Transport Policy 49 (2016) 68–77

Contents lists available at ScienceDirect

Transport Policy
journal homepage: www.elsevier.com/locate/tranpol

The importance of service quality attributes in public transportation:

Narrowing the gap between scientific research and practitioners' needs
Begoña Guirao a,n, Antonio García-Pastor b, María Eugenia López-Lambas a
a
Departamento de Ingeniería Civil: Transportes y Territorio, ETSI Caminos, Canales y Puertos, Universidad Politécnica de Madrid, Avenida Profesor Ara-
nguren, s/n., 28040 Madrid, Spain
b
AVANZA Grupo, C/San Norberto, 48, 28021 Madrid, Spain

art ic l e i nf o a b s t r a c t

Article history: Customer Satisfaction Surveys (CSS) have become an important tool for public transport planners, as
Received 5 January 2016 improvements in the perceived quality of certain service attributes can lead to greater use of public
Received in revised form transport and lower traffic pollution. The literature shows that the importance of quality attributes has
11 March 2016
until now been estimated indirectly, as they are derived from the Customer Satisfaction Index using
Accepted 4 April 2016
various different and complex techniques. Little work has been dedicated to its direct estimation (stated
importance) by designing ad-hoc surveys, an approach that represents a considerable reduction in the
Keywords: length of the questionnaire.
Public transport This paper contributes to the limited existing literature by developing a survey technique based on
Customer satisfaction Surveys (CSS)
hierarchy processes to estimate the stated importance of quality attributes, and compares the results
Service Quality (SQ)
with the derived importance obtained using conventional surveys with the same sample. The added
User Perception
Factorial Analysis value of this research is that it provides the first comparison between two quality survey methods using
MIMIC models the same real case study in Madrid (Spain). The results achieved using this pioneer survey method (293
valid questionnaires) were validated using conventional face-to-face surveys (520 valid questionnaires).
Factorial analysis, multiple regression analysis and Multiple Indicators Multiple Causes (MIMIC) models
were applied to the conventional survey sample to analyse and derive the importance of the attributes.
The results clearly show that, after a few teething troubles, the stated importance of quality attributes
can be estimated directly, thus providing transport management companies with a simple and useful tool
to implement in their Customer Satisfaction Surveys (CSS), and narrowing the gap between practitioners’
needs and scientific research.
& 2016 Elsevier Ltd. All rights reserved.

1. Introduction perception surveys, although a number of authors (Grönroos,

1988) differentiate between consumer expectations and percep-
The analysis of Service Quality (SQ) is of vital importance for tion of service during the trip, and maintain that the perception of
both operators and public transport authorities, as the increase in SQ is the result of the comparison of consumer expectations with
SQ in public transport has been shown to play a key role in at- actual service performance. Other authors such as Hu (2010) de-
tracting new passengers from private cars to the public transport fine service quality in terms of the difference between perceived
system and in reducing traffic pollution as a result (Transportation and tolerable quality.
Research Board, 1999). The literature reveals a significant gap be- Leaving aside this theoretical approach, the study of quality
tween the scientific research and practitioners’ needs. Scientific took a giant stride forwards in linking the fields of research and
research regards the concept of SQ as complex, fuzzy and abstract, practice with the QUATTRO project entitled “Quality Approach in
mainly due to the three aspects of service: intangibility, hetero- Tendering Urban Public Transport Operations” (European Com-
geneity for each individual, and the inseparability of production mission, 1998), whose objective was to define and introduce
and consumption (Parasuraman et al., 1985). Most scientific quality indicators into tendering and contracting in public trans-
methodologies for analysing SQ are applied only to customer port services. The QUATTRO project was also the basis for the
European Standard EN 13816 Quality of service in passenger
n
transport services (2003), but provided a more practical concept of
Corresponding author.
E-mail addresses: bguirao@caminos.upm.es (B. Guirao),
SQ. Four quality levels were identified in the QUATTRO project:
agarciapa@avanzagrupo.com (A. García-Pastor), expected quality, perceived quality, targeted quality and delivered
mariaeugenia.lopez@upm.es (M.E. López-Lambas). quality. The level of quality desired (expected) by passengers and

http://dx.doi.org/10.1016/j.tranpol.2016.04.003
0967-070X/& 2016 Elsevier Ltd. All rights reserved.
 B. Guirao et al. / Transport Policy 49 (2016) 68–77 69

citizens in general may be different from the perceived quality – adopted).

observed with varying degrees of objectivity – by the passengers The survey is usually designed by the operating companies, and
during their journeys. The level of quality the company wishes to the resulting database is used first by the companies and then
achieve (targeted quality) is determined by external and internal passed on to researchers. This highlights the need to narrow the
pressures, expected quality, budgetary constraints and competi- gap between theory and practice. There is currently no proper
tors’ performance. Finally, the delivered quality is the level of debate on the design and format of the survey, and in most si-
quality obtained on a daily basis in real operating conditions. tuations researchers use only part of the survey results, as their
Likewise, the emphasis in the US was also on measuring service modelling tools are only suited to a specific database. There is also
quality through customer satisfaction, as evidenced by the Hand- a problem from the scientific point of view due to the lack of
book for Measuring Customer Satisfaction and Service Quality critical comparison – using the same case study – between the
(TRB, 1999) and the Transit Capacity and Quality of Service Manual competing techniques in order to analyse user perception and
(TRB, 2004). attribute importance.
The only objective information for operating companies is the This paper contributes to the limited existing literature by de-
provided quality, normally established in the concession contracts; veloping a survey technique based on hierarchy processes to es-
however Customer Satisfaction Surveys (CSS) are also conducted timate the stated importance of quality attributes, and compares
on a yearly or six-monthly basis to monitor the users’ perception the results with the derived importance obtained with the same
of the service. The data collected from the CSS are used to analyse sample using conventional surveys. The added value of this re-
the company's operations, and provide useful information on search resides in the fact that it is the first comparison between
overall service quality. They are also used in research and by two quality survey methods using data from the same case study,
academics to focus on the mathematical analysis of perceived in Madrid (Spain). The article has been divided into the following
quality, and test a large number of indicators: from simple indices parts in order to describe the research as a whole: introduction
such as SERVQUAL (Chau and Kao, 2009; Chou et al., 2011), (Section 1); a review of the literature on the methods used to
SERVPERF (Sánchez et al., 2007), Customer Satisfaction Index CSI estimate attribute importance (Section 2); a description of the case
(Hill et al., 2003) and Heterogeneous Customer Satisfaction Index study and conventional survey campaign (Section 3); the proposed
HCSI (Eboli and Mazzulla, 2009) to other more complex indices new stated survey method and its application to the case study
obtained by applying econometric models to satisfaction rates, (Section 4); the validation of the stated importance survey using
such as Structural Equation Models (SEM) and discrete choice the conventional survey (Section 5); and finally, the presentation
models (Hensher at al., 2003; Román et al., 2014). Some authors of the most important conclusions and recommendations (Con-
(Del Castillo and Benitez, 2013; Celik et al., 2014) have used mixed clusions section).
methodologies to determine the quality of the bus service. Del
Castillo and Benitez (2013) used three models simultaneously
(weighted means, a multivariate discrete distribution and a gen- 2. Estimation of attribute importance using CSS
eralised linear model), while Celik et al. (2014) integrated statis-
tical analysis, SERVQUAL, interval type-2 fuzzy sets and VIKOR The literature shows that a considerable number of attributes
(Opricovic and Tzeng, 2004) to evaluate customer satisfaction with are used to evaluate SQ, so they are normally grouped into a
the rail transit network in Istanbul. smaller number, called dimensions. Although there is no general
The more complex the indicators (and the techniques to obtain agreement as to the nature or content of SQ dimensions, it is
them), the less likely it is that practitioners will be able to un- generally recognised that service quality is a multidimensional
derstand and use them in practice. Due to its simplicity, CSI, based (Lehtinen and Lehtinen, 1982), multilevel or hierarchical (Brady
on the importance of “attributes” and satisfaction rates, is the most and Cronin, 2001) construct. Various papers (Eboli and Mazzulla,
widely applied index – even by operating companies – to de- 2007) have pointed to several categories of attributes that have a
termine service quality in public transportation. From a marketing greater or lesser impact on SQ and satisfaction. In 2002 the Eur-
point of view, an attribute is a characteristic or feature of a product opean Committee for Standardization CEN (2003) established a
that is thought to appeal to customers. In the public transport quality standard – EN 13816 Service Quality Standard for Public
sector, the term service attribute is commonly used to refer to Transport – in connection with the QUATTRO research, and pub-
cleanliness, on-time performance, availability, comfort or security, lished a final report. The UNE-EN 13186 standard classifies the
constituting the criteria applied to assess customer service quality. characteristics of a service into basic, proportional and attractive,
As the main tools for analysing service quality in public trans- depending on how compliance and non-compliance affects cus-
port are based on CSS, the design of the questionnaire is absolutely tomer satisfaction. In the US, the Transit Capacity and Quality of
crucial and depends strongly on the service attributes to be con- Service Manual TCQSM (Transportation Research Board, 2004)
sidered and on the approach used to estimate the relative im- groups attributes into availability factors, and comfort and con-
portance of the attributes to the customers. This relative im- venience factors. The primary distinction made by the TCQS is
portance is another key point, as once a group of attributes is whether a transit service is offered, and if it is, customers then
selected for a specific survey, public transport operators and ser- consider both the type of availability (e.g. frequency or access), and
vice industries need to know not only how the users rate the its comfort and convenience. In practice, the choice of variables is
service in terms of detailed service attributes (attribute-perfor- far from straightforward and usually derives from exhaustive lists
mance rating), but also the relative importance of these attributes of attributes (for instance the one included in UNE EN 13816),
to their customers (attribute-importance measures). The CSS re- although some are chosen through other CSS. Some authors like
sults can help managers choose from a long list of service attri- Dell’Olio et al. (2010) recommend identifying the attributes to be
butes (e.g. cleanliness, on-time performance, availability, comfort included in the CSS independently. Focus-group methodologies
or security) so they can target their organisation's attention and are suitable for this objective, although they are costly and require
resources more effectively. The rates are normally expressed in a separate prior study. The heterogeneity of the users and services
two scales, numeric and linguistic. Numeric scales are more must also be taken into account when analysing CSS results, as
commonly used and have a wider range, from 3 to 11 points. demonstrated by Bordagaray et al. (2014) when modelling bus
Linguistic scales are used less, and have a narrower range, from transit quality in the city of Santander.
3 to 7 points (the 5-point Likert scales are the most widely Apart from the selection of attributes, the design of a CSS
70 B. Guirao et al. / Transport Policy 49 (2016) 68–77

survey depends strongly on the approach used to estimate the principles guide problem-solving using the AHP: decomposition,
relative importance of the attributes to the customers. In con- comparative judgments and synthesis of priorities. In our service
ventional CSS designed by companies to obtain a general sa- quality case study, the decomposition is based on the selection of
tisfaction index (CSI), it is necessary to consider both the attribute- the attributes to be ranked and on the comparative judgments
performance rating and attribute-importance measures when the given by the surveys. The AHP priorities are synthesised from the
operator's priority is to improve or sustain the current overall SQ. second level down by multiplying local priorities by the priority of
This dual target often requires a long questionnaire, although re- their corresponding criterion in the level above, and adding a level
searchers only use the results of the first part (attribute-perfor- for each element according to the criteria it affects. This gives the
mance rating), as the attribute importance can be indirectly de- composite or global priority of that element, which in turn is used
rived from the attribute-performance rating. This problem has to weight the local priorities of the elements in the level below.
already been debated in the literature, as described below. Aydin et al. (2015) recently used a type of AHP methodology,
Weinstein (2000) was the first author to clearly distinguish two FAHP (Fuzzy Analytic Hierarchy Process), to measure the perfor-
main approaches to estimate attribute importance: stated im- mance of rail transit lines in Istanbul; however the FAHP was
portance and derived importance. Stated importance involves applied to fix the weights of the main “criteria” (train comfort,
asking customers to rate each attribute on a scale of importance; ticketing, information system, accessibility, station comfort, fare
this is the more intuitive and direct of the two methods, but re- and time) based on the unbiased opinions of experts. The weights
quires a significant increase in the length of the questionnaire of the sub-criteria were simultaneously calculated by trapezoidal
(which can lower the overall response rate and the accuracy of the fuzzy numbers based on customer responses. The FAHP was
survey). It can also sometimes fail to differentiate sufficiently be- therefore not directly applied to the CSS itself, as in our case.
tween mean importance ratings; if customers score nearly all the In designing the survey questionnaire using an AHP process,
measures near the top of the scale, certain attributes may be rated some practical ideas have been borrowed from the stated pre-
as important even though they in fact have little influence on ferences experiments in transportation described by Saako (2001)
overall satisfaction. As this is the more intuitive and direct of the in order to collect useful data with as little bias as possible. Stated
two methods, operating companies have tended to use this type of preference surveys have been used in transportation to analyse
questionnaire, while the scientific research has focused on more alternative trip choices (each alternative is composed of various
complex methodologies using the derived importance approach. attributes), but we have found no literature that ranks simple
The derived importance approach is less intuitive and is based quality attributes, although statistically the problem to be solved is
on “deriving” a measure of attribute importance by statistically fairly similar. Ampt and Meyburg (1995) suggest a maximum of 9–
testing the strength of the relationship of individual attributes 16 options as acceptable in this type of stated preference surveys,
with overall satisfaction. A simple conventional attribute rating with most current designs now adopting the lower end of this
survey is needed to derive importance, and this type of ques- range. With a maximum of nine options for the respondent to
tionnaire is always included in the CSS. Recent literature is now set ponder, this severely limits the number of attributes that can be
on seeking other alternatives to the methods commonly used until considered. Our Customer Satisfaction surveys consider over 10–15
now to derive importance, namely; (a) bivariate Pearson correla- attributes, so this limitation must be overcome while allowing the
tions, (b) factor analysis, and (c) multiple regression analysis. consideration of more attributes and/or more attribute levels. One
These other alternatives include structural equation models (SEM), of the strategies proposed by Pearmain et al. (1991) in stated
based on a multivariate technique combining regression, factor preference surveys is to separate the options into “blocks”, so that
analysis and analysis of variance to estimate interrelated depen- the full choice set is completed by groups of respondents, but with
dence relationships simultaneously. This approach allows a phe- each group responding to a different sub-set of options. Each
nomenon to be modelled by considering both the unobserved group responds to a full-factorial design within each sub-set of
“latent” constructs and the observed indicators that describe the options, and the responses from the different sub-groups can be
phenomenon. SEM has also been adopted to measure customer assumed to be sufficiently homogeneous to provide the full picture
satisfaction in several public transport services such as me- when combined.
tropolitan public transport (Lai and Chen, 2011; Shen et al., 2016). As part of a research project led by the Madrid Polytechnic
More recently, de Oña et al. (2012) have used decision trees to University, the authors of this paper had the opportunity to design
derive attribute importance in public transport quality, and a new an ad-hoc CCS, based on this previous literature, in a Spanish case
methodology of “index numbers” has been developed to monitor study: the Madrid-Tres Cantos corridor, with four urban bus lines
the evolution of attribute importance throughout successive CSS (operated by the company ALSA). A new type of survey ques-
(de Oña et al., 2016). However, these last complex methodologies tionnaire (to state importance) was tested using a more sophisti-
are not based on stated attribute importance from the CSI, but on cated process of hierarchy, separating the options into blocks and
derived importance. As far as the authors are aware, there are no reducing the length of the survey questionnaire (not all users were
studies comparing the different methodologies for obtaining at- asked for the same attribute ranking). In order to validate this new
tribute importance using the same case study data (or even a stated importance method, a conventional survey was also re-
comparison between the most commonly used derived im- quired (designed to derive importance), and the whole campaign
portance methodologies). was based on face-to-face surveys (293 surveys to state attribute
The possibility of comparing techniques and estimating stated importance and 520 to derive importance). As the face-to-face
importance has been practically abandoned by academics, but survey campaign was starting to become very costly, additional
other survey formats could have been tested and studied, such as research based on Quick Response (QR) code surveys was also
ranking attributes using hierarchy process together with stated implemented in the study. A third type of questionnaire was
preference techniques. Analytic Hierarchy Process (AHP) is a gen- therefore designed for the QR survey (also derived-importance)
eral theory of measurement used to derive ratio scales from both and uploaded to the operating company's (ALSA) website. The QR
discrete and continuous paired comparisons (Saaty, 1987), which code is a simple way of providing the user with a virtual link to the
may be taken from actual measurements or from a basic scale that questionnaire in order to test how to reduce the cost of future SQ
reflects the relative strength of preferences and feelings. Pairwise survey campaigns using new Intelligent Transport Systems (ITS).
comparisons are fundamental in the use of AHP, although this The results of the QR research have been published recently
theory can be extrapolated to a three-option choice. Three (Guirao et al., 2015), and this article shows the results of the main
 B. Guirao et al. / Transport Policy 49 (2016) 68–77 71

part of the research project, based on stated importance.

3. The case study: CSS in a bus corridor in Madrid

The initiative to conduct surveys in different formats among

urban bus users came from a research project led by the Madrid
Polytechnic University (UPM). The campaign was carried out in
March 2013 in four periurban bus lines along the Madrid-Tres
Cantos corridor operated by the company ALSA. This corridor leads
towards the north of Madrid, starting from the interchange at
Plaza de Castilla and connecting the UAM (Universidad Autónoma
de Madrid), El Goloso and Tres Cantos along a length of 20 km, as
shown in Fig. 1. Bus lines 712, 713 and 716 connect the Madrid
public transport interchange hub in Plaza de Castilla to the city of
Tres Cantos along the M-607 corridor (M-607 is a dual carriageway
with two lanes in each direction). The last part of the route, now in
Tres Cantos, divides into different routes inside the city. Line 714 is
a special case, since it connects the interchange hub to a university
campus (Universidad Autónoma de Madrid – UAM) a few kilo-
metres outside the city, meaning this bus service is a specialised
line for trips for the purpose of study.
Suburban bus services in Spain are usually tendered in route
bundles according to factors like line proximity, feasibility of lines
or in order to avoid overlapping. Contracts are normally subject to
European Regulation 1370:2007 on public passenger transport
services by rail and by road, which envisages the concept of Public
Service Obligations (PSO) and other national requirements (LOTT,
1987). In the case study, the bundle is a combination of purely
metropolitan lines connecting the city of Madrid at one of the
mayor interchanges (Plaza de Castilla) with the municipality of
Tres Cantos, a city located 18 km north of Madrid with a popula-
tion of 50,000. Table 1 shows the main characteristics of the bus
lines included in this case study.
Two previous groups of questionnaires were designed in order
to achieve the objectives of the research project: one to determine
the derived attribute importance (group 1), and the other to find
the stated importance (group 2). Over 850 face-to face surveys
were finally conducted following various parameters of statistical
significance and maximum error, from which 813 observations
were drawn as valid (520 from group 1 and 293 from group 2).
These results allowed the quality analysis to be completed with a
sufficient sample size for the planned objectives. The pilot survey
was carried out on February 20, 2013 and definitive surveys were
made throughout the last two weeks of March from 6 am to 11 am
(18.3% of the sample), 11:01 am to 4:40 pm (64.8%) and 4:41–
11 pm (16.9%), at the main bus stops (Plaza de Castilla interchange
hub, La Paz Hospital, Ramón y Cajal Hospital, Einstein-Rectorado
UAM) and on board. Table 2 shows the sample rate for each line for
survey group 1 (designated “conventional survey” in our research
on QR codes). These sample rates present errors of around 5–7%
for high confidence intervals. Bus line 714 has a distinct student
dimension, and while the sample rate is low, the results are still Fig. 1. Location of the Madrid-Tres Cantos corridor (M-607 dual carriageway) in
Spain.
considered sufficient for the analysis. All the bus lines have a si-
milar age and gender distribution except for line 714, which – as it
is used mainly by students – has a higher percentage of young
users; it also carries more women than men. Table 2 shows the Table 1
number of valid questionnaires per user and trip profile (ticket Main characteristics of the suburban lines (Madrid-Tres Cantos corridor).
type, gender, activity, frequency, age and trip purpose) in the
Line Length Travel time (min Headway (min) Yearly passengers
conventional survey, with their percentages. An additional ques- (km) per way) (2012)
tionnaire (group 3) was designed to test the implementation of QR
codes in web-based surveys. Only group 1 was used to validate the 712 22.3 45 15 1,050,901
QR surveys, as the format was comparable. 713 21.3 45 15 879,525
714 11.5/13.0 35 12 687,099
In the conventional survey carried out at the time and in the 716 22.8 35 20 651,455
place mentioned above, the users were asked to rate the following
15 attributes in addition to the overall level of satisfaction with the
72 B. Guirao et al. / Transport Policy 49 (2016) 68–77

Table 2
Conventional survey collection per bus line. Sample rates and questionnaires collected per user and trip profile.

Sample rate estimation Bus Line Total

712 713 714 716

Workday demand (trips) 4106 3072 3250 3160 13,588

No. of surveys collected 207 116 91 106 520
Sample rate 5% 3.8% 2.8% 3.4% 3.8%

Number of valid questionnaires per user and trip profile

User activity
Working 112 (54.1%) 68 (58.6%) 17 (18.7%) 62 (58.5%) 259 (49.8%)
Unemployed 11 (5.3%) 6 (5.2%) 1 (1.1%) 2 (1.9%) 20 (3.8%)
Retired 26 (12.6%) 9 (7.8%) 6 (6.6%) 6 (5.7%) 47 (9.0%)
Student 43 (20.8%) 26 (22.4%) 67 (73.6%) 29 (27.4%) 165 (31.7%)
Other 15 (7.3%) 7 (6.0%) 0 (0.0%) 7 (6.6%) 29 (5.6%)

Ticket
Single 10 (4.8%) 6 (5.2%) 0 (0.0%) 7 (6.6%) 23 (4.4%)
10 trips 16 (7.7%) 10 (8.6%) 2 (2.2%) 5 (4.7%) 33 (6.3%)
Season ticket 176 (85.0%) 99 (85.3%) 89 (97.8%) 94 (88.7%) 458 (88.1%)
Other 5 (2.4%) 1 (0.9%) 0 (0.0%) 0 (0.0%) 6 (1.2%)

Frequency of trip
Z5 days 142 (68.6%) 84 (72.4%) 65 (71.4%) 73 (68.9%) 364 (70.0%)
3–4 days 22 (10.6%) 14 (12.1%) 13 (14.3%) 11 (10.4%) 60 (11.5%)
1–2 days 31 (15.0%) 9 (7.8%) 10 (11.0%) 13 (12.3%) 63 (12.1%)
Less than 1 d 12 (5.8%) 9 (7.8%) 3 (3.3%) 9 (8.5%) 33 (6.3%)

Trip purpose
Work 117 (56.5%) 65 (56.0%) 15 (16.5%) 63 (59.4%) 260 (50.0%)
Study 38 (18.4%) 23 (19.8%) 71 (78.0%) 25 (23.6%) 157 (30.2%)
Medical 11 (5.3%) 8 (6.9%) 0 (0.0%) 4 (3.8%) 23 (4.4%)
Leisure 10 (4.8%) 3 (2.6%) 0 (0.0%) 3 (2.8%) 16 (3.1%)
Other 31 (15.0%) 17 (14.7%) 5 (5.5%) 11 (10.4%) 64 (12.3%)

Age
r to 23 48 (23.2%) 22 (19.0%) 60 (65.9%) 30 (28.3%) 160 (30.7%)
From 23 to 35 59 (28.5%) 33 (28.4%) 19 (20.9%) 24 (22.6%) 135 (25.9%)
From 36 to 50 38 (18.4%) 30 (25.9%) 7 (7.7%) 29 (27.4%) 104 (20.0%)
Z50 62 (30.0%) 31 (26.7%) 5 (5.5%) 23 (21.7%) 121 (23.2%)

Gender
Male 66 (31.9%) 37 (31.9%) 33 (36.3%) 41 (38.7%) 177 (34.0%)
Female 141 (68.1%) 79 (68.1%) 58 (63.7%) 65 (61.3%) 343 (66.0%)

TOTAL 207 (39.8%) 116 (22.3%) 91 (17.5%) 106 (20.4%) 520 (100%)

service: the last three on the list (bus driving security, customer attention
from the bus driver and the possibility of sitting during the jour-
● Route (bus route). ney) were introduced at the request of the operating company and
● Connections (connection with other lines and transport modes). located in a different part of the conventional survey.
● Punctuality (on-time performance). The statistical mode and median of the results of the analysis of
● Frequency (timetable and headway). these abovementioned bus lines show that most of the variables
● Access (ease of access to the bus stop from origin –home, work, have an average and median with the semantic meaning “Good”.
university, etc.). Only the variable “Frequency” has a semantic value “Not Good” for
● Information-incidents (delays, breakdowns, changes in the line, the median, which indicates the importance of this variable and
etc.). how it is valued by respondents. The statistical analysis by line
● Cleanliness (cleanliness of the bus). does not reveal any substantial difference, except in the case of the
● Information-service (timetables, routes, etc.). valuation of ICTs by the users of bus 714, who describe it as “very
● Journey time (of the route). good”. A preliminary aggregated analysis of the conventional sur-
● Comfort (air conditioning, seating, etc.). vey is shown in Table 3, with the average rating of each attribute-
● Information and communication technologies (ICTs) (internet on performance. The three best rated attributes (over 7.0 out of 10.0)
board, mobile payment, real-time information screens both on are bus cleanliness, access to bus stops and the possibility of sitting
board and at stops). during the journey, while the three worst rated are ICTs, in-
● Shelters (along the route). formation about incidents and frequency. It should be noted that
● Bus driving security. global customer satisfaction on this line is high (7.0 out of a
● Customer attention from the bus driver. maximum score of 10.0). These data allow the importance of each
● Possibility of sitting during the journey. attribute to be estimated mathematically, although the most in-
Faced with the impossibility of developing a focus group of tuitive and direct method for operating companies would be to ask
corridor users, twelve attributes were selected based on a socio- the customers directly which attributes they consider more im-
logical study (on SQ attributes for periurban lines) carried out by portant from a general point of view – not necessarily linked to
the Madrid's Regional Transport Consortium (CRTM, 2005). Finally their trip experience – when answering the survey. Although
 B. Guirao et al. / Transport Policy 49 (2016) 68–77 73

Table 3 (A2), 74 from Card 3 (A3), and 61 from Card 4 (A4). The number of
Average rating of each attribute-performance in Madrid-Tres Cantos corridor. cards collected per bus line ranged from 41.98% on Line 712–
18.77% on Line 714 (with 19.45% on Line 713% and 19.8% on Line
Rated variables Rating (over 10)
716). This means that each survey has an error of around 11% for
Cleanliness 7.72 high confidence intervals. However, as each card has a scale with
Access 4.57 several attributes, we obtained three pairs of discrete choices per
Possibility of sitting during the journey 7.48 user, and the error per survey thus drops to 6% for a confidence
Journey time 7.36
Customer attention from the bus driver 7.28
interval of 95.5%. Obviously, this simplification would have been
Comfort 7.04 unnecessary had we collected a higher number of stated im-
Connections 7.00 portance surveys, but does not invalidate the results. Moreover,
Punctuality 6.96 the user profile registered in the stated survey is consistent with
Bus driving security 6.86
the one obtained through the conventional survey and with the
Route 6.85
Information-service 6.81 information on demand provided by the operating company.
Shelters 6.75 Table 4 shows the structure of each card and a preliminary
Frequency 5.64 analysis of the attribute importance results depending on the
Information-incidents 4.59 number of times the attribute is in first, second and third position.
Information and communication technologies (ICTs) 3.28
Each time an attribute is in first place in a survey, it is assigned a
value of 3.0. This value is 2.0 in second place, and 1.0 in third place.
Table 4 shows the score given to each attribute for each type of
operating companies have tended to use – and most continue to card. The number of valid surveys obtained per card must be taken
use – this type of “stated format”, the required length of the into account to guarantee statistically robust results. Each card
questionnaire is excessive and can lower the overall response rate contains seven or eight attributes and it is also necessary to
and the accuracy of the survey. The following section contains a average (or weight) the number of times an attribute appears in
proposed design for a new type of stated important questionnaire the top three positions. For example, in card 1 the score for
together with its application to the case study. punctuality has been divided by 474 (the sum of all the scores in
this card); this percentage (out of ten) is shown in the last column
of Table 4. Once these values have been calculated, the scores are
4. A proposal for a stated importance survey aggregated for each attribute from two different cards, but con-
sidering the total range of scores; the highest score corresponds to
The stated importance survey was carried out in the Madrid- the “punctuality” attribute on card 1 (4.43) while the lowest
Tres Cantos corridor in March 2013, but on a different date from corresponds to “ICTs” on card 2. We therefore assigned the value
the conventional survey, in order to avoid biases or “contamina- 10.0 to the highest score and 0.0 to the lowest, interpolating the
tion” between them. The new questionnaire was designed to in- intermediate scores (see the last column in the table). Table 5
clude the same 15 attributes as in the conventional survey but shows the final aggregation per attribute, and the ranking of at-
these were offered to the customers in four different sub-sets of tributes in terms of their importance for users.
attributes (blocks) according to the literature review and in order Punctuality, frequency and driving security can be seen to be
to reduce the length of the survey. The customers were asked to the three most important attributes for customers, while ICTs, bus
identify the three most important attributes in each sub-set, and driver attention and incident information appear at the bottom of
to rank them in descending order of importance. This solution the table. According to Table 3, two of the least important attri-
allowed the number of attributes to be reduced to a smaller butes for users are also the worst rated in the conventional survey
ranking, thereby improving the reliability of the survey process. (ICTs and incident information). After defining this pioneer survey
The first questions in the survey concerned user and trip profile, tool, we validated and analysed our results using the conventional
and these were common to all the users surveyed. In contrast, the survey database for the same corridor.
attribute importance questions were organised in four scale cards
and the customers were assigned only one, with no more than
eight attributes. One of the main problems with earlier long stated 5. Validation of the stated importance survey
preference surveys was that they sometimes failed to differentiate
sufficiently between mean importance ratings if customers rated The stated importance survey was validated based on the
nearly all the measures near the top of the scale. Certain attributes conventional survey analysis, in which the same 15 attributes were
could therefore be rated as important even though they in fact rated using a 5-point Likert scale and subsequently normalised
have little influence on overall satisfaction. To avoid this type of with a 0–10 scale during data processing. The number of valid
bias, the attributes in each card and their order of appearance questionnaires collected in the conventional survey (520) shows a
were selected according to the following guidelines: uniform error of 4.4% for a confidence interval of 95.5%.
Before deriving the attribute importance mathematically from
● Each card includes a total of seven or eight attributes (almost the conventional survey, the valid surveys were analysed in depth
half the total attributes). using different statistical techniques. An independence test was
● Each attribute appears only twice; that is, on only two of the first carried out considering the different bus lines to check that
four available cards. the samples were independent and unbiased. As the variables
● Each time an attribute appears, attempts were made to change were categorical, this test was done by estimating the Pearson Chi-
its order of appearance, alternating the top and bottom positions squared (χ2). The Chi-square goodness-of-fit test revealed that
in the cards. To achieve this target, all the attributes meet the there is sample independence for most of the variables; that is, the
requirement that the difference between their two appearances survey answers do not depend on the chosen segment although
is at least two positions (at the top or bottom of the scale). there are variables that show different behaviours. For example,
Table 4 shows the four scale cards used in the case study bus route perception, frequency and information-service depend
(namely A1, A2, A3 and A4). 293 valid surveys were collected with on the bus line considered. Age affects the perception of the
this type of questionnaire: 79 from Card 1 (A1), 79 from Card 2 connection to other transport modes, frequency and information-
74 B. Guirao et al. / Transport Policy 49 (2016) 68–77

Table 4
Structure of the four ranking cards in the stated important survey, and number of times an attribute appear in the first, second and third position of the ranking.

Card 1 First position Second position Third position Score 1 Score 2 (over the card) Score 3 (over all cards)

Punctuality 61 12 3 210.00 4.43 10.00

Information-service 4 26 9 73.00 1.54 3.15
Cleanliness 4 7 10 36.00 0.76 1.30
Shelters 1 4 5 16.00 0.34 0.30
Access 3 11 17 48.00 1.01 1.90
Journey Time 4 12 15 51.00 1.08 2.05
Route 2 7 20 40.00 0.84 1.50
Card 2
Frequency 47 7 5 160.00 3.38 7.50
Bus seating 12 15 7 73.00 1.54 3.15
Journey time 6 20 3 61.00 1.29 2.55
Comfort 2 11 10 38.00 0.80 1.40
Information-service 7 17 22 77.00 1.62 3.35
Information-incidents 2 1 19 27.00 0.57 0.85
Bus driver attention 3 5 9 28.00 0.59 0.90
ICTs 0 3 4 10.00 0.21 0.00
Card 3
Bus driving security 34 9 8 128.00 2.88 6.33
Information-incidents 4 5 8 30.00 0.68 1.10
Route 4 7 7 33.00 0.74 1.26
ICTs 4 1 6 20.00 0.45 0.57
Punctuality 22 27 8 128.00 2.88 6.33
Frequency 2 18 14 56.00 1.26 2.49
Connections 2 4 11 25.00 0.56 0.83
Comfort 2 3 12 24.00 0.54 0.78
Card 4
Shelters 14 0 3 45.00 1.23 2.41
Access 21 12 5 92.00 2.51 5.46
Bus driver attention 1 9 2 23.00 0.63 0.99
Connections 13 21 8 89.00 2.43 5.26
Bus driving security 5 13 23 64.00 1.75 3.64
Cleanliness 2 2 9 19.00 0.52 0.73
Bus seating 5 4 11 34.00 0.93 1.70

Table 5 Table 6
Final ranking of attributes according to importance. Cluster analysis results according to bus lines. Number of clusters in two stages.

Variable Ranking Points over 100 Line Estimation Clusters in two stages Total

Punctuality 16.33 20.47 Stage 1 Stage 2

Frequency 9.99 12.52
Bus driving security 9.98 12.50 712 Counting 43 164 207
Access 7.36 9.22 % inside the clusters (two 25.4 46.7 39.8
Information-service 6.50 8.15 stages)
Connections 6.10 7.64 713 Counting 26 90 116
Bus seating 4.85 6.08 % inside the clusters (two 15.4 25.6 22.3
Journey time 4.60 5.76 stages)
Route 2.76 3.46 714 Counting 70 21 91
Shelters 2.71 3.40 % inside the clusters (two 41.4 6.0 17.5
Comfort 2.18 2.73 stages)
Cleanliness 2.03 2.54 716 Counting 30 76 106
Information-incidents 1.95 2.45 % inside the clusters (two 17.8 21.7 20.4
Bus driver attention 1.89 2.37 stages)
ICTs 0.57 0.71 Total bus Counting 169 351 520
lines % 100.0 100.0 100.0

service. Gender also influences perception of cleanliness, bus the lines.

comfort, route and access to stops. Finally, trip purpose has an To exploit the opportunity to test other techniques using this
effect on perception of frequency and the presence of bus shelters case study sample, we looked for latent SQ attributes or a group of
along the route. attributes that could best explain user perception. As factor ana-
To complement the Pearson Chi-squared test and detect dif- lysis has already been used for this purpose in SQ studies on urban
ferent behaviours between bus lines, a cluster analysis was applied public transportation (D’Ovidio et al., 2014) and recently also in
to the sample (see Table 6), which revealed that the majority of the high-speed train HST services (Alpu, 2015), this tool was applied to
sample observations from lines 712, 713 and 716 belong to the the sample together with a SEM methodology called MIMIC
same group (2), while those obtained from line 714 were assigned (Multiple Indicators Multiple Causes). Three of the 15 attributes
to another group (1). This is consistent with the fact that Line 714 included in the conventional survey – suggested by the operating
is a special case, since it connects the interchange hub to a uni- company – were excluded from this analysis, as they were in a
versity campus (Universidad Autónoma de Madrid UAM), meaning separate block in the questionnaire and were directly correlated to
this bus service is a specialised line for trips for the purpose of comfort (driving security, bus driver attention and seating com-
study, and the socioeconomic user profile differs from the rest of fort). The main aim of the factorial analysis was to identify the
 B. Guirao et al. / Transport Policy 49 (2016) 68–77 75

underlying variables or factors that explain the pattern of corre- more than 70% of the users surveyed were workers (49.8%) or
lations within a set of observed variables. After several trials, three students (31.7%).
indicators were identified with high factor loading ( 40.5), and in In addition to these supplementary studies, the conventional
order to explain the variance of 56.1% with a KMO index (Kaiser- survey allowed us to derive the attribute importance and compare
Meyer-Olkin) of 0.803, all the 11 attributes were included in the the results with those obtained from the conventional survey.
analysis, except “shelters”, due to its loading factor (lower than Multiple regression analysis was used to design a model in which
0.5). The first factor identified was designated SERVICE, as it de- the dependent variable was overall service satisfaction (CSI) –
scribes the quality attributes associated to the characteristics of whose values were collected from the last question in the con-
the service operation such as punctuality, frequency, information- ventional survey – and the dependent variables were the 15
service and information-incidents. The second factor was called quality attributes. The coefficients for each attribute therefore re-
INTEGRATION as it captures concepts associated to the inclusion of presented the average weight (or importance) given by the users.
the bus in the transportation systems, such as access to bus stops, Table 7 shows the descriptive statistics of the variables from the
connections to other modes, journey time and route. Finally the multiple regression. Specifically it can be seen that the attributes
third factor, identified as SUPPLEMENTARY FEATURES, includes considered most important are “frequency”, “punctuality”, “route”
attributes that are usually secondary to the users such as comfort, and “bus driving security”, three of which are also the most im-
cleanliness and ICTs. portant attributes in the stated importance survey, providing a
Fig. 2 shows the path diagram with the significant parameters consistent validation of the top positions in the ranking. There is
and relations (p o0.1) from the best MIMIC model estimation also consistency in the worse positions for the rest of the attri-
obtained with the following modelling fit indexes: root mean butes (“ICTs” and “Information-incidents”) and in some inter-
square error of approximation (RMSEA ¼0.079), confirmatory fit mediate positions (“journey time” and “bus seating position), al-
index (CFI¼0.956) and adjusted goodness-of-fit statistic though some differences can be seen in the rest of the ranking
(AGFI ¼0.820). The MIMIC estimation was obtained using the positions.
AMOS program from the SPSS package (Arbuckle, 2013). This It is clear from the stated importance survey that attributes
diagram shows how the relations between the observable vari- linked to supplementary features (cleanliness, comfort, ICTs, bus
ables and the three main factors (Service, Integration and Sup- driving attention) are worse ranked than those associated to ser-
plementary Features) are weaker (0.3) than the relation between vice (punctuality, frequency, bus driving security and information
factors and quality attributes. The results show more clearly how service), which are in the top positions. Attributes included in the
gender mainly affects the perception of attributes associated to concept of integration (connections, access, route, journey time)
factors of Integration and Service, while users with work purpose are in the middle of the ranking list. These results are clearly
trips are more sensitive to Service and Integration attributes. Fi- connected to the MIMIC model designed to analyse the presence
nally, the age of the customers conditions the perception of the of latent variables in this case study, and also consistent with the
attributes linked only to the Supplementary Features of the ser- trip and user profile statistics (more than 49% work trips).
vice, and usually means that the older the users, the lower the In the derived importance ranking, the main differences in the
rating of comfort, cleanliness and ICTs. These results are consistent attribute position correspond to the integration category (“route”,
with those obtained using the frequency analysis of the attribute “access”, “connections”) and two in the “supplementary features”
ratings in the conventional survey, in which the work-based trips category (“comfort” and “bus driver attention”). As “route” was the
are not as sensitive to the comfort attribute, and Supplementary second most important attribute, we revised the survey format for
features and Integration attributes are worse rated by women than differences in the wording or interpretation of the route question
by men. in the conventional survey and the stated importance survey. We
The results obtained with the study of latent variables in- found one small difference in the wording that may have affected
evitably led to the issue of the disaggregation of attribute im- the results: while in the stated survey the concept of route was
portance according to user and trip profile. Two of the most im- explained in brackets (“itinerary and stops”), in the conventional
portant attributes in the stated importance survey were punctu- survey no explanation was given of this category. This means that
ality and frequency, and both attributes are directly link to work- the interpretation of what the route actually involves (location of
based trips. This is consistent with the survey's main statistics, as stops, distance between stops, adaptation to urban sprawl and

Fig. 2. Path diagram with the significant parameters and relations (po 0.1) of the best MIMIC model estimation (RMSEA¼ 0.079. CFI¼ 0.956. AGFI ¼ 0.820).
76 B. Guirao et al. / Transport Policy 49 (2016) 68–77

Table 7 directly from a stated preference survey of customers. Attribute

Multiple regression model to derive importance. Descriptive statistics of the importance is one of the main indicators of user perception used
variables.
by operating companies to estimate when an improvement in the
Main model indicators SQ is required. We maintain that the simplicity and potential of
the method are two of the strengths of our work, since this allows
R Squared R Adjusted Squared-R Standard error
of the estimate
it to be easily applied in CSS designed and conducted by operating
0.669 0.447 0.431 1.043 companies. Stated importance methods are more intuitive than
derived importance methods for these companies, but require a
Model coefficients
significant increase in the length of the questionnaire, which
Variable Non-standardised Standardised T Sig. lowers the overall response rate and the accuracy of the survey. As
coefficients coefficients shown in the literature, new techniques to obtain stated im-
portance through CSS have been practically abandoned by aca-
B Error Beta
demics, whereas it is becoming increasingly common to develop
Constant  0.915 0.532  1.719 0.086 new methods to derive importance from conventional CSS. There
Route 0.317 0.060 0.203 5.266 0.000 is also a lack of studies on comparative methodologies to obtain
Access 0.042 0.053 0.030 0.794 0.428
Punctuality 0.248 0.056 0.163 4.423 0.000 attribute importance using the same case study data (and even
Frequency 0.254 0.053 0.190 4.822 0.000 comparison between the most commonly used derived im-
Access 0.018 0.064 0.010 0.075 0.784 portance methodologies). This absence is also a problem for op-
Information- 0.008 0.030 0.010 0.267 0.790
incidents
erating companies, who also need practical guidelines and re-
Cleanliness 0.133 0.081 0.060 1.652 0.099 commendations to implement new approaches in their CSS.
Information- 0.099 0.058 0.066 1.717 0.087 The method proposed in this paper calculates attribute im-
service
portance on the basis of both stated preference and hierarchy
Journey time 0.113 0.077 0.054 1.457 0.146
Comfort 0.209 0.072 0.090 2.593 0.009 process theories, separating the ranking options into blocks and
ICTs 0.007 0.025 0.011 0.302 0.763 calculating the final score for each attribute. It reduces the length
Shelters 0.006 0.044 0.005 0.130 0.897 of the questionnaire, which is one of the main drawbacks of the
Driving 0.349 0.076 0.169 4.617 0.000
security
traditional ranking technique used by companies to state attribute
Seating 0.208 0.087 0.085 2.395 0.017 importance. This pioneer survey campaign (293 valid ques-
Driver 0.188 0.071 0.097 2.649 0.008 tionnaires) was validated using conventional face-to-face surveys
attention
(520 valid questionnaires). Using Madrid as a case study, the re-
liable survey database also offered a good opportunity to test
different traditional techniques to study attribute importance
even journey time were open to interpretation) was not delimited.
(factor analysis, MIMIC models and multiple regression analysis).
We agree that differences in wording should be avoided in any
The validation process revealed that the top and bottom ranking
repetition of these comparative surveys in order to strengthen the
positions are perfectly identified by the stated survey. Some dif-
validation process, but there is another argument that reinforces
ferences between the two ranking results were observed, mainly
the consistency of the results from the stated importance survey.
The study carried out in 2005 by the CRTM (Consorcio Regional de in intermediate positions, although the derived importance results
Transportes de Madrid, 2005) on bus SQ in the same city and for were less consistent with the MIMIC model analysis and even with
the same type of (periurban) corridors using focus group techni- SQ studies conducted in the same context (same area as the study
ques revealed a ranking of SQ attributes quite similar to our stated case and same type of periurban service).This pioneer experience
importance survey results, where the “route” attribute (when used could be improved by avoiding any differences in wording in a
to define itinerary and stops) was not in the top positions but in repetition of these comparative surveys (to strengthen the vali-
intermediate ones. dation process) and increasing the number of valid questionnaires
We can therefore conclude that the stated preference survey in order to allow a direct disaggregation study (according to trip
clearly identifies the top and bottom positions in the attribute and user profile).
ranking, while the rest of the positions are more consistent The authors of this paper acknowledge that the policy re-
(compared to the ranking derived from the attribute perception) commendations derived from this study are constrained by the
with the MIMIC model results, and even with the rankings ob- number of similar experiences and the sample size used (which
tained in other SQ studies in the same study case area/context. We only focuses on four periurban bus lines), but the findings cannot
agree that the disaggregation of attribute importance according to be ignored by operating companies. We encourage operating
user and trip profile is also recommended for the stated preference companies to include (in their CSS surveys) both attribute rating
survey and would give a better understanding of the ranking re-
questions and attribute importance questions together with the
sults; however the number of valid surveys (293) was insufficient
overall level of satisfaction with the service. This part of the survey
for a significant disaggregation, and the user and trip profile were
can be simplified by using a reliable hierarchy card system (like
fairly similar in both surveys. The results of this pioneer experi-
the one used in our experience), thus reducing the length of the
ence indicate that stated survey techniques (to estimate attribute
importance in SQ studies) should not be abandoned by academics, survey format. This recommendation would allow operating
and that transport planners and operators should also continue to companies to easily obtain their own results on attribute im-
use these techniques, guided by this experience and implementing portance, in addition to using the findings from more complex
this intuitive and simple survey method in their CSS design. scientific academic research.
The results of this paper clearly offer transport management
6. Conclusions and recommendations companies a simple and useful tool for use in their Customer Sa-
tisfaction Surveys (CSS), thereby narrowing the gap between
This paper proposes a method to estimate attribute importance practitioners’ needs and scientific research.
 B. Guirao et al. / Transport Policy 49 (2016) 68–77 77

Acknowledgments European Committee for Standardization, 2003. EN 13186. Transportation. Logistics

and Services. Public Passenger Transport. Service Quality Definition, Targeting
and Measurement.
The conclusions in this article are derived from a research work European Union, 1998. QUATTRO – 4th F.P. project Quality approach in tendering/
on service quality in public transport by the authors of this paper contracting urban public transport operations. Final Report. Synthesis and re-
as members of the Department of Transport at the Madrid Poly- commendations 1998. 〈http://www.transport-research.info/web/projects/pro
ject_details.cfm?id ¼ 636〉, (29.07.14).
technic University (UPM) and the Consorcio Regional de Trans- Grönroos, C., 1988. Service quality: the six criteria of good service quality. In: St.
portes de Madrid (Regional Transport Board – CTRM). The ALSA John’s University (Ed.), Review of Business. St. John's University Press, New
Group (the bus operator) made significant contributions to this York.
research, particularly with regard to designing the customer sa- Guirao, B., García, A., López, M.E., Acha, C., Comendador, J., 2015. New QR survey
methodologies to analyze user perception of service quality in public transport:
tisfaction survey, authorising on-board survey campaigns and the experience of Madrid. J. Public Transp. 18 (3), 71–88.
making the bus staff available to the research team throughout the Hensher, D.A., Stopper, P., Bullock, P., 2003. Service quality-developing a service
study. quality index in the provision of commercial bus contracts. Transp. Res. Part A
37, 499–517.
Hill, N., Brierley, G., MacDougall, R., 2003. How to Measure Customer Satisfaction.
Gower Publishing, Hampshire.
References Hu, K.C., 2010. Evaluating city bus service based on zone of tolerance of expectation
and normalized importance. Transp. Rev. 30 (2), 195–217.
Lai, W., Chen, C., 2011. Behavioral intentions of public transit passengers-the roles
Alpu, O., 2015. A methodology for evaluating satisfaction with high-speed train of service quality, perceived value, satisfaction and involvement. Transp. Policy
services: a case study in Turkey. Transp. Policy 44, 151–157. 18 (2), 318–325.
Ampt, E., Meyburg, A., 1995. Survey Methods for Transport Planning. Eucalyptus Lehtinen, U., Lehtinen, J., 1982. Service Quality – A Study of Quality Dimensions.
Press, United States. Service Management Institute, Helsingfors.
Arbuckle, J.L., 2013. IBM SPSS Amos 22 User's Guide. Amos Development Cor- de Oña, J., de Oña, R., Calvo, F.J., 2012. A classification tree approach to identify key
poration, Crawfordville, FL. factor of transit service quality. Expert Syst. Appl. 39, 11164–11171.
Aydin, N., Celik, E., Gumus, A.T., 2015. A hierarchical customer satisfaction frame- de Oña, J., de Oña, R., Eboli, L., Mazzulla, G., 2016. Index numbers for monitoring
work for evaluating rail transit systems of Istanbul. Transp. Res. Part A: Policy
transit service quality. Transp. Res. Part A 84, 18–30.
Pract. 77, 61–81.
Opricovic, S., Tzeng, G.H., 2004. Compromise solution by MCDM methods:a com-
Bordagaray, M., dell’Olio, L., Ibeas, A., Cecín, P., 2014. Modelling user perception of
parative analysis of VIKOR and TOPSIS. Eur. J. Oper. Res. 156 (2), 445–455.
bus transit quality considering user and service heterogeneity. Transp. A:
Parasuraman, A., Zeithaml, V.A., Berry, L.L., 1985. A conceptual model of service
Transp. Sci. 10 (8), 705–721.
quality and its implications for future research. J. Mark. 49, 41–50.
Brady, M.K., Cronin, J.J., 2001. Some new thoughts on conceptualizing perceived
Pearmain, D., Swanson, J., Kroes, E., Bradley, M., 1991. Stated Preference Techniques:
service quality: a hierarchical approach. J. Mark. 65 (3), 34–49.
A Guide to Practice. Steer Davies and Gleave and Hague Consulting Group,
Celik, E., Aydin, N., Gumus, A.T., 2014. A multiattribute customer satisfaction eva-
London.
luation approach for rail transit network: a real case study for Istanbul, Turkey.
Román, C., Martín, J.C., Espino, R., 2014. Using Stated preferences to analyze the
Transp. Policy 36, 283–293.
service quality of public transport. Int. J. Sustain. Transp. 8 (1), 28–46.
Chau, V.S., Kao, Y.Y., 2009. Bridge over troubled water or long and winding road?
Sánchez, M., Gázquez, J.C., Marín, G.M., Sánchez, R., 2007. Effects of service quality
Gap-5 in airline service quality performance measures. Manag. Serv. Qual. 19
(1), 106–134. dimensions on behavioural purchase intentions: a study in public‐sector
Chou, C.C., Liu, L.J., Huang, S.F., Yih, J.M., Han, T.C., 2011. An evaluation of airline transport. Manag. Serv. Qual. 17 (2), 134–151.
service quality using the fuzzy weighted SERVQUAL method. Appl. Soft Com- Saako, N., 2001. Guidelines for Stated Preference Experiment Design. Dissertation
put. 11, 2117–2128. submitted for publication for the degree of Master of Business Administration.
Consorcio Regional de Transportes de Madrid, 2005. Estudio Sociológico sobre Los School of International Management Ecole Nationale des Ponts et Chaussées
atributos utilizados En La evaluación DE La calidad del servicio En La Red DE (Class 2000/01).
líneas interurbanas. EMIC-ETIC Consultans. Madrid. Saaty, R.W., 1987. The analytic hierarchy process-what it is and how it is used.
D’Ovidio, F.D., Leogrande, D., Mancarella, R., Schinzano, A., Viola, D., 2014. A Mul- Math. Model. 9 (3–5), 161–176.
tivariate Analysis of the Quality of Public Transport Services. Procedia Econ. Shen, W., Xiao, W., Wang, X., 2016. Passenger satisfaction evaluation model for
Financ. 17, 238–247. Urban rail transit: a structural equation modeling based on partial least squares.
Del Castillo, J.M., Benitez, F.G., 2013. Determining a public transport satisfaction Transp. Policy 46, 20–31.
index from user surveys. Transp. A: Transp. Sci. 9 (8), 713–741. Transportation Research Board (TRB), 1999. A Handbook for Measuring Customer
Dell’Olio, L., Ibeas, A., Cecín, P., 2010. Modelling user perception of bus transit Satisfaction and Service Quality. TCRP Report 47.
quality. Transp. Policy 17, 388–397. Transportation Research Board, 2004. Transit Capacity and Quality of Service
Eboli, L., Mazzulla, G., 2007. Service quality attributes affecting customer satisfac- Manual, second edition.
tion for bus transit. J. Public Transp. 10 (3), 21–34. Weinstein, A., 2000. Customer satisfaction among transit riders. How customers
Eboli, L., Mazzulla, G., 2009. A new customer satisfaction index for evaluating rank the relative importance of various service attributes. Transp. Res. Rec.
transit service quality. J. Public Transp. 12 (3), 21–37. 1735, 123–132.