International Journal of Sustainable Transportation, 8:28–46, 2014

Copyright # Taylor & Francis Group, LLC

ISSN: 1556-8318 print=1556-8334 online
DOI: 10.1080/15568318.2012.758460

Using Stated Preferences to Analyze the Service

Quality of Public Transport

Concepción Román, Juan Carlos Martı́n, and Raquel Espino

Institute of Tourism and Sustainable Economic Development (TiDES),
University of Las Palmas de Gran Canaria, Spain

ABSTRACT
In this article we use discrete choice experiments to analyze service quality of
public transport bus services in Gran Canaria, Spain. Discrete choice experiments
are created to estimate travellers’ preferences and to obtain a measure of the
global service quality. The modeling strategy considers the combination of two
data sets, the analysis of the inertia effect as well as the correlation among
responses of the same individual. Results suggest the existence of behavioral differ-
ences for urban and interurban passengers, especially in their perception of cer-
tain attributes as well as in the willingness to pay for improving the level of service.
Key Words: public transport, quality of service, stated preferences, willingness
to pay

1. INTRODUCTION
Measuring the quality of public services is currently one of the major challenges
for economic analysis; this challenge is due to the importance of its results both for
companies that provide these services and for the supervisory public administra-
tion. One relevant variable in evaluating public transport is the level of services
offered. Typically, the demand side is measured in terms of passengers or passen-
gers per kilometer, and the supply side using vehicles per kilometer. As with the
other grouped output indicators, these aggregate measures implicitly assume the
quality of service to be homogeneous. However, in practice, passengers evaluate
the transport services in various ways, using criteria that are not associated with
the level of transport usage. For example, no relationship between the level of sat-
isfaction of passengers and the number of passengers per kilometer has been
demonstrated (Hensher and Prioni 2002).

Received 19 April 2011; revised 29 December 2011; accepted 11 April 2012.

Address correspondence to Concepción Román, Institute of Tourism and Sustainable
Economic Development (TiDES), University of Las Palmas de Gran Canaria, Juan de
Quesada, no. 30, Las Palmas de Gran Canaria 35001, Spain. E-mail: [email protected]

28
 Analysis of the Service Quality of Public Transport

Despite the maximization of the number of passengers per kilometer being used
as an indicator to represent the maximization of social welfare and passenger
accessibility, there is little research that analyzes how the passengers perceive the
quality of services. Among the available papers, attention should be drawn to the
pioneering article by Ortúzar, Ivelic, and Candia (1997), where the most relevant
attributes for public transport users were identified as well as the perceived level
of importance for these aforementioned variables. More recently, David Hensher
at the University of Sydney oversaw research that developed a methodology specifi-
cally designed for constructing quality indicators for public bus transport services in
Australia; the starting point was stated preferences (SP) designs (for a more detailed
reference see Hensher 2000, 2002a, 2002b; Hensher and Prioni 2002; Hensher and
Houghton 2002, 2004; Hensher, Stopher, and Bullock 2003). Among the most
recent studies adopting the Hensher approach, we can cite Cirillo, Ebola, and Maz-
zulla (2011), Eboli and Mazzulla (2008, 2010), and Gatta and Marcucci (2007).
Other studies based on the methodology proposed by Cook and Kress (1988)
aim to obtain the relative weightings of the attributes that determine the level of ser-
vice, by analyzing the ranking of the attributes (see e.g., Garrido and Ortúzar 1994).
This article has a double objective. On the one hand, to obtain indicators that
permit the aggregate measurement of the quality of the public transport bus ser-
vices in the island of Gran Canaria, Spain; and on the other, to obtain knowledge
of passengers’ willingness to pay to improve the service’s main defining factors.
In order to determine the level of satisfaction of the public transport services, pas-
sengers jointly evaluate the various attributes that represent different aspects of the
level of service; for example travel time, frequency, comfort, punctuality, infor-
mation at the bus stops, and so on. Some of these attributes are perceived positively
(e.g., service frequency), whereas others have a negative connotation (e.g., waiting
time). The idea is to find the weighting that individuals use to evaluate each attri-
bute within what is considered as the level of global satisfaction or utility. In this way,
it is possible to understand the contribution of each basic attribute to the global
measurement of quality. The analysis focuses on obtaining the preferences using
SP designs that face the individual to the choice between the current bus service
and a hypothetical service, obtained from a combination of changes in the most rel-
evant attributes. As a novel aspect regarding previous literature, this article consid-
ers the mixture of two SP data sets in order to account for the effect of a relatively
large number of attributes. The main motivation of this research is to provide pub-
lic transport authorities with valuable instruments that help in the management of
the bus transport network as well as in the evaluation of public transport policies.
With this information, discrete choice models are estimated1 in order to obtain
the weights for the (parameters hi) of the attributes (Xi) in the utility function,
P which
is generally represented by the linear-in-the-parameter expression U ¼ hi Xi . The
i
modeling strategy considers the study of inertia and panel effects, being the latter
analyzed through the specification of mixed Logit models that include an error
component to account for the correlation among choices of the same respondent.

1
See McFadden (1981), Domencich and McFadden (1975), and Ortúzar and Willumsen
(2001), in order to consult the theoretical foundations of the discrete choice models.

International Journal of Sustainable Transportation Vol. 8, No. 1, 2014 29

 C. Roman et al.

Once the parameters have been estimated, the service-quality indicators (SQI)
for each route, company, group of passengers, and so on are obtained by applying
the utility expression to the average value of the attributes; taking into consider-
ation the levels achieved in the current trip for the specific group. For example,
P Qj
P
j j j j
the SQI for route j is given by: SQj ¼ hi X i ; where X i ¼ Q1j Xiq , Xiq is the value
i q¼1
of the attribute i in the current trip for the individual q who travels route j and Qj is
the number of passengers in the sample for route j.

2. DATA
The public transport services in Gran Canaria (Spain) are provided by two com-
panies: Guaguas Municipales,2 which is in charge of urban transport; and Global,
which is responsible for interurban transport. The analysis is based on the infor-
mation provided by 551 urban and 1052 interurban passengers. Beforehand, in
order to determine the sample, a classification of the different routes using cluster
analysis was carried out. Afterwards, one representative route for each group was
chosen, and then passengers on that route were selected randomly at different stops.
The sample size for each route was proportional to the total number of passengers.
The surveys were carried out by a group of fully trained interviewers who used
laptop computers to track passenger’ responses.3 The questionnaire was com-
pleted by urban and interurban transport users and consisted of three blocks of
questions. The first brought together information about the journey. The second
block was composed of a stated-preferences experiment that presented a choice
between the current service4 and a hypothetical service that was designed using
a combination of the most relevant attributes. The third and final part of the ques-
tionnaire collected socioeconomic information at both the household and the
individual level; this group of questions was strategically placed at the end of the
interview. Had the order been different, then the confidential nature of many of
these questions may have produced a negative reaction and less information about
the service might have been obtained.
The final set of attributes was decided using the information obtained after sev-
eral focus group meetings, which were attended by a wide range of representatives
from companies and the island’s transport administration. The attributes and
levels used in the SP experiment are shown in Table 1.5 With the aim of making
2
In the Canary Islands the word ‘‘guagua’’ is used to refer to a bus.
3
To construct the experiments we used the WinMint program created by Rand Europe,
formerly known as The Hague Consulting Group.
4
The objective was to analyze the level of perceived global utility achieved by the current
public bus transport service.
5
The experiment was constructed with an orthogonal design, which only permitted the
estimation of the main effects. More recently, it has been demonstrated that the use of
efficient designs, based on minimizing the D-error, give rise to more robust estimations of
the parameters for a given sample size; see Bliemer and Rose (2006) for further details.
Nevertheless, as the sample size increases, the differences in the standard deviations of
the estimators obtained by the efficient and orthogonal designs decrease.

30 International Journal of Sustainable Transportation Vol. 8, No. 1, 2014

 Analysis of the Service Quality of Public Transport

Table 1. Attributes and levels used in the SP experiment.

Attribute Level 0 Level 1 Level 2

1 Travel time T þ 20% T T
20%

2 Price P þ 20% P P
20%
3 Time between two F þ 25% F F
25%
services (frequency)
4 Access time to Ta þ 5 min Ta Ta
5 min
bus stops
5 Delay (only in 10 min delay 5 min delay On time
interurban
transport)
6 Comfort Standing during Standing in part Seated during all
all the journey of the journey the journey
7 Information at the Itinerary maps Maps and Electronic panels,
bus stop (urban timetable maps and
transport) timetables
8 Information at the None Maps and Electronic panels,
bus stop (interurban timetable maps and
transport) timetables
9 Shelter No Yes –
10 Driver behavior Unpleasant Pleasant treatment Pleasant treatment
treatment and aggressive and smooth
and aggressive driving driving
driving
11 Cleanliness Poor Normal Good

Note. T ¼ current travel time; P ¼ current price; F ¼ current frequency; Ta ¼ current access
time.

the experiment realistic, the levels of the variables (1) travel time, (2) prices, (3)
frequency, and (4) access time were customised to respondents pivoting from the
attribute values associated to the current service. In the case of access time, in
order to avoid small variations, we considered deviations in absolute values rather
than in percentage changes. The number of attributes was found, in our context,
relatively high for this kind of experiment, and could give rise to the individual
having difficulties evaluating the information provided by the whole set of attri-
butes simultaneously (Caussade et al. 2005), so in each application there were
two games having in common the most relevant attributes.6 Each game consisted
of six hypothetical choices that were randomly selected by the computer, and the
individual had to choose between the current and the alternative service.7
With urban transport, the scenarios for the first game were based on the varia-
tions in the levels of the following attributes: (1) travel time, (2) price, (3) frequency,

6
This issue was extensively discussed with public transport authorities, bus companies’
managers and public transport users.
7
In recent research either modular arithmetics has been used to divide the design into
balanced blocks, or efficient designs whereby the researcher predetermines the number
of situations of choice to be employed.

International Journal of Sustainable Transportation Vol. 8, No. 1, 2014 31

 C. Roman et al.

Figure 1. Attributes included in the choice games.

(4) access time, (5) comfort, and (6) information. The second game for urban
transport included (1) travel time, (2) price, (3) frequency, (4) driver attitude,
(5) cleanliness, and (6) the bus shelter.
For the designed interurban transport application, the first game included (1)
travel time, (2) price, (3) frequency, (4) access time, comfort and punctuality. The
attributes included in the second game for interurban transport were (1) travel
time, (2) price, (3) information, (4) the driver, (5) cleanliness and (6) the bus
shelter. Figure 1 shows a schema of the attributes included in each choice game.
Attributes in bold are considered common to both data sets in the mixed data esti-
mation process.

3. MODELS AND RESULTS

3.1. Methodology
The analysis of passenger preferences is based on the exploitation of the two SP
databases obtained from the games presented in the questionnaire. The estimation
methodology with mixed data, as proposed by Bradley and Daly (1997), permits
databases with different characteristics to be combined. This way it is possible to
estimate the marginal effect of each of the attributes included in the experiment.
The combined use of different databases to estimate modal choice models is
based on the hypothesis that the difference between the stochastic errors of the
utilities can be represented in terms of its variances in accordance with the
expression (Ben-Akiva and Morikawa 1990):

r2e ¼ l2 r2g ð1Þ

where l is an unknown parameter and r2e and r2g represent the variances of the
random terms e and g for the utilities of the two respective SP databases. In order
to be able to combine the two databases, the following utility functions are set out
for an alternative j that is present in the two databases:

lUjSP1 ¼ lðVjSP1 þ gj Þ ¼ lðhXjSP1 þ xZjSP1 þ gj Þ

UjSP2 ¼ VjSP2 þ ej ¼ hXjSP2 þ aYjSP2 þ ej ð2Þ

32 International Journal of Sustainable Transportation Vol. 8, No. 1, 2014

 Analysis of the Service Quality of Public Transport

Figure 2. Artificial tree structure; estimation with mixed data.

where h, a and x are parameters to be estimated, XjSP1 and XjSP2 are common attri-
butes to the two databases, and ZjSP1 and YjSP2 are attributes that are held in only
one of the databases.
A prior analysis, based on the separate estimation of the two SP sets, permitted us
to verify that the parameters of the common attributes varied in the same pro-
portion, in accordance to the recommendations by Louviere et al. (2000). There-
fore, the application of this method allowed us to analyze the effect of a large
number of attributes without increasing the respondent burden. As the purpose
here is to account for scale differences in the error terms for the two data sets,
the artificial tree structure (known as the ‘‘Nested Logit trick’’) that was proposed
for the estimation with mixed data is depicted in Figure 2. The total number of
alternatives are twice as many alternatives as there are in reality and the alternatives
corresponding to the game two are placed just bellow the root of the tree; whereas
the alternatives corresponding to the game one are each placed in a
single-alternative nest with identical nest inclusive value. For every observation in
game one, the alternatives of game two are set unavailable and vice versa (Ortúzar
and Willumsen 2001).

3.2. Model Results

For the urban and interurban transport Multinomial Logit (MNL1 and MNL2)8
type models and Mixed Logit (ML1) were estimated by using the artificial tree
structure proposed in the methodology for estimations with mixed data. All the
variables were considered to be linear, except for the travel costs which were div-
ided by the expenditure rate (g); this was defined as R=(T
Wh), where R is the
individual income, T is the total time for the reference period and Wh are the work-
ing hours.9 In this way the effect of income upon the decisions of the passengers is
8
In fact, and properly speaking, MNL1 and MNL2 are Nested Logit models, and
consequently are subject to the assumption that observations of the same respondent are
treated as independent.
9
In this study one week is considered as the reference period, and also the per capita
indicator for the family income.

International Journal of Sustainable Transportation Vol. 8, No. 1, 2014 33

 C. Roman et al.

included, since the marginal utility of income (k), obtained as minus the marginal
utility and travel cost ð
@V =@C Þ, is different for each person.10 As each individual
responds to six different choice situations in each set, the effect of the inertia and
the correlation among the responses of a specific individual are analyzed. Model
MNL2 has an inertia variable (I) included, which groups together the effects that
previous decisions have upon the current choice. This variable is specified in the
utility of the current option11 and is defined as the number of times the current
alternative has been previously chosen (Adamowicz 1994).12 For example:

Choice situation Chosen option Inertia (Current option 2)

1 1 0
2 2 0
3 2 1
4 1 2
5 2 2
6 2 3

The model ML1 analyzes both the inertia and correlation effect among choices of the
same respondent. Following the recommendations of Hensher, Rose, and Greene
(2008) a Panel Mixed Logit13 specification with fixed parameters has been considered,
but with an error component (lSPi) in each set that follows a normal distribution
with zero mean. Differences in scale are also analyzed introducing the scale factor
l in the model specification.14 Thus the utility of the alternatives is represented by:

Current service SP1: lU1SP1 ¼ lðV1SP1 þ g1 Þ ¼ lðhX1SP1 þ xZ1SP1 þ g1 Þ

Alternative service SP1: lU2SP1 ¼ lðV2SP1 þ g2 Þ
¼ lðhX2SP1 þ xZ2SP1 þ lSP 1 þ g2 Þ; with lSP 1 ! N ð0; r1 Þ
Current service SP2: U1SP2 ¼ V1SP2 þ e1 ¼ hX1SP2 þ aY1SP2 þ e1
Alternative service SP2: U2SP2 ¼ V2SP2 þ e2 ¼ hX2SP2 þ aY2SP2 þ lSP 2 þ e2 ;
with lSP 2 ! N ð0; r2 Þ

10
Other studies have demonstrated the existence of systematic variation in tastes in the context
of urban transport; they consider the interactions of socioeconomic variables with the attributes
of the level of service; see Ortúzar et al. (1997) and Espino, Roman, and Ortúzar (2006). How-
ever, in this case the specification has not been set out, since the study’s main objective is to
obtain a measurement of the global perception of each attribute. Nevertheless, the variation
analysis of both systematic and random tastes remains pending for future research.
11
Alternative two in the first set and alternative four in second one.
12
Although the inertia effect has been widely analyzed in the literature, a more recent
research by Chorus and Dellaert (2010) reveals that inertia is due to a combination of risk
aversion and learning and suggesting for future research the incorporation of travellers’
strategic behavior.
13
See Train (2003) and Hensher et al. (2008) for a complete reference of the Mixed Logit
model.
14
In a recent paper, Collins, Rose, and Hess (2012) presents an innovative method to
account for scale heterogeneity between different data sets, imposing homogeneity for some
parameters after accounting for scale differences.

34 International Journal of Sustainable Transportation Vol. 8, No. 1, 2014

 Table 2. Urban transport; estimation results.
MNL1 MNL2 ML1
Attributes Parameter t-test Parameter t-test Parameter t-test

Travel time (T) hT
0.1337 (
8.4)
0.1480 (
8.86)
0.1510 (
8.38)

Travel cost=g (C=g) hC=g
4.2110 (
9.1)
4.4100 (
9.09)
4.7600 (
8.97)
Frequency (F) hF
0.1402 (
11.3)
0.1230 (
10.34)
0.1530 (
10.73)
Access time (TA) hTA
0.0717 (
5.2)
0.0646 (
4.84)
0.0754 (
4.95)
Comfort Level 1 (C1) hC1 0.6285 (5.0) 0.6350 (5.41) 0.7730 (5.06)
Comfort Level 2 (C2) hC2 1.2180 (7.5) 1.0900 (7.59) 1.4400 (7.35)
Information Level 1 (I1) hI1 0.0225 (0.3) 0.0499 (0.8)


Information Level 2 (I2) hI2 0.5581 (5.4) 0.5220 (5.55) 0.6470 (5.26)
Driver level 1 (CH1) hCH1 0.4215 (4.0) 0.3420 (3.24) 0.4870 (3.93)
Driver level 2 (CH2) hCH2 1.2120 (10.5) 1.1400 (9.81) 1.4300 (9.37)
Cleanliness level 1 (L1) hL1 0.5133 (5.2) 0.5080 (5.16) 0.5820 (4.57)
Cleanliness level 2 (L2) hL2 1.0020 (9.7) 1.0700 (10.35) 1.2300 (9.52)
Shelter (M) hM 0.3168 (3.7) 0.2350 (2.72) 0.3270 (2.94)
Constant alt 1 SP1 C1
0.1372 (
2.8) 0.1540 (2.6)
0.0727 (
0.88)
Constant alt 1 SP2 C2
0.5824 (
7.1)
0.1530 (
1.43)
0.5310 (
3.73)
Inertia (I) hI – – 0.2270 (7.28) 0.0650 (1.57)
Scale factor SP l 1.0610 (8.9) [0.5] 1.1400 (9.13) [1.12] 1.0600 (8.8) [0.5]
Sigma 1 r1 – – – – 0.7640 (6.44)

International Journal of Sustainable Transportation Vol. 8, No. 1, 2014

Sigma 2 r2 – – – – 0.9160 (8.16)
l (0)
4579.6234
4579.6234
4579.6234
l (h)
3043.1248
3010.512
2981.454
Observations 6607 6607 6607

Test t with respect to l ¼ 1.

35
36
Table 3. Interurban transport; estimation results.
MNL1 MNL2 ML1
Attributes Parameter t-test Parameter t-test Parameter t-test

Travel time (T) hT
0.0794 (
9.8)
0.0484 (
6.74)
0.0848 (
9.29)

Travel cost=g (C=g) hC=g
1.8250 (
12.4)
1.3900 (
9.26)
1.9500 (
11.46)
Frequency (F) hF
0.0598 (
5.6)
0.0270 (
6.15)
0.0501 (
6.16)
Delay (R) hR
0.4889 (
6.0)
0.2450 (
7.8)
0.4420 (
6.98)
Walking access time hTA W LP
0.0798 (
1.7)
0.0495 (
2.2)
0.1100 (
2.89)
in LP (TA W LP)
Comfort Level 1 (C1) hC1 2.2740 (3.8) 1.2900 (4.94) 1.6500 (3.76)
Comfort Level 2 (C2) hC2 7.8750 (5.6) 3.4900 (7.27) 6.4600 (6.45)
Information Level 1 (I1) hI1 0.7072 (8.8) 0.6020 (7.35) 0.7140 (6.9)
Information Level 2 (I2) hI2 1.2370 (11.7) 1.1700 (10.88) 1.3300 (10.76)
Driver level 1 (CH1) hCH1 0.3107 (2.8) 0.8430 (7.22) 0.6070 (4.57)
Driver level 2 (CH2) hCH2 1.4000 (13.4) 1.9200 (17.32) 1.8700 (14.95)
Cleanliness level 1 (L1) hL1 0.2844 (3.1) 0.2550 (2.75) 0.3890 (3.46)
Cleanliness level 2 (L2) hL2 0.8159 (9.0) 0.7650 (8.38) 1.0600 (9.01)
Shelter (M) hM 0.0704 (1.0) 0.1090 (1.55) 0.0962 (1.03)
Constant alt 1 SP1 C1 1.8530 (4.6) 1.1000 (7.26) 1.7100 (5.44)
Constant alt 1 SP2 C2
0.8243 (
10.2)
0.1310 (
1.36)
0.6050 (
4.68)
South corridor (S) hS 0.2955 (4.3) 0.1750 (2.98) 0.3100 (3.46)
Inertia (I) hI – – 0.3330 (13.27) 0.1550 (4.64)
Scale factor SP l 0.3815 (5.9) [–9.51] 0.8120 (7.75) [
1.79] 0.5570 (6.77) [
5.38]
Sigma 1 r1 – – 2.0600 (6.3)
Sigma 2 r2 – – 0.9180 (11.35)
l (0)
8747.5174
8747.5174
8747.5174
l (h)
5131.140
5045.493
4950.601
Observations 12620 12620 12620

Test t with respect to l ¼ 1.

International Journal of Sustainable Transportation Vol. 8, No. 1, 2014

 Analysis of the Service Quality of Public Transport

Where X represents the vector of common attributes, Z and Y are the vectors of
specific attributes to SP1 and SP2 respectively; h, a, x, l, r1, and r2 are parameters
to be estimated; and gj and ej are respectively iid Gumbel variables for j ¼ 1,2.
Tables 2 and 3 show the estimation results for urban and interurban transport,
respectively. Parameters were calibrated with the software Biogeme 1.8. (Bierlaire
2009). For urban transport, as shown in Table 2, 17 of the 6624 observations
corresponded to inconsistent responses; the worst alternative in all attributes was
chosen. These were eliminated from the sample, which resulted in a total of
6607 valid observations. In this case, all the parameters were estimated with the
correct sign, and only the level 1 of the information variable presented a low level
of significance.15 This means that the effect of having information with maps and
timetables at bus stops versus only providing information about the route and=or
itinerary was negligible for urban transport users. However, the effect of passing
from a zero level of information to a level two, that is to say electronic panels, maps
and timetables, was much greater and this variable resulted very significant. Para-
meters corresponding to discrete variables were estimated consistently according
to the definition of the respective levels.
The inertia variable was very significant and with a positive sign in model MNL2,
which shows evidence that the current choice is influenced by the decisions taken
by the individual in the situations presented in previous choices; this indicates iner-
tia behavior towards the use of the current service. Nonetheless, model ML1 shows
also the existence of correlation among the responses of the same individual in
each one of the sets (as r1 and r2 resulted highly significant), and the inertia
ceases to be significant at the 95% confidence level for urban trips. This indicates
that in model MNL2 the effect of the inertia is confounded with the correlation, so
we consider that ML1 is the statistically superior model.
For urban journeys access time at the bus stop generally produces less disutility
than travel time. This may be explained because the access times are not high (nor-
mally less than ten minutes), the majority of people walk to the stop and the
weather is generally pleasant. However, the effect of time between two services (fre-
quency), which is related to the waiting time at the bus stop, is more negatively per-
ceived than travel time. The qualitative attributes that have a greater importance
on the utility are level 2 of comfort, the attitude of the driver and cleanliness.
Specific constants are specified in the utilities of the current option. Except for
the set 1 in MNL2 model, both have a negative sign. This result demonstrates that
in general there is a preference for the current service, when the effect of the rest
of the attributes is zero.
In the case of interurban transport (see Table 3), all the parameters were esti-
mated with the correct sign and were significant in all the models at the 95% con-
fidence level, with the exception of the presence of bus shelters and access time
(TA). With respect to the latter variable, it is worth mentioning that it only turned
out to be significant, at the 90% confidence level in MNL1; and its effect was
only significant for those passengers that walked to the bus stop (W ¼ 1, and 0
otherwise) and lived in Las Palmas (LP ¼ 1, and 0 otherwise). In all other cases,

15
This variable was eliminated from the specification of model ML1 due to an erroneous
sign.

International Journal of Sustainable Transportation Vol. 8, No. 1, 2014 37

 C. Roman et al.

the contrary effect was observed with a positive sign. This result, apparently coun-
ter intuitive, can be explained because within the sample there were a significant
percentage of individuals that travelled to the bus stop by car. For these people the
bus stop access time could result more pleasant than the in-vehicle travel time. In
reality, some of these trips should be considered as multimodal car–bus journeys. If
we consider that for these people the total duration of the journey is fixed, the
individuals would prefer the options of greater access time since more time in
the car would involve less time in the bus; that is to say that the access time is a
better option than the travel time. This behavior could give rise to the estimation
of access time with a positive sign.
Compared to the urban journeys, the negative effect of time between the two
services is somewhat lower than the travel time. This is due to passengers on the
interurban routes being more accustomed to consulting the timetable, and of
being capable of better adjusting their arrival times at the bus stop. However, time-
table delays have a negative effect that is six times greater that the travel time.
With respect to the three defined discrete levels for qualitative variables, the esti-
mations follow the same model as urban transport; that is to say the estimated para-
meter for level two was superior to that for level one, which is consistent with the
experiment. In terms of the importance of the previously mentioned variables, in
level two we would like to draw attention to the comfort, followed by driver’s atti-
tude, information and cleanliness. In contrast to urban transport, the inertia vari-
able was significant in both models, and according to model ML1 there was even
correlation in the results of the same individual. The specific constant had a posi-
tive sign in set one, the variables included are: time, frequency cost, punctuality,
access time and comfort. All this demonstrates that there are aspects that have
not been measured, but which make individuals show their preference for the
alternative rather than the current service. As with urban transport, the specifi-
cation of the ML1 model was statistically superior. Finally, the analysis the para-
meter l allowed us to confirm the existence of differences in the scale
parameters of the SP1 and SP2 data sets only in the case of interurban transport.
3.3. Obtaining Service Quality Indicators in Public Transport
After obtaining the estimations for the best model, a hybrid utility function was
constructed by scaling (i.e., multiplying by the factor l) those parameters of the set
one attributes that were not common to both datasets (Louviere et al. 2000). Indi-
cators of the quality of service were obtained for each passenger by evaluating the
aforementioned utility function in the current service.16 The quality of the global
urban and interurban transport service was given by the average of all the individ-
ual utilities.
In Figure 3 it is possible to appreciate the relative importance that each of the
attributes has upon the construction of the global quality indicator. In both cases it
can be seen that the attributes whose effect is negative, that is travel time, walking

16
The study also provided representative indicators of the quality of service for each
researched route by evaluating the utility function of the average value for each attribute
for the passengers on the route, in accordance with the current service. These results are
not available in this article but interested readers can request them to us.

38 International Journal of Sustainable Transportation Vol. 8, No. 1, 2014

 Analysis of the Service Quality of Public Transport

Figure 3. Relative importance of all the dimensions considered in the service

quality. (Figure appears in color online.)

time, frequency and delay, are more important than the others. This is especially
important for urban transport, where these attributes have a weighting of 80%; the
time between two services (or service frequency), which in this type of trips is asso-
ciated to the waiting time at the bus stop is especially important. This attribute has
a much lower weighting for interurban journeys, which means that the disutility
produced by this variable is significantly lower. It is important to note that
although the frequency of service may be worse in some interurban routes, the
estimated parameter for this variable is almost three times less. With respect to

International Journal of Sustainable Transportation Vol. 8, No. 1, 2014 39

 C. Roman et al.

the other variables, attention should be drawn to the importance of comfort and
the driver’s attitude in interurban journeys in level 2.
Figure 4 shows the sample distribution of the SQI for the urban and interurban
journeys. In both cases the effect on the perception of the current quality of trans-
port (black line) has been analyzed. The application of consistent policies in

Figure 4. Density distribution of the service quality index. (Figure appears in

color online.)

40 International Journal of Sustainable Transportation Vol. 8, No. 1, 2014

 Analysis of the Service Quality of Public Transport

obtaining gradual discounts in terms of the distance covered has been analyzed
but only for interurban journeys (red line), and finally the 30% travel cost
reduction (blue line) has been analyzed for both cases. The application of this last
policy, for the interurban case, significantly increases the proportion of passengers
who have a perception of superior quality to a given value, and is represented by
the area beneath the line and to the right of the aforementioned value. Moreover,
it can be seen that this effect is much more frequent with individuals that have a
worse perception of the quality of service.
3.4. Willingness to Pay Measures
The willingness to pay (WTP) measures express, in monetary units, the changes
produced in the utility due to the changes in the attributes. For continuous attri-
butes, these are represented as the quotient of marginal utility for the correspond-
@V =@q
ing attribute and the marginal utility of the travel cost; that is to say @V =@Ck. When
dealing with discrete variables, the WTP to pay in order to improve the variable
from one given level to another, for example from zero to one, is equal to:
1 0
WTPqkj ¼ V
V k ; where k is the marginal utility of income, obtained as minus the
marginal utility of travel cost.
In the two applications the cost of the journey divided by the rate of expenditure
is specified. In this way the effect of income upon the decisions of the passengers is
taken into account, something that is not possible when the specification of the tra-
vel cost adopts the linear form. This specification gives rise to different value of the
h
willingness to pay among the passengers, since the marginal utility of income
C=g g
is different for each person. A common practice in these cases is to apply the sam-
ple enumeration method, in order to obtain a representative measure through the
average of the willingness to pay for all the individuals in the sample; see Ortúzar
and Willumsen (2001). However, when the distribution of the WTP is asymmetrical
(that is, when the mean and the median of the distribution do not coincide), then
using the mean as the representative measure for the WTP may result in an unsat-
isfactory policy for the majority of the individuals. For example, Figure 5 shows an
asymmetrical distribution of the WTP for travel time savings that is skewed to the
right; i.e., the median (Me) is less than the mean (
x ) of the distribution. If for each
unit of time saved passengers were charged the price p ¼ x
, then P(WTP  p)  0.5.
This would mean that over 50% of the passengers would be dissatisfied with this
policy, since P(WTP < p) > 0.5; that is to say that their willingness to pay would
be lower than p. As such, charging a higher than median price would not fulfil
the objective of satisfying at least 50% of the individuals. A similar analysis could
be carried out, if the distribution is skewed to the left.
In the case of discrete variables, for example those categorized into three levels,
we spread the distribution of the willingness to pay from level 0 to level 1 and from
level 0 to level 2. In the example shown in Figure 6 both distributions are skewed to
the right, moreover Me1 < Me2. If by improving the level of service in this attribute
the price p > Me2 was charged, then over 50% of individuals would be dissatisfied
even when offered the highest level of quality (i.e., level 2). Note that P(WTP0
2
> p) < 0.5. If Me1 < p  Me2, over 50% of the passengers would be satisfied, but only
if the company offered the highest level of quality. However, the majority of the
passengers would be dissatisfied if they received level 1 quality, in this

International Journal of Sustainable Transportation Vol. 8, No. 1, 2014 41

 C. Roman et al.

Figure 5. Asymmetric distribution of WTP for a continuous variable.

Figure 6. Asymmetric distribution of WTP for a discrete variable. (Figure

appears in color online.)

42 International Journal of Sustainable Transportation Vol. 8, No. 1, 2014

 Analysis of the Service Quality of Public Transport

case P(WTP0
2 > p) > 0.5 but P(WTP0
1 > p) < 0.5. If p  Me1, the majority of the
passengers would be happy, even if they received level 1 quality. This analysis reiter-
ates that the shape of the distribution of the willingness to pay should be borne in
mind when these measures are used for policy purposes.
Table 4 shows the values for the willingness to pay for improvements in quality of
service in urban and interurban transport for the specification of the best model
(ML1). It is pertinent to point out that in all the cases analyzed the distributions
were skewed to the right and, that in accordance with the previous discussion,
to indicate the mean and median values of the distributions.
The results of the analysis demonstrate that public transport users in Gran
Canaria suggest the existence of differentiated behavior, depending on whether
they use urban or interurban transport. However, to confirm this hypothesis a
detailed comparison of the whole distribution of the WTP measures should be
required. By taking the mean reference value into account, for urban transport
users the greater willingness to pay is for saving waiting time between the two ser-
vices (2.14 euros=hr), followed by travel time (2.12 euros=hr) and access time (1.12
euros=hr). However, for interurban transport, the greater willingness to pay is to
reduce delays (7.93 euros=hr), which is more than double what individuals are pre-
pared to pay for saving travel time (2.73 euros=hr). On the other hand, the willing-
ness to pay to save waiting time is not as significant as for urban routes.
In terms of qualitative variables, for urban routes the greater willingness to pay
to pass to the optimal level of quality, that is to pass from level 0 to level 2, is given
by improving comfort (0.36 euros per trip to be seated during all the journey), the
driver’s attitude (0.33 euros per trip for pleasant treatment and smooth driving)
and cleanliness (0.29 euros per trip so that the bus is clean). Passengers are pre-
pared to pay 0.16 euros per trip for electronic panel to be placed at the stops,
but only 0.08 euros for bus shelters. Willingness to pay to improve the qualitative

Table 4. Willingness-to-pay measures.

Interurban transport Urban transport

Attributes Mean Median Mean Median

Travel time (4=hr) 2.73 2.36 2.12 1.93

Access time (4=hr) 2.11 1.71 1.12 1.02
Time between two services (4=hr) 0.90 0.78 2.14 1.96
Delay time (4=hr) 7.93 6.86 – –
Comfort From level 0 to level 1 (4) 0.49 0.43 0.19 0.17
Comfort From level 0 to level 2 (4) 3.47 3.00 0.36 0.33
Information From level 0 to level 1 (4) 0.38 0.33 – –
Information From level 0 to level 2 (4) 0.71 0.62 0.16 0.15
Driver From level 0 to level 1 (4) 0.33 0.28 0.11 0.10
Driver From level 0 to level 2 (4) 1.00 0.87 0.33 0.31
Cleanliness From level 0 to level 1 (4) 0.21 0.18 0.14 0.12
Cleanliness From level 0 to level 2 (4) 0.57 0.49 0.29 0.26
Shelter (4) 0.05 0.04 0.08 0.07

International Journal of Sustainable Transportation Vol. 8, No. 1, 2014 43

 C. Roman et al.

aspects were in general greater for interurban routes than for urban ones, in parti-
cular the willingness to pay for improving comfort (3.47 euros to be seated during
the whole route) and for improving the driver’s attitude (0.80 euros per trip for
receiving pleasant treatment and smooth driving).

4. CONCLUSIONS
Providing a public transport system that is capable of satisfying the growing
mobility needs of citizens currently constitutes an irrefutable need for the economic
development of any society. Despite this, there is great controversy over what the
future role of public transport should be, and over how public administration should
be involved in satisfying the real needs of the demand versus other possible priorities.
By observing the tendency in most cities worldwide, we can certainly see that pub-
lic transport has not done well in assuring its long term sustainability and that too
much trust has been placed in the help provided by both local and national govern-
ments. In recent decades institutional reforms focused on reducing productive
inefficiency due to inadequate state intervention have also taken place. This new
regulation shows the possibility of improving intervention, in order to achieve a
public transport policy that benefits society as a whole. Thus, some net welfare gains
can be achieved that the traditional schemes would otherwise not produce, and
these gains may be reflected in the prices. This is because the productive inefficien-
cies, which are the consequence of inadequate public involvement, are reduced.
Alternatively, other rules that are compatible with incentives may be followed, in
order to get transport companies to follow the signals emitted by the users.
The majority of public transport companies have either their own productivity
indicators or quality parameters for some of their services. These indicators are
usually based on the calculation of an index, such as the number of vehicle kilo-
meters per hr worked by the drivers or the number of passengers per kilometer
for each worker. The service quality parameters are usually based on the measure-
ments of some attributes that are considered important, such as travel time, the
average frequency of the services, the average delay, and the average age of the
buses that serve a route. Despite the internal utility of these self-evaluation pro-
cesses, in the majority of the cases they are not valid for formulating either trans-
port policies or for specific actions to improve user welfare. This is due to the fact
that the studies on quality must be based not only on setting certain standards
within the industry with spatial or temporary criteria for a series of processes,
but also on the degree of relative importance that the combined attributes have
upon the level of passenger satisfaction. For this reason, the public authorities,
be they regulators or providers, must go about establishing levels of quality that
maximize user utility.
The starting point of the method applied in this article is the need to analyze
users’ perceptions in order to evaluate the transit service quality as well as to esti-
mate how much they are willing to pay for improvements in the level of service.
The aim is to obtain a monetary measure for the benefits and social costs associa-
ted with the implementation, or otherwise, of specific courses of action. Addition-
ally, given the ever present existence of budgetary restrictions, the monetary

44 International Journal of Sustainable Transportation Vol. 8, No. 1, 2014

 Analysis of the Service Quality of Public Transport

valuation of user preferences offers an objective guide to how to prioritize differ-

ent quality enhancing policies in the public transport sector.
The use of stated preference designs has enabled us to estimate, with a reason-
able margin of error, the level of importance and the monetary value that transport
users assign to certain service variables. The method used allowed us to combine
different data sets and to analyze the joint effect of a potentially large number
of attributes. Other issues such as the analysis of the inertia effect and the corre-
lation induced by repeated observations are also addressed in a consistent way.
However there is still scope for a more detailed study of these technical aspects
following the trends published in the more recent literature.
The WTP figures obtained in the article constitute an essential input in the wel-
fare analysis of transport policies. This way, it is possible to socially evaluate the dis-
tinct performances of public transport services in cities, and determine which
policies have better results. Therefore, the main findings emerged from this
research, highlight the appropriateness of this methodology to evaluate public
transport performance.
When the social costs of carrying out a transport policy measure are lower than
the social benefits, we have an objective criterion that guarantees transport users
will improve in terms of welfare. Nonetheless, this does not mean that all the poli-
cies whose net balance is positive should automatically be carried out, since
budgetary restrictions or the need to account for the benefits and costs for
non-transport users may establish different priorities.

REFERENCES
Adamowicz W. 1994. Habit formation and variety seeking in a discrete choice model of rec-
reation demand. Journal of Agricultural and Resource Economics 19:19–31.
Ben-Akiva ME, Morikawa T. 1990. Estimation of travel demand models from multiple data
sources. Proceedings of the 11th International Symposium on Transportation and Traffic
Theory, July 1990, Yokohama.
Bierlaire M. 2009. Estimation of discrete choice models with BIOGEME 1.8. Lausanne:
École Polytechnique Fédérale de Lausanne EPFL.
Bliemer MCJ, Rose JM. 2006. Designing stated choice experiments: State-of-the-art. Proceed-
ings of the 11th International Conference on Travel Behavior Research, Kyoto.
Bradley MA, Daly AJ. 1997. Estimation of logit choice models using mixed stated preference
and revealed preference information. In: Stopher PR, Lee-Gosselin M (eds.), Under-
standing Travel Behavior in an Era of Change. Oxford: Pergamon, pp. 209–232.
Caussade S, Ortúzar JD, Rizzi LI, Hensher DA. 2005. Assessing the influence of design dimen-
sions on stated choice experiment estimates. Transportation Research Part B 39(7):621–640.
Cirillo C, Eboli L, Mazzulla G. 2011. On the asymmetric user perception of transit service
quality. International Journal of Sustainable Transportation 5(4):216–232.
Collins AT, Rose JM, Hess S. 2012. Interactive stated choice surveys: A study of air travel
behavior. Transportation 39:55–79.
Cook WD, Kress M. 1988. Deriving Weights from pairwise comparisons ratio matrices: An
axiomatic approach. European Journal of Operational Research 37:355–362.
Chorus CG, Dellaert B. 2010. Travel choice inertia: The joint role of risk aversion and
learning. Research paper from Erasmus Research Institute of Management (ERIM),
ERS-2010–040-MKT. http://econpapers.repec.org/paper/dgreureri/1765021097.htm

International Journal of Sustainable Transportation Vol. 8, No. 1, 2014 45

 C. Roman et al.

Domencich T, McFadden D. 1975. Urban Travel Demand: A Behavioral Analysis. Amsterdam:

North Holland.
Eboli L, Mazzulla G. 2008. A stated preference experiment for measuring service quality in
public transport. Transportation Planning and Technology 31(5):509–523.
Eboli L, Mazzulla G. 2010. How to capture the passengers’ point of view on a transit service
through rating and choice options. Transport Reviews 30(4):435–450.
Espino R, Román C, Ortúzar JD. 2006. Analyzing demand for suburban trips: A mixed RP=
SP model with latent variables and interaction effects. Transportation 33:241–261.
Garrido RA, Ortúzar JD. 1994. Deriving public transport level of service weights from mul-
tiple comparison of latent and observable variables. Journal of the Operational Research
Society 45(10):1099–1107.
Gatta V, Marcucci E. 2007. Quality and public transport service contracts. European Trans-
port 36:92–106.
Hensher DA. 2000. A service quality index for area-wide contract performance assessment
for commercial bus contracts. Sidney: The University of Sydney Institute of Transport
Studies.
Hensher DA. 2002a. Urban public transport delivery in Australia: Issues and challenges in
retaining and growing patronage. Sidney: The University of Sydney Institute of Transport
Studies.
Hensher DA. 2002b. Contract areas and integrated fares for public transit provision. Sidney:
The University of Sydney Institute of Transport Studies.
Hensher DA, Houghton E. 2002. Growth after transition under performance-based
contracts. Sidney: The University of Sydney Institute of Transport Studies.
Hensher DA, Houghton E. 2004. Performance-based quality contracts for the bus sector:
Delivering social and commercial value for money. Transportation Research Part B
38:123–146.
Hensher DA, Prioni P. 2002. A service quality index for area-wide contract performance
assessment regime. Journal of Transport Economics and Policy 36(1):93–114.
Hensher DA, Rose JM, Greene WH. 2008. Combining RP and SP data: Biases in using the
nested logit trick contrasts with flexible mixed logit incorporating panel and scale effects.
Journal of Transport Geography 16(2):126–133.
Hensher DA, Stopher P, Bullock P. 2003. Service quality: Developing a service quality index
in the provision of commercial bus contracts. Transportation Research Part A
37:499–517.
Louviere JJ, Hensher D, Swait J. 2000. Stated Choice Methods: Analysis and Applications.
Cambridge, UK: Cambridge University Press.
McFadden D. 1981. Econometric models of probabilistic choice. In: Manski C, McFadden D
(eds.), Structural Analysis of Discrete Choice Data with Econometric Applications.
Cambridge, MA: MIT Press, pp. 198–272.
Ortúzar JD, Ivelic AM, Candia A. 1997. User perception of public transport level-of-service.
In: Stopher PR, Lee-Gosselin M (eds.), Understanding Travel Behavior in an Era of
Change. Oxford: Pergamon, pp. 123–142.
Ortúzar JD, Willumsen LG. 2001. Modeling Transport. 3rd ed. Chichester: John Wiley.
Train K. 2003. Discrete Choice Methods with Simulation. Cambridge, UK: Cambridge
University Press.

46 International Journal of Sustainable Transportation Vol. 8, No. 1, 2014

Copyright of International Journal of Sustainable Transportation is the property of Taylor &
Francis Ltd and its content may not be copied or emailed to multiple sites or posted to a
listserv without the copyright holder's express written permission. However, users may print,
download, or email articles for individual use.