Transport Policy 63 (2018) 157–164

Contents lists available at ScienceDirect

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

Modelling user perception of taxi service quality

Borja Alonso, Rosa Barreda, Luigi dell’Olio *, Angel Ibeas
University of Cantabria, Grupo de Investigacion de Sistemas de Transportes (GIST), Av. De los Castros 44, 39005, Santander, Spain

A R T I C L E I N F O A B S T R A C T

Keywords: This article presents a research study on modelling taxi user perceived quality. Following a thorough review of the
Taxi available international literature, relevant variables are individualised using focus groups of taxi users, and a
Service quality satisfaction survey is designed. The resulting data are used to estimate two types of ordered probit models, which
Ordered probit consider systematic and random variations in taste: on the one hand, a model that provides attributes and their
Random effect influence on the perceived quality of taxi service when no previous information about the system is available and,
Systematic variation in taste
on the other hand, a model that considers their changed perception after being informed about the attributes that
could affect the system's perceived quality. The results mention the attributes that the users normally value and
those that they would value after being asked to reflect on them.
It is worth noting that waiting time is the most important factor for frequent users, and journey time is highly
valued by almost all users, along with safety, accessibility and comfort. Surprisingly, the fare is not usually as
important as at the first thought. The results of this type of study are valuable because they allow us to understand
and improve the system depending on user preferences, thereby improving the quality of services and increasing
the demand.

1. Introduction
The constant growth in the number of private vehicles travelling on
urban roads causes problems such as noise, congestion, pollution and
traffic jams. All administrations address these issues by promoting sus-
tainable mobility, a policy fundamentally based around the use of public
transport. These goals can be achieved by using management tools to
support policy application while optimising the use of available re-
sources. The first step in guaranteeing sound resource management
consists of applying user and journey characterisation studies to deter-
mine how users perceive the quality of service when deciding to make
their journeys by public transport.
From a point of view of sustainable mobility, the taxi is not the best
way to travel around an urban area; nevertheless, it has its environmental
advantages in that people will generally use it to avoid problems asso-
ciated with parking and for its speed and ease of use. Taxi use avoids the
creation of ‘parasitic traffic’ due to cars cruising for a parking place, and
it frees up public space because one vehicle provides a service to many
users who have not used their own vehicle and who would otherwise be
occupying street space and ultimately a parking space. To encourage
continued public use, the service needs to operate in accordance with
certain quality standards that would be better under public control.
This article proposes an overall methodology for studying the quality
of service perceived by taxi users. The methodology includes the design
of a satisfaction survey and the use of ordered probit models on the data
collected to propose marketing strategies aimed at increasing taxi use.
The paper is structured as follows: the introduction is followed by an
analysis of the state of the art, and then, the methodology used is pre-
sented, which includes the design of the questionnaire and an explana-
tion of the models used in the analysis of the data collected. A case study
is then described, and the results are analysed. This is followed by the
most important conclusions drawn from the work.
2. State of the art
The perception and understanding of user behaviour relative to service
quality is vital in researching any mode of transport (Tam et al., 2008;
dell’Olio et al., 2011). An abundance of literature is available on the
methodologies for evaluating the perceived quality of service of transport
modes. These were based mainly on the debate and application of two of
the most well known and most widely used methodologies: SERVQUAL
(Parasuraman et al., 1988) and SERVPERF (Cronin and Taylor, 1992).
Alternative approaches were applied later: the use of neuronal net-
works (Behara et al., 2002; Garrido et al., 2014), discrete choice models

* Corresponding author.
E-mail address: delloliol@unican.es (L. dell’Olio).

https://doi.org/10.1016/j.tranpol.2017.12.011
Received 11 January 2016; Received in revised form 11 October 2017; Accepted 6 December 2017
Available online 3 January 2018
0967-070X/© 2017 Elsevier Ltd. All rights reserved.
B. Alonso et al.
(Hensher and Prioni, 2002; dell’Olio et al., 2010; Eboli and Mazzulla,
2008, 2014; Echaniz et al., 2017), the use of mixed data from passenger
perceptions (subjective and objective measures) and transit agency per-
formance measurements (Eboli and Mazzulla, 2011), structural equa-
tions (De O~ na et al., 2013) and classification tree techniques (de O~na and
de O~ na, 2015). An extensive review can be found in the studies by de O~ na
and de O~ na (2014) and Barabino and Di Francesco (2016).
However, few works have concentrated on evaluating the importance
of perceived quality on the demand for taxis. Yao and Ding (2011) stated
that despite the scarcity of publications about the subject, many cities
around the world (e.g. Singapore, Shanghai, Hangzhou or Hong Kong)
periodically check the prevailing perception of taxi service quality.
However, they also explain that no research has shown that the variables
being measured in these studies really show the true quality perceived by
users.
Despite being in a different context, we found a clear example of the
importance of choosing the right variables for measuring the quality of
service in the research study by Tan et al. (2014). They held focus groups
(FGs) to find the key factors and attributes for measuring the quality of
service in the Chinese fast food restaurant sector before applying the
widely used DINESERV method (Stevens et al., 1995). They found that
additional variables should be added to the standard method to adapt it
to a different cultural context. As will be explained later, in the taxi
service case, the cultural and economic context has a significant bearing
on the importance of the variables perceived by users as well.
The first studies the authors are aware of that refer to the quality of a
taxi service were from the early 1970s, and all of them were related to
regulating or liberalising the service (Douglas, 1972; De Vany, 1975; Abe
and Brush, 1976; Foerster and Gorman, 1979; Paratransit Services, 1983;
Frankena and Pautler, 1984; Moore and Balaker, 2006; Yang and Yang,
2011). In most of these studies, quality of service is expressed in terms of
taxi user waiting time, which is a relevant variable used in the design,
modelling and market research of the sector (Yang and Wong, 1998; Yang
et al., 2000; Wong et al., 2001; Wong et al., 2008; Kim et al., 2011);
however, the perception of the users themselves and the consideration of
certain other variables other than the waiting time have not been so widely
studied. This fact was confirmed by Tam et al. (2010) who incorporated
satisfaction as a latent variable among the explanatory variables in a mode
choice model on the entrances to the Hong Kong international airport. A
noteworthy improvement in specification and estimation was found,
showing that satisfaction makes an important contribution.
The importance of customer satisfaction has increased recently
because of the appearance of new competing players in the taxi sector.
Wallsten (2015) analysed the competition between taxi services and
ride-sharing services such as Uber. He used two case studies, the city of
New York and Chicago, and analysed the evolution of the number of
complaints. He found that this new sharing mode has considerably
reduced the number of complaints in both cities, or, at least, the partic-
ular complaints related to service facilities. Furthermore, the author
found that the results from both cities were consistent with the idea that
taxis respond to new competition by improving quality.
One of the first studies to cover a wide range of variables for
measuring the quality of service was done by Yao and Ding (2011). They
evaluated perceived quality and user satisfaction by weighting the
SERVPERF method using the data from 230 questions asked to taxi users
in Hangzhou (China). The 22 variables analysed, adapted for the taxi
service, were grouped into five dimensions: tangible, assurance, reli-
ability, responsiveness and driver empathy. The drivers' knowledge of
routes was a determining factor (they explain that this was to be expected
as it determines how long the journey will take or how much it will cost),
and all the variables related to driver behaviour were highly valued by
the users.
The importance of these variables is consistent with the research of Li
and Song (2011). They applied entropic weighting to find the relative
weight of each quality variable and the well-known TOPSIS method
(Hwang and Yoon, 1981; He et al., 2009) to find the satisfaction index of
158
Transport Policy 63 (2018) 157–164
each explanatory variable. Twenty-two variables were analysed, and the
quadrant method was used to combine the two factors and evaluate the
results. According to this method, the authors recommended the
improvement of factors related to information, vehicle performance and
driver behaviour. Dachyar and Rusydina (2015) added the corporate
image as a significant variable to improve customer satisfaction and
loyalty in Jakarta.
Recent studies have correlated the user perception of performance
and quality of service variables with their socioeconomic characteristics.
Techarattanased (2015) and Shaaban and Kim (2016) analysed the taxi
services in Bangkok and Doha, respectively. The former applied the
SERVQUAL-based approach and the latter proposed a SEM-based
approach. The results of both studies are conditioned to the socioeco-
nomic context of case studies; however, both studies found driver
behaviour to be an important variable, among others, to improve both
the quality of service and user loyalty, and they stated that travel cost was
statistically significant only for low-income level users. However, in Doha
case study, Shaaban and Kim also highlighted accessibility to taxi ranks
as a key variable.
As can be seen, there are several approaches for modelling the quality
of taxi services. The importance of the variables analysed is strongly
dependent on the cultural and socioeconomic context. However, the
variations in the user perception of these variables have not been studied
in detail. The present research tends to fill this gap by combining (i) a
previous analysis based on FGs for choosing the right variables to mea-
sure and to provide the input for the final user survey (revealed prefer-
ence survey) and (ii) an approach based on applying ordered probit
models that consider systematic and random variations in taste.
The aim of this paper is to make progress on all the previous studies in
the following ways:
a. Propose a robust quality evaluation method for finding the attributes
valued by users and the attributes they would value if they had pre-
vious knowledge about them.
b. To create a method not based on synthetic indices with the incon-
venience of not being able to determine the statistical relevance of the
quality attributes, but on an ordered probit model, which is also a
compartmentalised model.
c. A method that can capture systematic variations in user tastes, an
important factor given that it allows administrations to implement
direct marketing policies aimed at certain sections of users.
3. Methodology
When modelling the quality of service as perceived by users, it is
important to remember that perceptions are qualitative, and therefore,
the determinant variables for consideration when designing the satis-
faction survey were found to involve the qualitative methodology of FGs
formed of taxi users.
The main objective of holding the FG was to help the planners design
the surveys. The FG is one of the qualitative tools used in social research
for the process of citizen involvement. The FG allows social discourse to
be reproduced on a small scale by selecting suitable participants (eight or
nine members) defined according to the objectives of the study (Ibeas
et al., 2011).
For the questionnaire designed in this research, the judgement of
quality varied on a scale of five levels ranging from ‘very bad’ to ‘very
good’. The model proposed in this article is based on the McKelvey and
Zavoina (1975) specification, who defined the ordered model as a latent
regression adapted to ordinal outcomes. In the taxi user perceived quality
study, the semantic and qualitative responses of the scale defining the
overall service are converted into four discrete but ordinal numerical
values (‘very bad-bad’, ‘regular’, ‘good’ and ‘very good’, and as there
were very few replies scoring ‘very bad’, this particular category was
merged with ‘bad’) and introduced as the dependent variable following
expressions (1) and (2) below (Greene and Hensher, 2010):
B. Alonso et al.
q*i ¼ θki ⋅vki þ εi ; εi  FðE½εi j Var½εi Þ; E½εi  ¼ 0; Var½εi  ¼ 1 (1)
qi ¼ j if μj1 < q*i  μj (2)
where j represents each response from the numerical scale.
Expressions (1) and (2) imply that the unobservable, dependent and
continuous variable, q*, which represents the overall score for service
quality, is transformed into an ordinal and discrete variable q containing
the user responses. In the ordered probit model, the error term εi in (1) is
assumed to be normally distributed.
By applying the maximum likelihood method, we can estimate the
threshold parameters μj and the parameters θki associated with the var-
iables vki, which represent the perceived quality scores of taxi users for
the specific variables relative to the quality attributes in discrete choice
models. The parameters can be identified by fixing the value of μ1 by
making (μ1 ¼ 0). This method produces the most probable answer for
each individual according to the responses they provided for the
explanatory variables and their socioeconomic characteristics and
mobility constraints.
The generalisation of the ordered models means that the parameters
θki are random variables and not constants and are estimated by aver-
aging the collected data (Bordagaray et al., 2014, 2015). As the param-
eter θki that is associated with the specific quality variables perceived by
the taxi users vki is not fixed but varies from individual to individual, the
perceived overall quality qi is also an individual variable. This allows the
model to more closely represent reality by considering the different tastes
found between different users.
The random coefficients correspond to the following definition:
θki ¼ θ*ki þ Δzi þ Γrki (3)
where θ*ki is the average value of the parameter in the population, and the
systematic heterogeneity in the mean is induced by the variables zi,
which are normally socioeconomic in nature or journey constraints. The
variability across the population in the perception of each explanatory
variable k of alternative i is considered through the unobservable random
component rki, presenting the distribution which best fits the existing
heterogeneity.
Γ is the lower triangular matrix that introduces the correlation be-
tween random parameters.
Similarly, in the ordered model, the variability in the parameter
thresholds μi may be studied by taking different values for each indi-
vidual. Systematic variations that can be studied by introducing in-
teractions between the vki variable and the socioeconomic variable ski can
also be added to the random variations.
The quality perceived by taxi users is modelled with calibrated
models, representing the process of valuing quality from a limited group
of predefined variables vki collected from each surveyed user. The choice
of these variables vki is of great importance as they help to explain the
choice process to be modelled. This choice was made with the support of
a thorough international bibliographic review and the information pro-
vided by the taxi user FG. The collected data were analysed and modelled
to check the behaviour of the different variables. Finally, the definitive
survey was designed to provide the final input data for the calibrated
models. The Revealed Preferences survey consisted of a series of ques-
tions about the interviewee's current experiences with the service being
provided. Each user i was asked about their overall perception of service
quality qi, then their score for each quality attribute vki, and finally, after
providing the scores for these specific attributes (vki), they were asked
whether their final overall score for quality (qfi ) had changed with respect
to the initial score they gave, qi.
In a previous study on the perceived quality of urban public transport
users, dell'Olio et al. (2010) concluded that the overall quality of a system
is perceived differently in the beginning (qi) of the interview from after
having valued different specific aspects of the service (qfi ). The potential
159
Transport Policy 63 (2018) 157–164
for explaining choice mechanisms using different perceptions at different
times during the survey is the reason why both scores (qi and qfi ) were
asked for in this survey.
4. Results
The methodology explained in section 3 is applied to the real case of
Santander, the capital city of the Autonomous Community of Cantabria
(northern Spain), one of the 17 cities in the Spanish state. Santander is a
medium-sized city covering an area of 36 km2, with a population of
around 200,000. Public transport is represented by a network of bus
lines, a taxi service and a recently installed public bicycle system.
Two FGs were held. One specifically comprised taxi users to assess
their perceived and desired qualities and to detect the problems they
perceived about the taxi mode in the urban environment. This was a
mixed FG that contained 11 people, 6 women and 5 men, of varied ages
(from 18 to 67 years) with average income levels. The members used
taxis at different frequencies (rarely, with little frequency, frequently or
several times a week).
The second FG consisted of 8 male taxi drivers with ages ranging from
35 to 60 years, and 80% of them were aged between 45 and 60 years.
This reflects the clear male domination of the sector, where very few
female taxi drivers are active, although their numbers are starting to
increase. Taxi drivers in Santander are mainly self-employed workers
with a taxi licence; a few are employees (mostly immigrants in this case).
The FGs identified the variables that determined the taxi as the chosen
mode of transport, thereby providing knowledge about the occasions, the
most frequent journeys, the reasons for using the taxi and the most
common problems that the users were associated with the taxi service.
Knowledge about the stops, waiting times, fares, available information
and the level of understanding about the current fare system is forth-
coming. The FG provided the perceived acceptable journey time by taxi
compared to other modes of transport and the feasibility of combining the
different available modes of public transport: bus-taxi-public bicycle.
One of the main conclusions from the FG was that the Santander taxi
service is well regarded with high satisfaction levels. The main variables
provided by the FG were as follows: journey time, waiting time, fare, taxi
rank signage, taxi rank location, payment method, accessibility, on board
comfort and, finally, occupancy.
The most common occasions for using the taxi by the people
participating in the FG were trip to the airport or bus and train stations or
to return home after a night out on the weekend. The main barriers
against using the taxi as a mode of transport were as follows:
 High cost. Transport by taxis is thought to be of higher cost than by
any other mode.
 Lack of taxi availability at certain locations during the night and when
it is raining.
 Uncertainty about waiting times at the taxi rank.
Interestingly, the users expressed a preference for calling a taxi rather
than going to the rank to find one. The main reasons for this preference
were the following suppositions: they do not take long time to arrive, the
taxi comes to the passenger's location, the locations of the taxi ranks are
unknown and the taxi ranks are often empty or there is a queue.
The result of the FG is very interesting because it is used as the
starting point for designing the survey, but as will be seen, the results of
the survey are much more detailed and statistically representative than
those produced by the FG. For example, the survey shows that despite the
thoughts of the FG members, most taxi users find a taxi at the rank
(77.2%) and not by telephone. Furthermore, most trips do not start at the
airport or the train station as thought by the FG, only 10% of all trips start
at these locations.
A total of 215 questions were asked to taxi users. The survey had two
parts: one for characterising the user and the other to study their
perception of quality (see Fig. 1).
B. Alonso et al.
The user and journey characterisation data are presented below
(Table 1).
After analysing the data collected during the survey, the next step is to
model the perception taxi users have of the service provided. As
mentioned above, the users were initially asked about their overall
perception of the quality of service qi, and after asking about specific
quality attributes vik, they were once again asked for their final overall
score for service quality qfi .
The following table (Table 2a) shows the frequencies of the users'
replies for the initial (qi) and final (qfi ) scores. The results show us that the
initial and final scores have very similar frequencies in the total column
for qi and qfi . However, this changes substantially if the differences in the
initial and final replies for individual users are studied.
Of the interviewees, 21% changed their opinion with respect to the
initial and final scores; the change was both positive and negative. The
differences between qfi and qi were analysed for each type of user and
journey and were determined as ηi ¼ qfi - qi. The data associated with
these changes are presented in Table 2b and Fig. 2.
The figures show a tendency for the overall quality score to worsen,
independently of the population category or the type of journey being
considered.
The only exception is in the case of daily taxi users who increase their
final score for overall quality. This result contrasts greatly with that of
dell'Olio et al. (2010) who carried out a similar study for an urban bus
service and found a tendency to improve the final overall score once the
service users had time to individually reflect on all the important attri-
butes affecting service quality.
This difference could initially be put down to the different ways that
each mode of transport is perceived; for most people, the bus is a
frequently used public transport service, while the taxi is only occa-
sionally used. Therefore, the taxi will be perceived as more of a luxury
Fig. 1. Survey.
160
Transport Policy 63 (2018) 157–164
item, and its users will be more critical when making their final quality
evaluation.
Both studies clearly show that frequent users are less critical than
occasional users. This is probably because they depend on the particular
service for their daily journeys, and as they have little or no alternative,
they accept it as it is without further criticism. As they use the service so
frequently, they have much greater experience of a great variety of sit-
uations (compared to less frequent users); therefore, when they make
their final judgement on quality, they appreciate that the service is in fact
even better than they first thought.
5. Ordered probit models of perceived quality for the taxi mode
The two best ordered probit models that we estimated are presented
below. The first one uses the initial quality score qi as the independent
variable and the second uses the final score qfi .
The first model relates the overall perception of quality qi with spe-
cific aspects of quality vik, before the user is provided with any infor-
mation about these attributes. The second model relates the quality score
qfi with specific attributes of quality vik, after the user has been given
information about them.
Normally, a model that adds more explanatory variables is expected
to have a better fit than others. In this case, despite the second model
having fewer variables, it has a better fit if the Akaike Information Cri-
terion (AIC) is considered. Although this result may appear unlikely, it
was to be expected because the way the dependent variable qfi was ob-
tained (see previous paragraph) means it should be more correlated with
the scores of the specific quality attributes vik.
The second model is expected to fit better than the first, given that the
specific quality attributes vik should provide a better explanation for qfi
than for qi. The utility provided by estimating the first model is to see
B. Alonso et al. Transport Policy 63 (2018) 157–164

Table 1 Table 2b
Statistical analysis of the data. % change in ƞi ¼ qfi - qi.

User Characterisation Category % change in the value of ƞi

variable Nº % 1 0 1 2

GENDER Female 146 67.9 Women 12% 80% 6% 1%

Male 69 32.1 Men 16% 77% 6% 1%
AGE OF INTERVIEWEE <¼ 24 9 4.2 Age <35 18% 77% 5% 0%
25–34 35 16.3 Age 35-55 12% 77% 9% 3%
35–44 44 20.5 Age >55 13% 83% 4% 0%
45–54 50 23.3 Driving license YES 16% 75% 6% 2%
55–64 30 14 Driving license NO 8% 86% 5% 0%
>¼ 65 47 21.9 Car owner YES 13% 78% 6% 3%
OWNER OF DRIVING LICENSE YES 141 65.6 Car owner NO 14% 80% 6% 0%
NO 74 34.4 Reduced mobility YES 14% 72% 10% 3%
VEHICLE OWNER YES 116 54 Reduced mobility NO 13% 80% 5% 1%
NO 99 46 Low income 17% 83% 0% 0%
REDUCED MOBILITY YES 29 13.5 Middle income 14% 77% 6% 3%
NO 186 86.5 High income 13% 80% 6% 1%
INCOME <¼ 900 € 6 2.8 Daily use 8% 67% 19% 6%
900 - 1500 € 18 8.4 Weekly use 14% 80% 6% 0%
1500 - 2500 € 59 27.4 Monthly-occasional use 15% 82% 3% 1%
>¼ 2500 € 55 25.6 Reason Home 12% 80% 6% 2%
NS/NC 77 35.8 Reason work-study 25% 71% 4% 0%
OCCUPANCY 1 person 129 60 Reason medical 27% 55% 18% 0%
2 people 64 29.8 Reason shopping leisure, other 10% 84% 5% 1%
3 people 18 8.4 Sunny 15% 77% 6% 2%
4 people 4 1.9 Cloudy 13% 83% 3% 0%
CHARGE FOR LUGGAGE YES 16 7.4 Raining 11% 79% 9% 2%
NO 199 92.6
ORIGIN Call 49 22.8
Taxi rank 166 77.2 Systematic variations in taste were introduced into the model in an
FREQUENCY OF JOURNEY Daily 36 16.7
attempt to explain part of the random variations by considering in-
Weekly 35 16.3
Monthly 24 11.2 teractions with certain random parameters. In these cases, we also found
Occasional 120 55.8 negative signs, which can be immediately interpreted if we consider the
DESTINATION Home 98 45.6 differential effect caused by the non-interacting parameter (please see
Study 1 0.5
‘Heterogeneity in the means’ in Table 3).
Work 23 10.7
Health 11 5.1
In the case of the first model, the interactions of the systematic var-
Leisure 53 24.7 iations in taste of the fare with gender (vFare;gender) and people with
Shopping 8 3.7 reduced mobility (vFare;Reduced Mobility) were significant (both negative),
Other 21 9.8 showing that the weight of the fare variable decreases in the final score
for the categories of women and people with reduced mobility who are
more critical of the fares. This is because the interaction reduces the
which variables the users concentrate more on without having previous weight of the parameter associated with the fare (θFare ¼ 1.762) by 1.000
knowledge about vik. (θFare;gender) in the case of women and by 1.027 (θFare;Reduced Mobility) in the
Their impact on the overall service score is very important from a case of people with reduced mobility, indicating that these user types
marketing point of view as it helps us to understand which attributes are place less weight on the fare than men and the remaining people who do
most valued by users before and after they receive any kind of advertising not have reduced mobility.
or information leaflets promoting the use of public transport. The interaction of waiting time with the possibility of taking luggage
Table 3 shows that for the first model, the fare is significant along (vWaiting time;luggage) was significant. The same occurs with the interaction of
with the location of the rank, method of payment, safety, journey time waiting time in the case of rain (vWaiting time;rain), where the reason for
and cleanliness. The parameters of the last two attributes had a vari- journey is medical (vWaiting time;reason for journey medical) and with gender
ability, which can be explained using a normal probability distribution. (vWaiting time;sex). A positive sign in all cases indicates that waiting time has a
All the parameters were expected to have a positive sign because even higher weighting because of the uncomfortable situation (having luggage
though the scores for quality could be low, they were never represented and rain) and haste (reason for journey medical) and being female.
by negative values, and they use the same scale as the dependent vari- A similarly significant and positive interaction was vehicle comfort
able, making it easier to read the results of the models in terms of which with driving license (vComfort in vehicle;driving license) and driver behaviour
attributes contribute more or less to the overall score for service quality. when the taxi is hired at a rank rather than through a call to radiotaxi
In this case, it was not necessary to study the marginal effects of the (vDriver behaviour;waiting time at stop).
variables. Relative to the heterogeneity in the means of the random parameters,
the interactions with journey time scores for users aged between 35 and
44 years (vJourney time;Age35-44), in the case of rain (vJourney time;rain), for other
Table 2a
reasons for journey (vJourney time;Other reasons) and waiting time at stop
Frequencies of replies to initial (qi) and final (qfi ) scores.
(vJourney time;Waiting time at stop) were significant and negative. This is because
qfi Total users expect longer journey times when it is raining and therefore do not
Very bad/bad regular good Very good place such importance on this variable. The same thing happens with trip
qi Very bad/bad 2 2 1 0 5 reasons other than work or studies and when the taxi is hired at a rank
Regular 1 34 7 2 44 because more waiting time is assumed than when the taxi is called for by
Good 0 13 117 4 134 phone.
Very good 0 0 15 17 32 The specific score that had the most effect on the overall score for
Total 3 49 140 23 215
quality was the journey time followed by safety and waiting time. The

161
B. Alonso et al. Transport Policy 63 (2018) 157–164

Fig. 2. Frequency diagram by category of user.

162
B. Alonso et al. Transport Policy 63 (2018) 157–164

Table 3 Table 4
Ordered Probit model for qi. Ordered Probit Model for qfi .

Ordered Probit Model for qi Ordered Probit Model for qfi

Variable (vk) Coefficient (θk) Standard Error (SE) t-Test Variable (vk) Coefficient (θk) Standard Error (SE) t-Test

NonRandom Parameter NonRandom Parameter

constant 9.495 2.049 4.63 constant 16.484 4.436 3.72
vFare 1.762 0.484 3.64 vJourney time 2.454 0.814 3.01
vLocation 1.231 0.468 2.63 vWaiting time 2.227 0.644 3.46
vPayment method 1.524 0.526 2.90 vFare 1.158 0.486 2.38
vsafety 2.766 0.741 3.73 vSignage 4.518 1.285 3.52
Heterogeneity in the means vlocation 1.636 0.634 2.58
vFare;gender 1.000 0.377 2.65 vpayment method 2.299 0.794 2.90
vFare;Reduced Mobility 1.027 0.427 2.40 vAccessibility 1.900 0.671 2.83
vDriver behaviour;waiting time at stop 0.878 0.405 2.17 Heterogeneity in the means
vWaiting time;luggage 2.676 1.111 2.41 vWaiting time;Frequency daily 3.327 0.999 3.33
vWaiting time;rain 2.217 0.690 3.21 vWaiting time;E25-34 2.208 0.731 3.02
vWaiting time;reason for journey medical 2.914 1.107 2.63 vJourney time;Frequency occasional 0.587 0.388 1.51
vWaiting time;sex 1.122 0.315 3.57 vAccessibility;E24 2.233 1.091 2.05
vComfort in vehicle;driving license 0.799 0.266 3.01 vPayment method;reason leisure 0.874 0.440 1.99
Random Parameter vAccessibility;Reason work 1.240 0.627 1.98
vJourney time 3.288 0.959 3.43 vDriver behaviour;posession of license 0.931 0.428 2.17
vCleansliness 1.347 0.493 2.73 Random Parameter
Scale Parameters for dists. of random parameters vVehicle comfort 3.405 0.976 3.49
vJourney time 1.204 0.233 5.16 Scale Parameters for dists. of random parameters
vCleansliness 0.572 0.146 3.91 vVehicle comfort 1.814 0.450 4.03
Heterogeneity in the means of random parameter Heterogeneity in the means of random parameter
vJourney time;Age35-44 1.278 0.368 3.47 vVehicle comfort:occupancy>2 0.557 0.366 1.52
vJourney time;rain 2.712 0.660 4.11 Threshold parameters for probabilities
vJourney time;Other reasons 1.363 0.493 2.76 μ1 14.140 3.630 3.90
vJourney time;Waiting time at stop 2.403 0.867 2.77 μ1 31.586 7.684 4.11
Threshold parameters for probabilities
Log Likelihood 53.862
μ1 6.496 1.356 4.79
AIC 147.7
μ2 16.577 2.974 5.57
AIC/N 0.687
Log Likelihood 70.547
Number of observations 215
AIC 187.1
AIC/N 0.870
Number of observations 215

variables with the least weight, but still significant, were vehicle comfort
and driver behaviour.
In the case of the second model where qfi was used as the dependent
variable, the results were different, as was expected (see Table 4).
The specific attribute with the most influence on the final overall
score was waiting time for frequent users, followed by signage for the
rank, vehicle comfort and journey time. Fare is one of the attributes that
had the lowest influence on the final score.
The inclusion of the parameter for the vehicle comfort score as a
random variable and its interaction with the level of occupancy (vVehicle
comfort:occupancy>2 more than 2 passengers in the taxi) was significant in this
case with a negative sign because it reduces the variable's weight on the
overall score. Users travelling together believe that sharing a vehicle is a
disutility, and they are prepared for it.
The interactions of the score for waiting time with whether a pas-
senger is a frequent user or not (vWaiting time;Frequency daily) were significant,
increasing the weight those users give to waiting time in the final score.
Waiting time interacting with the age variable 25–34 (vWaiting time;E25-34)
always significantly reduced its impact on the overall score.
The score given to journey time increased its weight when it inter-
acted with occasional users (vJourney time;Frequency occasional). These users
appear to be more demanding about this attribute.
Accessibility or ease of access to the system interacted with users less
than 24 years old (vAccessibility;E24) and those travelling for reasons of work
(vAccessibility;Reason work). For those less than 24 years old, this attribute was
very important, whereas, strangely, it was not so for people using the taxi
for work, which could be because they have an in-depth knowledge about
what to expect from the service that they tend to place greater impor-
tance on other attributes.
The payment method interacted with users travelling for reasons of
leisure (vPayment method;reason leisure), in this case, reducing the weight of this
variable and, as in the previous case, the behaviour of the driver had a
positive interaction with the users who possess a driving licence.
6. Conclusions
The success of this mode of public transport demands a thorough
analysis of its characteristics and what the users think of them. This
article proposes a methodology for modelling the perceived quality of a
taxi service to identify the influential variables and their relevance within
the overall score given to the service.
It is important to point out the changes in opinion for perceived
quality given by service users once they have reflected on the different
influential variables (models estimated with qi versus models estimated
with qfi ).
The results allow different marketing policies to be set in motion
depending on each case. The first model can be used to determine the
most important variables to be improved when the user does not have
enough information about the system, and the second model provides
information to evaluate the effect of advertising campaigns and policies
aimed at encouraging taxi use.
The article confirms that users have a significantly different percep-
tion about the quality of service, which depends on both the journey
constraints such as the method of payment, the way of accessing the
service and journey time and socioeconomic characteristics such as
possession of a driving licence, reason for journey or age.
With regard to systematic variations in perceptions, the greatest
impact on perceived quality would be generated by improving the
waiting time for daily users (model 2). An analysis of all the results
concludes that the factors that require improvement that are most
influential on the final score are safety, comfort, waiting time and
journey time. This suggests the importance of working towards a more
even and coherent distribution of the taxi ranks and on calculating the
optimal number of taxi licenses that are allowed, which significantly
affects this variable. The use of new more comfortable vehicles equipped
with improved safety features also needs to be encouraged as well as the
provision of reserved lanes for taxis or taxis and buses to improve journey
times.

163
B. Alonso et al.
Although tangible factors such as fare have some bearing on the
quality perceived by users of other modes of transport, in the case of the
taxi, it is not among the most influential variables that affect the overall
system quality.
The model in Table 3 shows that the most important variable for taxi
users is journey time. This means that taxi companies have to guarantee
acceptable journey times and, wherever possible, try to obtain exclusive
lanes that would allow for faster journeys. However, they also have to
guarantee a safe service by correctly choosing their drivers as safety is a
very important variable for users. Waiting time after making a call is also
important, especially when the reason for journey was medical. User
expectations are considerably lowered when it is raining, especially in
terms of journey time; therefore, it is important to guarantee adequate
quality standards without the need to reduce journey time too much.
The models in Table 4 show that if the user is made to reflect on other
factors, then the hierarchy changes substantially with respect to model 3.
Journey time continues to be important, but other variables appear to
attract the users' attention to a greater degree. These variables are the
signage of the taxi ranks and comfort. This shows the importance of
improving signage throughout the city showing where the taxi ranks are
located, possibly even locating maps around the city and using more
comfortable vehicles. As expected, the more frequent users are prepared
to wait for shorter times than the other users.
The authors consider that for marketing policies, the recommenda-
tions drawn from the model considering qi as a dependent variable are
important and should be introduced to guarantee a better quality service.
They are the variables that the users consider important without having a
profound knowledge of the system.
The recommendations relative to the model considering qfi as a
dependent variable are those that could have an unexpected impact in
favour of the operators, given that the users realise about them only after
they are suggested. It is clear that a combination of improvements drawn
from applying both models would help in providing the best possible
service to the users.
The methodology presented in this article provides valuable results to
further our knowledge about the user perception of an existing service
and the aspects they consider to be most important. The application of
this methodology to any taxi service would provide guidelines for service
improvements and their effects on the overall perceived quality.
Acknowledgements
This study was supported by research funding from MINECO through
the projects TRA2015-69903-R (co-financed with FEDER funds) and
TRA2013-48116-R. The authors would also like to thank the City Council
of Santander, which has allowed this research to take place, and the
anonymous reviewers for their useful comments.
References
Abe, M.A., Brush, B.C., 1976. On the regulation of price and service quality: the taxicab
problem. Q. Rev. Econ. Bus. 16 (autumn), 105–111.
Barabino, B., Di Francesco, M., 2016. Characterizing, measuring, and managing transit
service quality. J. Adv. Transport. 50, 818–840.
Behara, R.S., Fisher, W.W., Lemmink, J.G.A.M., 2002. Modelling and evaluating service
quality measurements using neural networks. Int. J. Oper. Prod. Manag. 22 (10),
1162–1185.
Bordagaray, M., dell'Olio, L., Ibeas, A., Cecín, P., 2014. Modelling user perception of bus
transit quality considering user and service heterogeneity. Transportmetrica:
Transport. Sci. 10 (8), 705–721.
Bordagaray, M., dell'Olio, L., Ibeas, A., Barreda, R., Alonso, B., 2015. Modelling the
service quality of public bicycle schemes considering user heterogeneity. Int. J.
Sustain. Transport. 9 (8), 580–591.
Cronin, J.J., Taylor, S.T., 1992. Measuring service quality: a reexamination and extension.
J. Market. 56 (July), 55–58.
Dachyar, M., Rusydina, A., 2015. Measuring customer satisfaction and its relationship
towards taxi's service quality around Capital City Jakarta. Int. J. Eng. Technol. IJET-
IJENS 16 (1), 24–27.
164
Transport Policy 63 (2018) 157–164
de O~ na, J., de O~na, R., 2014. Quality of service in public transport based on customer
satisfaction surveys: a review and assessment of methodological approaches.
Transport. Sci. 49 (3), 605–622.
de O~ na, R., de O~na, J., 2015. Analysis of transit quality of service through segmentation
and classification tree techniques. Transportmetrica: Transport. Sci. 11(5) (23),
365–387.
De O~ na, J., de O~na, R., Eboli, L., Mazzulla, G., 2013. Perceived service quality in bus
transit service: a structural equation approach. Transport Pol. 29, 219–226.
De Vany, A.S., 1975. Capacity utilization under alternative regulatory constraints: an
analysis of taxi markets. J. Polit. Econ. 83 (1), 83–94.
dell'Olio, L., Ibeas, A., Cecín, P., 2010. Modelling user perception of bus transit quality.
Transport Pol. 17 (6), 388–397.
dell'Olio, L., Ibeas, A., Cecín, P., 2011. The quality of service desired by public transport
users. Transport Pol. 18 (1), 217–227.
Douglas, G.W., 1972. Price regulation and optimal service standards: the taxicab industry.
J. Transport Econ. Pol. 20, 116–127.
Eboli, L., Mazzulla, G., 2008. A stated preference experiment for measuring service
quality in public transport. Transport. Plann. Technol. 31 (5), 509–523.
Eboli, L., Mazzulla, G., 2011. A methodology for evaluating transit service quality based
on subjective and objective measures from the passenger's point of view. Transport
Pol. 18 (1), 172–181.
Eboli, L., Mazzulla, G., 2014. Investigating the heterogeneity of bus users' preferences
through discrete choice modelling. Transport. Plann. Technol. 37 (8), 695–710.
 2017. Modelling perceived quality for urban public
Echaniz, E., dell'Olio, L., Ibeas, A.,
transport systems using weighted variables and random parameters. Transport Pol.
1–9 in press. https://doi.org/10.1016/j.tranpol.2017.05.006.
Foerster, J.F., Gorman, G., 1979. Taxicab deregulation: economic consequences and
regulatory choices. Transportation 8 (4), 371–387.
Frankena, Mark. W., Pautler, Paul A., 1984. An Economic Analysis of Taxicab Regulation.
Federal Trade Commission, Washington, D.C.
Garrido, C., De O~ na, R., De O~na, J., 2014. Neural networks for analyzing service quality in
public transportation. Expert Syst. Appl. 41 (15), 6830–6838.
Greene, W.H., Hensher, D.A., 2010. Modeling Ordered Choices: a Primer. Cambridge
University Press, Cambridge.
He, X., Zhu, H.-Q., Gao, C., 2009. Risk evaluation of real estate project based on entropy
weight model and TOPSIS method. Commerc. Res. 105 (3), 105–107.
Hensher, D.A., Prioni, P., 2002. A service quality index for area-wide contract
performance assessment. J. Transport Econ. Pol. 36 (1), 93–113.
Hwang, C.L., Yoon, K., 1981. Multiple Attribute Decision Making: Methods and
Applications. Springer-Verlag, New York.
Ibeas, A., Dell'Olio, L., Montequín, R.B., 2011. Citizen involvement in promoting
sustainable mobility. J. Transport Geogr. 19 (4), 475–487.
Kim, H., Yang, I., Choi, K., 2011. An agent-based simulation model for analyzing the
impact of asymmetric passenger demand on taxi service. KSCE J. Civ. Eng. 15 (1),
187–195.
Li, Q., Song, W., 2011. Service quality evaluation model for taxi industry based on
entropy weight and TOPSIS method. Paper presented at the Proceedings -
International Conference on Machine Learning and Cybernetics, 3, 1043–1047.
McKelvey, R., Zavoina, W., 1975. A statistical model for the analysis of ordinal level
dependent variables. J. Math. Sociol. 4 (1), 103–120.
Moore, A., Balaker, T., 2006. Do economists reach a conclusion on taxi deregulation?
Econ. J. Watch 3 (1), 109–132.
Parasuraman, A., Zeithaml, V.A., Berry, L.L., 1988. SERVQUAL: a multiple-item scale for
measuring customer perceptions of service quality. J. Retailing 64 (1), 12–40.
Paratransit Services, 1983. The experiences of U.S. Cities with Taxicab Open Entry.
International Taxicab Association, Chapel Hill, NC.
Shaaban, K., Kim, I., 2016. Assessment of the taxi service in Doha. Transport. Res. Pol.
Pract. 88, 223–235.
Stevens, P., Knutson, B., Patton, M., 1995. Dineserv: a tool for measuring service quality
in restaurants. Cornell Hotel and Restaurant Administration Quarterly, April, 36 (2),
56–60.
Tam, M.L., Lam, W.H.K., Lo, H.P., 2008. Modeling air passenger travel behaviour on
airport ground access mode choices. Transportmetrica 4 (2), 135–153.
Tam, M.L., Lam, W.H.K., Lo, H.P., 2010. Incorporating passenger perceived service
quality in airport ground access mode choice model. Transportmetrica 6 (1), 3–17.
Tan, Q., Oriade, A., Fallon, P., 2014. Service quality and customer satisfaction in Chinese
fast food sector: a proposal for CFFRSERV. Adv. Hospit. Tourism Res. 2 (1), 30–53.
Techarattanased, N., 2015. Service quality and consumer behavior on metered taxi
services. Int. J. Soc. Behav. Educ. Econ. Bus. Ind. Eng. 9 (12), 4235–4239.
Wallsten, S., 2015. The Competitive Effects of the Sharing Economy: How Is Uber
Changing Taxis? Technology Policy Institute.
Wong, K.I., Wong, S.C., Yang, H., 2001. Modeling urban taxi services in congested road
networks with elastic demand. Transport. Res. Part B 35 (9), 819–842.
Wong, K.I., Wong, S.C., Yang, H., Wu, J.H., 2008. Modeling urban taxi services with
multiple user classes and vehicle modes. Transport. Res. Part B 42 (10), 985–1007.
Yang, H., Wong, S.C., 1998. A network model of urban taxi services. Transport. Res. Part
B 32 (4), 235–246.
Yang, H., Yang, T., 2011. Equilibrium properties of taxi markets with search frictions.
Transport. Res. Part B 45 (4), 696–713.
Yang, H., Lau, Y.W., Wong, S.C., Lo, H.K., 2000. A macroscopic taxi model for passenger
demand, taxi utilization and level of services. Transportation 27 (3), 317–340.
Yao, Z.G., Ding, X.D., 2011. Measuring Passenger's perceptions of taxi service quality with
weighted SERPERF. Appl. Mech. Mater. 97–98, 1181–1184.