Transport Policy 46 (2016) 20–31

Contents lists available at ScienceDirect

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

Passenger satisfaction evaluation model for Urban rail transit: A

structural equation modeling based on partial least squares
Weiwei Shen a, Weizhou Xiao a,n, Xin Wang b
a
School of Urban Rail Transportation, Soochow University, Suzhou 215137, Jiangsu Province, China
b
University of Tennessee, Knoxville, TN 37996, USA

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

Article history: The rail transit has played an important role in economic vitality of the urban area. Providing services
Received 19 March 2015 with high levels of quality is essential in order to promote public transportation by customizing the users
Received in revised form of the services, and to reduce traffic congestion by shifting people away from private car use. For this
10 September 2015
reason, it is essential to understand passenger satisfaction with urban rail transit from a quantitative and
Accepted 28 October 2015
systematic perspective. This paper borrows the fundamental concept of the American Customer Sa-
tisfaction Index (ACSI) model to establish a passenger satisfaction evaluation model for urban rail transit
Keywords: in China. A structural equation modeling (SEM) method and its parameter estimation method: Partial
Rail transit Least Squares (PLS), are applied to estimate the proposed model. An evaluation indicator system in-
Passenger satisfaction
cluding three levels of indicators is established to measure passengers’ satisfaction on the services of-
Evaluation indicator system
fered by the rail transit operation companies. The satisfaction index is obtained to quantize the degree of
Partial least square method
Structural equation modeling passenger satisfaction. The IPA matrix is used as an assist tool to show the advantages and disadvantages
of the services of rail transit. Suzhou rail transit line 1 was used as a case study, four models with
different latent constructs or estimation methods were built and compared, to demonstrate the proposed
PSI model based on PLS estimation method was reliable and the sign and magnitude of parameters were
reasonable. The causality between passenger satisfaction and its influence factors were confirmed by
path coefficients of the model.
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction
The expanding economy has transformed and reshaped the
urban landscapes of many Chinese cities, while it also challenges
the urban transport systems. As an emerging tier-2 city, Suzhou
(50 miles west of Shanghai), has its population increased from 3.37
million in 2004 to roughly 5.1 million in 2014 (this is 50% increase
within ten years). Meanwhile the deterioration of urban trans-
portation and air quality are following the expanding economy.
The share of private automobile in transportation mode has in-
creased from 6% to 23% in past decade, which contributes to severe
confliction between motorized and non-motorized transportation,
transportation social equity issues as well as heavy air pollution.
To deal with these problems, the rail transit has been adopted by
Suzhou to provide reliable, efficient, spatial-economic, and en-
vironmental friendly service to urban passengers. The first rail
transit line in city of Suzhou began operation in 2012 and the
n The paper is structured in six sections. After a literature review
Corresponding author.
E-mail addresses: 20134246017@stu.suda.edu.cn (W. Shen),
xiaoweizhou@suda.edu.cn (W. Xiao), wxin_chn@hotmail.com (X. Wang).
second line began operation in Dec 2013, with other five lines
planned to be completed before 2020. The ridership of rail transit
line 1 has doubled within two years with near 1.5 million pas-
sengers per week.
As suggested by many literatures, the success of a public
transport system depends on the number of passengers which the
system is able to attract and retain (de Oña et al., 2013). Therefore,
it is critical for transit management agencies and rail transit op-
eration companies to assess how they are meeting the needs of
their customers by investigating whether rail transit passengers
are satisfied with the products and services the rail transit in-
dustry offered. To that end, a strategic tool is needed for assessing
the current passenger satisfaction level and identifying the man-
agement strategies which can be potentially used to improve
passenger satisfaction, match passenger desires and promote the
use of rail transit system. Also, a standard passenger satisfaction
evaluation model can be a handy tool for assessing and comparing
the rail transit system performance against peer or nationwide.
in Section 2, Section 3 introduces the methodology of PSI model.
Section 4 takes Suzhou rail transit as a case study, introduces the

http://dx.doi.org/10.1016/j.tranpol.2015.10.006
0967-070X/& 2015 Elsevier Ltd. All rights reserved.
 W. Shen et al. / Transport Policy 46 (2016) 20–31
Fig. 1. The Conceptual framework of the PSI model for Urban rail transit.
questionnaire design and passenger survey, and then demon-
strates the feasibility of proposed PSI model. Section 5 gives sev-
eral limitations in this study, and finally Section 6 presents some
conclusions about the study.
2. Literature review
The customer satisfaction theories and methodologies have
been utilized in private industry. The Swedes proposed the theory
of satisfaction called customer satisfaction at the earliest, later
Fornell summarized Customer Satisfaction Index (CSI) and came
up with the American customer satisfaction index (ACSI) model
(Fornell et al., 1996). The ACSI model measured the causality be-
tween the antecedents of customer satisfaction (customer ex-
pectation, perceived service quality, and perceived value) and its
consequences (customer complaints and customer loyalty).
Moreover, studies have point out that CSI can be used to predict
profitability and market value of a company (Anderson et al., 1994,
1997; Eklof et al., 1999). Currently based on the ACSI model, lots of
countries in the world (e.g. Germany, South Korea, China and other
countries in Europe) have established CSI models customized for
their contextual situation.
In the field of transportation, researchers are increasingly re-
cognizing the importance of passenger satisfaction. Hensher
(1990) considered the key of measuring bus service quality was to
identify the important dimensions of service quality that passen-
ger perceived. By studying the preferences of bus passengers, they
put forward a bus preference model. Later in 2003, Hensher et al.
(2003) established an in-depth methodology to quantify each
service quality measurement and identify their relative im-
portance in the overall satisfaction calculation using stated pre-
ference approach. Friman and Gärling (2001) proposed an eva-
luation model by studying public transportation passenger sa-
tisfaction and found that overall satisfaction was positively corre-
lated with accumulative satisfaction. Stradling et al. (2007) pre-
sented a six-step method for measuring satisfaction to combine
measures of performance and importance. Friman et al. (2013)
used confirmatory factor analyses to examine the psychometric
properties of the satisfaction with travel scale. Celik et al. (2014)
proposed a novel framework which includes SERVQUAL, statistical
analysis, interval type-2 fuzzy sets and Vlsekriterijumska
21
Optimizacija I Kompromisno Resenje (VIKOR) methods to evaluate
the customer satisfaction for the rail transit network of Istanbul.
Various theories and methodologies have been applied in sa-
tisfaction research, including hierarchical process, fuzzy mathe-
matics and factor analysis etc. These methods (e.g. Delphi, analytic
hierarchy process and fuzzy clustering method) focus on using
subjective methods for determining the weights of variables for
assessment. At the same time, studies focusing on analyzing the
relationship between satisfaction, or service quality, and different
attributes are more and more popular by using SEM models or
path analysis. Some studies (Stuart et al., 2000; Karlaftis et al.,
2001; Eboli and Mazzulla, 2007; Chen, 2008; Ngatia et al., 2010;
Irfan et al., 2011; de Oña et al., 2013) have estimated SEM models
which can represent the causality between variables. However,
they mainly focused on the service attributes impacting overall
passenger satisfaction but neglected the degree of satisfaction.
Moreover for the parameter estimation in SEM model, traditional
method is linear structural relationships (LISREL), which assumes
that all observations are independent and the manifest variables
obey the multivariate normal distribution, but this is not usually
the case in passenger satisfaction survey (Chin and Newsted, 1999;
Lin et al., 2005). Alternative method is Partial Least Squares (PLS),
which relaxes the assumption of normal distribution and can ob-
tain explicitly estimated latent variable scores directly in the
process of parameter estimation, therefore PLS is more suitable for
satisfaction research. In this study, SEM method based on PLS es-
timation is used to build the passenger satisfaction index (PSI)
model for urban rail transit, and the causality between passenger
satisfaction and its influence factors (e.g. passenger expectation,
service quality and service value) is addressed.
Prior studies also explored the relationships between custo-
mers' expectation, perceived quality, perceived value, satisfaction,
complaints and loyalty (Fornell et al., 1996; Zeithaml et al., 1996;
Cronin et al., 2000; Kuo et al., 2009). Expectation, perceived
quality, and perceived value are assumed to be three antecedents
of satisfaction, and complaints and loyalty are two behavioral in-
tentions or consequences of satisfaction.
Expectation represents both prior consumption experience
with the company’s offering and a forecast of the company’s
ability to provide quality in the future, and it is assumed to have a
direct and positive association with the perceived performance,
such as service quality, perceived value, and customer satisfaction
22 W. Shen et al. / Transport Policy 46 (2016) 20–31

(Fornell et al., 1996). Perceived quality is the service quality that
customer perceived, while the concept of service quality is defined
as a comparison between expectation and actual service perfor-
mance (Parasuraman et al., 1985). Lovelock (2000) presented that
perceived value comes from a trade-off between perceived bene-
fits and perceived costs. Cronin et al. (2000) confirmed that service
quality positively affects perceived value and perceived value is
positively related to satisfaction.
Satisfaction is an overall affective response to a perceived dis-
crepancy between prior expectations and perceived performance
after consumption (Oliver, 1980). It refers to the feeling of pleasure
or disappointment formed by comparison between service quality
and expectation. Wen et al. (2005), Chou and Kim (2009), and
Chou et al. (2011) proved that satisfaction is positively related to
loyalty. Apparently there is a negative relationship between sa-
tisfaction and complaints, and the complaints decrease the loyalty.
Complaints and loyalty are two different behavioral intentions.
Complaints are generally considered to comprise a set of responses
to purchase dissatisfaction (Singh, 1988). Loyalty implies repeat
purchasing based on cognitive, evaluative and dispositional factors
that are the classic primary components of attitude (Jacoby, 1971).
In the field of transportation, few studies talk about the com-
plaints and loyalty, because they are difficult to measure. Thus
most researchers use the behavioral intentions instead (Chen,
2008; Chen and Chen, 2010; Kuo and Jou, 2014). In this paper,
several observable variables are set to measure the loyalty, and the
complaint rate is adopted to measure the complaints.
3.2. Model solution
3.2.1. SEM methodology
To obtain the latent variables scores, the weights and the
causality between variables, the SEM method based on PLS esti-
mation is used to solve the PSI model for urban rail transit.
As a powerful multivariate technique mixing factor analysis
and path analysis together, SEM has advantage in the quantitative
study of interactive relationships between variables. SEM consists
of two components, a measurement model describing the re-
lationships between latent and manifest variables, and a structural
model describing the causal relationships between endogenous
and exogenous latent variables.
The basic equation of the structural model is defined as (Bollen,
1989):
η = βη + Γξ + ζ (1)
where η is an m × 1 vector of endogenous latent variables, ξ is an
n × 1 vector of exogenous latent variables, β is an m × m matrix of
path coefficients associated with η , Γ is an m × n matrix of path
coefficients associated with ξ and η , and ζ is an m × 1 residual
vector of the equation.
The basic equations of the measurement model are the fol-
lowing:
x = Λx ξ + δ (2)

3. Methodology-PSI model y = Λy η + ε (3)

3.1. Conceptual framework of the PSI model where x is an q × 1 vector of exogenous manifest variables, Λx is an
q × n factor loading matrix for the effects of the exogenous man-
To evaluate the rail transit passenger satisfaction of a city, this ifest variables on exogenous latent variables, and δ is an q × 1
paper establishes a PSI model for urban rail transit. Fig. 1 shows vector of measuring error; Correspondingly y is an p × 1 vector of
the conceptual framework. Based on the literature, several latent
endogenous manifest variables, Λy is an p × m factor loading
variables were identified as critical variables of satisfaction eva-
matrix for the effects of endogenous manifest variables on en-
luation, including passenger expectation, perceived quality, per-
dogenous latent variables, and ε is an p × 1 vector of measuring
ceived value, satisfaction, complaints and loyalty. Some variable is
error.
exogenous, e.g. passenger expectation, which means it is not sys-
Assumed there are six latent variables is PSI model, as shown in
tematically affected by changes of other variables. The other
variables are endogenous latent variables since they can be im- Fig. 1. According to the path relationships of PSI model, the
pacted by other latent variables. Since the latent variables are ty- equations of structural model and measurement model are spe-
pically hard to measure directly, manifest variables are used to cified, that is formulas (1)–(3) are correspondingly written as
describe and quantify the latent variables. Accordingly the variable formulas (4)–(6):
used to describe exogenous latent variable is named exogenous
manifest variable and the variables used to describe endogenous ⎡ Q ⎤ ⎡⎢ 0 0 0 0 0⎤ ⎡ Q ⎤ ⎡ γ ⎤
⎥ 11
⎡ ζ1 ⎤
⎢ ⎥ ⎢ β21 0 0 0 0⎥ ⎢ ⎥ ⎢ γ ⎥ ⎢ ⎥
latent variables are endogenous manifest variables. ⎢ ζ2 ⎥
⎢V ⎥ ⎢ ⎢ V ⎥ ⎢ 21⎥
The variables in circles are latent variables, among which the ⎢ S ⎥ = ⎢ 31 32
β β 0 0 0⎥ ⎢ S ⎥ + ⎢ γ31⎥ E + ⎢ ζ3 ⎥
⎢C ⎥ ⎢ 0 ⎥ ⎢ ⎥
variable in dashed circle is exogenous and the variables in solid 0 β43 0 0⎥ ⎢ C ⎥ ⎢ 0 ⎥ ⎢ ζ4 ⎥
circles are endogenous variables. The arrows with þ/  signs re- ⎢ ⎥ ⎢ ⎢ ⎥ ⎢
⎥⎣ H ⎦ ⎣ 0 ⎦ ⎥
⎣ H⎦ ⎣ 0 0 β53 β54 0⎦ ⎢⎣ ζ5 ⎥⎦
present the path relationships between latent variables, the þ (4)
represents the positive correlation and the-represents the negative
where E , Q , V , S , C , H respectively denote passenger expectation,
correlation. The variables in solid line box are the manifest vari-
perceived quality, perceived value, passenger satisfaction, pas-
ables which measure the latent variables. Note that, subordinate
senger complaints and passenger loyalty. βij and γij are the path
observed variables (the apostrophes in Fig. 1) are used to describe
the nine service dimensions of perceived quality furthermore, but coefficients, and ζi is the residual vector of the equation.
they are not shown in Fig. 1 due to limited space. A complete list of ⎡ δ E1 ⎤
⎡ E1 ⎤ ⎡ α E1 ⎤
these variables is presented in Table 4. The arrows connecting ⎢ ⎥ ⎢ ⎥ ⎢ ⎥
α
⎢ E2 ⎥ = ⎢ E2 ⎥ E + ⎢ δ E2 ⎥
circles and boxes represent the loading relationships between la-
⎢⎣ E3 ⎥⎦ ⎣ α E3 ⎦ ⎢⎣ δ E ⎥⎦
tent variable and their corresponding manifest variables. 3 (5)
Fig. 1 is the conceptual framework of the PSI model. Several
models with different latent constructs will be compared in the where Ei is the i-th manifest variable of E , α Ei is the corresponding
following to find the most suitable model which can fit the survey coefficient which is called loading coefficient, and δ Ei is the mea-
data. suring error.
 W. Shen et al. / Transport Policy 46 (2016) 20–31 23

Table 1
Differences between PLS and LISREL (Chin and Newsted, 1999; Efron, 1979; Huo, 2006).

Differences PLS LISREL

Objective Prediction oriented Parameter oriented

Approach Variance based Covariance based
Distributional No distribution requirements Typically multivariate normal distribution and independent
Hypothesis observation
Model Identification No model identification problems The identifiability of proposed model is needed to be checked.
Latent variable scores Explicitly estimated Indeterminate
Hypothesis Testing PLS usually use Stone-Geisser Q2, R-square, AVE, CR, Cronbach’s Alpha, etc to The RMSEA, GFI, chi-squared, df, etc. are the usual fit indices
test the predictive ability of the model. Bootstrap or Jacknife method to test used to determine how well the SEM model fits the sample
the significance of parameters. data.
Minimum sample size Recommendations range from 30-100 cases. Recommendations range from 200–800 cases.

⎡ αQ 0 0 0 0 ⎤ the researchers concern more about the explicitly estimated latent

⎡ Q1 ⎤ ⎢ ⎥
1
variable scores (i.e. satisfaction index in PSI model), while the la-
⎢ ⎥ ⎢ Q2 α 0 0 0 0 ⎥
tent variable scores in LISREL estimation are closely related to the
⎢ Q 2⎥ ⎢ αQ 3 0 0 0 0 ⎥
⎥ ⎡ εQ 1 ⎤
⎢ Q 3⎥ ⎢ ⎢ εQ ⎥ calculation method and lack of comparison, in that case the PLS
⎢ ⎥ ⎢ αQ 4 0 0 0 0 ⎥ ⎢ 2⎥ parameter estimation method is more suitable.
⎢ Q 4⎥ ⎢ ⎥ ⎢ εQ 3 ⎥ The PLS parameter estimation method of SEM obtains the es-
⎢ Q ⎥ ⎢ αQ 5 0 0 0 0 ⎥ ⎢ εQ 4 ⎥
⎢ 5⎥ ⎢ αQ 6 0 0 0 0 ⎥
⎥ ⎢ εQ ⎥ timated values of latent variables by iteration, and the estimated
⎢ Q 6⎥ ⎢ ⎢ 5⎥ value of a latent variable is the weighted sum of its corresponding
⎢ Q ⎥ ⎢ αQ7 0 0 0 0 ⎥ ⎢ εQ 6 ⎥ manifest variables. In the PSI model, this method extracts principal
⎢ 7⎥ ⎢ ⎥ ⎢ εQ 7 ⎥
⎢ Q 8⎥ ⎢ α Q 8 0 0 0 0 ⎥ ⎡ Q ⎤ ⎢ εQ 8 ⎥ component from the subset of manifest variables for diffident la-
⎢ Q ⎥ ⎢ αQ 9 0 0 0 0 ⎥
⎥ ⎢ ⎥ ⎢ε ⎥ tent variables (i.e. passenger expectation, perceived quality, per-
⎢ 9⎥ ⎢ ⎢ V ⎥ ⎢ Q 9⎥ ceived value, passenger satisfaction and passenger loyalty) to es-
⎢ V1 ⎥ = ⎢ 0 α V1 0 0 0 ⎥ ⎢ S ⎥ + ⎢ εV1 ⎥
⎢V ⎥ ⎢ ⎥ ⎢ C ⎥ ⎢ εV2 ⎥ timate the value of latent variables, and then estimates the loading
⎢ 2 ⎥ ⎢ 0 α V2 0 0 0 ⎥ ⎢ ⎥ ⎢ε ⎥ coefficients and path coefficients using ordinary least squares.
⎢ S1 ⎥ ⎢ 0 0 α S1 0 0 ⎥
⎥ ⎣ H ⎦ ⎢ S1 ⎥
Suppose that the passenger satisfaction evaluation contains k
⎢ ⎥ ⎢ ⎢ ε S2 ⎥
⎢ S2 ⎥ ⎢ 0 0 α S2 0 0 ⎥ ⎢ εS ⎥ independent variables named x1, x2, .. . , xk and m dependent
⎢ S3 ⎥ ⎢ ⎥ ⎢ 3⎥ variables named y1, y2 , ... , ym , the sample size is n, from this we
⎢ ⎥ ⎢ 0 0 α S3 0 0 ⎥ ⎢ εC1 ⎥ can respectively obtain the data table X and Y of independent
⎢ C1 ⎥ ⎢ 0 0 0 α C1 0 ⎥
⎥ ⎢ εC2 ⎥
⎢ C2 ⎥ ⎢ ⎢ εH ⎥ variables and dependent variables, among them:
⎢ ⎥ ⎢ 0 0 0 α C2 0 ⎥ ⎢ 1⎥ X = [x1, x2, ... , xk ]n × k and Y = [y1, y2 , ... , ym ]n × m . The PLS regression
⎢ H1 ⎥ ⎢ ⎥ ⎢ ε H2 ⎥ consists of two steps: component extraction and linear regression.
⎢ H2 ⎥ ⎢ 0 0 0 0 α H1 ⎥ ⎢⎣ ε H3 ⎥⎦
Step one: component extraction is a process of improving the
⎢ ⎥ ⎢ 0 0 0 0 α H2 ⎥
⎥
⎣ H3 ⎦ ⎢ accuracy constantly and it won’t stop until achieving satisfactory
⎢⎣ 0 0 0 0 α H3 ⎥⎦ accuracy. In the process, we respectively extract the first compo-
(6)
nent of PLS t1 and u1 from X and Y every time, the two components
where Q i , Vi , Si , Ci and Hi and respectively denote the i-th manifest t1 and u1 respectively include the largest compile information of X
variable of Q , V , S , C and H , α yi is the corresponding coefficient and Y , at the same time, the degree of correlation between t1 and
which is also called loading coefficient, and ε yi is the measuring u1 requires to reach the maximum. Later the second round of
error. component extraction will be proceeded by making use of the rest
information separately explained by t1 and u1 of X and Y and will
3.2.2. Fundamental principle of PLS not stop until achieving satisfactory accuracy. Step two: linear
The conventional estimation method of SEM is LISREL which is regression supposes that finally X extracts a components named
also called covariance modeling method. It estimates the para- t1, t2, .. . , ta , severally establishes regression equations of
meters of SEM by using maximum likelihood, generalized least t1, t2, .. . , ta for yi (i = 1, 2, ... , m), and then translates into the re-
squares or some other methods to construct a model to estimate gression equations of yi for x1, x2, .. . , xk (Li, 2011).
the fitting function of covariance ∑ (θ ) and sample covariance (S )
(Joreskong and Dag, 1993). 3.3. The satisfaction index
The PLS parameter estimation method is a mathematical opti-
mization technique. It searches for a set of best-matching data by The satisfaction index of a latent variable is an index to quantify
minimizing the sum of squares of error. It is an iterative estimation satisfaction level and measure service quality from the perspective
combined with multiple linear regression analysis, canonical cor- of consumers. Generally, the satisfaction index is calculated using
relation analysis, principal component analysis and causal weighted manifest variables (Fornell et al., 1996) and the value
modeling. varies according to the methods used to calculate the index. As the
The main differences between PLS and LISREL are shown in value of latent variable is calculated in the process of parameter
Table 1. It can be found that the two estimation methods have estimation in PLS, the satisfaction index of each latent variable can
their own advantages and application conditions. LISREL focuses be obtained directly from the estimated values of latent variables
on model validation, so the data distribution generally has strict by PLS estimation.
assumptions. While PLS focuses on the explanation of variance, so
it can obtain moderate estimation results even if the data is de- 3.4. Indicator system of passenger satisfaction evaluation
viating from the normal situation. Considering the characteristics
of passenger satisfaction evaluation work, the data from satisfac- The system is divided into three levels of indicators. As shown
tion survey usually do not obey multivariate normal distribution, in Fig. 1, the latent variables (in circles) are the first-level
24 W. Shen et al. / Transport Policy 46 (2016) 20–31
indicators, the nine service dimensions in boxes are the second-
level indicators, and the other variables in boxes are the third-level
indicators. Considering the perceived quality reflects passen-
gers' comprehensive assessment of rail transit service quality and
using nine second-level indicators of perceived quality are not
enough to capture a complete picture of rail transit service quality,
third-level indicators are used to provide more detailed and direct
measurements of each of the nine second-level indicators. The
detailed indicators from different levels can be found in Table 4.
The selection of third-level indicators combines with literature
review and survey of rail transit operation companies. The in-
dicators used to characterize the latent variables in addition to
perceived quality refer to the measurement variables used in the
ACSI model of Fornell (Fornell et al., 1996). While considering the
characteristics of rail transit services, numbers of literatures pre-
sent the attributes and dimensions used to evaluate service
quality.
Eboli and Mazzulla (2011) evaluated the transit service quality
from 11 dimensions with 26 attributes which include route char-
acteristics, service characteristics, service reliability, comfort,
cleanliness, fare, information, safety and security, personnel, cus-
tomer services, and environmental protection. Mouwen and
Rietveld (2013) used 15 service attributes from 3 dimensions (i.e.,
general aspects of transit system, terminals and stops, and ve-
hicles), and finally revealed that service frequency, on-time per-
formance, travel speed, and vehicle tidiness contribute the most to
the effect on satisfaction. Liou et al. (2014) evaluated the bus ser-
vice quality in Taipei city based on 4 dimensions (i.e., driver atti-
tude, equipment, convenience, and reliability) with 12 attributes.
de Oña et al. (2013) considered totally 14 items from 3 aspects
which are service, comfort and personnel. Hassan et al. (2013)
listed the most common attributes of transit service as frequency,
capacity, reliability, price, cleanliness, comfort, security, staff, in-
formation and ticketing system.
Even if the categories can be different in different literatures,
the most common service aspects discussed can be concluded as
reliability, cleanliness, comfort, safety, fare, personnel, information
and so on. Based on literatures, nine service dimensions totally
containing fifty attributes are formed in this paper to characterize
perceived quality.
3.4.1. Direction and guidance
The indicators belonging to this dimension capture the char-
acteristics of rail transit in term of signs of direction and guidance.
As the rail transit has fixed stations, so the clarity of the signs of
direction and guidance is very important. It is divided into three
types of signs (i.e., signs outside the station, signs in the station,
and signs of direction of rail transit).
3.4.2. Cleanliness and comfort
Cleanliness and comfort are importance service aspects directly
influencing passengers' impression on rail transit. It is related to
the stations and compartments. Cleanliness and comfort mainly
capture the environmental comfort including customer feelings
about brightness, indoor air quality, temperature suitability and
environment cleanliness in station and in the compartment.
3.4.3. Speediness and convenience
As a public transport, the speediness of rail transit affects travel
time, and convenience affects the accessibility of rail transit. So
they are also critical attributes affecting peoples' perception on
service quality. The speediness involves passenger assessment for
the transit punctuality, station dwell time, departure interval,
service time of first and last train, while convenience refers to the
convenience of the bus transfer.
3.4.4. Safety and security
Safety is about whether the elevator and escalator, the shielded
gate, and the train can run smoothly and safely, and the frequency
of equipment failure is at a relatively low level. Security mainly
focuses on the order in public space at stations and the boarding/
alighting security.
3.4.5. Ticket service
Ticket is the voucher for taking rail transit, so the ticket service
is a crucial part of operations. It is related to all the service aspects
about ticket system which contains the ticket types, quantity of
ticket vending machines, introduction about buying tickets, fre-
quency of failure of ticket vending machines, and convenience of
changing money.
3.4.6. Equipment and facilities
There are lots of equipment and facilities in transit stations and
compartments, such as ticket vending machines, ticket gate, seats,
and armrests. The location, quantity and material of these equip-
ments may affect passengers' satisfaction on services.
3.4.7. Staff service
The staffs not only contain regular employees of rail transit
operation companies but also include security personnels and
cleaners. They have close contact with passengers, so their ap-
pearance, service attitude and work efficiency of solving problems
can have great influence on passenger satisfaction on services.
3.4.8. Information distribution/disclosure
Information distribution/disclosure is divided into two aspects
of publicity. The first aspect is about the publicity of rail transit
including introducing lines, fare policy, service time of rail transit.
The other aspect is the publicity of taking rail transit with civility
because it is needed to broadcast the passenger regulations.
3.4.9. Convenient facilities for passengers
Apart from the necessary facilities to ensure the normal op-
eration of rail transit, convenient facilities for passengers can sa-
tisfy passengers' diversified needs. The indicators belonging to this
dimension include the volume and updating frequency of TV, the
comfort of advertising layout and images, the attraction of rail
transit newspaper and signal strength of cell phone.
3.5. Importance–performance analysis
Importance–performance analysis (IPA) is the most widely
used quadrant analysis in transportation or other fields (de Oña
and de Oña, 2014). It can identify indicators which have potential
to improve the effectiveness of the service in rail transit industry
and meet the most urgent demands of passengers, so this method
is widely used by transport company managers in the me-
tropolitan transport sector (Figler et al., 2011). IPA is formed
through a two-dimensional space with X-axle denoting satisfac-
tion performance (or satisfaction index) of indicators and Y-axle
denoting the importance (or the weight) of indicators, and the
space is divided into four quadrants with the average satisfaction
index and weight. Characteristic service indicators referring to the
nine service dimensions were assigned to the four quadrants
based on their X and Y values. Then the corresponding improve-
ment strategies can be identified according to the matrix. Fig. 2
shows the IPA matrix.
The four quadrants are labeled as I, II, III and IV, and separately
named as ‘Keep up the good work’, ‘Concentrate here’, ‘Lower
priority’ and ‘Possible overkill’. The indicators located in Quadrant
I have relatively higher importance and satisfaction index, so
maintaining the current status continually is desirable; the
 W. Shen et al. / Transport Policy 46 (2016) 20–31
indicators located in Quadrant II do not have high satisfaction
index but their importance are relatively high, so attentions need
to concentrate here and prompt improvement is required. The
indicators located in Quadrant III do not have too much im-
portance nor high satisfaction index, so these indicators have
lower priority and gradually improve the performance of these
indicators may be beneficial to obtaining higher public attention.
The indicators located in Quadrant IV have higher satisfaction in-
dex but low importance, so they are possible overkill and main-
taining the current status moderately is acceptable.
4. Case study
Suzhou rail transit system is used as a case study in this paper.
A passenger satisfaction survey was conducted annually in Suz-
hou. This study uses the survey conducted in Sept 2013. During
that time, line one was the only running line which connecting
Mudu Station and Zhongnan Jie Station, with a total length of
25.7 km and 24 stations. To evaluate the rail transit passenger
satisfaction in Suzhou, a questionnaire was designed, and the
passenger satisfaction survey was conducted.
4.1. Questionnaire design
The design of the questionnaire is based on the evaluation in-
dicator system and specific service quality characteristics in urban
rail transit. The questionnaire is structured into four major sec-
tions: screening and identify, personal background information,
riding and travel information, satisfaction survey. The first section
is to exclude those undesirable respondents who or whose family
members are related to rail transit or bus companies. The second
section aims to know about respondents' demographical in-
formation such as the gender, level of education, job, income and
so on. The third section is to know about respondents' riding in-
formation such as travel time, trip purpose, ticket category and so
on. The fourth section of the survey is the main body of the
questionnaire and designed to capture the manifest variables
based on evaluation indicator system. In the questionnaire, a 10-
point scale, with “1” denoting strongly dissatisfied and “10” de-
noting strongly satisfied, is used to measure how respondents
agree or disagree with a particular statement. Particularly for the
second-level indicator ‘passenger complaints’, it is measured by
asking respondents whether or not have they had minor com-
plaint or formally filed a complaint about the rail transit, so the
binary variable is used with “1” denoting no and “2” denoting yes.
4.2. Passenger survey
Two types of surveys were conducted, including on-site inter-
cept interview at stations and screening questionnaire handout.
The number of questionnaires sent out at each of the 24 rail transit
stations was proportional to the passenger flow volume at those
stations. Eventually 840 questionnaires containing 813 valid
questionnaires were collected, and the effective rate is 96.8%.
To test whether the collected data obey the multivariate normal
distribution, the one-sample Kolmogorov–Smirnov test was con-
ducted in software SPSS. The test results show that values of
Kolmogorov–Smirnov Z are between 4.202 and 15.190, and bi-
lateral asymptotic significances are all 0.000, which indicate that
the data do not conform to normal distribution and it is not sui-
table to estimate the proposed model by using conventional esti-
mation approaches of SEM.
25
4.3. Model results
4.3.1. Model comparison
To reveal which latent variables affecting passenger satisfaction
through path analysis, different models with different number of
latent variables were specified, and to justify the superiority of the
PLS compared with conventional estimation approaches, two
kinds of estimation methods of SEM (i.e. the PLS and the LISREL)
were used. Thus four models with different constructs or estima-
tion methods were proposed. These models were calibrated se-
parately by using the software AMOS17.0 and the software
SmartPLS2.0 (Ringle et al., 2005).
For comparing the goodness of each model and choose the best
one, a number of fit indices assessing the goodness-of-fit of SEM
models were used. The chi-squared, degree of freedom (DF),
goodness of fit index (GFI), comparative fit index (CFI), root mean
square error of approximation (RMSEA) etc. are the usual fit in-
dices used to determine how well the model fits the sample data.
As chi-squared is extremely sensitive to sample size, the chi-
squared/DF replaced. The recommended value of chi-squared/DF
needs to be lower than 5. The GFI and CFI above 0.9 indicate a
close fit of the model to the data, and the RMSEA of 0.08 or less
indicates a good fit (Hair et al., 2006).
However, the conventional fit indices of SEM such as RMSEA
and chi-squared need the precondition of distribution, so they can
not be used to assess the estimation results of PLS. To assess the
goodness of SEM models estimated by PLS, two types of indices
are given to separately test the predictive ability of measurement
model and structural model. The predictive ability of measure-
ment model can be assessed by loading coefficients, communality,
Average Variance Extracted (AVE) and redundancy, while the
predictive ability of structural model mainly focuses on R-Square
(Jin and Liang, 2005) and bootstrapping technique (Efron, 1979).
Loading coefficients greater than 0.5 can usually warrant reliability
of each manifest variable (Hair et al., 2006). The communality is
required to be greater than 0.5 (Hair et al., 2006). AVE greater than
0.5 means that the latent variable can explain more than 50% total
variances of its manifest variables (Fornell and Larcker, 1981). The
redundancy, used to test the overall predictive ability of the
measurement model and structural model, is required to be less
than 0.25 (Hair et al., 2006). R-Square of endogenous latent vari-
ables is an essential criterion for assessment of the structural
model, the larger R-Square indicating the better goodness-of-fit.
Chin (1998) describes R-Square values of 67%, 33%, and 19% in PLS
estimation as substantial, moderate, and weak respectively. Be-
sides of these indices, the bootstrapping method is nonparametric
test applied to test the validity of parameter estimation. In addi-
tion, The Cronbach's Alpha (CA) and Composite Reliability (CR) are
also used to test the reliability of the model. The CA greater than or
equal to 0.7 indicates the reliability is high, while the CA between
0.4 and 0.7 indicates the reliability is general (Hair et al., 2006).
Composite Reliability is required to be greater than 0.7 (Hair et al.,
2006).
The fit indices of different models used to compare the good-
ness of different models are displayed in Table 2.
In the first model (M1), six latent variables including passenger
expectation, perceived quality, perceived value, passenger sa-
tisfaction, passenger complaints and passenger loyalty were built
(as shown in Fig. 1). Expectation, perceived quality, and perceived
value were assumed to be three antecedents of satisfaction, and
complaints and loyalty were two behavioral intentions or con-
sequences of satisfaction. The estimated results show that R-
square of passenger complaints is 0.028 which means bad ex-
planation of the variance in this latent variable, at the same time,
the AVE of passenger complaints is 0.4293 (o0.5), the CR is 0.1487
(o0.7), the CA is 0.2913 (o0.4), most of the fit indices relating to
26 W. Shen et al. / Transport Policy 46 (2016) 20–31

Fig. 2. Sketch map of IPA matrix.

Table 2 portance of each service dimension can not be obtained and the
Fit Indices of different models. dimensions which mostly influence the perceived quality and
passenger satisfaction can not be identified. In the third model
Fit Indices M1 M2 M3 M4
(M3), nine service dimensions (which include direction and gui-
R-square Perceived quality 0.297 0.297 0.947 1.070 dance, cleanliness and comfort, speediness and convenience, etc.)
Perceived value 0.330 0.330 0.350 0.115 are treated as exogenous latent variables and have direct path
Passenger satisfaction 0.413 0.413 0.419 0.413
relations with perceived quality, the fifty third-level indicators are
Passenger loyalty 0.217 0.209 0.209 0.400
Passenger complaints 0.028 separately treated as manifest variables of nine service dimen-
Chi-squared /DF 8.765 sions, and the overall satisfaction on nine service dimensions are
treated as manifest variables of perceived quality. The structure of
RMSEA 0.098
GFI 0.508 M3 is shown in Fig. 3. According to Table 2, the R-square of per-
CFI 0.699 ceived quality significantly improved and the R-square of pas-
senger satisfaction is the highest among all the models.
The fourth model (M4) has the same structure with M3, but the
different estimation methods of SEM. M3 used the PLS method
passenger complaints do not meet the test criterion. This might and M4 used the LISREL method. The R-square of passenger sa-
due to the low response probability since most respondents did tisfaction and perceived value in M4 are lower than M3, the chi-
not formally file a complaint about rail transit. So in the second squared/DF is 8.765 (4 5), the RMSEA is 0.098 (4 0.08), and the
model (M2), the unreliable construct ‘passenger complaints’ is GFI and CFI are both lower than the criterion value 0.9, which
excluded from M1. indicates that M4 can not fit the sample data very well and the
In M1 and M2, all the fifty third-level indicators of nine services conventional estimation method of SEM is not appropriate in the
are treated as manifest variables of perceived quality, so the im- passenger satisfaction evaluation work.

Passenger
Expectation 0.121
Direction and
guidance 0.085 0.106

Cleanliness
0.140 0.025 Perceived 0.365 Passenger 0.458 Passenger
and comfort
Value Satisfaction Loyalty

Speediness and 0.092 0.528

convenience Perceived
0.079 Quality 0.282
Safety and
security
0.118

Ticket service

0.186 0.105
0.168 0.153
Information Convenient
Equipment distribution/ facilities for
and facilities Staff service
disclosure passengers

Fig. 3. Path coefficients of the PSI model (M3).

 W. Shen et al. / Transport Policy 46 (2016) 20–31 27

Table 3
Overall associations between latent variables.

Latent variables Passenger loyalty Passenger satisfaction Perceived value Perceived quality

Passenger expectation 0.079 0.172 0.119 0.025

Passenger satisfaction 0.458 0 0 0
Perceived value 0.167 0.365 0 0
Perceived quality 0.217 0.475 0.528 0
Direction and guidance 0.018 0.040 0.045 0.085
Cleanliness and comfort 0.030 0.066 0.074 0.140
Speediness and convenience 0.020 0.044 0.049 0.092
Safety and security 0.017 0.038 0.042 0.079
Ticket service 0.026 0.056 0.062 0.118
Equipment and facilities 0.037 0.080 0.089 0.168
Staff service 0.033 0.073 0.081 0.153
Information distribution/disclosure 0.041 0.088 0.098 0.186
Convenient facilities for passengers 0.023 0.050 0.055 0.105

Then, M3 is selected as the most suitable model compared with
other three models and named as PSI model. The structure of M3
is shown in Fig. 3 and the results of M3 are discussed in the
following.
4.3.2. Model results of the selected model M3
For the model specification in SmartPLS2.0, the ‘Path Weighting
Scheme’ was chosen as the Weighting Scheme, and the Data Me-
tric was set as ’Mean 0, Var 1’ to standardize the data. The outer
weight of indicators, path coefficients between latent variables
and loading coefficients between latent variables and their corre-
sponding manifest variables were obtained.
Fig. 3 shows the estimated path coefficients of the model. The
path coefficients present the direct relationships between latent
variables. Table 3 summarizes the relationships including the direct
and the indirect impacts between latent variables. Fig. 3 shows that
the direct impact of perceived quality on perceived value has the
largest magnitude (with a coefficient of 0.528), following by the
direct impact of passenger satisfaction on passenger loyalty (with a
coefficient of 0.458). The largest direct impact on passenger sa-
tisfaction comes from perceived value, following by perceived
quality. Among the nine service dimensions, ‘Information distribu-
tion/disclosure’ has the largest direct influence on perceived quality,
and followed by ‘Equipment and facilities’. ‘Safety and security’ has
the least influence on perceived quality. For the overall influences
shown in Table 3, passenger satisfaction has the greatest positive
influence on passenger loyalty compared with other latent vari-
ables. Moreover, perceived quality has the largest overall influence
on passenger satisfaction which means that passengers’ perception
on service quality has a great positive influence on passenger sa-
tisfaction. In this sense, improving service quality is particular cri-
tical for rail transit providers to satisfy passengers.
Table 4 shows the estimated outer weights and loading coef-
ficients of the PSI model. In the evaluation indicator system, the
outer weight is to reflect the importance of a subordinate indicator
relative to its superior indicator. Higher outer weight of an in-
dicator denotes more importance compared with other indicators
under the same superior indicator. In PLS path modeling, the
loading coefficients are used to evaluate the reliability of each
manifest variable. According to the test criterion introduced above,
the loading coefficient usually needs to be greater than 0.5. It can
be found that all the loading coefficients are above 0.6, which
indicates the reliability of the PSI model.
Table 5 shows the fit indices accessing the validity of the model
and all the fit indices conform to the requirements of the test
criterion.
Table 6 shows the bootstrapping (with 600 sub-samples) test
results of path coefficients. According to the standard of
significance testing, the T Statistics greater than 1.96 indicates that
it is significant at the 0.05 level when the sample size is large (Hair
et al., 2006). Only the relation between passenger expectation and
perceived quality is not supported by statistical significance test,
but the T Statistics is within the acceptable limits.
4.4. The satisfaction indexes of Suzhou rail transit
The satisfaction indexes of latent variables are shown in Table 7
and the indexes are translated into centesimal system. To measure
the degree of satisfaction, a standard of satisfaction evaluation
(measured using centesimal system) is given, i.e. a score lower
than 60 means very dissatisfied, a score between 60 to 70 means
relatively dissatisfied, a score between 70 to 80 means moderate
satisfied, a score between 80 to 90 means relatively satisfied and
the score higher than 90 means great satisfied.
From Table 7, it can be concluded that all the satisfaction in-
dexes of latent variables are relatively high (with satisfaction index
more than 80). Passenger satisfaction has a satisfaction index of
83.27 (i.e. the PSI), indicating that the service quality of Suzhou rail
transit line 1 is at a good level, but improvement is still needed.
Among the nine service dimensions, ‘Safety and security’ has the
highest satisfaction index and then the ‘Direction and guidance’,
most of the nine service dimensions perform well (with satisfac-
tion index more than 85), while there is a little gap between ‘In-
formation distribution/disclosure’, ‘Convenient facilities for pas-
sengers’, and other service dimensions.
4.5. Problems in rail transit service and countermeasures
4.5.1. IPA analysis
The IPA was used as a tool to assisting identifying the im-
provement based on nine service dimensions. The performance is
the satisfaction index of nine dimensions, while the importance is
described by the path coefficients between nine dimensions and
perceived quality (as shown in Fig. 3).
According to Fig. 4, it is find that equipment and facilities and
information distribution/disclosure are in Quadrant II which
means they are the problems require great attention and the rail
transit operation companies should work on improvement mea-
sures immediately. At the same time, ticket service, speediness
and convenience, direction and guidance, and safety and security
are in Quadrant IV which indicates that these are the factors that
rail transit operation companies perform well and they just need
to maintain the current status moderately. Convenient facilities for
passengers is in Quadrant III which indicates that this service as-
pect has lower priority. Cleanliness and comfort and staff service
are in Quadrant I, so they are the advantages of Suzhou rail transit
28 W. Shen et al. / Transport Policy 46 (2016) 20–31

Table 4
Outer weights and loading coefficients of the PSI model.

The first-level and second-level indicators The third-level indicators (manifest variables) Outer weights Loading coefficients
(latent variables)

Passenger expectation Overall expectation before taking rail transit 0.385 0.883
Expectation on the degree of personalized service to meet the demand before 0.401 0.895
taking rail transit
Expectation on the reliability of service before taking rail transit 0.347 0.868
Perceived quality Overall satisfaction on direction and guidance 0.122 0.766
Overall satisfaction on cleanliness and comfort 0.124 0.769
Overall satisfaction on speediness and convenience 0.131 0.813
Overall satisfaction on safety and security 0.137 0.867
Overall satisfaction on ticket service 0.134 0.853
Overall satisfaction on equipment and facilities 0.140 0.867
Overall satisfaction on staff service 0.141 0.840
Overall satisfaction on information distribution/disclosure 0.142 0.829
Overall satisfaction on convenient facilities for passengers 0.141 0.808
Direction and guidance The signs of transit station location outside the station 0.224 0.849
Export instructions in the station 0.226 0.883
The signs of ticket gates in the station 0.230 0.882
The clarity of direction of rail transit on the platform 0.238 0.885
The hints of rail transit’s arriving at the station 0.235 0.837
Cleanliness and comfort The brightness in the station 0.248 0.828
Air liquidity and temperature suitability in the station 0.238 0.859
Environment cleanliness in the station 0.250 0.885
Air liquidity and temperature suitability in the compartment 0.204 0.748
Environment cleanliness in the compartment 0.251 0.861
Speediness and convenience The punctuality of the rail transit’s arriving 0.243 0.816
The acceptability of station dwell time 0.244 0.857
The acceptability of departure interval 0.236 0.839
The acceptability of service time 0.233 0.765
The convenience of bus transfer 0.262 0.828
Safety and security The order in the station 0.112 0.836
The safety of stairs, escalators and elevators 0.118 0.895
The frequency of failure of escalators 0.109 0.858
The induction sensitivity and frequency of failure of gate 0.112 0.848
The frequency of failure of shielding barrier 0.115 0.891
The timeliness of advance notice about the closing of car door and shielding 0.117 0.881
barrier
The safety and stability of train running 0.121 0.886
The readiness of the security alarm facilities 0.120 0.901
The method of informing passengers and handling the problem is appropriate 0.117 0.852
when abnormal situation happens
The maintenance of order of getting on or off the rail transit when it is the rush 0.115 0.798
time
Ticket service The degree of satisfaction on ticket types 0.212 0.789
The quantity of ticket vending machines which can satisfy needs 0.227 0.880
The clarity of introduction about buying tickets in the ticket vending machines 0.245 0.918
The frequency of failure of ticket vending machines 0.229 0.876
The convenience of changing money in the call center 0.236 0.879
Equipment and facilities The rationality of locations of vending machines 0.236 0.903
The quantity of ticket gate which can meet the needs 0.228 0.884
The convenience of locations of waiting seats 0.230 0.912
The comfort of materials of waiting seats 0.215 0.865
The rationality of settings of armrest and rings in the compartment 0.220 0.867
Staff service The appearance of staffs 0.219 0.884
The service attitude of staffs 0.213 0.920
The efficiency of solving passengers’ problems 0.222 0.913
The call center provides service in the whole service time 0.218 0.897
The overall image and service attitude of security personnels 0.228 0.934
Information distribution/disclosure The degree of publicity of rail transit 0.247 0.845
The degree of satisfaction on the executive condition about the measure ‘No food 0.198 0.807
or drink in the compartment’
The degree of satisfaction on the executive condition about the measure ‘Get on 0.203 0.826
the train on both sides, and get off in the middle’
The convenience and efficiency of knowing about rail transit in the station 0.255 0.927
The convenience and efficiency of knowing about rail transit outside the station 0.249 0.913
Convenient facilities for passengers Suitable volume of TV in the station or compartment 0.224 0.853
The updating frequency of television media in the station or compartment 0.225 0.893
The comfort of advertising layout and images in the station or compartment 0.250 0.905
The attraction of rail transit newspapers 0.234 0.880
Signal strength of cell phone in the station or compartment 0.221 0.800
Perceived value Evaluation of the service according to current ticket price 0.561 0.927
Evaluation of the ticket price according to current condition of service 0.524 0.916
Passenger satisfaction Overall satisfaction 0.446 0.834
Compared with the service expectation, the gap between service expectation and 0.364 0.838
reality
Compared with the perfect service of rail transit, gaps between perfection and 0.387 0.836
 W. Shen et al. / Transport Policy 46 (2016) 20–31 29

Table 4 (continued )

The first-level and second-level indicators The third-level indicators (manifest variables) Outer weights Loading coefficients
(latent variables)

reality
Passenger loyalty The possibility of choosing rail transit if it connects with desired destination 0.353 0.653
The possibility of choosing rail transit if it does not connects with desired 0.439 0.676
destination
The possibility of recommending rail transit to your family and friends 0.559 0.846

Table 5
Validity of the model.

Latent variables Average Variance Extracted (AVE) Composite Reliability (CR) R-Square Cronbach's Alpha (CA) Communality Redundancy

Passenger loyalty 0.533 0.772 0.210 0.560 0.533 0.110

Passenger expectation 0.778 0.913 0.000 0.858 0.778 0
Passenger satisfaction 0.698 0.874 0.419 0.785 0.698 0.060
Perceived value 0.849 0.919 0.350 0.823 0.849 0.061
Perceived quality 0.680 0.950 0.947 0.941 0.680 0.018
Direction and guidance 0.752 0.938 0.000 0.918 0.752 0
Cleanliness and comfort 0.701 0.921 0.000 0.893 0.701 0
Speediness and convenience 0.675 0.912 0.000 0.879 0.675 0
Safety and security 0.749 0.968 0.000 0.963 0.749 0
Ticket service 0.756 0.939 0.000 0.919 0.756 0
Equipment and facilities 0.786 0.948 0.000 0.932 0.786 0
Staff service 0.827 0.960 0.000 0.948 0.827 0
Information distribution/disclosure 0.748 0.937 0.000 0.915 0.748 0
Convenient facilities for passengers 0.752 0.938 0.000 0.917 0.752 0

Table 6 Table 7
Bootstrapping Test Results of Path Coefficients. Satisfaction indexes of latent variables

Path relation Standard error T Statistics Latent variable Satisfaction index Degree of satisfaction

Passenger expectation - 4 Passenger satisfaction 0.041 2.932 Passenger Expectation 83.50 relatively high
Passenger expectation - 4 Perceived value 0.048 2.223 Perceived Quality 85.52 relatively satisfied
Passenger expectation - 4 Perceived quality 0.013 1.872* Perceived Value 80.20 relatively satisfied
Passenger satisfaction - 4 Passenger loyalty 0.045 10.184 Passenger Satisfaction 83.27 relatively satisfied
Perceived value - 4 Passenger satisfaction 0.047 5.951 Passenger Loyalty 82.35 relatively loyal
Perceived quality - 4 Passenger satisfaction 0.045 11.637 Direction and guidance 86.51 relatively satisfied
Perceived quality - 4 Perceived value 0.025 3.454 Cleanliness and comfort 85.55 relatively satisfied
Direction and guidance - 4 Perceived quality 0.020 6.889 Speediness and convenience 86.27 relatively satisfied
Cleanliness and comfort - 4 Perceived quality 0.021 4.462 Safety and security 86.79 relatively satisfied
Speediness and convenience - 4 Perceived 0.026 3.106 Ticket service 85.62 relatively satisfied
quality Equipment and facilities 84.37 relatively satisfied
Safety and security - 4 Perceived quality 0.022 5.451 Staff service 86.15 relatively satisfied
Ticket service - 4 Perceived quality 0.025 6.699 Information distribution/disclosure 81.91 relatively satisfied
Equipment and facilities - 4 Perceived quality 0.023 6.527 Convenient facilities for passengers 81.36 relatively satisfied
Staff service - 4 Perceived quality 0.025 7.554
Information distribution/disclosure - 4 Perceived 0.018 5.988
quality
Convenient facilities for passengers - 4 Perceived 0.041 2.932 information distribution/disclosure, compared with the publicity
quality of taking rail transit with civility, introducing lines, fare policy,
* service time of rail transit is more important because this aspect
indicates the value is not significant at the 5% level.
has higher weights.
For the indicator ‘convenient facilities for passengers’ in
line 1. With the help of IPA matrix, the problems in existing service
Quadrant III, advertising layout and images in the station or
aspects can be identified in an intuitive way and a specific im-
compartment has the highest weight and should have the greatest
provement plan can be established.
priority, and then the attraction of rail transit newspapers needs to
be improved. The enhancement of signal strength of cell phone is
4.5.2. Detailed improving suggestions for rail transit operation
relatively hard and has the lowest priority.
companies
For the indicators in Quadrant I, staff service has great im-
To help the rail transit operation companies use the findings to
portance and good performance, the service of security personnels
satisfy the passengers and further enhance passenger loyalty. A cannot be ignored and the efficiency of solving passengers' pro-
number of operational implications are explicitly described. blems has higher priority compared with other indicators. For
For the indicators in Quadrant II, the improvement in equip- cleanliness and comfort, environment cleanliness in the station or
ment and facilities is relatively hard to make as the location or compartment is more important than brightness, air liquidity and
material of equipment and facilities is difficult to change, the temperature suitability.
things that operator can do is to increase the number of equip- For the indicators in Quadrant IV, ticket service has the higher
ment and facilities properly to satisfy passengers' demand. For importance but worse performance, so it needs to be improved first
30 W. Shen et al. / Transport Policy 46 (2016) 20–31
Fig. 4. IPA matrix of Suzhou rail transit line 1.
compared with other three indicators in Quadrant IV. The clarity of
introduction about buying tickets in the ticket vending machines
needs prompt improvement, and then the convenience of changing
money in the call center. The ticket types and quantity of ticket
vending machines can be added to meet the diversified demands. For
speediness and convenience, the convenience of bus transfer is most
important, so the rail transit operation companies can promote co-
operation with bus companies to optimize the transfer between rail
transit and bus lines. Among the three types of signs in direction and
guidance, the signs of direction of rail transit have the highest priority
to make improvements, signs in the station have the second priority,
and the signs outside the station rank the third. As the recognition of
safety and security of Suzhou rail transit, this service aspect performs
very well. Among ten subordinate indicators of safety and security, the
safety and stability of train running is most important and needs to be
ensured.
5. Limitation
Several limitations exist in this study. ‘Passenger complaints’ is
an important indicator reflecting passengers’ dissatisfaction on
services. As the limit of survey data, the indicator used to measure
passenger complaints is not adequate, which result in the insig-
nificant parameters of passenger complaints and its subordinate
indicators. So ‘Passenger complaints’ is excluded from the final
model. In that case, further investigation and better survey design
are needed to improve the PSI model in the future.
6. Conclusions
In the context of increasing traffic congestion and goals to re-
duce automobile transportation by promoting public transit, it is
essential for railway transit providers to assess how the current
transit system performs in order to retain and attract more pas-
sengers. This study focuses on a comprehensive evaluation
methodology to assess the current rail transit passenger satisfac-
tion level in a quantifiable way and identify factors impacting
passenger satisfaction. This methodology is practical which can be
potentially utilized as standard evaluation procedure for rail
transit operation companies to assess passenger satisfaction reg-
ularly and track the effects of the improvement strategies.
The evaluation methodology includes the evaluation model (i.e.
PSI model), evaluation indicator system of three levels and the IPA
matrix, which allows in-depth investigation of the degree of pas-
senger satisfaction and the associated factors contribute to pas-
senger satisfaction. An intuitive picture of detailed causality be-
tween variables can be revealed by the evaluation model. Then
improvement strategy can be designed more effectively. The case
study shows that Suzhou rail transit line 1 has a moderate satisfied
level (with PSI of 83.27), but more importantly, perceived value
followed by perceived quality are found to have the greatest direct
influence on passenger satisfaction. Therefore, the operation
companies must make a great effort to improve service quality.
In addition, the causal relationships between other latent
variables (e.g. passenger expectation and passenger loyalty) were
proved to be consistent with previous studies (Fornell et al., 1996;
Cronin et al., 2000; Wen et al., 2005). Besides, this study extended
the previous studies by enriching the concept of perceived quality
to detailed service dimensions. Specifically, a first-level indicator
perceived quality, nine second-level indicators and fifty third-level
indicators were used to measure the performance of services.
These captures provide a complete picture of the different service
dimensions that can impact the passenger satisfaction. Moreover,
instead of using the LISREL estimation method which has been
used for previous studies (Ngatia et al., 2010; Irfan et al., 2011; de
Oña et al., 2013), a more sophisticated structural equation mod-
eling method based on partial least squares has been used to solve
the PSI model. This estimation method was proved to have the
advantage of relaxing the normal distribution assumption and
obtaining explicitly estimated latent variable scores which are
important for operation companies to gain satisfaction scores.
Meanwhile the PLS avoids the poor comparison of satisfaction
indexes calculated by different methods in the LISREL estimation.
Practically speaking, the findings of this study can aid in
identifying rail transit operation and management strategies to
help them making more effective improvement decisions towards
better service quality and higher passenger satisfaction. Clearly,
the rail transit operation and management require systematic
perspective of system design and optimization and understanding
passengers' satisfaction level is fundamental to this process.
Acknowledgements
The authors would like to thank Roger Vickerman, Greg
Marsden, and two anonymous reviewers for their valuable
 W. Shen et al. / Transport Policy 46 (2016) 20–31
comments and suggestions that have substantially improved this
paper. We also wish to thank the research assistants who help us
obtain the survey data.
References
Anderson, E.W., Fornell, C., Lehmann, D.R., 1994. Customer satisfaction, market
share, and profitability: findings from Sweden. J. Mark. 58, 53–66.
Anderson, E.W., Fornell, C., Rust, R.T., 1997. Customer satisfaction, productivity, and
profitability: differences between goods and services. Mark. Sci. 16 (2),
129–145.
Bollen, K.A., 1989. Sturctural Equations with Latent Variables. Wiley, New York.
Celik, E., Aydin, N., Gumus, A.T., 2014. A multiattribute customer satisfaction eva-
luation approach for rail transit network: a real case study for Istanbul, Turkey.
Transp. Policy 36, 283–293.
Chen, C.F., Chen, F.S., 2010. Experience quality, perceived value, satisfaction and
behavioral intentions for heritage tourists. Tour. Manag. 31 (1), 29–35.
Chen, C.F., 2008. Investigating structural relationships between service quality,
perceived value, satisfaction, and behavioral intentions for air passengers:
evidence from Taiwan. Transp. Res. A 42, 709–717.
Chou, C.C., Liu, L.J., Huang, S.F., Yih, J.M., Han, T.C., 2011. An evaluation of air line
service quality using the fuzzy weighted SERVQUAL method. Appl. Soft Com-
put. 11 (2), 2117–2128.
Chou, J.S., Kim, C.W., 2009. A structural equation analysis of the QSL relationship
with passenger riding experience on high speed rail: an empirical study of
Taiwan and Korea. Expert Syst. Appl. 36, 6945–6955.
Chin, W.W., 1998. The partial least squares approach to structural equation mod-
eling. Mod. Methods Bus. Res. 295 (2), 295–336.
Chin, W.W., Newsted, P.R., 1999. Structural equation modeling analysis with small
samples using partial least squares. Statistical strategies for small sample re-
search. Sage Publications, London, pp. 307–341.
Cronin, J.J., Brady, M.K., Hult, G.T.M., 2000. Assessing the effects of quality, value and
customer satisfaction on consumer behavioral intentions in service environ-
ments. J. Retail. 76 (2), 193–218.
de Oña, J., de Oña, R., 2014. Quality of service in public transport based on customer
satisfaction surveys: a review and assessment of methodological approaches.
Transp. Sci. 49 (3), 605–622.
de Oña, J., de Oña, R., Eboli, L., Mazzulla, G., 2013. Perceived service quality in bus
transit service: a structural equation approach. Transp. Policy 29, 219–226.
Eboli, L., Mazzulla, G., 2007. Service quality attributes affecting customer satisfac-
tion for bus transit. J. Public Transp. 10 (3), 21–34.
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.
Transp. Policy 18 (1), 172–181.
Efron, B., 1979. Bootstrap methods: another look at the jackknife. Ann. Statist. 7 (1),
1–26.
Eklof, J.A., Hackl, P., Westlund, A., 1999. On measuring interactions between cus-
tomer satisfaction and financial results. Total Qual. Manag. 10 (4), 514–522.
Figler, S.A., Sriraj, P.S., Welch, E.W., Yavuz, N., 2011. Customer loyalty and Chicago,
Illinois, transit authority buses. Transp. Res. Rec. 2216, 148–156.
Fornell, C., Johnson, M.D., Anderson, E.W., Cha, J., Bryant, B.E., 1996. The American
customer satisfaction index: nature, purpose and findings. . J. Mark. 60, 7–18.
Fornell, C., Larcker, D.F., 1981. Evaluating structural equation models with un-
observable variables and measurement error. J. Mark. Res. 18, 39–50.
Friman, M., Gärling, T., 2001. Frequency of negative critical incidents and satisfac-
tion with public transport services. J. Retail. Consum. Serv. 8, 105–114.
Friman, M., Fujii, S., Ettema, D., Gärling, T., Olsson, L.E., 2013. Psychometric analysis
of the satisfaction with travel scale. Transp. Res. A 48, 132–145.
31
Hair, J.F., Anderson, R.E., Tatham, R.L., Black, W.C., 2006. Multivariate Data Analysis,
6th ed. Prentice Hall, New Jersey.
Hassan, M.N., Hawas, Y.E., Ahmed, K., 2013. A multi-dimensional framework for
evaluating the transit service performance. Transp. Res. A. 50, 47–61.
Hensher, D.A., Stopher, P., Bullock, P., 2003. Service quality-developing a service
quality index in the provision of commercial bus contracts. Transp. Res. A. 37,
499–517.
Hensher, D.A., 1990. Hierarchical stated response designs-an application to bus user
preferences. Logist. Transp. Rev. 26 (4), 299–321.
Huo, Y.B., 2006. Analysis and comparison of the path modeling principles between
LISREL and PLS. Stat. Dec. 10, 19–21.
Irfan, S.M., Mui, H.K.D., Shahbaz, S., 2011. Service quality in rail transport of Paki-
stan: a passenger perspective. In: Proceedings of the 3rd SAICON: International
Conference on Management, Business Ethics and Economics (ICMBEE), Lahore,
Pakistan.
Jacoby, J., 1971. A model of multi-brand loyalty. J. Adv. Res. 11, 25–30.
Jin, Y.J., Liang, Y., 2005. Fit index of partial least square (PLS) and its application to
the customer satisfaction model. Appl. Stat. Manag. 24 (2), 40–44.
Joreskong, K.G., Dag, S., 1993. New Features in LISREL. IL: Scientific Software, Chi-
cago, p. 8.
Karlaftis, M.G., Golias, J., Papadimitriou, E., 2001. Transit quality as an integrated
traffic management strategy: measuring perceived service. J. Public Transp. 4, 1.
Kuo, C.W., Jou, R.C., 2014. Asymmetric response model for evaluating air line service
quality: an empirical study in cross-strait direct flights. Transp. Res. A. 62,
63–70.
Kuo, Y.F., Wu, C.M., Deng, W.J., 2009. The relationships among service quality,
perceived value, customer satisfaction, and post-purchased intention in mobile
value-added services. Comput. Hum. Behav. 25 (4), 887–896.
Li, W.J., 2011. Evaluation of Service Quality and Satisfaction of Urban Rail Transit.
China Railway Publishing House, Beijing.
Lin, S., Liu, J.L., Han, X.W., 2005. Method on customer satisfaction assessment based
on the partial least square for structural equation. J. Syst. Eng. 20 (6), 653–656.
Liou, J.J.H., Hsu, C.C., Chen, Y.S., 2014. Improving transportation service quality
based on information fusion. Transp. Res. A. 67, 225–239.
Lovelock, C.H., 2000. Service Marketing, 4th ed. Prentice Hall International, New
Jersey.
Mouwen, A., Rietveld, P., 2013. Does competitive tendering improve customer sa-
tisfaction with public transport? A case study for the Netherlands. Transp.
Res. A. 51, 29–45.
Ngatia, G.J., Okamura, T., Nakamura, F., 2010. The structure of users' satisfaction on
urban public transport service in developing country: the case of Nairobi. J.
East. Asia Soc. Transp. Stud. 8, 1288–1300.
Oliver, R., 1980. A cognitive model of the antecedents and consequences of sa-
tisfaction decisions. J. Mark. Res. 17, 460–469.
Parasuraman, A., Zeithaml, V., Berry, L., 1985. A conceptual model of service quality
and its implications for future research. J. Mark. 49, 41–50.
Ringle, C.M., Wende, S., Will, A., 2005. SmartPLS–Version 2.0. Universität Hamburg,
Hamburg.
Singh, J., 1988. Consumer complaint intentions and behavior: definitional and
taxonomical issues. J. Mark. 52, 93–107.
Stradling, S.G., Anable, J., Carreno, M., 2007. Performance, importance and user
disgruntlement: a six-step method for measuring satisfaction with travel
modes. Transp. Res. A 41, 98–106.
Stuart, K.R., Mednick, M., Bockman, J., 2000. Structural equation model of customer
satisfaction for the New York City subway system. Transp. Res. Rec. 1735,
133–137.
Wen, C.H., Lan, L.W., Cheng, H.L., 2005. Structural equation modeling to determine
passenger loyalty toward intercity bus services. Transp. Res. Rec. 1927,
249–255.
Zeithaml, V.A., Berry, L.L., Parasuraman, A., 1996. The behavioral consequences of
service quality. J. Mark. 60 (2), 31–46.