Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul

Abstract

Providing a high quality of service in public transportation is essential to reduce dissatisfactions stemming from traffic congestion and noise. Public transport providers need to find ways to dilute the effects of immoderate use of private cars in big cities while maintaining a sufficient level of customer satisfaction. This study aimed to identify the key service quality (SQ) factors that drive passenger satisfaction in Istanbul’s rail transit (RT) system using data obtained from an extensive survey conducted by the Istanbul Public Transportation Co. A total of 11,116 passengers who used rail transport from May 15–June 3, 2012, and June 17–July 3, 2013, were interviewed in person. The relative importance of the SQ factors was assessed so that service provision could be prioritized and the enhancement of passenger satisfaction can be achieved employing several social choice techniques. The results indicate that, from an overall perspective, waiting time, crowdedness in cars, and fare are the SQ factors that best reflect the public good.

Full text

Journal of Public Transportation, Vol. 20, No. 1, 2017 63
Identifying Key Factors of
Rail Transit Service Quality:
An Empirical Analysis for Istanbul
Erkan Isikli
Istanbul Technical University
Nezir Aydin
Yildiz Technical University
Erkan Celik
Munzur University
Alev Taskin Gumus
Yildiz Technical University
Abstract
Providing a high quality of service in public transportation is essential to reduce
dissatisfactions stemming from traffic congestion and noise. Public transport providers
need to find ways to dilute the effects of immoderate use of private cars in big cities while
maintaining a sufficient level of customer satisfaction. is study aimed to identify the
key service quality (SQ) factors that drive passenger satisfaction in Istanbul’s rail transit
(RT) system using data obtained from an extensive survey conducted by the Istanbul
Public Transportation Co. A total of 11,116 passengers who used rail transport from
May 15–June 3, 2012, and June 17–July 3, 2013, were interviewed in person. e relative
importance of the SQ factors was assessed so that service provision could be prioritized
and the enhancement of passenger satisfaction can be achieved employing several social
choice techniques. e results indicate that, from an overall perspective, waiting time,
crowdedness in cars, and fare are the SQ factors that best reflect the public good.
Keywords: Service quality; public transportation; rail transit systems; stated preferences;
fallback voting; Istanbul

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 64
Introduction
Public transportation (PT) is a cost-effective solution for traffic congestion, especially
in crowded areas, and its improvement is of critical importance to city governance and
decisionmakers. As with many other PT services, rail transit (RT) systems should also
hear the voice of the customer since decisionmakers need to create an efficient system
to promote public transport use (Gronau and Kagermeier 2007; Le-Klähn et al. 2014).
With its 14.3 million inhabitants and a high level of socio-economic development,
Istanbul is the heart of Turkey. e daytime population of this metropolis increases
as many people commute from neighboring cities to Istanbul, which increases traffic
congestion. e city’s population is expected to increase to nearly 15 million by 2019
and 16 million by 2023, according to the Turkish Statistical Institute (2015). Economic
recovery and improvement in the standard of living makes passengers expect better PT
services (bus rapid transit, rail transit, etc.). People prefer PT to avoid traffic congestion,
noise, and long waiting times, especially during rush hours.
When all drawbacks of traveling by a private car are considered, RT has been one of the
most appropriate modes of travel for public transport users in Istanbul. e city’s six RT lines
(M1, M2, M4, T1, T4, F1), which are operated by Istanbul Public Transportation Co., total
145.5 kilometers in length and carry more than 1.3 million passengers daily (www.metro.
istanbul/en). Figure 1 shows the network maps and characteristics of RT lines in Istanbul.

Journal of Public Transportation, Vol. 20, No. 1, 2017 65
Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
FIGURE 1. Istanbul rail transit line network map

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 66
To increase the number of people using the city’s RT system, it is critical to gather
information from the recent users of the system regarding how much their expectations
are met so decisionmakers can make changes to meet the passenger needs better
(Andaleeb 2007; Le-Klähn et al. 2014). Customer surveys are especially important
nowadays considering the depth and amount of information they can provide so public
transport providers can understand which service aspects play a more critical role in
passenger satisfaction (Le-Klähn et al. 2014).
e Istanbul Public Transportation Co. conducts a customer satisfaction survey annually
to determine the needs and problems of its RT passengers to improve the system based
on their demand. In this study, we analyzed the results of the 2012 and 2013 surveys
using several voting rules to evaluate the priority of each service quality (SQ) factor for
RT passengers in Istanbul by line and year. A total of 11,116 passengers completed the
surveys, which were distributed among 6 RT lines.
e remainder of this paper is structured as follows.Section 2 reviews the literature
highlighting stated importance methods,Section 3 reports the details on survey
data, and Section 4 provides a summary of the voting methods employed with
exemplification. In Section 5, results from different categories of methods are clustered
and interpreted. Finally, Section 6 concludes with an overall discussion emphasizing
possible avenues for future research.
Related Literature
Proposing higher SQ levels in PT to reduce dissatisfaction (traffic congestion, noise, etc.)
resulting from immoderate use of private cars in big cities is one of the most important
issues for public transport providers. us, PT service planners pursue lessening the
use of private cars by developing quality improvement plans that will initiate higher
customer satisfaction. Increasing customer satisfaction or SQ levels results in a higher
use of the service, involvement of new customers, and a better public image (de Oña et
al. 2012; Çelik et al. 2013).
To reach an appropriate SQ level, service providers should consider several SQ
factors associated with PT. Mouwen and Rietveld (2013) considered several factors
to determine if competitive tendering increases SQ for PT in the Netherlands and
determined that frequency of service, time accuracy, travel speed, and vehicle tidiness
were the most effective. Waiting time, cleanliness, and comfort were observed to be
the most valued PT factors in a study by dell’Olio et al. (2011). Redman et al. (2013)
presented a comprehensive review on SQ factorss in PT and determined that reliability,
frequency, price, speed, access, comfort, and convenience were the factors that attract
car users to use PT. Hassan et al. (2013) asserted that the most desirable SQ factors of
PT services were reliability, frequency, capacity, price, cleanliness, comfort, security, staff,
information, and ticketing system, with loading/ridership, travel time, travel distance,
and service duration indicated as “efficiency” indicators.
Currently, in big and crowded cities, RT systems are preferred as one of the easiest
ways of avoiding traffic congestion and noise. erefore, analyzing service quality in

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 67
RT systems has gained importance. Gerçek et al. (2004) evaluated three alternative RT
networks based on four main factors—financial, economic, system planning, and policy.
Awasthi et al. (2011) integrated SERVQUAL and TOPSIS to evaluate the SQ of Montreal
metro services. Nathanail (2008) evaluated the performance of Hellenic Railways based
on 22 factors group into six major factors—itinerary accuracy, system safety, cleanliness,
passenger comfort, servicing, and passenger information. e author concluded that
the RT systems that paid attention to itinerary accuracy and system safety would
perform best.
Brons et al. (2009) aimed to determine the significance level of the access-to-the-station
effect on passenger overall satisfaction and the balance between the factors of the RT
services. ey concluded that, in several parts of the RT network, improving access
services to the railway stations could substitute for improving the services provided
on the rail network, which would attract passengers who used other transportation
modes. Eboli and Mazzulla (2012) analyzed how RT passengers perceived different
SQ factors, noting that promptness, consistency, frequency, and cleanliness had the
highest positive influence for RT services. However, dell’Olio et al. (2010) noted that
passenger perceptions on SQ might change depending on the type of passengers under
consideration. Cascetta and Cartenì (2014) provided a comparison between perceived
and calculated SQ for a metro line servicing in the Campania region of Italy. In a study
by de Oña et al. (2014b), passengers were clustered to determine the most important
SQ factors, concluding that different factors may be determined as the most important
for different groups of passengers. Punctuality was selected as the most important SQ
factor for the first group (young female students who do not have a private car), and
frequency was selected for the second group (women of medium age who frequently
use public transport service for reaching jobs). From a general perspective, comfort,
personnel, information, and service were determined as the most important factors (de
Oña et al. 2014a).
As mentioned in Berry et al. (1990), since passengers are the only rulers of the systems in
terms of SQ, their perception on SQ factors should be contemplated when evaluating
the SQ level of a system. (Tyrinopoulos and Antoniou 2008; Filipović et al. 2009; Eboli
and Mazzulla 2009, 2011). A study by de Oña et al. (2012) classified the methods
proposed to evaluate the perceived importance of SQ factors into two main categories:
“stated importance methods and derived importance methods. In the former,
customers were asked to rate each factor on an importance scale, whereas in the latter,
the importance of factors was determined by analyzing the relationship of each factor
with the overall customer satisfaction via statistical testing.
In this study, the stated importance approach was adopted; however, as discussed
in the related literature, it has several drawbacks (Eboli and Mazzulla 2008a, 2008b,
2010; Cirillo et al. 2011; Dell’Olio et al. 2011). First, stated importance methods may
greatly suffer if passengers rate almost all of the criteria/items close to the top scale
(e.g., 5 on a 5-point Likert scale). is results in an inadequate differentiation among
mean importance ratings. In addition, such methods require that the survey cover a
relatively longer period, which may reduce the overall response rate and the accuracy

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 68
of the survey. Some criteria found important may, in fact, have little effect on overall
satisfaction (de Oña et al. 2012).
Despite the notable increase in the number of studies employing derived importance
methods (Eboli and Mazzulla 2007; Dell’Olio et al. 2010; Jen et al. 2011) due to the issues
summarized above, the stated importance approach has advantages over the derived
importance approach. First, it is understood by decisionmakers and public policymakers
more easily. It also requires fewer analytical skills and less expertise to employ (Van Ryzin
and Immerwahr 2007). Nevertheless, interested readers are referred to Van Ryzin and
Immerwahr (2007), Eboli and Mazzulla (2007), Dell’Olio et al. (2010), and Jen et al. (2011)
to gain more insight on different applications of the derived importance methods.
Although many studies have focused on the evaluation of criterion-wise satisfaction
levels or overall satisfaction level, few have paid attention to the relative importance of
service quality. When determining the key SQ factors, research to date has neglected
to consider customer preference rankings that are information-rich and can be easily
processed and interpreted. If customers rate their satisfaction with only a specific SQ
factor, the path followed by research done so far is inevitable; however, when customers
order SQ factors based on their preferences, they provide more information regarding
on what decisionmakers should focus. erefore, this study contributes to the existing
literature by providing a different aspect to analyze passenger satisfaction using a
considerably large sample and comparing results between years and RT lines with the
help of different voting procedures that are easy to implement. Highly-prioritized SQ
factors were determined using a representative sample consisting of 11,116 individuals.
e joint investigation of traditional and non-traditional voting methods for ranking
the most important SQ factors also added value. In addition, determining high-priority
SQ factors for each line separately provides more insight on potential differentiation
between the lines considered. Finally, the procedures provide valuable information
regarding SQ factors that should be primarily focused on to provide a better service in
RT lines for future investments.
Survey Data
e survey was composed of four parts: Station and Ticketing, Rail Transit Usage,
Overall and Criterion-Based Satisfaction, and Demographics. e survey questions
measured each SQ factor on a 6-point Likert scales with “extremely satisfied” reflecting
the highest favorable response and “extremely dissatisfied” indicating the least favorable
response to each statement.
To determine the importance of SQ factors for RT lines in Istanbul, we analyzed
passenger satisfaction surveys that were conducted among 4,966 passengers in 2012
(from May 15 to June 3), and 6,150 passengers in 2013 (from June 17 to July 3). e
distribution of the 11,116 survey participants across years and lines are shown in Table
1. Since the M4 line was not open during the time the survey was conducted in 2012,
there were no data available regarding that year.

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 69
Year Rail Transit Line
T1 T4 M1 M2 M4 F1
2013 1,560 1,129 1,044 1,145 1,084 188
2012 1,575 1,047 1,076 1,069 N/A 199
All passengers were interviewed in person. A multistage stratified sampling procedure
was employed in which tiers were formed considering the differences at rush hours and
off-peak traffic hours among the transit lines. e quotas for the tiers were set according
to the following criteria:
1. Day of use: weekdays, Saturday, Sunday
2. Time slot: morning rush, morning, noon, evening rush, evening
3. Station-wise crowdedness
4. Ticket type: token, full fare, discount fare, free
e participants in the survey were selected as follows to achieve randomness: A
pollster waiting at an exit asked the 6th (5th in 2013) passenger who passed the turnstiles
to participate in the survey; if that passenger was not willing, then the next passenger
was asked to participate, and so on. At that point, the sequence of passengers was no
longer important. Note that the pollster was not asked to follow a systematic sampling
procedure in the first place, as it would be impractical to select every nth passenger for
the survey, especially during rush hours.
Table 2 shows details on the survey data regarding demographics and travel
characteristics. Note that median monthly household income of the survey participants
was 1,782 Turkish lira (TL) (approx. $970 based on the Central Bank of Turkey’s exchange
rate in May 28, 2012) in 2012, and the full transit fare was 1.65 TL during that time. is
increased to 2,431 TL (approx. $1,250 based on the Central Bank of Turkey’s exchange
rate in June 24, 2013) in the next survey year, and the full transit fare increased to 1.95
TL. Some notable differences between survey years regarding demographics appear
in education level and household income level. e percentage of participants who
had a primary school degree significantly decreased in contrast to the percentage of
participants with an undergraduate degree. e frequency distribution of household
income also changed; it was right-skewed in 2012, but was fairly symmetric in 2013
(with a higher median value compared to the previous survey year). is might be
attributed to the introduction of a new line (M4) into the RT system by the time survey
was conducted in 2013 since this line provides service in the Anatolian part of Istanbul,
unlike the other five.
TABLE 1.
Subsample Sizes by Line
and Year

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 70
TABLE 2. Respondent Profile
Demographic Characteristics 2012 2013 Travel Profile Characteristics 2012 2013
Gender Male 77.4% 74. 6% Car
Ownership
Yes 40.2% 43.0%
Female 22.6% 25.4% No 59.8% 57.0%
Age
15-25 46.7% 48.4%
Time to
Station
Less than 5 min 31.3% 26.7%
26-35 26.4% 28.6% 6 -10 m in 25.8% 25.2%
36-45 13.6% 13.9% 11-15 mi n 13.6% 13 .1%
46-55 8.0% 5.9% 16-20 min 9.1% 12.5%
Older than 55 5.3% 3.3% 21-25 min 2.8% 2.8%
26-30 min 6.3% 5.5%
More than 30 min 11.0% 14.2%
Educational
Level
Primary school not completed 0.6% 0.5%
Total Time
of Travel
Less than 10 min 8.0% 14.6%
Primary school degree 11.4% 8 .1% 11-20 min 26.6% 37.9%
Secondary school degree 9.9% 9.8% 21-30 min 23.1% 18.7%
High school student 12.4% 9.2% 31-40 min 13.3% 10.6%
High school degree 22.1% 23.5% 41-50 min 10.2% 7.3%
Vocational school student/
degree 2.5% 3.0% 51- 60 m in 7.1% 4.6%
Undergraduate student 19.1% 17.3% 61-70 m in 2.8% 1.7%
Undergraduate degree 18.7% 23.6% 71-80 min 2.0% 1.9%
Post graduate student/degree 3.3% 4.8% 81-90 min 2.6% 0.9%
More than 90 min 4.3% 1.7%
Monthly
Household
Income
Less than 500 TL 1.2% 0.6%
Frequency
Of Use
At least once a week 25.7% 24.3%
501-1000 TL 13.9% 5.3% Once a day 16.1% 9.7%
1001-1500 TL 22.0% 11.0% Twice a day 48.0% 53.8%
1501-2000 TL 20.7% 15.2% ree tımes a day 2.9% 4.5%
2001-2500 TL 12.5% 12.9% More than three tımes a day 7.2% 7.7%
2501-3000 TL 10.2% 12.1%
3001-3500 TL 4.0% 7.2%
Ticket Type
Full 52.5% 58.3%
3501-4000 TL 3.3% 5.5% Discount (student, teacher, or
social) 34.6% 34.0%
4001-4500 TL 1.9% 3.4% Token or Free 13.0% 7. 7%
4501-5000 TL 3.1% 3.6% Full 52.5% 58.3%
More than 5001 TL 4.6% 9.5%
Median 1782 TL 2431 TL
Marital
Status
Single 62.5% 66.6%
Main
Purpose
of Travel
Commute 44.8% 53.1%
Married 37.5% 33.4% Go to or return from school 20.9% 12.2%
Work-related activities 13.3% 8.1%
Employment
Status
Unemployed or student 35.7% 30.2% Entertainment or social activities 12 .1% 20.2%
Employed 64.3% 69.8% Other 8.9% 6.4%

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 71
Methodology
As mentioned earlier, we employed various voting procedures from the social choice
literature to find the criterion that had the highest importance for Istanbul Public
Transportation Co. passengers by RT line and year. One question in the surveys asks
passengers to rank, in order of importance, five of the SQ factors given. ese SQ factors
were treated as candidates aiming to be the winner of an election and the passengers
as voters. Given a set of SQ factors, each passenger prioritized the five based on their
preferences.
e most appropriate method in the social choice literature to analyze such data is
Fallback Voting, mainly because customers are almost always allowed to provide partial
preference; however, one can argue that Approval Voting, in which a voter may cast
one vote for as many candidates as desired without ranking them, or even Condorcet-
consistent methods, which work mostly with complete preferences, also could work.
In this section, we briefly cover the relevant voting procedures without going into too
much detail. Interested readers should refer to Tideman (1987) and Plassmann and
Tideman (2014) for more information on the fundamentals of voting systems and social
choice functions.
Assume that individual j is endowed with a preference relation ≿j that is defined over C,
a finite set of candidates competing in the election. A voting system is a function that
takes an election as input and produces a set of winners, a subset of C. e preference
relation ≿j is desired to have some characteristics such as completeness, reflexivity, and
transitivity. Completeness requires that given two different candidates, say A and B,
from C, either A ≿j B (A beats B according to j’s preference or they are tied) or B ≿j A. In
other words, a voter’s ranking contains all available candidates. Incompleteness (partial
preferences) corresponds to the case where the voters rank only a subset of candidates
contained in C. Reflexivity states that any candidate A ∈ C is as preferable as itself; that
is, A ≿j A. Finally, given three candidates A, B, and C ∈ C such that A ≿j B and B ≿j C,
transitivity implies that A ≿j C. Note that the preference relation ≿j is called weak since
it allows for ties (indifference). A strict preference relation, which can be denoted by ≿j,
is irreflexive and individual j is assumed to rank one of the two arbitrary candidates in C
over the other; that is, individual j is never indifferent between any two candidates.
Bulk literature exists on consumer choice modeling that is based on the utility concept
and is directly related to these preference relations with the aforementioned properties.
ey primarily assume that an individual gains an economic utility when he/she selects
an alternative. However, these models are mostly probabilistic and make certain
assumptions regarding individual characteristics and/or candidate characteristics. is
stream of research is not covered here; however, interested readers should refer to
Anderson et al. (1992) for a comprehensive review and detailed discussion on the origin
and the evolution of statistics-oriented choice models and utility maximizing voters.
e preference relation explained above should carry extra properties to have a fair
voting system; however, note that there is no ideal scheme to decide a winner in an
election, as Arrow’s Impossibility eorem proves (Kelly 1978).

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 72
1. Pareto optimality (unanimity): For all pairs of candidates A and B, if A is preferred
to B by all the voters, then B should not be declared as the winner.
2. Monotonicity: Increasing (decreasing) the number of votes for a winning (losing)
alternative cannot make it a loser (winner).
3. Anonymity: Voters are treated the same.
4. Independence of Irrelevant Alternatives (IIA): Suppose that a group of individuals
decided that A should be ranked before B. If a new candidate, say N, which is
outside C, was introduced, then the group decision would not change, provided
that the relative ordering of A and B was preserved.
5. Non-dictatorship: “No voter should decide the outcome of an election” (Menton
2013).
Traditional voting rules can be grouped mainly into two categories according to their
starting point: Condorcet-consistent methods and Condorcet-inconsistent methods. In
the former, the main objective is to find a Condorcet winner if one exists; in the latter,
the winner may be determined by “the points allocated to candidates according to their
ranking on individual voters’ ballots” (Cox 1989). Such methods are called scoring-based
methods. e rest of this section provides an overview on traditional methods along
with recently-proposed voting methods, followed by main assumptions.
Condorcet (1789) asserted that the candidate that is preferred pairwise to every
other candidate by a majority of voters wins the election. Such a candidate is called a
Condorcet winner. If no such winner exists, all candidates tie for the win (Mattei 2012).
As an illustrative example, adapted from Schulze (2003), suppose that there are four
cities (A, B, C, D) vying to host a special event and the 30 members of the international
organizing committee are asked to rank each of these cities from the most favorable
to the least favorable in terms of suitability to stage the event. e aggregated ranked
ballots are as follows:
A≻C≻D≻B 3 B≻C≻D≻A 5 C≻D≻A≻B 5 D≻A≻B≻C 2
A≻D≻B≻C 5 B≻A≻C≻D 4 C≻A≻D≻B 2 D≻B≻A≻C 4
Most of the methods discussed here use a pairwise preference matrix that shows how
many times candidates were preferred over one another. e original matrix is shown
below (Table 3), and the first line reads: City A was preferred to cities B, C, and D in 17,
18, and14 instances, respectively.
TABLE 3.
Pairwise Comparison Matrix
for Illustrative Example
AGAINST
FOR A B C D
A17 18 14
B13 20 9
C12 10 19
D16 21 11

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 73
City A wins over cities B and C, but loses to city D. City D wins over cities A and B,
but loses to city C. us, there is no Condorcet winner since none of the candidates
won every comparison with all the other candidates. Note that if two of the voters
who preferred the ranking D≻B≻A≻C changed to A≻C≻D≻B, city A would be the
Condorcet winner since it would win all of its pairwise comparisons.
On the other hand, de Borda (1781) argues that a majority winner may not always
exist and proposes a paired-comparisons procedure that assigns points to candidates
in head-to-head elections. Each candidate’s total score is calculated based on points
associated with each rank. e winner that has the highest score is then declared the
winner.
For the illustrative example above, city A has a Borda score of 49 ( = 8x3 + 8x2 + 9x1),
city B has a Borda score of 42 ( = 9x3 + 4x2 + 7x1), city C has a Borda score of 41 ( = 7x3
+ 8x2 + 4x1), and city D has a Borda score of 48 ( = 6x3 + 10x2 + 10x1). us, city A wins
the election.
e Borda method may elect a candidate that was not ranked first by any of the voters
(Mattei 2012) and it does not satisfy the IIA property. Eğecioğlu and Giritgil (2011)
addressed the difficulty encountered when one aims to implement this method in case
of partial preferences.
Condorcet-Consistent Methods
e rules summarized below assume complete linear orderings and select the
Condorcet winner if one exists.
1. Baldwin’s iterative procedure employs the Borda count and eliminates the
candidate(s) with the lowest Borda score(s) at each step and recalculates Borda
scores for the remaining candidates; the procedure proceeds until a group of
candidates with the same Borda score can be formed (Hwang and Lin 1987).
For the illustrative example mentioned earlier, at the first step, city C is eliminated
since it has the lowest Borda score. Following the elimination, the Borda scores
of cities A, B, and D become 31, 22, and 37, respectively. us, city B is eliminated.
City D wins eventually as its reduced Borda score (16) is greater than that of city
A’s (14).
2. Black (1958) elects a Condorcet winner if one exists; otherwise, the Borda count
winner is elected.
3. Copeland’s rule (Copeland 1951) works with pairwise comparisons; it counts
the number of wins and losses for each candidate competing in the election.
For each win (loss), a candidate gains (loses) one point. e candidate with the
highest total score wins the election. It allows for ties (no points assigned to the
candidates that are tied), but it may be indecisive.
In the pairwise comparison matrix for the illustrative example above, we see that
cities B and C are eliminated immediately since the former wins only over city
C, whereas the latter wins only over city D. Cities A and D are tied since they

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 74
both have a Copeland score of 2. A tie-breaking rule is necessary at this point to
determine the winner.
4. Dodgson’s method determines the fewest number of pairwise interchanges
needed to make a candidate the Condorcet winner. e candidate with the fewest
interchanges is declared the winner (Black 1958). Determining a Dodgson winner
is a rather complicated procedure that is NP-hard and in which the time required
to determine the winner is polynomial (Caragiannis et al. 2012). Like Bucklin’s, this
method also works with full preference information.
City A needs only 2 swaps to win the election, whereas cities B and C both need
16 swaps, and city D needs 8. City A is the Dodgson winner as it needs the least
number of swaps to win the election.
5. Similar to Baldwin’s, Nanson’s rule eliminates at each step the candidates with a
Borda score smaller than the average Borda score (calculated considering all the
candidates at a step). e Borda scores are then revised, taking only the remaining
candidates into account. e procedure repeats until a Borda winner can be
determined (Nanson 1883).
Condorcet-Inconsistent Methods
e rules summarized below are not guaranteed to select the Condorcet winner if one
exists.
1. Bucklin’s method elects the candidate that was ranked first by the majority of
voters as the winner. If there exists no such candidate, the candidate that was
ranked either first or second by the majority of voters is declared the winner. e
procedure continues, expanding the number of levels to consider every time a
majority winner cannot be determined, until one of the candidates has more than
half the number of votes. Bucklin requires complete preference information as
well (Hoag and Hallett 1926).
Consider the illustrative example above. According to Bucklin’s rule, one of the
cities would need to be ranked first by at least 16 of the committee members to
win the election. However, the number of times cities A, B, C, and D preferred as
the organizer is 8, 9, 7, and 6, respectively. us, at the second stage, we count the
total number of times a city was ranked either first or second. In the end, cities A
and D are tied as they take the first or second places 16 times, whereas city B (city
C) appeared in the top two only 13 (15) times.
2. Coombs (1964) proposed a recursive elimination method that discards at each
step the candidate who was ranked last the most number of times. is rank
scoring procedure repeats until someone can be declared winner.
Approval Voting is also a rank scoring rule that allows individuals to vote for a
predetermined number of candidates available. For instance, under k-Approval Voting,
each ballot contains at most k candidates, but the voter is not asked to rank them. e
candidate that appears the most in the ballots wins the election.

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 75
Note that Felsenthal and Tideman (2014) report that Nanson, Dodgson, and Coombs
are all vulnerable to monotonicity failure.
Non-traditional Methods
1. As an attempt to avoid cloning in elections, Schulze (2011) introduced a new
Condorcet-consistent method based on a weighted majority graph using a
best-path finding algorithm, which is solved in polynomial time. On a majority
graph, vertices represent the candidates and edges correspond to the relative
performance of pairs of candidates (Menton 2013). e method describes “the
strong paths from each winning candidate to every other candidate” (Menton
2013). One needs to first determine the number of voters who strictly prefer one
candidate over another. en, all possible paths to reach from one candidate to
another must be identified. e weakest link in a path determines the strength
of that path. If there are more than one possible path from one candidate to
another, then the path with the largest strength is chosen and it is called the
strongest path. If the path from one candidate (X) to another (Y) is stronger than
(no ties) the path backwards (Y to X), then X disqualifies Y. If X disqualifies every
other candidate, then X wins the election outright (Schulze 2003).
Consider the example introduced earlier. e directed graph given in Figure 2 is
constructed using the pairwise comparison matrix. ere are two possible ways
to reach B in this case: a direct path from A to B (with a strength of 17) and an
indirect path from A to C to D to B (with a strength of 18). e strength of the
latter is determined by the weakest link, which is A to C. e strongest path is the
one with the largest relative pairwise performance; that is, A~C~D~B. Here, “~”
denotes a direct link from one candidate to another. ere is only one path from
B to A with a strength of 16: B to C to D to A. Since the strength of the path from
A to B is larger than that of B to A, A disqualifies B.
FIGURE 2.
Weighted majority graph
for illustrated example
A B
D C
17
20
19
16
21
18
Table 4 compares the strongest beatpaths. City A wins the election since it loses none of
the beatpath comparisons.

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 76
From To
A B C D
A
A~B: 17
A~C~D~B: 18
= min(18,19,21)
A~C: 18
A~B~C: 17
= m in(17, 20)
A~C~D: 18
= min(18,19)
A~B~C~D: 17
= m in(17, 20,19)
BB~C~D~A: 16
= min(20,19,16) B~C: 20 B~C~D: 19
= min(20,19)
CC~D~A: 16
= m in(19,16)
C~D~B: 19
= min(19,21)
C~D~A~B: 16
= min(19,16,17)
C~D : 19
DD~A: 16
D~B: 21
D~A~B: 16
= m in(16,17)
D~A~C: 16 =
min(16 ,18)
D~B~C: 20 =
min(21,20)
D~A~B~C: 16 =
min(16 ,17, 2 0)
2. Tideman’s ranked pairs method is very similar to that of Schulze’s and tells one
“what edges are considered in what order, and whether and how the edges are
set in the election graph.” (Menton 2013). e method “requires the collective
ranking of the candidates to be consistent with the paired comparisons decided
by the largest and second largest margins, and then, if possible, with the paired
comparison decided by the third largest margin, and so on.” (Tideman 1987). e
candidates are first ordered from top to bottom based on margin of victory in
head-to-head elections. e ranking with the largest margin is determined and
locked. en, all rankings that contradict it is eliminated. e procedure continues
with the next largest margin of victory until one ranking remains (Levin and
Nalebuff 1995). For the example above, we start with B and D since they have
the largest margin of victory (21 – 9 = 12). e ranking DB is locked. e second
largest margin is between B and C (20 – 10=10) which lets us lock BC. Since DB
and BC, the ranking DC is also locked. Finally, we lock AC, AB, and DA. erefore,
D wins the election based on the final ranking: DABC.
3. Fallback Voting (FV) is an extension of Bucklin’s procedure that does not need
complete orderings, yet it does not allow for ties. FV combines Bucklin’s method
with approval voting (Erdélyi et al. 2015), and, as Brams and Sanver (2009)
summarized, it proceeds as follows.
First, voters rank a set of candidates they approve in order of preference. e set
of approved candidates is allowed to be empty or to consist of all the candidates
competing in the election. If a candidate was ranked first by a majority of voters,
this candidate is called a level 1 FV winner. If no candidate can be declared a level
1 winner, the candidate that is ranked either first or second by a majority of voters
is considered, and this candidate is declared the winner. If there are more than one
such candidates, then the candidate with the largest majority is called a level 2 FV
winner. If there is no level 2 winner, the voters descend—one level at a time—to
TABLE 4.
Strong Paths between
Each Candidate

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 77
lower and lower ranks of approved candidates, stopping when, for the first time,
one or more candidates are approved of by a majority of voters, or no more
candidates are ranked. If exactly one candidate receives majority approval, this
candidate is the FV winner. If more than one candidate receives majority approval,
then the candidate with the largest majority is the FV winner. If the descent
reaches the lowest rank of all voters and no candidate is approved of by a majority
of voters, the candidate with the most approval is the FV winner.
In the illustrative example, since none of the candidates take the majority of the
votes, a level 1 FV winner is not found. Next, we add the second-rank counts
and see that cities A and D are tied. Descending to the third level leads city D to
victory since it appears in the top three 26 times, once more than city A. Hence,
city D is a level 3 FV winner.
Black, Copeland, Dodgson, Schulze, Tideman, Nanson, and Baldwin choose the
Condorcet winner if one exists. One disadvantage of Condorcet arises when the group
decision is not transitive, even though the individual preferences are (Mattei 2012). Most
of the traditional methods enjoy completeness; however, it is often highly impractical
to ask individuals compare alternatives in pairwise fashion (a preference or a tie). us,
in customer satisfaction surveys, where there are too many alternatives, respondents
are usually asked to rank a subset of them. is avoids cognitive complexity and
waste of time, yet results in incomplete preferential votes. FV is designed to work with
incomplete information and asks voters to select a set of candidates they approve and
then rank them (Brams and Sanver 2009). e social choice literature on voting rules is
expanding continually and alternative methods are being introduced. Recently, Camps
et al. (2013) provided a continuous rating method for the social acceptance of different
alternatives in case the individuals do not express a comparison between every pair of
alternatives available or they provide an ordered list restricted to a subset of the most
preferred options.
Finally, we list below our main assumptions that will provide us flexibility when
interpreting the results in Section 5:
1. e respondents did not choose strategically; he/she is not be interested in what
other respondents think or how they decide. In short, the voters are assumed to
be sincere.
2. When employing the traditional methods, we assumed that the ballots are
completely filled.
3. Since multiple winners would not be an issue, we did not work through a
tiebreaking procedure.
Results and Discussion
e survey question we considered asked passengers to rank, based on their
preferences, the five most important SQ factors listed in Table 5. ey were allowed to
report incomplete rankings; however, fewer than 2% of participants provided a ranking

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 78
with less than five SQ factors. In total, 26 SQ factors were assessed to determine those
of higher priority—based on individual rankings—for each line and year.
TABLE 5.
SQ Factors
Waiting time Lighting
Access to stations Fares
Security at stations Ventilation systems in cars
Security in cars Moving stairways/walkways and escalators
Attitude of security personnel Token machines
Travel (in-vehicle) time Working conditions of turnstiles
Crowdedness in cars Comfort level at stations
Cleanliness of stations Transport information systems
Timeliness of cars Notifications at stations in case of delay
Information systems at stations Notifications in cars in case of delay
Token s ale Transfer fares
Cleanliness of cars Notifications on website
Noise and vibration in cars Notifications from phone line
We used several traditional voting procedures and three recently-proposed voting
procedures (Schulze, Tideman, and Fallback Voting) to determine the highly-prioritized
criteria by line and year. We reported the first, second, and third priorities identified
using the methods explained in Section 4. As mentioned earlier, traditional procedures
can be mainly grouped into two categories with respect to Condorcet-consistency. It
would not be surprising to see that two traditional methods from different categories
chose different candidates as winners. Yet, they agreed with each other at almost
every instance, as seen in Tables 6 and 7, when only the primary (first, second, or third)
priorities were considered. In contrast with other traditional methods, Dodgson and
Simpson chose security at stations as the third priority for the M2 line in 2012. Hence,
we combined the results of the traditional methods other than Borda in one table.
Note that the ballots in our study are truncated. e passengers ranked, at most, five
of the SQ factors available in order of importance. When employing the traditional
methods, we assumed that the voters strictly ranked the first five candidates and
they were indifferent with the rest, which let us work with completely-filled ballots in
return. Investigating the second and third priorities reveals that the Borda method is
significantly affected by this assumption; there are nine such instances on which Borda
and the other traditional methods do not agree. Regarding the first priorities, the Borda
method and the traditional methods disagree only for M1 in 2013. e former favors
fares, whereas the latter favors waiting time in that case. at is, when the first priorities
are considered, these two clusters of methods differ from each other less significantly
compared to the case when the second-ranked or third-ranked priorities are taken into
account.
Waiting time appears to be a consistent problem for M2 line. In both years, this criterion
is observed as a first priority for M2 passengers. Another interesting finding belongs to
F1 and T4 lines. e priorities of F1 and T4 passengers changed through survey years.

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 79
Waiting time and crowdedness in cars appear to be the most important SQ factors on
which decisionmakers should focus given the survey results of 2013. As for the second
and third priorities, Borda and the other traditional methods point out different criteria
at almost every instance for T1, M4, and F1 lines.
TABLE 6.
Passenger
Priorities by Line
and Year – Borda
Priority Yea r Rail Transit Line
T1 T4 M1 M2 M4 F1
First
2013 Crowdedness
in cars
Crowdedness
in cars Fares Waiting time Waiting time Crowdedness
in cars
2012 Crowdedness
in cars Waiting time Fares Waiting time Fares
Second
2013 Fares Fares Waiting time Crowdedness
in cars
Security at
stations Fares
2012 Fares Security at
stations
Crowdedness
in cars Tra vel ti me Crowdedness
in cars
ird
2013 Ventilation
systems in cars Waiting time Travel time Fares Crowdedness
in cars
Access to
stations
2012 Waiting time Crowdedness
in cars Waiting time Cleanliness of
cars Waiting time
TA B L E 7.
Passenger
Priorities by
Line and Year –
Other Traditional
Methods
Priority Yea r Rail Transit Line
T1 T4 M1 M2 M4 F1
First
2013 Crowdedness
in cars
Crowdedness
in cars Waiting time Waiting time Waiting time Crowdedness
in cars
2012 Crowdedness
in cars Waiting time Fares Waiting time Fares
Second
2013 Fares Fares Fares Crowdedness
in cars
Access to
stations Fares
2012 Waiting time Security at
stations
Crowdedness
in cars Tra vel ti me Waiting time
ird
2013 Waiting time Waiting time Tr avel time Fares Security at
stations Waiting time
2012 Fares Crowdedness
in cars Waiting time Cleanliness of
cars*
Crowdedness
in cars
* Excluding Dodgson and Simpson procedures.

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 80
As seen in Tables 8 and 9, in 13 instances, Tideman, Schulze, and FV disagree with each
other when determining a high-priority SQ factor. In 2013, in contrast with FV, Tideman
and Schulze favored waiting time and crowdedness in cars over fares as the first priority,
for the M1 and F1 lines, respectively. On the other hand, FV and the Borda procedure
addressed different SQ factor as a priority at a total of five instances (T1, M2, and F1
lines); however, only the orderings differ for M2 and F1 lines. us, these two methods
refer to the same set of SQ factors for both of these lines in a given year when ordering
is overlooked. is is not the case when the results of FV are compared with those
from the Tideman and Schulze methods. Distinctions occur mostly for the second and
third priorities in 2013 (M1, M4, and F1 lines). e major difference in the set of highly-
prioritized criteria in 2012 is observed for M2 line; FV elects cleanliness of cars (travel
time) as the second (third) priority, whereas both Tideman and Schulze elect travel time
(security at stations) as the second (third) priority.
In 2013, the methods are quite consistent regarding first priorities. For M1 line, FV and
Borda elect fares, whereas Tideman and Schulze favor waiting time along with the
other traditional methods. Similarly, for F1 line, all the methods excluding FV elects
crowdedness in cars. On the other hand, the methods lead exactly to the same set of
primary priorities with subtle differences in the ordering for F1 in 2012.
As mentioned in Section 4, it is not always possible to find an FV winner. For example,
distinctions were detected between the traditional methods and the recent methods
in 2012 for M1 and F1 lines. e passengers of these lines prioritized fares above others.
However, we were not able to determine an FV winner for M1 line even after the
first five ranks were considered in 2012. us, fares is the SQ factor that has the most
approvals among the others in that case and is also the 5-approval winner.
TABLE 8.
Passenger
Priorities by
Line and Year
– Fallback
Priority Yea r Rail Transit Line
T1 T4 M1 M2 M4 F1
First
2013 Crowdedness
in cars
Crowdedness
in cars Fares Waiting time Waiting time Fares
2012 Crowdedness
in cars Waiting time Fares Waiting time Fares
Second
2013 Fares Fares Waiting time Crowdedness
in cars
Security at
stations
Crowdedness
in cars
2012 Fares Security at
stations
Crowdedness
in cars
Cleanliness of
cars
Crowdedness
in cars
ird
2013 Waiting time Waiting time Tr avel time Fares Crowdedness
in cars
Access to
stations
2012 Waiting time Crowdedness
in cars Waiting time Tra vel ti me Waiting time

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 81
TABLE 9.
Passenger
Priorities by
Line and Year
– Tideman and
Schulze
Priority Yea r Rail Transit Line
T1 T4 M1 M2 M4 F1
First
2013 Crowdedness
in cars
Crowdedness
in cars Waiting time Waiting time Waiting time Crowdedness
in cars
2012 Crowdedness
in cars Waiting time Fares Waiting time Fares
Second
2013 Fares Fares Fares Crowdedness
in cars
Access to
stations Fares
2012 Waiting time Security at
stations
Crowdedness
in cars Tra vel ti me Waiting time
ird
2013 Waiting time Waiting time Tr avel time Fares Security at
stations Waiting time
2012 Fares Crowdedness
in cars Waiting time Security at
stations
Crowdedness
in cars
e average level of satisfaction for each SQ factor can be seen in Table 10. e number
in parenthesis shows the rank of an SQ factor among all the others based on its
satisfaction level for a given line and year. Crowdedness in cars, with which customers
from all lines are dissatisfied, was found to be a high-priority SQ factor in all lines
with two exceptions: M1 in 2013 and M2 in 2012. us, service providers should focus
resources on improving this SQ factor to significantly increase ridership. On the other
hand, even though the satisfaction level for waiting time was relatively high for F1 line
in 2012, it was addressed as a third priority for this line that year. is SQ factor has a
lower satisfaction level whenever it is highly prioritized. us, spending time and/or
money on its improvement can also enhance the overall passenger satisfaction. Even
though waiting time was perceived as a highly important SQ factor by passengers, travel
time (in-vehicle time) was prioritized in 2012 and 2013 by only M2 and M1 passengers,
respectively. Yet, the passengers in these cases seem to be satisfied with this SQ factor
as seen in Table 10.
e frequency distributions for overall customer satisfaction levels, which should
definitely be taken into account to make a better conclusion, are given in Figure 3. e
percentage of extremely satisfied passengers shows a significant increase in both M2
and T4 lines from 2012 to 2013. erefore, one should compare not only the rankings,
but also the average satisfaction levels of an SQ factor from different years since the
change in the overall satisfaction may be attributed to sample-based differences in these
years rather than a significant increase in customer satisfaction. For instance, waiting
time should be improved in M2 line as it is a high-priority SQ factor with an increasing
satisfaction level (from 4.82 to 5.04) but a decreasing relative satisfaction (from 7th to
14th). As for T4 line, crowdedness in cars definitely needs attention since it appeared as
a high-priority SQ factor with a decreasing satisfaction level despite the increase in its
overall satisfaction level.

Journal of Public Transportation, Vol. 20, No. 1, 2017 82
Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
TABLE 10. Satisfaction Levels for Highly-Prioritized SQ Factors
Rail Transit Line
T1 T4 M1 M2 M4 F1
2012 2013 2012 2013 2012 2013 2012 2013 2012 2013 2012 2013
Waiting time 4. 31 (19) 4.41 (14) 4.58 (19) 4.9 (18) 4.6 3 (19) 4 .67 (16) 4.82 (7) 5.04 (14) N/A 5.53 (8) 5.12 (5) 4.92 (1)
Access to stations 4.61 (8) 4.53 (9) 4.81 (7) 5.13 (11) 4.82 (12) 4.75 (9) 4.79 (13) 5.06 (11) N/A 5.34 (20) 5.12 (6) 4.73 (9)
Security at stations 4.58 (10) 4. 33 (16) 4.74 (14) 5.16 (8) 4.75 (16) 4.63 (17) 4.69 (19) 5 .01 (16) N/A 5 .49 (14) 4.91 (14) 4.71 (12)
Travel (in-vehicle) time 4. 37 (17) 4.2 6 (18) 4. 67 (18) 4.82 (19) 4.95 (6) 4.87 (2) 4.96 (1) 5 .06 (10) N/A 5.52 (1 0) 5.14 (3) 4.57 (18)
Crowdedness in cars 2.40 (24) 2.94 (22) 4.00 (25) 3.69 (26) 3.47 (26) 3.92 (23) 3.93 (25) 4.09 (25) N/A 5.11 (24) 3.58 (25) 3.79 (25)
Cleanliness of cars 4.57 (11) 4.58 (3) 4.79 (11) 5 .10 (13) 4. 76 (15) 4.73 (11) 4.81 (8) 5.0 0 (18) N/A 5.4 8 (15) 4.92 (13) 4. 62 (17)
Fares 3. 54 (23) 3.65 (21) 4.19 (24) 3.97 (25) 3.68 (25) 3.83 (24) 4.11 (23) 4.16 (24) N/A 5.38 (18) 3.64 (24) 3.76 (26)
Ventilation sys. in cars 3.71 (22) 4. 21 (20) 4.51 (20) 4.73 (20) 4. 37 (21) 4.49 (20) 4 .45 (21) 4. 68 (21) N/A 5.25 (22) 4.42 (22) 4.51 (20)

Journal of Public Transportation, Vol. 20, No. 1, 2017 83
Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
FIGURE 3. Distribution of overall satisfaction levels by year and line

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 84
As stated previously, the purpose of this study was to provide a relatively simple way
to decide which SQ factors perceived more important by RT passengers considering
preference rankings. In line with dell’Olio et al. (2011) and Celik et al. (2014), we found
that waiting time is a highly-prioritized SQ factor on which the service provider should
focus resources. Similarly, crowdedness in cars was also addressed as a highly important
SQ factor in de Oña et al. (2014) and Aydin et al. (2015). Givoni and Rietveld (2007) and
Brons et al. (2009) emphasized the importance of access to stations, whereas Bhat and
Sardesai (2006) highlighted travel (in-vehicle) time as an important SQ factor. However,
in our case, these two factors were favored by the respondents at considerably few
instances. However, the service provider should never overlook their importance.
Assuming that the respondents had no knowledge of the ranking patterns, there was no
concern about the effects of manipulation or bribery on the overall conclusion (Mattei
2012). On the other hand, even though the results of the traditional methods coincide
with those obtained from the FV, the reader should note that the complete preferential
votes could change the outcome.
Incorporating preference rankings into the analysis avoids missing valuable information.
Such information should not easily be disposed, as this may distort the overall
conclusion. Employing k-approval voting could also have been considered; however, the
relative ranking of the five most favored SQ factors would have been ignored in that
case. For instance, one could employ 5-approval voting in this case, which would ignore
the ordering of SQ factors unlike FV and different SQ factors might be declared winners.
Note also that we do not report the priorities by year using the overall data, as some
of the voting rules employed might suffer from the multiple districts paradox, which
describes the case in which a candidate that won an election in distinct electoral
districts is not declared the winner when the districts are joined together (Young 1974;
Plassmann and Tideman 2014).
Conclusions
Offering high-quality service in PT allows passengers to avoid traffic congestion and
noise, especially in big and crowded cities such as Istanbul. Hence, determining the
key SQ factors that passengers value most is an essential task for PT service providers
and policymakers. is paper reports on the results obtained by analyzing data from
a passenger satisfaction survey conducted annually by Istanbul Public Transportation
Co. Several voting rules available in the literature were employed and compared to
decide which SQ factors would be perceived as more important by RT passengers.
e findings indicate that improving waiting time, crowdedness in cars, and fares can
increase passenger satisfaction with RT services. Since a considerably high percentage
of passengers prefer RT to commute or for work-related activities, they ranked
waiting time and crowdedness in cars higher than most of the other SQ factors; they
would rather get to work on time, comfortably. Hence, policymakers should focus on
improving the comfort in cars and increasing the frequency of cars to decrease waiting
time.

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 85
In addition, customers reported that they were dissatisfied with fares and they favored
it as one of the most important SQ factors. However, the median household income
level appears to increase through survey years. Even though this does not clearly reflect
purchasing power, the service provider should either find a way to decrease fares or
increase the level of service customers are provided to a level such that price paid is not
seen as a problem. is would be a rational decision considering more than half of the
respondents pay full fare.
Arrow’s Impossibility eorem states that there is no ideal scheme to decide a winner
in an election (Kelly 1978); however, following the footsteps of Camps et al. (2013) and
re-analyzing the data in hand may lead decisionmakers to more reliable results even
though the overall conclusion presented here repeats. is is highly recommended,
especially when the decisionmaker would like to learn about how much social
acceptance a certain criterion is provided rather than whether it is a primary priority
for RT passengers. Note that passengers are one of the stakeholders in transportation
planning. Other stakeholders such as motorists and transit agencies should also be
brought into the discussion to make better decisions.
A possible avenue for future research is to investigating the vulnerability of the methods
employed to sampling procedures. We assumed that the samples in each year are the
best representations of the population. However, one might be interested in checking
this, especially whether monotonicity property is violated, since Coombs, Nanson, and
Dodgson are all vulnerable to monotonicity failure (Felsenthal and Tideman 2014).
Another possibility for future research lies within a machine-learning setting that
finds rank orderings, as mentioned in Dobrska et al. (2011). Investigating the effect
of demographics on the priorities did not provide an enhancement of the results
mentioned in Section 5, mainly due to the similitude of RT lines from this perspective.
Further investigation using multivariate techniques such as multiple discriminant
analysis might be considered to assess the importance of demographics.
Acknowledgments
e authors express their utmost gratitude to Istanbul Public Transportation Co.
(İstanbul Ulaşım A.Ş.) for their understanding, support, and the data provided.
References
Andaleeb, S. S., M. Haq, and R. I. Ahmed. 2007. “Reforming Innercity Bus Transportation
in a Developing Country: A Passenger-Driven Model.” Journal of Public
Transportation, 10 (1) .
Anderson, R. D., R. D. Mackoy, V. B. ompson, and G. Harrell. 2004. “A Bayesian
Network Estimation of the Service-Profit Chain for Transport Service Satisfaction.”
Decision Sciences, 35(4), 665-689.
Anderson, S.P., A. de Palma, and J. F. isse. 1992. Discrete Choice eory of Product
Differentiation. e MIT Press, London, England.

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 86
Awasthi, A., S. S. Chauhan, H. Omrani, and A. Panahi. 2011. “A Hybrid Approach Based
on SERVQUAL and Fuzzy TOPSIS for Evaluating Transportation Service Quality.”
Computers & Industrial Engineering, 61: 637–646.
Aydin, N., E. Celik, E., and A. T. Gumus. 2015. “A Hierarchical Customer Satisfaction
Framework for Evaluating Rail Transit Systems of Istanbul.” Transportation Research
Part A: Policy and Practice,77: 61-81.
Berry, L. L., V. A. Zeithaml, and A. Parasuraman. 1990. “Five İmperatives for İmproving
Service Quality.” Sloan Management Review, Summer: 9-38.
Bhat, C. R., and R. Sardesai. 2006. “e İmpact of Stop-Making and Travel Time
Reliability on Commute Mode Choice.” Transportation Research Part B:
Methodological, 40(9): 709-730.
Black, D. 1958. e eory of Committees and Elections. Cambridge University Press,
Cambridge.
Brams, S., and R. Sanver. 2009. “Voting Systems that Combine Approval and Preference.”
In e Mathematics of Preference, Choice, and Order: Essays in Honor of Peter C.
Fishburn. Springer: 215–237.
Brons, M., M. Givoni, and P. Rietveld. 2009. “Access to Railway Stations and İts Potential
in İncreasing Rail Use.” Transportation Research Part A: Policy and Practice, 43(2):
136 -149.
Camps, R., X. Mora, and L. Saumell. 2013. “A Continuous Rating Method for Preferential
Voting. e İncomplete Case.” Social Choice and Welfare,4 0 : 1111–42 .
Caragiannis, I., J. A. Covey, M. Feldman, C. M. Homan, C. Kaklamanis, N. Karanikolas, A.
D. Procaccia, and J. S. Rosenschein. 2012. “On the Approximability of Dodgson and
Young Elections.” Artificial Intelligence,187: 31–51.
Cascetta, E., and A. Cartenì. 2014. “A Quality-Based Approach to Public Transportation
Planning: eory and a Case Study.” International Journal of Sustainable
Transportation, 8(1) : 8 4-10 6 .
Celik, E., N. Aydin, and A. T. Gumus. 2014. “A Multi Attribute Customer Satisfaction
Evaluation Approach for Rail Transit Network: A Real Case Study for Istanbul,
Turkey.” Transport Policy, 36: 283-293.
Celik, E., O. N. Bilisik, M. Erdogan, A. T. Gumus, and H. Baracli. 2013. “An İntegrated
Novel İnterval Type-2 Fuzzy MCDM Method to İmprove Customer Satisfaction in
Public Transportation for Istanbul.” Transportation Research Part E: Logistics and
Transportation Review, 58: 28-51.
Cirillo, C., L. Eboli, and G. Mazzulla. 2011. “On the Asymmetric User Perception of
Transit Service Quality.”International Journal of Sustainable Transportation, 5(4):
216-232.
Condorcet, M. J. A. N. de Caritat. 1847. “Sur la forme des Zlections 1789.” Oeuvres
9: 285–330. English translation in McLean, I., and F. Hewitt (eds.) “Condorcet—
Foundations of Social Choice and Political eory”: 169–189.

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 87
Coombs, C. H. 1964. A eory of Data. Wiley, New York
Copeland, A. H. 1951. A Reasonable Social Welfare Function. University of Michigan,
Department of Mathematics, Seminar on Applications of Mathematics to the Social
Sciences.
Cox, G. W. 1989. “Undominated Candidate Strategies under Alternative Voting Rules.”
Mathematical and Computer Modelling, 12(4/5): 451–459.
de Borda, J. ., 1781. “Mémoire sur les élections au scrutin.” Histoire de l’Académie Royale
des Sciences, Paris.
de Oña, J., R. de Oña, and F. J. Calvo. 2012. “A Classification Tree Approach to İdentify
Key Factors of Transit Service Quality.” Expert Systems with Applications, 39(12):
11164–71.
de Oña, R., L. Eboli, and G. Mazzulla. 2014. “Key Factors Affecting Rail Service Quality in
Northern Italy: A Decision Tree Approach.” Tra n spor t , 29(1): 75-83.
de Oña, R., G. López, F. J. D. de los Rios, and J. de Oña, J. 2014. “Cluster Analysis
for Diminishing Heterogeneous Opinions of Service Quality Public Transport
Passengers.”Procedia–Social and Behavioral Sciences,162: 459-466.
dell'Olio, L., A. Ibeas, and P. Cecín. 2010. :Modelling User Perception of Bus Transit
Quality." Transport Policy, 17 (6): 388-397.
dell'Olio, L., A. Ibeas, and P. Cecín. 2011. “e Quality of Service Desired by Public
Transport Users.” Transport Policy, 18(1): 217-227.
Dobrska, M., H. Wang, and W. Blackburn. 2011. “Data-driven Rank Ordering – A
Preference-based Comparison Study.” International Journal of Computational
Intelligence Systems, 4(2): 142–152.
Eboli, L., and G. Mazzulla. 2007. “Service Quality Attributes Affecting Customer
Satisfaction for Bus Transit.” Journal of Public Transportation, 10(3): 21-34.
Eboli, L., and G. Mazzulla, G. 2008a. “A Stated Preference Experiment for Measuring
Service Quality in Public Transport.” Transportation Planning and Technology, 31(5):
509-523.
Eboli, L., and G. Mazzulla. 2008b. “Willingness-to-Pay of Public Transport Users for
Improvement in Service Quality.” European Transport, 38: 107–118.
Eboli, L., and G. Mazzulla. 2009. “A New Customer Satisfaction İndex for Evaluating
Transit Service Quality.” Journal of Public Transportation, 12(3): 21–37.
Eboli, L., and G. Mazzulla. 2010. How to Capture the Passengers’ Point of View on a
Transit Service through Rating and Choice Options.” Transport Review, 30: 4.
Eboli, L., and G. Mazzulla. 2011. “A Methodology for Evaluating Transit Service Quality
Based on Subjective and Objective Measures from the Passenger’s Point of View.”
Transport Policy 18: 172–181.

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 88
Eboli, L., and G. Mazzulla. 2012. “Structural Equation Modelling for Analysing Passengers’
Perceptions about Railway Services.”Procedia–Social and Behavioral Sciences,54:
96-10 6 .
Eğecioğlu, Ö., and A. E. Giritligil. 2011. “e Likelihood of Choosing the Borda-Winner
with Partial Preference Rankings of the Electorate.” Journal of Modern Applied
Statistical Methods, 10(1): 349-61.
Erdélyi, G., M. R. Fellows, J. Rothe, and L. Schend. 2015. “Control Complexity in Bucklin
and Fallback Voting: A eoretical Analysis.” Journal of Computer and System
Sciences, 81: 632-660.
Felsenthal, D.S., and N. Tideman. 2014. “Interacting Double Monotonicity Failure with
Direction of İmpact under Five Voting Methods.” Mathematical Social Sciences, 67:
57-66.
Filipović, S., S. Tica, P. Živanović, and B. Milovanović. 2009. “Comparative Analysis of
the Basic Features of the Expected and Perceived Quality of Mass Passenger Public
Transport Service in Belgrade.” Transport , 24(4): 265-273 .
Gerçek, H., B. Karpak, and T. Kılınçaslan. 2004. “A Multiple Criteria Approach for the
Evaluation of the Rail Transit Networks in Istanbul.” Transportation, 31(2): 203-2 28.
Givoni, M., and P. Rietveld. 2007. “e Access Journey to the Railway Station and its Role
in Passengers’ Satisfaction with Rail Travel.”Transport Policy,14(5): 357-365.
Gronau, W., and A. Kagermeier. 2007. “Key Factors for Successful Leisure and Tourism
Public Transport Provision.” Journal of Transport Geography 15(2): 127-135.
Hassan, M. N., Y. E. Hawas, and K. Ahmed. 2013. “A Multi-Dimensional Framework for
Evaluating Transit Service Performance.” Transportation Research Part A: Policy and
Practice, 50: 47-61.
Hoag, C. G., and G. H. Hallett. 1926. Proportional Representation. Macmillan, New York.
Hwang C. L., and M. Lin. 1987. Group Decision-Making: Mathematical Models. Springer-
Verlag, Berlin.
Jen, W., R. Tu, and T. Lu. 2011. “Managing Passenger Behavioral İntention: An İntegrated
Framework for Service Quality, Satisfaction, Perceived Value, and Switching
Barriers.”Transportation,38(2): 321-342.
Kelly, J. S. 1978. Arrow Impossibility eorems. Academic Press, New York.
Le-Klähn, D. L., C. M. Hall, and R. Gerike. 2014. “Analysis of Visitor Satisfaction with
Public Transport in Munich.” Journal of Public Transportation, 17(3).
Levin, J., and B. Nalebuff. 1995. “An Introduction to Vote-Counting Schemes.” Journal of
Economic Perspectives, 9(1): 3-26.
Mattei, N., 2012. “Decisionmaking under Uncertainty: eoretical and Empirical
Results on Social Choice, Manipulation, and Bribery.” Doctoral dissertation,
ProQuest (AAI3579270).

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 89
Menton, C. 2013. “Attacking and Defending Popular Election Systems.” Doctoral
dissertation, ProQuest (AAI3561011).
Mouwen, A., and P. Rietveld. 2013. “Does Competitive Tendering Improve Customer
Satisfaction with Public Transport? A Case Study for the Netherlands.”
Transportation Research Part A: Policy and Practice, 51: 29-45.
Nanson, E. 1883. “Methods of Elections.” Transactions and Proceedings of the Royal
Society of Victoria, 19: 197-240.
Nathanail, E. 2008. “Measuring the Quality of Service for Passengers on the Hellenic
Railways.” Transportation Research Part A: Policy and Practice, 42: 48-66.
Nwachukwu, A. A. 2014. “Assessment of Passenger Satisfaction with Intra-City Public
Bus Transport Services in Abuja, Nigeria.”Journal of Public Transportation,17(1): 5.
Plassmann, F., and T. N. Tideman. 2014. “How Frequently Do Different Voting Rules
Encounter Voting Paradoxes in ree-Candidate Elections?” Social Choice and
Welfare, 42: 31-75.
Redman, L., M. Friman, T. Gärling, and T. Hartig. 2013. “Quality Attributes of Public
Transport that Attract Car Users: A Research Review.” Transport Policy, 25: 119-127.
Schulze, M. 2003. “A New Monotonic and Clone-İndependent Single-Winner Election
Method.” Voting Matters, 17: 9-19.
Schulze, M., 2011. “A New Monotonic, Clone-İndependent, Reversal Symmetric, and
Condorcet-consistent Single-Winner Election Method.” Social Choice and Welfare,
36(2): 267-303.
Tideman, T. N. 1987. “Independence of Clones as a Criterion for Voting Rules.” Social
Choice and Welfare, 4(3): 185-206.
Turkish Statistical Institute. 2015. www.tuik.gov.tr, accessed July 9, 2015.
Tyrinopoulos, Y., and C. Antoniou. 2008. “Public Transit User Satisfaction: Variability and
Policy İmplications.” Transport Policy, 15: 260–272.
Van Ryzin, G. G., and S. Immerwahr. 2007. “Importance-Performance Analysis of Citizen
Satisfaction Surveys.” Public Administration, 85(1): 215-226.
Young, P. 1974. An Axiomatization of Borda’s Rule.” Journal of Economic eory, 9: 43-52.
About the Authors
E I (isiklie@itu.edu.tr) is currently an instructor in the Department of
Industrial Engineering at the Istanbul Technical University in Turkey. He received BSc.
and MSc. degrees in Mathematical Engineering and Business and Administration from
Istanbul Technical University in 2005 and 2007, respectively, and a Ph.D. degree in
Industrial and Systems Engineering from Wayne State University in Detroit. He taught
several graduate and undergraduate level courses during his Ph.D. studies and currently

Identifying Key Factors of Rail Transit Service Quality: An Empirical Analysis for Istanbul
Journal of Public Transportation, Vol. 20, No. 1, 2017 90
is conducting projects related to decisionmaking, statistics, data mining, mathematics,
and public transportation.
N A (nzraydin@yildiz.edu.tr) is currently an Assistant Professor in the
Department of Industrial Engineering at the Yildiz Technical University in Istanbul,
which he joined in 2014. He received BSc. and MSc. Degrees in Industrial Engineering
from Yildiz Technical University in 2005 and 2007, respectively, a Ph.D. degree in
Industrial and Systems Engineering from Wayne State University. He worked as a
researcher andtaught several graduate and undergraduate level courses during his Ph.D.
studies and currently is conducting projects related to optimization, mathematical
modeling, humanitarian logistics, decisionmaking, supply chain management, and
public transportation.
E C (erkancelik@munzur.edu.tr) is an Assistant Professor in the Department
of Industrial Engineering at Munzur University in Turkey. He received BSc. and MSc.
degrees in Industrial Engineering from Selcuk University in 2008 and 2011, respectively,
and a Ph.D. degree from the Department of Industrial Engineering at Yildiz Technical
University in 2015. His research interests are in decision analysis, public transportation,
humanitarian logistic, and fuzzy sets.
A T G (ataskin@yildiz.edu.tr) is an Associate Professor in the
Department of Industrial Engineering at Yildiz Technical University. She received
BSc. and MSc. degrees in Industrial Engineering from Yildiz Technical University,
an MBA from Istanbul Technical University, and a Ph.D. in Industrial Engineering
from Yildiz Technical University. She completed postdoctoral research at Zaragoza
Logistics Center in Spain, and her research interests are in supply chain management,
public transportation, production and inventory systems, decision analysis, artificial
intelligence, and fuzzy logic applications in industrial engineering and management
sciences.