See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/289500577

Passenger Perspectives in Railway Timetabling: A Literature Review

Article  in  Transport Reviews · January 2016

DOI: 10.1080/01441647.2015.1113574

CITATIONS READS

82 2,785

3 authors:

Jens Parbo Otto Anker Nielsen

Technical University of Denmark Technical University of Denmark
3 PUBLICATIONS   163 CITATIONS    150 PUBLICATIONS   2,121 CITATIONS   

SEE PROFILE SEE PROFILE

Carlo Giacomo Prato

The University of Queensland
100 PUBLICATIONS   3,556 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Optimal pricing model and strategy of shared autonomous vehicles considering accessibility effect View project

TRANSPORT (ISSN 1648-4142 / eISSN 1648-3480) View project

All content following this page was uploaded by Otto Anker Nielsen on 06 November 2017.

The user has requested enhancement of the downloaded file.

 Transport Reviews

ISSN: 0144-1647 (Print) 1464-5327 (Online) Journal homepage: http://www.tandfonline.com/loi/ttrv20

Passenger Perspectives in Railway Timetabling: A

Literature Review

Jens Parbo, Otto Anker Nielsen & Carlo Giacomo Prato

To cite this article: Jens Parbo, Otto Anker Nielsen & Carlo Giacomo Prato (2016):
Passenger Perspectives in Railway Timetabling: A Literature Review, Transport Reviews, DOI:
10.1080/01441647.2015.1113574

To link to this article: http://dx.doi.org/10.1080/01441647.2015.1113574

Published online: 05 Jan 2016.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=ttrv20

Download by: [DTU Library] Date: 06 January 2016, At: 14:00

 TRANSPORT REVIEWS, 2016
http://dx.doi.org/10.1080/01441647.2015.1113574

Passenger Perspectives in Railway Timetabling: A Literature

Review
Jens Parbo, Otto Anker Nielsen and Carlo Giacomo Prato
Department of Transport, Technical University of Denmark, Bygningstorvet 116B, 2800 Kgs, Lyngby, Denmark

ABSTRACT ARTICLE HISTORY

When looking at railway planning, a discrepancy exists between Received 11 February 2015
planners who focus on the train operations and publish fixed Accepted 25 October 2015
railway schedules, and passengers who look not only at the
KEYWORDS
schedules but also at the entirety of their trip, from access to railway planning; railway
Downloaded by [DTU Library] at 14:01 06 January 2016

waiting to on-board travel and egress. Looking into this operations; passengers’
discrepancy is essential, as assessing railway performances by perceptions; timetable
merely measuring train punctuality would provide an unfair picture robustness; timetable
of the level of service experienced by passengers. Firstly, optimisation
passengers’ delays are often significantly larger than the train
delays responsible for the passengers to be late. Secondly, trains’
punctuality is often strictly related to too tight schedules that in
turn might translate into knock-on delays for longer dwelling times
at stations, trip delays for increased risk of missing transfer
connections, and uncertain assessment of the level of service
experienced, especially with fluctuating passenger demand. A key
aspect is the robustness of railway timetables. Empirical evidence
indicates that passengers give more importance to travel time
certainty than travel time reductions, as passengers associate an
inherent disutility with travel time uncertainty. This disutility may
be broadly interpreted as an anxiety cost for the need for having
contingency plans in case of disruptions, and may be looked at as
the motivator for the need for delay-robust railway timetables.
Interestingly, passenger-oriented optimisation studies considering
robustness in railway planning typically limit their emphasis on
passengers to the consideration of transfer maintenance. Clearly,
passengers’ travel behaviour is far more complex and multi-faceted
and thus several other aspects should be considered, as becoming
more and more evident from passenger surveys. The current
literature review starts by looking at the parameters that railway
optimisation/planning studies are focused on and the key
performance indicators that impact railway planning. The attention
then turns to the parameters influencing passengers’ perceptions
and travel experiences. Finally, the review proposes guidelines on
how to reduce the gap between the operators’ railway planning
and performance measurement on the one hand and the
passengers’ perception of the railway performance on the other
hand. Thereby, the conclusions create a foundation for a more
passenger-oriented railway timetabling ensuring that passengers
are provided with the best service possible with the resources
available.

CONTACT Jens Parbo [email protected]

© 2016 Taylor & Francis
 2 J. PARBO ET AL.

1 Introduction
Firstly, the purpose of the present literature review is to review the vast amount of papers
focusing on different aspects of designing a timetable with a train-oriented focus. Sec-
ondly, the review compares these approaches to the way passengers perceive and
value railway operations. Based on the gap between passengers’ perception of railway
operations and the way timetables are designed, suggestions for future research direc-
tions are outlined.
A discrepancy exists between how train-oriented railway operations are planned with
the main focus being on the trains and how passengers actually perceive and respond
to railway performances. The ideal and simplistic vision of railway operations is that
trains run according to the planned schedule. Transportation science has dedicated a
great deal of attention to methods guaranteeing that trains run on time regardless of
what passengers do. In reality however, disturbances to the timetable are frequent and
affect passengers. Contradicting desires to an effective usage of resources between oper-
Downloaded by [DTU Library] at 14:01 06 January 2016

ators and passengers mean that the optimal timetable for the operators could be far from
optimal for the passengers (Medeossi, Marchionna, & Longo, 2009; Schöbel & Kratz, 2009).
Addressing this discrepancy is crucial as passenger delays are often larger than the train
delays responsible for the passengers being delayed (Nielsen, Landex, & Frederiksen,
2008; Vansteenwegen & Van Oudheusden, 2007). Empirical evidence showed that passen-
ger on-time performance was up to ten percentage points below train punctuality during
peak hours, with the reasons being cancelled trains, missed transfers and/or route choice
adaptations (Nielsen et al., 2008).
For the passengers to rely on the timetable, travel time should be stable and kept at a
minimum. Typically, contracts with railway operators have explicit targets for trains’ on-
time performance and railway operators not providing punctual service face financial
penalties (Noland & Polak, 2002). To meet the punctuality goals, planners and researchers
have tried to make railway operations more robust. Several definitions of robustness sup-
porting timetable planning have been proposed, although a general definition of timeta-
ble robustness was recently noted as missing (De-Los-Santos, Laporte, Mesa, & Perea, 2012;
Dewilde, Sels, Cattrysse, & Vansteenwegen, 2011). Early robustness definitions focused on
the ability to absorb minor disruptions and the recoverability of the schedule (Bush, 2007;
Cacchiani, Caprara, & Fischetti, 2009; Medeossi et al., 2009; Salido, Barber, & Ingolotti, 2008;
Vromans, 2005). Later definitions also considered the trade-off between having a very tight
(nominal) timetable, with only minimum safety headway between subsequent trains, and
having a (robust) timetable that could absorb minor delays, thus reducing travel time
uncertainty (Dewilde et al., 2011, 2013; Schöbel & Kratz, 2009). The difference between
a nominal and a robust timetable was defined as the price of robustness (Schöbel &
Kratz, 2009).
On-time performance of the trains is not always a major passenger concern and it is
highly dependent on the network characteristics. In low-frequency networks, passengers
perceive on-time performance to be the most important service characteristic (Milan,
1996). However, in high-frequency networks, regularity (i.e. the ability to keep equal head-
ways between trains) is perceived as more important (Sun & Xu, 2012; Weston et al., 2006).
Although on-time performance is known to be among the most important factors influen-
cing mode choice (Carrasco, 2012), transit users are very unlikely to change mode choice in
 A LITERATURE REVIEW 3

response to delayed trains (Batley, Dargay, & Wardman, 2011). Instead, passengers start
adapting the route choice or departure time choice accordingly (Benezech & Coulombel,
2013).
Passengers’ perceptions of railway performances affect their travel behaviour (Nielsen,
2000) and hence addressing all attributes perceived as important by passengers is essen-
tial to improve railway timetables. Failing to do so and considering only a subset of attri-
butes could potentially result in an only apparent timetable enhancement obtained at the
expense of non-measured attributes highly relevant to passengers. In fact, it has been
shown that when travellers experience transit vehicles leaving early one out of ten
times, they tend to perceive the probability of transit vehicles leaving early as larger
than the de facto 10% (Rietveld, 2005). A reason that travellers often underestimate the
experienced quality is that low quality often coincides with peak periods, and they typi-
cally emphasise punctuality more during peak periods due to less flexible arrival times.
Treating peak and non-peak delays equally is thus not fair when looking at the number
of passengers affected and their arrival time flexibility (Vansteenwegen & Van Oudheus-
Downloaded by [DTU Library] at 14:01 06 January 2016

den, 2007). Planners designing timetables often focus on minimising travel time. Passen-
gers are taken into account by using passenger counts (e.g. Liebchen Schachtebeck,
Schöbel, Stiller, & Prigge 2010; Schöbel & Kratz, 2009), but in the design process, passen-
gers’ adapted travel behaviour as a result of a changed timetable is generally neglected.
Therefore, using old passenger counts yields an inaccurate picture of what can be
expected by the passengers. Passengers’ travel behaviour should be considered explicitly
in order to assess what impacts passengers can expect from future timetables. Recent
studies have emphasized that exhaustive passenger-oriented measurements are needed
to avoid that passengers are given a low priority in railway planning (Andersson, Peterson,
& Krasemann, 2013; Carrasco, 2012; Medeossi et al., 2009). For example, measurements of
passengers’ average delay would ensure that cases where some trains are delayed on
purpose will be avoided if they imply reductions in the level of service.
The current review is structured as follows. Section 2 presents research focused on
improving robustness-related attributes in railway planning from the planners’ perspec-
tive. Section 3 first outlines passenger-oriented optimisation of railway operations and
then examines how passengers actually perceive railway performances. Finally, Section
4 summarises the gap between the planners’ and the passengers’ perspectives and pro-
poses directions for future research within passenger-oriented railway planning.

2 Train-Oriented Railway Timetabling

Traditionally, the focus in railway planning is train-oriented. To meet the on-time require-
ments of the trains, railway plans need to be robust against delays. Since future delays are
unknown, planners try to approximate them, for example, by using historical data (Bene-
zech & Coulombel, 2013; Hooghiemstra & Teunisse, 1998; Kroon, Maróti, Helmrich,
Vromans, & Dekker, 2008; Takeuchi, Tomii, & Hirai, 2007; Tsuchiya et al., 2006). However,
Sels, Dewilde, Vansteenwegen, and Cattrysse (2012) suggested that only historical
delays that are expected to be present in a future timetable should be included.
Yaghini, Khoshraftar, and Seyedabadi (2013) used a neural network model and applied
it to the Iranian railways. This delay prediction made it easier for operators to create suit-
able timetables minimising delays, errors and problems for future railways. Milinković,
 4 J. PARBO ET AL.

Marković, Vesković, Ivić, and Pavlović (2013) developed a model based on a fuzzy petri net
model aimed at estimating delays through train dispatchers’ experience for railway net-
works where delays were not recorded regularly.
Planners have certain “tools” to make timetables robust against delays, for example,
adding time supplements, lowering heterogeneity (i.e. similarity in stopping patterns
and headways), finding optimal speed and reducing interdependencies between trains
(Goverde, 2010; Salido, Barber, & Ingolotti, 2012; Schöbel & Kratz, 2009). In the following,
the usage of these “tools” is described in detail as well as their impact on operations and
performance measurements. Table 1 provides a bird’s eye overview of which “tools” have
been used in the literature, and for each “tool” details whether it is explicitly considered, ✓,
implicitly considered (✓), or not considered at all, ✗. In Table 1, it is distinguished whether
the particular study applies an optimisation approach or the aim is more practice-oriented
and descriptive. The distinction should make it easier to choose which reference to consult
based on whether the reader is interested in thorough descriptions of certain character-
istics or particular algorithmic aspects.
Downloaded by [DTU Library] at 14:01 06 January 2016

2.1 Time Supplements

Time supplements are the extra time trains are assigned to the running time between two
stations, the dwelling time at intermediate stations or the layover time at the terminal
station. Time supplements are not to be confounded with buffer time, which is the time
added to the minimum headway between subsequently running trains (see Figures 1
and 2).

2.1.1 Travel time determination

Determining the published travel time is not straightforward. In practice, running a certain
trip would never result in the same travel time. Bush (2007) advocated for the use of the
85th percentile (according to historical travel time data) for the scheduled travel time. This
specific percentile was chosen because the railway company in question was said to
provide a decent level of service when being on-time in 85% of all trips. Choosing a

Figure 1. Time supplements.

 Table 1. Overview of reviewed literature on train-oriented railway planning
Train-related service characteristics addressed
Buffer Time Inter-
Studies Purpose Measure/definition developed time supplements Heterogeneity Speed dependencies Flexibility
Optimisation
Higgins et al. (1997) Reducing ✓ ✗ ✗ ✗ ✓ (✓)
interdependencies
Downloaded by [DTU Library] at 14:01 06 January 2016

Burkolter et al. (2005) Routing and train ✗ ✗ ✗ ✗ ✓ ✗

precedence
Sun and Hickman (2005) Recovery strategies (✓) ✗ (✓) (✓) ✗ (✓)
Liebchen et al. (2007) Allocation of time Price of robustness and ratio of delay ✓ ✓ ✗ ✗ ✗ ✗
supplements
Salido et al. (2008) Robustness of timetable Robustness indicator ✓ ✓ (✓) ✗ (✓) ✗
Cacchiani et al. (2009) Allocation of time ✓ ✓ ✗ ✗ (✓) ✗
supplements
Fischetti and Monaci (2009) Allocation of time Light robustness ✓ ✓ ✗ ✗ ✗ ✗
supplements
Schöbel and Kratz (2009) Delay absorption Price of robustness ✓ ✓ ✗ ✗ ✗ ✗
Caprara et al. (2010) Recovery strategies ✓ ✗ ✗ ✗ ✓ ✓
Goverde (2010) Delay propagation Delay propagation of initial delays ✓ ✓ ✗ ✗ ✓ ✗
Dewilde et al. (2011) Delay absorption ✓ ✗ ✗ ✗ ✗ ✗
Armstrong et al. (2012) Improve reliability ✓ ✗ ✗ ✗ ✗ ✗
De-Los-Santos et al. (2012) Stability against failures Network stability against intentional random ✗ ✗ ✗ ✗ ✓ ✓
infrastructure attacks
Gestrelius et al. (2012) Reducing ✗ ✗ ✗ ✗ ✓ ✗
interdependencies
Sels et al. (2012) Assigning trains to ✗ ✗ ✗ ✗ ✓ (✓)
platforms
Dewilde et al. (2013) Train spread Spread of trains ✓ ✓ ✗ ✗ ✓ (✓)

A LITERATURE REVIEW
Planning and Describing characteristics
Kikuchi and Vuchic (1982) Skipping stops ✗ ✗ ✓ (✓) (✓) ✗
Carey (1999) Train spread ✓ ✓ ✗ ✗ ✗ ✗
Huisman and Boucherie (2001) Punctuality Mean train delay and train delay probability ✓ ✓ ✗ ✗ ✗ ✗
Vromans (2005) Heterogeneity SSHR and SAHR ✓ ✗ ✓ ✗ ✗ ✗
Bush (2007) On-time performance (✓) ✗ ✗ ✓ ✗ ✗
Hofman et al. (2006) Recovery strategies ✓ ✗ ✗ ✗ (✓) (✓)
Kroon et al. (2007) Allocation of time Weighted Average distance of the allocated ✓ ✓ ✗ ✗ ✗ ✗
supplements time supplement

(Continued)

5
 6
J. PARBO ET AL.
Downloaded by [DTU Library] at 14:01 06 January 2016

Table 1. Continued.
Train-related service characteristics addressed
Buffer Time Inter-
Studies Purpose Measure/definition developed time supplements Heterogeneity Speed dependencies Flexibility
Yuan and Hansen (2007) Delay absorption ✓ ✓ ✗ ✗ ✓ ✗
Fischetti et al. (2009) Allocation of time ✓ ✓ ✗ ✗ ✗ ✗
supplements
Flier, Gelashvili, Graffagnino, and Detecting delays ✗ ✗ ✗ ✗ ✓ ✗
Nunkesser (2009)
Medeossi et al. (2009) Punctuality Delay frequency index ✓ ✓ ✗ ✗ ✗ ✗
Goerigk and Schöbel (2010) Allocation of time ✓ ✓ ✗ ✗ (✓) ✗
supplements
Andersson et al. (2011) Allocation of time ✓ ✓ ✗ ✗ ✗ ✗
supplements
Oort (2011) Travel time minimisation ✗ ✗ ✗ ✓ ✗ ✗
Forsgren et al. (2012) Adding flexibility (✓) ✓ ✗ ✗ ✗ ✓
Salido et al. (2012) Delay absorption ✓ ✓ ✓ ✓ ✗ ✗
Yamamura et al. (2012) Detecting delays ✗ ✗ ✗ ✗ ✓ ✗
Goverde and Hansen (2013) Defined timetabling levels Definitions of four different timetable quality ✓ ✓ (✓) ✗ ✓ ✓
levels
Andersson et al. (2013) Robustness of timetable RCP ✓ ✓ ✗ ✗ ✓ ✓
 A LITERATURE REVIEW 7
Downloaded by [DTU Library] at 14:01 06 January 2016

Figure 2. Buffer time.

sufficiently high percentile is equivalent to implicitly adding running time supplements.

Oort (2011) suggested using the 35th percentile in low-frequency systems to minimise
passengers’ travel time. The reason that two different recommendations were given
was a result of different objectives: Bush (2007) focused on providing a certain mean
level of service in high-frequency networks, while Oort (2011) tried to minimise passen-
gers’ total travel time for long-headway services in a system where small delays did not
impact other trains.

2.1.2 Robust optimisation

One paradigm in the optimisation of timetable robustness (focused on adding time sup-
plements) is robust optimisation. Robust optimisation derives a feasible timetable includ-
ing time supplements large enough to absorb the most frequently occurring delays. The
amount of time supplements to allocate is a compromise between travel time extension
and delay propagation risk. Having determined the total amount of time supplements, the
question is where to allocate it. A proportional allocation of the running time supplements
does not minimise average delay (Rietveld, 2005). Supplements on the earliest and in par-
ticular the last part of the line should be below the average supplement size. Delays
incurred in the beginning are often relatively small, thus running time supplements are
lost. Placing less time supplements in the end of a line may be done because only
delays on the last stations could be relieved from this, hence not affecting as many stations
as possible (Vromans, 2005). Andersson, Peterson, and Krasemann (2011) outlined the
importance of being familiar with frequently occurring delays and, by comparing different
train types with different distributions of allocated time supplement (early departures pro-
hibited), they recommended allocating time supplements shortly after the points where
the most frequent delays occurred.
 8 J. PARBO ET AL.

Generally, robust optimisation had a tendency to add too many time supplements,
resulting in a significant reduction of capacity utilisation (Fischetti & Monaci, 2009). With
the typical definition of robustness focusing on absorbing minor delays, most definitions
failed to describe the size of “minor” delays (e.g. Bush, 2007; Cacchiani et al., 2009;
Medeossi et al., 2009; Salido et al., 2008; Vromans, 2005). Schöbel and Kratz (2009)
applied robust optimisation and developed dynamic measures, which in theory could
absorb even large delays. They measured the maximum number of passengers missing
a connection and the sum of delays when all delays were below a certain threshold
value. To account for the conservative aspects of robust optimisation, Fischetti and
Monaci (2009) introduced light robustness. From the nominal timetable (without time sup-
plements), a maximum deterioration of the objective function was fixed, thereby limiting
the amount of added time supplements. Goerigk and Schöbel (2010) treated the robust
timetabling problem differently. Having generated several feasible timetables with differ-
ent amounts of time supplements, each of these was tested against different delay
scenarios.
Downloaded by [DTU Library] at 14:01 06 January 2016

2.1.3 Measurements
To overcome the limitations of discrete service measurements (e.g. measurements of
whether the train is punctual or not), continuous measurements are typically better to
reflect the level of service passengers experience. Huisman and Boucherie (2001) intro-
duced two performance measures related to punctuality: the mean delay of a train and
the delay probability of a train. Although being measured only at the final station, the
mean delay reflects the actual performance of the train more accurately than a discrete
punctuality (on-time) measure. Instead, the delay probability of a train reflects the risk
of a particular train being delayed.
To provide guidelines on how different distributions of time supplements impacted the
level of service, Kroon, Dekker, and Vromans (2007) developed a measure to assess at
which part of the line time supplements were added:


N
2t − 1
WAD = ∗st , (1)
t=1
2N

where WAD is the weighted average distance of the allocated time supplement from
the starting point, N is the number of consecutive trips t between stations, and st is the
amount of time supplement on a particular trip t. This measure takes a value between 0
and 1, indicating at which part of the line time supplements are allocated, with 0.5 indi-
cating a uniform allocation. This measure was later used by Fischetti, Salvagnin, and
Zanette (2009), who focused on minimising cumulative delays and found that allocating
the majority of time supplements towards the first part of the railway line minimised
delay propagation because it allowed trains to use it throughout the entire trip, thus
increasing the chance of being on-time at all subsequent stops (early departures
allowed).
In Italy, the punctuality level was solely measured at terminal stations (Medeossi et al.,
2009). Consequently, large time supplements were placed on track segments leading up
to that station. Additionally, the aggregated effect of infrastructure failures and failures
caused by operating companies became indistinguishable. An underway delay measure
 A LITERATURE REVIEW 9

was thus more useful when investigating performance on certain track sections (Anders-
son et al., 2011; Nyström & Söderholm, 2005). To make up for the weaknesses outlined,
Medeossi et al. (2009) proposed a continuous punctuality measure “Delay Frequency
Index”, taking into account both running time deviation and whether or not trains were
on time at the final station:


n
F= (Ni /N∗Di /P∗f ), (2)
i=1

where N is the number of trains, Ni is the number of trains arriving in delay interval i, Di is
the magnitude of the delay in interval i, P is a threshold value indicating whether a train is
on-time, f is a weight coefficient, and F is a percentage indicator.

2.2 Stability
Downloaded by [DTU Library] at 14:01 06 January 2016

Stability is defined as the inherent ability of a timetable to limit propagation of minor delays.
The most common approach to enhance timetable stability is to add buffer time between
subsequently running trains. Buffer time is often imposed similarly as adding time sup-
plements (see Section 2.1). Adding buffer time between trains is a trade-off between
capacity utilisation and delay propagation. The mean knock-on delay of all trains passing
a station was shown to increase exponentially with the decreasing amount of scheduled
buffer time between train paths; on the other hand, allocating too much buffer time
could remain unused, thus increasing passengers’ travel time (Yuan & Hansen, 2007).
Armstrong, Preston, Potts, Paraskevopoulos, and Bektas (2012) tried to adapt the spread
of trains running in a corridor with the aim to maximise reliability and potentially release
capacity for additional railway services. Forsgren, Aronsson, Gestrelius, and Dahlberg
(2012) addressed timetable stability by solely publishing a subset of the arrival and depar-
ture times, thus imposing flexibility by allowing traffic managers to re-distribute buffer
time during operations. The same idea was addressed by Goverde and Hansen (2013),
who elaborated on different levels of delay resistance for timetables. The highest level
was a resilient timetable which was robust against delays and flexible enough to handle
disturbances.
Another approach aiming at maximising the spread of trains is rerouting trains, assign-
ing alternative platforms or changing schedules (Dewilde, Sels, Cattrysse, & Vansteenwe-
gen, 2013). In high-frequency networks, the aim would be to increase the smallest
headway as much as possible (Carey, 1999). Sels et al. (2014) extended the idea behind
assigning alternative platforms. Assigning trains to alternative platforms could create
inconsistency and cause longer walking distances for transferring passengers (Dollevoet,
2013). Dewilde et al. (2013) and Sels et al. (2014) did not ensure that frequency was main-
tained and consistent when assigning alternative platforms, although it was known to
impact customer satisfaction (Sun & Xu, 2012).

2.3 Heterogeneity
Heterogeneity refers to the dissimilarity in the way different trains are operated. Having
low heterogeneity, namely a high degree of dissimilarity of trains’ stopping patterns,
 10 J. PARBO ET AL.

headway, and speed, allows more trains to run on the tracks when overtaking is prohib-
ited. Heterogeneous operations are typically seen on tracks where different types of
trains are run. Kikuchi and Vuchic (1982) covered the impacts of imposing skip-stop ser-
vices, namely changes in access time, in-vehicle time, fleet size requirements and operat-
ing costs. Access time increases on stops experiencing a less-frequent service. System area
coverage is reduced, thus affecting the number of potential customers. Riding time
decreases since dwell time, acceleration and deceleration time are saved on skipped
stops. ***Reduced in-vehicle time implies shorter round trip time, reducing requirements
for rolling stock, hence also the operating costs. To compare heterogeneity of railway
operations in a corridor, Vromans (2005) developed the following two analytical measures:

n
1
SSHR = −, (3)
h
i=1 i

n
1
SAHR = , (4)
Downloaded by [DTU Library] at 14:01 06 January 2016

hA
i=1 i

when measuring Shortest Headway Reciprocals (SSHR), hi is the smallest scheduled

headway between trains i and i + 1 on a track segment. This measure turned out to fall
short because the departure headway was considered as equally important as the
arrival headway. Vromans (2005) found arrival headways more important when trains
were blocking each other. To cope with this, the Sum of Arrival Headway Reciprocals
(SAHR) only accounted for arrival headways between subsequent trains, with hAi being
the headway between train i and train i + 1 at the arrival station. SAHR is a single point
measure not taking into account the entire track segment. Therefore, a more elaborate
heterogeneity measure averaging SSHR and SAHR was suggested (Vromans, Dekker, &
Kroon, 2006). A straightforward way to change heterogeneity is to change stopping pat-
terns on railway lines (see Parbo, Nielsen, & Prato, 2014 for a list of references on changing
stopping patterns).

2.4 Network Interdependencies

Network interdependencies refer to train movements where precedence constraints are
enforced. Precedence constraints state the train order and whether trains should be
held to maintain connections. Network interdependencies, although important for
network connectivity, increase the risk of propagating delays. The more constraints
imposed on trains, the more vulnerable to disruptions the system becomes. Although facil-
itating smooth transfers between several trains, synchronised networks create “snowbal-
ling” delays, when trains are held to maintain connections (Vansteenwegen & Van
Oudheusden, 2007). Consequently, synchronised networks experience larger increases
in passengers’ travel time when connections are not maintained (Finger, Haller, Martins,
& Trinkner, 2014).

2.4.1 Changing network layout

The intuitive way to reduce network interdependencies is to avoid conflicting train move-
ments. A “cheap” approach towards reducing the interdependencies for a single track
 A LITERATURE REVIEW 11

network without extending it fully to double track lines is to position a set of sidings to
limit conflicting train movements (Higgins, Kozan, & Ferreira, 1997). Also, running
railway lines independently of each other, building bridges, crossovers, side tracks or
extending the network layout from single track to double track will reduce the number
of interdependencies (Gestrelius, Aronsson, Forsgren, & Dahlberg, 2012; Landex, 2008).
Besides the limited capacity, delays propagated easier on single track lines because of
interdependent train movements both from ahead and behind (Landex, 2008).

2.4.2 Planning
Burkolter, Herrmann, and Caimi (2005) proposed a two-level method determining train
precedence constraints and routing, respectively. The higher level created a tentative
dense timetable by applying Petri Nets modelling on an aggregated track topology. After-
wards, the lower level verified the tentative timetable on local exact topologies.
Salido et al. (2008) ***suggested using the following robustness indicator R(x) on
Downloaded by [DTU Library] at 14:01 06 January 2016

double track lines explicitly considering interdependencies between trains. The following
robustness value was assigned to the timetable:
NT


NS
R(x) = Buff TS ∗%FlowST ∗TTS ∗NSucTT ∗(NS − S)/NS, (5)
T=1 S=1

where BuffTS is the buffer time a given train T has on a given station S, %FlowST is the
percentage of passenger flow in train T and station S, TTS is the percentage of tightness of
track between stations S and S + 1, NSucTT is the number of trains that may be disrupted
by train T, NS and NT are the number of stations and trains, respectively. The robustness
measure was developed to compare timetable quality among different timetables for the
same track layout (Salido et al., 2008).
Flier et al. (2009) presented efficient algorithms to detect the interdependencies that
occurred due to precedence constraints and due to maintained connections. Yamamura,
Koresawa, Adachi, and Tomii (2012) developed an algorithm based on daily recorded
traffic data to identify frequently occurring and widely influential delays. A backwards
tracing algorithm was applied to find the primary delay causing the secondary delays.
The approach identified notorious delays as well as delays that had not been recognised
by the timetable planners.
De-Los-Santos et al. (2012) addressed interdependencies by developing two robustness
measures where the stability of the network was considered when random failures and
intentional attacks, respectively, were imposed on a track segment in the network. The
two measurements were as follows:
T(K|N| )
dR(N, E) = , (6)
max DT((N, E), e)
e[E

T(K|N| )
mR(N, E) =  , (7)
e[E DT((N, E), e)/|E |

where δR is the ratio between the total travel time in the complete network T(K|N|) and
the total travel time DT(*) when the track segment ē of the network (N,E) that increases the
overall travel time the most is blocked, and µR is the ratio between total travel time in the
 12 J. PARBO ET AL.

complete network and the average total travel time in case of blocked track segments. This
way of considering stability goes well in hand with the statistical definition of robustness
and in particular the derived reliability and resilience. In this context, reliability is the prob-
ability that the network is connected given a failure probability for every edge and resili-
ence is the probability that the network disconnects after exactly i failures (Klau &
Weiskircher, 2005).
Andersson et al. (2013) developed robustness indicators deducible from the timetable
before operations had taken place, Robustness in Critical Points (RCP). Critical points were
points in the network with interdependencies between trains. RCP reflected the flexibility
determined by the time supplements as well as the buffer time. The more time sup-
plements and buffer time, the larger the flexibility to reschedule the trains in the case
of delays.

2.5 Recovery Strategies

Downloaded by [DTU Library] at 14:01 06 January 2016

Having planned the timetable to be robust against delays does not guarantee 100% punc-
tuality. The idea behind recovery strategies is to recover from the disruptions to the
planned schedule fast and smoothly, deteriorating the service as little as possible. If
time is not a limitation, planners could reset and restart the system. The challenge,
though, is to recover while maintaining a proper level of service for the passengers.
Sun and Hickman (2005) formulated a nonlinear 0–1 integer programming skip-stop
problem as a real-time decision support tool with the binary integer variables representing
whether or not to skip a stop. The idea was to allow vehicles to catch up on their delay by
skipping stops. Sun and Hickman (2005) outlined that skipping stops in real time should be
imposed with care, infrequently and never on subsequent trains from the same line.
Hofman, Madsen, Jespersen Groth, Clausen, and Larsen (2006) considered the following
recovery strategies:

. platform changes,
. allowing overtaking,
. skipping stops,
. early turning,
. reducing dwell time, headway and running time to a minimum,
. changing train status,
. inserting trains,
. cancellation of lines.

Simulation was used to test the recovery strategies. Hofman et al. (2006) found that
strategies yielding a large increase in headways resulted in the largest punctuality
increase. When disruptions occurred and no recover strategies were imposed, the capacity
utilisation had a huge impact on the regularity. Liebchen et al. (2010) created a two-stage
model, which, in the first stage, computed a robust timetable and, in the second stage,
solved the delay management problem. The timetable quality was evaluated through
simulation of railway operations. Caprara, Galli, Kroon, Maróti, and Toth (2010) presented
a robust scheduling approach in combination with three different recovery strategies to
mitigate knock-on delays. Firstly, they let delays propagate by keeping the nominally
 A LITERATURE REVIEW 13

assigned train order. Secondly, they chose alternative platforms, that is, exploited non-uti-
lised resources. Thirdly, they tried all possible strategies, e.g. allowing completely different
strategies than nominally assigned. The results indicated that incorporating robustness
considerations into the train routing problem, together with appropriately chosen
online re-scheduling algorithms, led to better train punctuality. However, neither of the
studies considered the effect of recovery strategies on the passengers.

2.6 Summary
Studies addressing robustness against delays of railway operations by improving train-
related service characteristics show that several approaches have been developed.
Despite the vast amount of different approaches, the basic idea behind enhancing robust-
ness can be boiled down to reducing the risk that delays affect subsequently running
trains by using the following “tools”:
Downloaded by [DTU Library] at 14:01 06 January 2016

. adding time supplements,

. adding buffer time,
. cancelling trains,
. reducing network interdependencies,
. lowering heterogeneity.

None of the studies reviewed in Section 2 takes the passengers explicitly into account.
The “tools” are in all the reviewed studies applied with the aim to ensure the on-time per-
formance of the trains without knowing or monitoring the exact impact on the passengers’
travel experience.

3 Passenger Perspectives
While planned railway schedules are published and fixed, passengers’ planned itineraries
are private and affected by their perception of several different attributes. Accordingly,
measuring passengers’ on-time performance is much more complex than evaluating
whether trains are on time. At the same time, neglecting passengers in the planning
may result in suboptimal railway plans. Table 2 outlines an overview of passenger-oriented
railway planning studies. For each study, the passenger-related attributes emphasised are
outlined in their being explicitly considered, ✓, implicitly considered (✓) or not considered
at all, ✗. This section starts by reviewing the passenger-oriented railway optimisation
studies focusing on robustness and then examines the way passengers’ perceive railway
operations and how their perception affects their travel behaviour.

3.1 Optimisation and Planning

3.1.1 Transfer maintenance
From the passengers’ perspective, a crucial part of a well-functioning railway system is
maintaining transfers. In Denmark, every third transit trip includes at least one transfer
(DMC, 2015). Maintaining transfers is a compromise between passengers’ travel time
and delay propagation. Holding trains to maintain transfers increases the risk of
 14
Table 2. Overview of reviewed literature on passenger-oriented railway planning
Passenger-related variables addressed
Measure In-vehicle Waiting Transfers/ Exact route Data source used to reveal

J. PARBO ET AL.
Studies Purpose developed time time Connections choice Delays passenger demand
Optimisation and Planning
Vansteenwegen and Emphasising passenger heavy trains in the ✓ ✗ (✓) ✗ ✗ Passenger counts
Oudheusden (2006) planning
Downloaded by [DTU Library] at 14:01 06 January 2016

Weston et al. (2006) Successful transfers through dispatching ✓ ✓ ✓ ✓ ✓ Simulating operations

Network (2008) Redistribute trains to maximise passenger ✓ ✓ (✓) ✗ ✗ Passenger counts
satisfaction
Oort et al. (2010) Dispatching focusing on holding times to ✓ ✓ ✓ (✓) ✗ Applied to a tram line
ensure successful transfers
Goerigk et al. (2011) Successful completion of transfers ✗ (✓) ✓ ✗ ✓ Approximated
Kanai et al. (2011) Determining whether to keep or drop ✗ ✗ ✓ ✓ (✓) Simulating operations
connections when delays occur
Corman et al. (2012) Delay management ✗ ✗ ✓ (✓) ✓ Approximated
Dollevoet et al. (2012) Dynamic delay management ✓ ✓ ✓ (✓) ✓ Approximated
Sels et al. (2012) Minimise passengers’ travel time ✓ ✓ ✓ (✓) ✗ Deterministic shortest path
Dewilde et al. (2013) Reduce delay propagation NR ✓ ✓ ✓ ✗ (✓) Passengers’ route choice calculated
by shortest path algorithm
Transport models and passenger surveys
Noland and Small (1995) Addressing delay uncertainty ✓ (✓) (✓) ✗ ✓ Passenger survey
Nielsen (2000) Accounting for seat availability in ✓ ✓ ✓ ✓ Passenger counts
assignment models
Bates et al. (2001) Estimating value-of-time ✓ (✓) (✓) ✗ ✓ Passenger survey
Nuzzolo et al. (2001) Examining passengers travel behaviour as a day-to-day ✓ ✓ ✓ ✓ ✓ Transit assignment model
result of historical delays learning process
Chen et al. (2002) Estimating value-of-time ✓ (✓) (✓) ✗ ✓ Passenger survey
Noland and Polak (2002) Passengers’ perception of travel time vs. Reliability ratio ✓ ✓ ✗ ✗ ✓ Passenger survey
reliability
Rietveld (2005) Passengers’ overestimation of train delays (✓) ✗ ✓ ✗ ✓ Passenger survey
Tsuchiya et al. (2006) Passenger information systems used to ✓ (✓) (✓) (✓) ✓ Passenger survey
guide passengers when delays occurred
Takeuchi et al. (2007) Describing passengers’ perceived level of ✓ ✓ ✓ ✓ ✓ Passenger survey
service
Nielsen et al. (2008) Distinguishing train and passenger delays (✓) (✓) (✓) ✓ ✓ Transit assignment model
Sumalee et al. (2009) Examining passengers travel behaviour as a ✓ ✓ (✓) ✓ ✓ Transit assignment model
result of historical seat availability
Börjesson and Eliasson Assessing average delay as reliability ✓ ✓ ✗ (✓) ✓ Passenger survey
(2011) measure
 Hensher et al. (2011) Addressing risk and uncertainty of delays ✗ ✓ ✓ ✗ ✓ Passenger survey
Sels et al. (2011) Derive passenger flows from ticket sales data ✓ (✓) (✓) (✓) ✗ Ticket sales data
✗
Downloaded by [DTU Library] at 14:01 06 January 2016

Sun and Xu (2012) Estimating passengers’ route choice from (✓) (✓) (✓) (✓) Travel card data
travel card data
Börjesson et al. (2012) Estimating value-of-time ✓ ✓ (✓) ✗ ✓ Passenger survey
Jiang et al. (2012) Distinguishing train and passenger delays (✓) (✓) (✓) ✓ ✓ Transit assignment model
Nuzzolo et al. (2012) Accounting for seat availability in schedule- ✓ ✓ ✓ ✓ ✓ Passenger counts
based assignment models
Shi et al. (2012) Assess how transfers impact route choice ✗ (✓) ✓ ✓ (✓) Transit assignment model
Wardman et al. (2012) Meta study on value-of-time studies ✓ ✓ ✓ ✗ ✓ Passenger survey

A LITERATURE REVIEW
15
 16 J. PARBO ET AL.

propagating train delays and increases the travel time for on-board passengers, while can-
celled or missed connections impose extra delays on transferring passengers. In this
regard, distinguishing trains as either punctual or not based on an (to some extent) arbi-
trary threshold can distort the picture of the service provided. Passengers on trains
delayed by less than the threshold value can also miss a transfer (Carrasco, 2012). Conse-
quently, allocating time supplements on stations with negligible transfer loads is ineffec-
tive (Dewilde et al., 2013; Oort, Wilson, & Van Nes, 2010). An approach emphasising the
passengers is to determine paths on which connections will be maintained even when
delays occur. This can be done by allocating additional time supplements on stations or
by modifying the timetable, so that subsequently running trains are not affected if
minor delays occur (Vansteenwegen & Oudheusden, 2006).
In real time, missed transfers can be reduced through dispatching. Weston et al. (2006)
compared actual arrival time at the destination station to the planned arrival time. After-
wards, it was checked if the transfer was completed. A microscopic simulation of a con-
gested part of the rail network in UK. was performed. Weston et al. (2006) concluded
Downloaded by [DTU Library] at 14:01 06 January 2016

that, due to missed connections, minimising train delays did not necessarily minimise pas-
senger delays.
The problem of maintaining transfers is sometimes referred to as the timetable infor-
mation problem, where light and strict robustness may be distinguished. A strict robust
path is defined as a path where all transfers are maintained under every delay scenario.
Similar to the timetabling problem, light robustness was found superior to strict robust-
ness since it ensured a modest level of robustness while only deteriorating passengers’
travel time marginally (Goerigk, Knoth, Müller-Hannemann, Schmidt, & Schöbel, 2011).
Kanai, Shiina, Harada, and Tomii (2011) developed a delay management plan minimis-
ing passengers’ dissatisfaction (the trade-off between additional in-vehicle time and extra
waiting time due to missed transfers) by combining simulation and optimisation. Passen-
gers behaved as if trains were on time. The decision of keeping or dropping connections
was solved by a tabu search algorithm and evaluated by a passenger simulation model.
Kanai et al. (2011) found that travellers were rarely willing to wait until the next service
arrived, but would rather seek alternative routes.
Corman, D’Ariano, Pacciarelli, and Pranzo (2012) developed a bi-objective delay man-
agement strategy. The aim was to minimise the number of missed connections (weighted
by the number of passengers) and avoid train conflicts when re-scheduling trains by using
a heuristic algorithm (Corman et al., 2012). No recommendations on how to handle these
issues were given; instead it was proposed that different stakeholders should agree on a
“practical optimum” between maintaining and cancelling connections.

3.1.2 Emphasising travel behaviour

In transit planning, the impacts of timetable changes on passengers’ travel behaviour
should be considered explicitly to accurately quantify the derived impacts (Sun, Jin, Lee,
Axhausen, & Erath, 2014). One way to obtain passengers’ route choice is by ticket sales
data. However, not all ticket sales reveal ending and transfer points, thus different approxi-
mation methods are needed. Sels, Dewilde, Cattrysse, and Vansteenwegen (2011) distrib-
uted passengers on stations according to existing demand data and calculated passenger
flows by shortest path algorithms. Sun and Xu (2012) described how travel card data from
 A LITERATURE REVIEW 17

the Beijing metro was used to analyse travel time variability and estimate passengers’
route choice behaviour.
Sels et al. (2012) applied Flow Allocation Periodic Event Scheduling Problem introducing a
feedback loop between (passenger) flow allocations and timetabling. The aim was to mini-
mise passengers’ travel time. Dollevoet, Huisman, Schmidt, and Schöbel (2012) elaborated
the flow allocation and proposed a delay management model allowing passengers to
change their route choice when disruptions occurred.
Dewilde et al. (2013) addressed the trade-off between delay propagation and passen-
ger travel time explicitly through Normalised Robustness (NR):
TT + WaitCostEx
NR = , (8)
NomTT
where TT is passengers’ realised travel time, WaitCostEx is passengers’ perceived extra
waiting cost, and NomTT is passengers’ nominal travel time. WaitCostEx addresses the
unused time supplements, which are assumed to be an annoyance factor when trains
Downloaded by [DTU Library] at 14:01 06 January 2016

run according to the planned schedule.

3.2 Passengers’ Perspectives

When passengers plan a journey, their perception of quantifiable attributes (e.g. valuation
of in-vehicle, waiting time, transfers), non-quantifiable attributes (e.g. aversion towards
being late, comfort perception) and whether they are frequent or occasional users have
a huge impact on their travel behaviour (Carrasco, 2012; Takeuchi et al., 2007; Vij, Carrel,
& Walker, 2013). In transport models, a generalised travel cost function may be used to
reflect the impedance travellers face when travelling by public transport, thus enabling
the estimation of travellers’ route choice. One example was presented by Nielsen (2000):
 

b(t)i
GenCost = Costa + (b(t) + j(t) ) · · tai + Seati + 1a , (9)
t
b(t)

where Tai is the time spent on link a when having boarded line i, Costa is the cost on link
a, Seati is the disutility associated with not being able to get a seat on line i (Nielsen, 2000),
ξ is an error component capturing taste variation for different attributes among passen-
gers, and εa captures the differences in passengers’ perception of different routes. The
β’s were related to access/egress time, waiting + transfer time, in-vehicle time, headway
(hidden waiting time), and delay (simulated based on discrete distributions), respectively.
According to the generalised travel cost function outlined above, both the hidden
waiting time and the delays impact passengers’ perception of railway performance.
Hypothetically, an operator could be interested in optimising railway operations for the
passengers by minimising knock-on delays, for example, by adding more buffer time.
Despite increased punctuality, this might not be unambiguously good, since hidden
waiting time has now increased. Similar interdependencies between other attributes
mean that exhaustive performance measurements are required to reveal the actual
impacts of certain initiatives (Yuan & Hansen, 2007).
One approach to investigate the difference between passenger delays and train delays
is to use a traffic assignment model. Assignment models consider quantifiable attributes
that impact passengers’ route choice, hence based on a certain plan, the outcome of the
 18 J. PARBO ET AL.

assignment model reveals how passengers are assumed to travel. Nielsen et al. (2008) ana-
lysed this and found that passengers’ on-time performance was significantly lower than
that of the trains.
As outlined in Figure 3, also more recent data show that on average train punctuality is
higher than passenger punctuality. Passenger punctuality is defined as the number of pas-
sengers reaching their destination within a certain time threshold. Passenger punctuality
evaluates the on-time performance of passengers’ entire journey from origin to destina-
tion including transfers, while train punctuality only evaluates the on-time performance
of individual train trips. The fact that passenger punctuality was generally worse than
the train punctuality was primarily explained by passengers missing transfer connections
and the fact that peak hour trains (having larger passenger loads) more often were
delayed. Although passengers were significantly more delayed than the trains responsible
for their delays, minor train delays did not necessarily cause passenger delays. In fact, some
passengers were able to take a connecting train earlier than planned (Jiang, Li, Xu, & Gao,
2012).
Downloaded by [DTU Library] at 14:01 06 January 2016

3.2.1 Travel time variability

Passengers’ preferences are generally obtained by either observing behaviour or asking
hypothetical questions. The first approach is referred to as revealed preferences (RP)
and the latter as stated preferences (SP), and the two approaches can also be combined.
Regarding travel time variability, RP surveys are hardly applicable since observations are
needed for long periods to deduce travellers’ reactions to disruptions. Consequently,
studies considering travel time variability traditionally use SP approaches.
Bates, Polak, Jones, and Cook (2001) used SP to examine travellers’ departure time
accounting for the relevant variables. According to the SP-study, travellers departed
earlier when trips were important and travellers were risk-averse. Below, an example of

Figure 3. Passenger punctuality vs. train punctuality on the suburban railway network in the Greater
Copenhagen area (Weekly data from 2010 to 2014).
 A LITERATURE REVIEW 19

a utility function from Bates et al. (2001) is shown:

E[U(th )] = aE[T(th )] + bE[SDE(th )] + gE[SDL(th )] + u pL (th ), (10)

where the expected utility U(*) is dependent only on the departure time th, E[*] is the
expected value, T(*) is the travel time, SDE(*) is the early deviation from the planned arrival
time, SDL(*) is the late deviation from the planned arrival time, and pL is the probability of
arriving later than the planned arrival time. α, β, γ and θ are parameters to be estimated
under the assumption that travellers are utility maximisers.
Noland and Polak (2002) measured the importance of reliability relative to travel time
by defining a reliability ratio RR:
 
b g
RR = ln 1 + , (11)
a b

where α is a parameter associated with in-vehicle time, β relates to Schedule Delay Early
Downloaded by [DTU Library] at 14:01 06 January 2016

(SDE) and γ to Schedule Delay Early (SDL). Due to the discrete nature of public transport
services, the disutility associated with low reliability is large.
Examining how the extent of train delays was perceived by the travellers, a few large
delays proved to be more hurtful than several minor delays (same total extent)
(Vromans et al., 2006). This was explained by the ability for several trains to utilise their
buffer time to catch up on smaller delays, while one large delay was more likely to
affect subsequent trains. Using average delay as performance indicator was thus mislead-
ing. A much larger disutility was associated with a 2% risk of being 50 minutes late than a
10% risk of being 10 minutes late, thus forcing risk-averse travellers to take an earlier train
(Börjesson & Eliasson, 2011). Although disturbances are highly disregarded, passengers
generally consider minor delays acceptable. In a survey among transit users from
Britain, 87% of the passengers were satisfied with a five-minute delay at their departure
station. This number fell to 77% when delays were between six and nine minutes (Trans-
port Focus, 2014).
The ratio between in-vehicle time and reliability for public transport was found by Bates
et al. (2001) to be above the ones for car (app. 1.3) and below 2. In a SP-survey from the
Netherlands, this value was estimated to be 1.4, that is, passengers considered a one
minute reduction in travel time variability 1.4 times higher than a one minute of travel
time reduction (Oort, 2011). From a meta-analysis of European studies (conducted
between 1963 and 2011 by Wardman, Chintakayala, de Jong, & Ferrer, 2012), it
emerged that the four variables used to reflect passengers’ perception of travel time varia-
bility were:

. Schedule Delay Early (SDE)

. Schedule Delay Late (SDL)
. Late arrival
. Standard deviation of travel time (StdDev)

Relative to in-vehicle time, the values for SDE, SDL, Late arrival and StdDev were on
average 0.8, 1.68, 3.29 and 0.66, respectively. The uncertainty related to late arrival and
standard deviation of travel time was thus considered significantly less attractive than
 20 J. PARBO ET AL.

using one minute in a vehicle. The travel distance did not affect the variables remarkably
and time multipliers were quite similar between the different studies (Wardman et al.,
2012).

3.2.2 Delay uncertainty and delay risk

When travelling, passengers have a perception (from past experiences) of the travel time
based on their route, mode and departure time choice, respectively. Travel time deviations
can be either expected (risk) or unexpected (uncertainty) by the travellers (Hensher,
Greene, & Li, 2011). Noland and Small (1995) added an extra disutility term to the travel
time uncertainty, with the discrete lateness penalty emphasising passengers aversion
towards being late. This was refined by Bates et al. (2001) and Chen, Ji, and Recker
(2002), who considered travellers’ perceived uncertainty distinguishing between risk-
averse and risk-seeking travellers. Conclusions were identical: risk-seeking travellers
were less stressed by travel time uncertainty than risk-averse travellers, and they did
not consider reliability, but rather chose the path with minimum travel time, while risk-
Downloaded by [DTU Library] at 14:01 06 January 2016

averse passengers chose the most reliable route (Finger et al., 2014).
Tsuchiya et al. (2006) examined passengers’ perception of a support system informing
about optimal routes in case of disruptions. The information was based on predicted
resumption time from the disturbance estimated from historical data. The information
helped passengers decide whether to wait for resumption or not and, if not, which
detour to choose: 94% preferred to have this piece of information as soon as possible,
although subject to uncertainty, rather than waiting until the information was certain.
Additionally, passengers appreciated being informed about the cause of the delay.
When delays were caused by external factors, travellers’ negative emotions were alleviated
compared to the situation where the operator was responsible for the delay (Cheng & Tsai,
2014; Transport Focus, 2014).
Börjesson, Eliasson, and Franklin (2012) studied passengers’ response to travel time vari-
ations by testing the equivalency between scheduling models and reduced-form models.
Passengers’ valuation of expected delay was significantly larger for reduced-form models.
In scheduling models, information about being late was associated with less uncertainty
than in reduced-form model. An inherent disutility (interpreted as an anxiety cost or a
cost associated with the need for contingency plans) was associated with the uncertainty
(Börjesson et al., 2012). Passengers disliked the stress they felt, when onward planned con-
nections looked doubtful. However, if the onward train was part of a frequent service,
missed connections were less damaging (Cheng & Tsai, 2014; Weston et al., 2006). Further-
more, passengers reported that they were more tolerant towards delays when being in the
vehicle, because they could typically see the cause of the problem causing the delay
(Transport Focus, 2014). This finding supports that it is in particular the uncertainty
aspect that passengers dislike.

3.2.3 Recovery strategies

When passengers face disruptions, they seek to find alternative routes. If disruptions start
occurring frequently, travellers start adapting their departure time or mode choice. To
examine passengers’ recovery strategies, Nuzzolo, Russo, and Crisalli (2001) developed a
doubly dynamic schedule-based assignment model. Doubly dynamic refers to the explicit
consideration of within-day and day-to-day variations in travel behaviour. Nuzzolo et al.
 A LITERATURE REVIEW 21

(2001) formulated the day-to-day learning process mathematically:

Xtfo = g · Xt−1
ex
+ (1 − g) · Xt−1
fo
, (12)

where the forecasted value Xt fo for an attribute on day t is expressed by a convex com-
bination of the attribute forecast Xt−1fo and the value realised Xt−1ex from the day before (t
−1), and γ is a weight between 0 and 1. Nuzzolo et al. (2001) found that the number of
transfers during a journey did not have a significant impact on short-term adaptations.
Sumalee, Tan, and Lam (2009) developed a dynamic stochastic assignment model
taking seat availability into account. Information about seat availability prior to their trip
made travellers adapt their departure time and/or route choice to maximise their prob-
ability of getting a seat.
Shi, Zhou, Yao, and Huang (2012) developed an equilibrium-based rail passenger flow
model explicitly taking the probability of successful transfers into account. Passengers
tended to choose a path where the probability of a successful transfer was higher.
Additionally, Shi et al. (2012) found that waiting time mostly affected within-day variations
Downloaded by [DTU Library] at 14:01 06 January 2016

in travel behaviour. Attributes affecting day-to-day variations in travel behaviour were in-
vehicle time, transfer time and comfort level.
Nuzzolo, Crisalli, and Rosati (2012) developed a schedule-based dynamic assignment
model for transit networks that jointly simulated departure time, boarding stop and run
choices on the basis of mixed pre-trip/en-route choice behaviour. Pre-trip choices con-
cerned departure time and boarding stops, since they were mainly influenced by past
experiences of congestion, while en-route choices occurred at boarding stops and con-
cerned the decision whether to board a specific vehicle. Vehicle capacity impacted passen-
gers’ departure time choice significantly. Passengers tried to board an earlier vehicle to
reduce failure-to-board probability.
van der Hurk (2015) developed a model that, in the case of large disruptions, person-
alised passenger information on alternative routes. The model took into account the prob-
ability of boarding and the uncertainty in the duration of the disruption. For all tested
disruption scenarios, the conclusion was that when passengers are provided personalised
information on alternative routes taking into account the probability of boarding, the
average expected delay a passenger may experience is reduced significantly compared
to the case where information is not personalised or the case where information is person-
alised but does not account for boarding probability.

3.3 Summary
To reduce the impact of train disruptions on passengers, maintaining transfers is often the
main concern among optimisation studies. Passenger loads from the existing system are
used as weights in order to prioritise specific transfer connections. Reducing the risk of
missing transfer connections is typically done by adding additional time supplements to
the involved stations. Doing so is a trade-off between delays passing on to subsequently
running trains and imposing additional passenger travel time and should thus be imposed
with care.
From the body of literature, it is clear that several quantifiable attributes impact passen-
gers’ travel behaviour. Unfortunately, all these attributes are rarely considered at the same
 22 J. PARBO ET AL.

time when planning railway operations. Comparisons between train and passenger on-
time performance, respectively, reveal that due to, for example, missed transfers and
demand variability, train punctuality is often significantly higher. These studies highlight
the importance of having delay-resistant timetables, thus minimising uncertainty and pro-
viding a punctual service to the passengers. Travel time uncertainty is highly disliked by
travellers, not only when disruptions occur, but also the need for having contingency
plans. Consequently, travellers value reduced travel time uncertainty higher than
reduced travel time.

4 Conclusions and Future Research

The present literature review provides an analysis of the work that, on the one hand, has
aimed at enhancing operational characteristics related to the robustness of railway oper-
ations and, on the other hand, has focused on passengers’ perception of railway perform-
Downloaded by [DTU Library] at 14:01 06 January 2016

ance. Table 1 reveals that the service characteristics considered most often regarding
robustness are time supplements and buffer times. Although all studies agreed on lower-
ing capacity utilisation to improve robustness (delay resistance), their recommendations
vary a lot based on the network layout, how performance is measured and whether or
not trains are allowed to depart earlier than scheduled.
When focusing explicitly on the passengers, Table 2 shows that maintaining transfers
has been the objective pursued most frequently. From the review of passengers’ perspec-
tives in railway operations, it is evident that their perception of the service level affects
their travel behaviour. Passengers’ travel behaviour (i.e. mode choice, route choice, depar-
ture time choice) depends on several attributes, for example, in-vehicle time, transfer time,
waiting time, access/egress time, crowding level and delays. Using on-time performance of
trains as performance measurement thus turns out to be inadequate. Taking passengers’
travel patterns into account shows that passengers’ on-time performance in some cases is
10 percentage points lower than for the trains. To close the gap between how railway
planning is performed and measured and, on the other hand, passengers’ perception of
railway performance and their actual experiences, the following directions for future
research are identified.
Understanding passengers’ preferences and being able to address these preferences
explicitly in the planning is the basis for a more passenger-oriented railway planning.
Accurate and disaggregate passenger travel data facilitates a more passenger-oriented
planning, especially when transfer patterns are revealed. Passenger-oriented key perform-
ance indicators (taking all relevant attributes into account) should be applied by research-
ers and operators. Exhaustive performance measurements ensure that enhancing a single
parameter is not realised at the expense of non-measured parameters, thus leading to a de
facto deterioration or status quo in service level.
Regarding travel time variability, studies show that passengers rate schedule adherence
higher than travel time reductions. Therefore, a definition of robustness and related
measurements need to capture the system performance as well as the efficiency, that
is, travel time and capacity utilisation. When the price of robustness is disregarded,
enhanced robustness may be achieved by allocating disproportionally large time sup-
plements and buffer times.
 A LITERATURE REVIEW 23

Since robustness is a trade-off between several attributes, improving railway operations

should take a more holistic and generic view, thus addressing all relevant attributes rather
than only a subset. The same goes for the performance measurements. Coupling generic
optimisation approaches and planning of railway operations with knowledge of passen-
gers’ travel behaviour (e.g. through transport models) will enhance the reliability and
applicability of the results, thereby decreasing the gap between what railway planners
provide and how it is perceived and experienced by the passengers.
Finally, providing real-time information to the passengers about their itinerary when
delays occur is shown to have a stress-relieving effect. There is a huge potential in indivi-
dualising information on the current state of the transit system (e.g. travel time uncer-
tainty, delayed and cancelled services). Providing passengers with information about
route alternatives in case of large disruptions taking into account seat availability has a sig-
nificant impact on the average expected delay and thus also on their travel experience.
Downloaded by [DTU Library] at 14:01 06 January 2016

Disclosure statement
No potential conflict of interest was reported by the authors.

References
Andersson, E., Peterson, A., & Krasemann, J. T. (2011). Robustness in Swedish railway traffic timetables.
In Railrome 2011. Book of Abstracts 4th International Seminar on Railway Operations Modelling
and Analysis.
Andersson, E. V., Peterson, A., & Krasemann, J. T. (2013). Quantifying railway timetable robustness in
critical points. Journal of Rail Transport Planning & Management, 3(3), 95–110.
Armstrong, J., Preston, J., Potts, P., Paraskevopoulos, D., & Bektas, T. (2012). Scheduling trains to maxi-
mize railway junction and station capacity. Proceedings of 12th Conference on Advanced Systems
for Public Transport (CASPT), Santiago, Chile.
Bates, J., Polak, J., Jones, P., & Cook, A. (2001). The valuation of reliability for personal travel.
Transportation Research Part E: Logistics and Transportation Review, 37(2), 191–229.
Batley, R., Dargay, J., & Wardman, M. (2011). The impact of lateness and reliability on passenger rail
demand. Transportation Research Part E: Logistics and Transportation Review, 47, 61–72.
Benezech, V., & Coulombel, N. (2013). The value of service reliability. Transportation Research Part B:
Methodological, 58, 1–15.
Börjesson, M., & Eliasson, J. (2011). On the use of “average delay” as a measure of train reliability.
Transportation Research Part A: Policy and Practice, 45(3), 171–184.
Börjesson, M., Eliasson, J., & Franklin, J. P. (2012). Valuations of travel time variability in scheduling
versus mean–variance models. Transportation Research Part B: Methodological, 46(7), 855–873.
Burkolter, D., Herrmann, T., & Caimi, G. (2005). Generating dense railway schedules. Advanced OR and
AI Methods in Transportation. Publishing House of Poznan University of Technology, 10th EWGT
Meeting and 16th Mini-Euro Conference, pp. 290–297.
Bush, R. (2007). Does every trip need to be on time? Multimodal Scheduling Performance Parameters
with an application to Amtrak Service in North Carolina. In Transportation Research Board 86th
Annual Meeting (No. 07-2791), Washington DC, USA.
Cacchiani, V., Caprara, A., & Fischetti, M. (2009). Robustness in train timetabling. 8th Cologne-Twente
Workshop on Graphs and Combinatorial Optimization, Cologne, Germany, pp. 171–174.
Caprara, A., Galli, L., Kroon, L. G., Maróti, G., & Toth, P. (2010). Robust train routing and online re-sche-
duling. In 10th workshop on algorithmic approaches for transportation modelling, optimization,
and systems, Liverpool, United Kingdom, pp. 24–33.
Carey, M. (1999). Ex ante heuristic measures of schedule reliability. Transportation Research Part B:
Methodological, 33(7), 473–494.
 24 J. PARBO ET AL.

Carrasco, N. (2012). Quantifying reliability of transit service in Zurich, Switzerland. Transportation

Research Record: Journal of the Transportation Research Board, 2274, 114–125.
Chen, A., Ji, Z., & Recker, W. (2002). Travel time reliability with risk-sensitive travelers. Transportation
Research Record: Journal of the Transportation Research Board, 1783, 27–33.
Cheng, Y. H., & Tsai, Y. C. (2014). Train delay and perceived-wait time: Passengers’ perspective.
Transport Reviews, 34(6), 710–729.
Corman, F., D’Ariano, A., Pacciarelli, D., & Pranzo, M. (2012). Bi-objective conflict detection and resol-
ution in railway traffic management. Transportation Research Part C: Emerging Technologies, 20,
79–94.
De-Los-Santos, A., Laporte, G., Mesa, J. A., & Perea, F. (2012). Evaluating passenger robustness in a rail
transit network. Transportation Research Part C: Emerging Technologies, 20, 34–46.
Dewilde, T., Sels, P., Cattrysse, D., & Vansteenwegen, P. (2011). Defining robustness of a railway time-
table. In Railrome 2011: Book of abstracts 4th international seminar on railway operations model-
ling and analysis, Rome, Italy.
Dewilde, T., Sels, P., Cattrysse, D., & Vansteenwegen, P. (2013). Robust railway station planning:
An interaction between routing, timetabling and platforming. Journal of Rail Transport Planning
& Management, 3(3), 68–77.
Downloaded by [DTU Library] at 14:01 06 January 2016

DMC. (2015). The Danish National Travel Survey. Retrieved June 30, 2015, from http://www.
modelcenter.transport.www7.sitecore.dtu.dk/english/TU
Dollevoet, T., Huisman, D., Schmidt, M., & Schöbel, A. (2012). Delay management with rerouting of
passengers. Transportation Science, 46, 74–89.
Dollevoet, T. A. B. (2013). Delay management and dispatching in railways (Doctoral dissertation).
Rotterdam School of Management, Erasmus University.
Finger, M., Haller, A., Martins, S. S., & Trinkner, U. (2014). Integrated timetables for railway passenger
transport services. Competition and Regulation in Network Industries, 15(1), 78–107.
Fischetti, M., & Monaci, M. (2009). Light robustness. Robust and online large-scale optimization
(pp. 61–84). Heidelberg: Springer.
Fischetti, M., Salvagnin, D., & Zanette, A. (2009). Fast approaches to improve the robustness of a
railway timetable. Transportation Science, 43(3), 321–335.
Flier, H., Gelashvili, R., Graffagnino, T., & Nunkesser, M. (2009). Mining railway delay dependencies in
large-scale real-world delay data. In R. K. Ahuja, R. H. Möhring, & C. D. Zaroliagis (Eds.), Robust and
online large-scale optimization (pp. 354–368). Berlin: Springer.
Forsgren, M., Aronsson, M., Gestrelius, S., & Dahlberg, H. (2012). Using timetabling optimization pro-
totype tools in new ways to support decision making. Computers in Railways XIII: Computer System
Design and Operation in the Railway and Other Transit Systems, 127, 439–450.
Gestrelius, S., Aronsson, M., Forsgren, M., & Dahlberg, H. (2012). On the delivery robustness of train
timetables with respect to production replanning possibilities. 2nd International Conference on
Road and Rail Infrastructure in Dubrovnik, Croatia.
Goerigk, M., Knoth, M., Müller-Hannemann, M., Schmidt, M., & Schöbel, A. (2011). The price of robust-
ness in timetable information. OASIcs-OpenAccess Series in Informatics, 20. Schloss Dagstuhl-
Leibniz-Zentrum fuer Informatik.
Goerigk, M., & Schöbel, A. (2010). An empirical analysis of robustness concepts for timetabling. OASIcs-
OpenAccess Series in Informatics, 14. Schloss Dagstuhl-Leibniz-Zentrum fuer Informatik.
Goverde, R. M. (2010). A delay propagation algorithm for large-scale railway traffic networks.
Transportation Research Part C: Emerging Technologies, 18(3), 269–287.
Goverde, R. M., & Hansen, I. (2013). Performance indicators for railway timetables (pp. 301–306).
Intelligent Rail Transportation (ICIRT), 2013 IEEE International Conference.
Hensher, D. A., Greene, W. H., & Li, Z. (2011). Embedding risk attitude and decision weights in non-
linear logit to accommodate time variability in the value of expected travel time savings.
Transportation Research Part B: Methodological, 45(7), 954–972.
Higgins, A., Kozan, E., & Ferreira, L. (1997). Modelling the number and location of sidings on a single
line railway. Computers & Operations Research, 24(3), 209–220.
 A LITERATURE REVIEW 25

Hofman, M., Madsen, L., Jespersen Groth, J., Clausen, J., & Larsen, J. (2006). Robustness and recovery in
train scheduling-a case study from DSB S-tog a/s. OASIcs-OpenAccess Series in Informatics,
5. Schloss Dagstuhl-Leibniz-Zentrum für Informatik.
Hooghiemstra, J. S., & Teunisse, M. J. (1998). The use of simulation in the planning of the Dutch railway
services. IEEE Simulation Conference Proceedings, Washington DC, USA, pp. 1139–1145.
Huisman, T., & Boucherie, R. J. (2001). Running times on railway sections with heterogeneous train
traffic. Transportation Research Part B: Methodological, 35(3), 271–292.
van der Hurk, E. (2015). Passengers, information, and disruptions. (PhD dissertation, Rotterdam School
of Management). Erasmus University.
Jiang, Z. B., Li, F., Xu, R. H., & Gao, P. (2012). A simulation model for estimating train and passenger
delays in large-scale rail transit networks. Journal of Central South University, 19, 3603–3613.
Kanai, S., Shiina, K., Harada, S., & Tomii, N. (2011). An optimal delay management algorithm from pas-
sengers’ viewpoints considering the whole railway network. Journal of Rail Transport Planning &
Management, 1, 25–37.
Kikuchi, S., & Vuchic, V. R. (1982). Transit vehicle stopping regimes and spacings. Transportation
Science, 16(3), 311–331.
Klau, G. W., & Weiskircher, R. (2005). Robustness and resilience (pp. 417–437). Network Analysis.
Downloaded by [DTU Library] at 14:01 06 January 2016

Heidelberg: Springer.
Kroon, L., Dekker, R., & Vromans, M. (2007). Cyclic railway timetabling: A stochastic optimization
approach. Algorithmic Methods for Railway Optimization, 16, 41–66.
Kroon, L., Maróti, G., Helmrich, M. R., Vromans, M., & Dekker, R. (2008). Stochastic improvement of
cyclic railway timetables. Transportation Research Part B: Methodological, 42(6), 553–570.
Landex, A. (2008). Methods to estimate railway capacity and passenger delays. (PhD dissertation).
Technical University of Denmark.
Liebchen, C., Schachtebeck, M., Schöbel, A., Stiller, S., & Prigge, A. (2010). Computing delay resistant
railway timetables. Computers & Operations Research, 37(5), 857–868.
Medeossi, G., Marchionna, A., & Longo, G. (2009). Capacity and reliability on railway networks: A simu-
lative approach (PhD dissertation). University of Trieste.
Milan, J. (1996). The Trans European railway network: Three levels of services for the passengers.
Transport Policy, 3(3), 99–104.
Milinković, S., Marković, M., Vesković, S., Ivić, M., & Pavlović, N. (2013). A fuzzy Petri net model to esti-
mate train delays. Simulation Modelling Practice and Theory, 33, 144–157.
Network, B. P. T. (2008). More passengers and reduced costs—the optimization of the Berlin public
transport network. Journal of Public Transportation, 11(3), 57–76.
Nielsen, O. A. (2000). A stochastic transit assignment model considering differences in passengers
utility functions. Transportation Research Part B: Methodological, 34(5), 377–402.
Nielsen, O. A., Landex, A., & Frederiksen, R. D. (2008). Passenger delay models for rail networks.
Schedule-Based Modeling of Transportation Networks: Theory and applications, 46, 27–49, Springer.
Noland, R. B., & Polak, J. W. (2002). Travel time variability: A review of theoretical and empirical issues.
Transport Reviews, 22, 39–54.
Noland, R. B., & Small, K. A. (1995). Travel-time uncertainty, departure time choice, and the cost of
morning commutes. Transportation Research Record, 1493, 150–158.
Nuzzolo, A., Crisalli, U., & Rosati, L. (2012). A schedule-based assignment model with explicit capacity
constraints for congested transit networks. Transportation Research Part C: Emerging Technologies,
20, 16–33.
Nuzzolo, A., Russo, F., & Crisalli, U. (2001). A doubly dynamic schedule-based assignment model for
transit networks. Transportation Science, 35(3), 268–285.
Nyström, B., & Söderholm, P. (2005). Improving railway punctuality by maintenance. Research project
of Luleå University of Technology.
Oort, N. V. (2011). Service reliability improvement by enhanced network and timetable planning.
Proceedings of 12th Conference on Advanced Systems for Public Transport (CASPT), Santiago,
Chile.
Oort, N. V., Wilson, N. H., & Van Nes, R. (2010). Reliability improvement in short headway transit ser-
vices. Transportation Research Record: Journal of the Transportation Research Board, 2143, 67–76.
 26 J. PARBO ET AL.

Parbo, J., Nielsen, O. A., & Prato, C. G. (2014). User perspectives in public transport timetable optim-
isation. Transportation Research Part C: Emerging Technologies, 48, 269–284.
Rietveld, P. (2005). Six reasons why supply-oriented indicators systematically overestimate service
quality in public transport. Transport Reviews, 25(3), 319–328.
Salido, M. A., Barber, F., & Ingolotti, L. (2008). Robustness in railway transportation scheduling
(pp. 2880–2885). Intelligent control and automation. WCICA. 7th World Congress on IEEE.
Salido, M. A., Barber, F., & Ingolotti, L. (2012). Robustness for a single railway line: Analytical and simu-
lation methods. Expert Systems with Applications, 39(18), 13305–13327.
Schöbel, A., & Kratz, A. (2009). A bi-criteria approach for robust timetabling. Robust and Online Large-
Scale Optimization. Lecture Notes in Computer Science, 5868, 119–144.
Sels, P., Dewilde, T., Cattrysse, D., & Vansteenwegen, P. (2011). Deriving all passenger flows in a railway
network from ticket sales data. In Railrome 2011: Book of Abstracts 4th International Seminar on
Railway Operations Modelling and Analysis.
Sels, P., Vansteenwegen, P., Dewilde, T., Cattrysse, D., Waquet, B., & Joubert, A. (2014). The train plat-
forming problem: The infrastructure management company perspective. Transportation Research
Part B: Methodological, 61, 55–72.
Sels, P. H. A., Dewilde, T., Vansteenwegen, P., & Cattrysse, D. (2012). Automated, passenger time
Downloaded by [DTU Library] at 14:01 06 January 2016

optimal, robust timetabling, using integer programming. Proceedings of the 1st International
Workshop on High-Speed and Intercity Railways, 87–92. Springer: Heidelberg.
Shi, F., Zhou, Z., Yao, J., & Huang, H. (2012). Incorporating transfer reliability into equilibrium analysis
of railway passenger flow. European Journal of Operational Research, 220(2), 378–385.
Sumalee, A., Tan, Z., & Lam, W. H. (2009). Dynamic stochastic transit assignment with explicit seat
allocation model. Transportation Research Part B: Methodological, 43(8), 895–912.
Sun, A., & Hickman, M. (2005). The real–time stop–skipping problem. Journal of Intelligent
Transportation Systems, 9(2), 91–109.
Sun, L., Jin, J. G., Lee, D. H., Axhausen, K. W., & Erath, A. (2014). Demand-driven timetable design for
metro services. Transportation Research Part C: Emerging Technologies, 46, 284–299.
Sun, Y., & Xu, R. (2012). Rail transit travel time reliability and estimation of passenger route choice
behavior: Analysis using automatic fare collection data. Transportation Research Record: Journal
of the Transportation Research Board, 2275, 58–67.
Takeuchi, Y., Tomii, N., & Hirai, C. (2007). Evaluation method of robustness for train schedules. Railway
Technical Research Institute, Quarterly Reports, 48(4), 197–201.
Transport Focus. (2014). How late is late – What bus passengers think about punctuality and timetables.
Retrieved June 30, 2015, from http://www.transportfocus.org.uk/research/publications/how-late-
is-late-what-bus-passengers-think-about-punctuality-and-timetables-full-report
Tsuchiya, R., Sugiyama, Y., Yamauchi, K., Fujinami, K., Arisawa, R., & Nakagawa, T. (2006). Route-choice
support system for passengers in the face of unexpected disturbance of train operations. Computers
in Railways X. In Proceedings of the tenth international conference. WIT, Boston.
Vansteenwegen, P., & Oudheusden, D. V. (2006). Developing railway timetables which guarantee a
better service. European Journal of Operational Research, 173(1), 337–350.
Vansteenwegen, P., & Van Oudheusden, D. V. (2007). Decreasing the passenger waiting time for an
intercity rail network. Transportation Research Part B: Methodological, 41(4), 478–492.
Vij, A., Carrel, A., & Walker, J. L. (2013). Incorporating the influence of latent modal preferences on
travel mode choice behavior. Transportation Research Part A: Policy and Practice, 54, 164–178.
Vromans, M. (2005). Reliability of railway systems. (PhD dissertation). Rotterdam School of
Management, Erasmus University.
Vromans, M. J., Dekker, R., & Kroon, L. G. (2006). Reliability and heterogeneity of railway services.
European Journal of Operational Research, 172(2), 647–665.
Wardman, M., Chintakayala, P., de Jong, G., & Ferrer, D. (2012). European wide meta-analysis of values
of travel time. Report prepared for the European Investment Bank.
Weston, P., Goodman, C., Takagi, R., Bouch, C., Armstrong, J., & Preston, J. (2006). Minimising train
delays in a mixed traffic railway network with consideration of passenger delay. 7th World
Congress on Railway Research, Montréal, Canada.
 A LITERATURE REVIEW 27

Yaghini, M., Khoshraftar, M. M., & Seyedabadi, M. (2013). Railway passenger train delay prediction via
neural network model. Journal of Advanced Transportation, 47(3), 355–368.
Yamamura, A., Koresawa, M., Adachi, S., & Tomii, N. (2012). Identification of causes of delays in urban
railways. Computers in Railways XIII. In Proceedings of the thirteenth international conference.
Boston: WIT.
Yuan, J., & Hansen, I. A. (2007). Optimizing capacity utilization of stations by estimating knock-on train
delays. Transportation Research Part B: Methodological, 41(2), 202–217.
Downloaded by [DTU Library] at 14:01 06 January 2016

View publication stats