reflection of
personal taste
1.72
1.00
1.81
1.05
-0.09
-0.73
AVERSION TO PUBLIC
TRANSPORT
B1. Travel times are too long
3.52
1.11
3.44
1.14
0.08
0.62
B2. Services are not reliable in that they do not guarantee
regularity and certainty of travel times
3.44
1.09
2.92
1.25
0.52
3.81
B3. Comfort is poor (overcrowding, carrying bulky goods, etc.)
3.47
1.10
3.17
1.18
0.30
2.28
B4. The service is not compatible with daily commitments (work,
school runs, shopping, etc.)
3.36
1.16
3.45
1.22
-0.09
-0.63
B5. Travelling on public transport is not a pleasurable experience
2.69
1.14
2.48
1.11
0.21
1.65
B6. Public transport is unpopular because people do not like
depending on others to get around
2.99
1.27
3.03
1.24
-0.04
-0.28
B7. Only those who do not have alternatives use public transport
as they are obliged to do so.
2.84
1.27
2.41
1.14
0.43
3.07
B8. Public transport use is commonly associated with modest
social and economic condition
2.03
1.24
1.91
1.13
0.13
0.93
WILLINGNESS TO USE
THE LIGHT RAIL
C1. I would use the light rail if travel times were shorter
4.23
1.09
3.86
1.25
0.37
2.72
C2. I would use the light rail if fares were cheaper
4.33
0.93
3.83
1.22
0.50
4.01
C3. I would use the light rail if CO2 emissions were reduced
4.30
0.97
3.96
1.14
0.34
2.79
C4. I would use the light rail if it was less stressful than driving
4.52
0.81
4.31
0.97
0.21
2.01
C5. I would use the light rail if the network was extended and the
number of lines increased
4.67
0.67
4.56
0.85
0.11
1.29
C6. I would use the light rail if the service was free
4.31
1.15
3.93
1.35
0.39
2.68
C7. I would use the light rail if there was a free wi-fi service on
board
3.94
1.23
3.41
1.40
0.54
3.51
PRO ENVIRONMENTAL
BEHAVIOUR
D1. I unplug electronic devices when they are not in use (e.g. TV,
phone charger, etc.)
3.98
1.23
3.82
1.16
0.15
1.11
D2. I use low-energy light bulbs
4.36
0.94
4.40
0.90
-0.04
-0.38
D3. I do not waste water
4.20
1.03
4.30
0.98
-0.10
-0.87
D4. I buy local fruit and vegetables, which are not transported by
plane or lorries
3.95
1.15
4.12
1.05
-0.17
-1.36
D5. When shopping, I use my own reusable bag instead of the
plastic bag provided by the supermarket
4.46
0.98
4.40
0.91
0.06
0.55
D6. I use public transport to deliberately reduce the air pollution
caused by car use
2.98
1.40
3.04
1.27
-0.06
-0.39
D7. For short trips, I cycle or walk, rather than taking the car
3.97
1.23
4.03
1.15
-0.06
-0.44
ENVIRONME
NTAL
AWARENESS
E1. It is very important to be aware of how one’s own actions can
impact the environment
4.56
0.73
4.50
0.85
0.05
0.58
E2. Environmental awareness is a very important personal
characteristic
4.49
0.75
4.42
0.90
0.07
0.77
E3. Human activities are seriously abusing the environment and
its resources
4.62
0.71
4.67
0.72
-0.05
-0.57
Sottile, Piras, Meloni 18
E4. Pro-environmental behaviour is very satisfying
4.32
0.90
4.33
0.89
-0.01
-0.06
E5. Daily use of the car is one the most environmentally harmful
human activities
3.91
0.97
3.91
1.07
0
0
E6. Using public transport for daily trips helps considerably to
improve our environment
4.23
0.84
4.20
0.90
0.03
0.33
TABLE 4 Factor analysis
Attachment to the car
A1. A2. A3. A4. A5.
Aversion to public transport
B1. B2. B3. B4. B5.
Willingness to use the light rail
C1. C2. C3. C4. C5.
Pro-environmental behaviour
D1. D2. D3. D4. D5.
Environmental awareness
E1. E2. E3. E4. E5. E6.
Sottile, Piras, Meloni 19
TABLE 5 Model results
DCM alone
ATTACHMENT TO THE CAR
HCM - AC1
HCM - AC2
Attributes
values
R t-test
values
R t-test
values
R t-test
Discrete part
Constant car
-5.05
-3.21
-4.14
-1.50
-4.65
-1.65
Constant bus
-3.61
-2.85
-2.16
-1.28
-2.24
-1.31
Car attributes
Travel time
-0.08
-1.70
-0.07
-2.02
-0.07
-2.05
Travel Cost
-0.34
-1.39
-0.22
-0.99
-0.22
-0.98
Walking Time from/to parking area
-0.02
-0.40
-0.02
-0.68
-0.01
-0.58
Bus attributes
Travel Time
-0.05
-2.66
-0.04
-2.47
-0.01
-2.53
Walking Time from/to Bus stop
-0.06
-1.70
-0.06
-1.84
-0.06
-1.80
Light rail attributes
Travel Time
-0.10
-3.39
-0.07
-2.22
-0.07
-2.23
Walking Time from/to Light rail station
-0.18
-2.82
-0.12
-1.50
-0.12
-1.57
Bus and Light rail attributes
Cost
-0.81
-1.87
-0.63
-1.54
-0.60
-1.45
Socio-economic characteristics (specific to car)
Personal income
0.28
1.30
0.22
1.01
0.24
1.06
Number of cars per driver in the household
0.87
1.73
0.45
0.84
0.48
1.00
Car ownership (Yes = 1; No = 0)
0.85
2.03
0.68
1.35
0.68
1.38
Other characteristics of the trip (specific to car)
Frequency of trips from origin to Cittadella
0.003
1.57
0.002
1.23
0.002
1.31
Scale factor θ (R t-test against 1)
1.30
1.03
2.10
0.79
2.09
0.79
LV attachment to car wave A
-
-
0.40
1.49
0.52
1.28
LV attachment to car wave C
-
-
0.20
0.74
0.31
2.08
Latent variable model
Latent variable Attachment to the CAR wave A structural equation
Intercept
-
-
3.74
10.50
3.83
13.19
Standard deviation of error term
-
-
0.80
5.95
0.78
5.78
Gender (Man = 1; Woman = 0)
-
-
0.33
2.00
0.30
1.90
Number of cars per driver in the household
-
-
0.44
1.40
0.46
1.48
Worker dummy (Yes = 1; No = 0)
-
-
-0.70
-1.90
-0.78
-2.63
Student dummy (Yes = 1; No = 0)
-
-
-0.36
-1.02
-0.45
-1.73
Latent variable Attachment to the CAR wave C structural equation
Intercept
-
-
-
-
4.01
13.53
Standard deviation of error term
-
-
-
-
0.54
3.47
Age
-
-
-
-
-0.01
-1.69
Number of cars per driver in the household
-
-
-
-
0.28
1.39
Initial log-likelihood
-251.16
-1,563.60
-2,846.83
Final log-likelihood
-134.74
-1,211.96
-2,278.60
Adjusted ρ2
0.41
0.20
0.18
Sottile, Piras, Meloni 20
TABLE 6 Probability and elasticity
Probability of choosing car
Elasticity of the demand respect to cost by car
MODEL AC1
MODEL AC2
MODEL AC1
MODEL AC2
First wave
83.3%
83.2%
-0.075
-0.075
Female
83.5%
83.1%
-0.075
-0.076
Student
81.0%
81.1%
-0.084
-0.083
Worker
85.4%
85.0%
-0.064
-0.065
Third wave
77.9%
78.3%
-0.197
-0.194
Female
75.9%
77.2%
-0.207
-0.194
Student
67.9%
68.4%
-0.315
-0.310
Worker
81.9%
82.7%
-0.151
-0.144
