Big Data To Improve Strategic-1
Big Data To Improve Strategic-1
Big Data To Improve Strategic-1
Maximilian Schosser
LEIPZIG
GRADUATE SCHOOL
OF MANAGEMENT
Schriftenreihe der HHL Leipzig
Graduate School of Management
Springer Gabler
© Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2020
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part
of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission
or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, expressed or implied, with respect to the material contained
herein or for any errors or omissions that may have been made. The publisher remains neutral with
regard to jurisdictional claims in published maps and institutional affiliations.
This Springer Gabler imprint is published by the registered company Springer Fachmedien
Wiesbaden GmbH part of Springer Nature.
The registered company address is: Abraham-Lincoln-Str. 46, 65189 Wiesbaden, Germany
For my wife Karin who filled the days of my PhD studies with pure joy.
For my mother Jutta who taught me the most important trait as a re-
searcher – curiosity.
For my father Rudolf whose perfectionist mind helped with great sugges-
tions for this thesis.
Foreword
Big data not just evolved as a popular buzzword over time but is meanwhile
seen as a high potential field for improving business processes and deci-
sions taken by responsible persons in different industry sectors, such as
airlines.
Nevertheless, available advantages of big data have not yet been ade-
quately measured from an economic perspective and have very often not
yet been exploited at all or to a reasonable level. Additionally, correspond-
ing theoretical concepts and scientific developments focusing on the area
of network planning in airlines are currently more or less missing. Those
challenging deficits make the topic considered by Mr. Schosser highly rel-
evant both from a theoretical as well as a practical perspective.
Thus, the main objective of the thesis consists in closing both the scientific
research gap and providing a solution to practitioners focusing on the as-
sessment of big data opportunities in the network planning context of air-
lines.
The author provides with the step-by-step, theoretically and empirically
based development of the framework, its elements and procedures in the
present doctoral thesis an outstanding analytical as well as conceptual per-
sonal contribution. The framework is from a content-related point of view
to be honored as a pioneering achievement and the dissertation contains
a lot of new findings that represent a starting point for further work predom-
inantly in the research and practice field of big data evaluation focused on
airline network planning that can be transferred to further use cases re-
spectively fields of applications.
The book, which is based on a dissertation at the HHL Leipzig Graduate
School of Management, is aimed equally at readers from science and prac-
tice, dealing with (big) data collection, economic evaluation of big data as
well as big data analytics.
The idea for this PhD thesis was born during a consulting project at a Eu-
ropean airline, where different departments were at completely different
maturity stages of using Big Data. While at some departments network
planning was done the old-fashioned way, with legacy IT-systems, data
and processes, other departments had already implemented a much more
agile way of integrating Big Data.
The network planning department was frequently approached by data pro-
viders with concrete offers, but there was no method to evaluate the impact
of the data for airline network planning, and most of the offers were turned
down for this reason. At the same time, there was no research available to
answer this question. The main objective of this PhD thesis is hence to
alleviate the lack of evaluation methods and provide a concise answer/ap-
proach/framework of how Big Data can create value for individual network
planning steps.
The target audience of this thesis comprises airline network planners of all
seniority levels and fellow researchers working on Big Data applications
for the transportation industry, on network optimization problems or on
commercial topics in airlines. I tried to find a middle ground between con-
tent of pure scientific interest (e.g., chapters 2 and 3), and presenting rel-
evant findings for practitioners (chapters 4, 5 and 6).
Finally, I want to thank everyone who contributed to this PhD thesis, in
particular my thesis advisor Prof. Dr. Iris Hausladen, who provided guid-
ance and challenge whenever necessary. Further, I want to thank all inter-
view partners of our airline case study group, who need to stay anonymous
due to non-disclosure agreements. Philipp Behrends, Karin Garcia and Ru-
dolf Schosser made an invaluable contribution by reviewing and comment-
ing on various drafts of this thesis. My thanks go also to all staff and stu-
dents of HHL Leipzig Graduate School of Management who made my time
there so enjoyable. Finally, I want to thank Maximilian Rothkopf for helping
to shape the idea and David Speiser for his support of my PhD project. My
PhD thesis would not have been possible without the financial support dur-
ing my educational leave granted by the Zurich Office of McKinsey & Com-
pany.
1 Introduction............................................................................................ 1
2 Methodology .......................................................................................... 5
3 Theoretical foundation.........................................................................29
7.3 Revisiting the RBV for big data in airline network planning ......333
8.1.4 How can the impact of big data opportunities for airline
network planning be quantified [RQ 4]? .......................................345
References .............................................................................................411
List of Figures
Big data has become one of the most prominent buzzwords of the modern
corporate world, which is a remarkably fast feat for a new phenomenon. It
became fashionable at the onset of the 2010s, fueled by popular articles in
the Harvard Business Review (McAfee & Brynjolfsson, 2012) and the MIT
Sloan Management Review (Davenport, Barth, & Bean, 2012). Its broad
scope makes it directly or indirectly crucial for all industries, and as early
as 2011, Manyika et al. (2011, p. 1) estimated a global big data (BD) value
generation potential of more than 1 trillion US Dollar (USD).
Since airlines had been producing vast amounts of data even before the
term big data was coined, its potential value in the industry was noticed
early on (Noyes, 2014). In a recent survey, more than 60% of aviation ex-
ecutives named big data as a top priority for their company (Hodgson &
Waldmeir, 2018). The airline business is receptive to a wide range of big
data applications: the single customer view enables hyper-personalized
customer experience and marketing activities (Chen et al., 2017; Hodgson
& Waldmeir, 2018; Noyes, 2014); higher granularity of customer insights
can be utilized for customized pricing and revenue management (Boin et
al., 2017); real-time information on transfer passengers and irregular situ-
ations can help reduce delays and smoothen flight operations (Bowcott &
Dichter, 2014; Chen et al., 2017); finally, predictive maintenance can re-
duce unplanned aircraft groundings and lower maintenance cost (Brad-
bury, 2018; Chen et al., 2017).
Although airlines are at the forefront of big data adoption, not all business
processes and departments in the industry participate at the same speed.
Bertram (2017) proposes various big data applications for network plan-
ning (NP), including mobile location data and real-time flight radar data.
However, most NP departments in airlines have not yet adopted any big
data application in their NP process.
Although a holistic big data value chain has been proposed by Gandomi
and Haider (2015, p. 141), little research has been conducted on the
sources and types of big data (Akoka et al., 2017, p. 111). While infor-
mation systems architecture and big data analytics techniques are under
1
Govindan et al. (2018) found an 8-fold increase of academic publications with big data in
title, abstract or full-text between 2014 and 2017.
Objective of the study and research questions 3
academic focus, big data sources are mostly treated as input factors and
not as key value contributors.
This study aims to bridge this gap in two ways. First, it shifts the attention
from big data architecture and analytics towards the opportunities evolving
from specific big data sources. Even the most advanced infrastructure and
analytic tools cannot create value if they use meaningless data. Second,
the study provides a perspective on big data applications for airline NP,
and thus complements research on other domains of application in airlines.
The objectives of this study aim to close the scientific research gap laid out
in the previous section and address the practical problem discussed previ-
ously in order to evaluate big data opportunities for airline NP. As the first
step, the most recent research on big data and on NP in airlines needs to
be assessed. Based on the existing body of literature, a clear definition of
big data in the context of airline NP must be developed to build a solid
foundation to find solutions to the practical problem.
Based on the status quo of airline NP, the information requirements of air-
line network planners can be identified. BDOs should address these infor-
mation needs to create value for airline NP. The definition of concrete
4 Introduction
Finally, the development of a cost and benefit evaluation scheme for BDOs
in airline NP should address the practical problem described in the previ-
ous section. This requires the determination of relevant key performance
indicators to measure the contribution of big data opportunities in terms of
financial value.
2
Source: Own illustration.
2 Methodology
The choice of the research methodology does not depend only on the char-
acteristics of the research topic or the individual preferences of the re-
searcher. It should also be underpinned by a sound philosophical research
paradigm and must be suitable to answer the research questions and fulfil
the objectives defined in the previous chapter. Sub-chapter 2 presents the
paradigm that drives the research. Detailed descriptions of the literature
review process (sub-chapter 2.2) and of the multi-stage comparative case
study (sub-chapter 2.3) introduce the major techniques employed for the
rest of this study.
3
Source: Own illustration.
This study adopts the pragmatic view as the research paradigm. Pragma-
tism incorporates elements of both post-positivism and constructivism by
accepting the co-existence of subjective, inter-subjective, and objective re-
ality (Antwi & Hamza, 2015, p. 223). Research problems in management
sciences, especially emerging phenomena, are rarely ever completely
quantifiable or solely based on human interactions (Wynn & Williams,
2012, p. 789). Hence, pragmatists emphasize on the consequences of ac-
tions that cause real-world problems, which are addressed in practical re-
search (Creswell, 2013, p. 6).
The four research questions defined in chapter 1 are clearly derived from
a practical problem – the vagueness of BDO benefits for airline NP. The
objective is not only to understand the causal effects of big data, mani-
fested in the expected economic impact, but also to analyze the current
processes and beliefs followed by airline network planners. Furthermore,
the recency of big data calls for a holistic investigation, as there may not
be sufficient material available to answer the research questions empiri-
cally. The combination of (1) the practical research focus, (2) an emergent
research topic, and (3) the complexity of the four research questions re-
quire the adoption of a pragmatist perspective for this research project.
4
Own illustration based on Creswell (2013, pp. 157–211).
8 Methodology
5
Own illustration following Yin (2017, p. 8).
Development of research design 9
Figure 2.2 schematizes the four phases of the research design. The de-
tailed methodology, including the chosen techniques, are presented in the
following sub-chapters. The set-up and execution of the structured litera-
ture review is explained in section 2.2. Section 2.3 is dedicated to explain-
ing the case study set-up, including the data collection and survey design
methodology. Chapters 3–6 of this study present the theories and findings
from each respective research question (see Figure 2.2).
6
Source: Own illustration.
10 Methodology
Figure 2.3 – Tactics to ensure the quality criteria for selected research methods7
7
Own illustration following Yin (2017, p. 41), Creswell (2013, pp. 155–157), Cooper and
Schindler (2000, pp. 215–218), and Vom Brocke et al. (2009, p. 2212).
Literature review 11
discussed quality criteria. The tactics are implemented in the research de-
sign and will be further described in the sections dedicated to them.
Paré et al. (2015, pp. 185–189) identified nine literature review archetypes
based on the research objective and the dominant research methods of a
project8. According to this classification, a realist review is most suited for
this study, since the research questions are already defined, and the over-
all research strategy follows a qualitative approach with quantitative ele-
ments. A realist review should follow an iterative and purposive search
strategy and include both conceptual and empirical works. The studies can
be down-selected based on pre-defined formal criteria (e.g., journal quality
or publication date) or qualitative criteria (e.g., abstract reviews).
8
Please refer to Appendix 1 for a complete overview of the typology of literature review.
12 Methodology
tions and databases that pertain to the relevant literature. To define appro-
priate search terms, Vom Brocke et al. (2009, p. 2214) proposed a struc-
tured decomposition of titles or research questions as well as the inclusion
of intended theories and research paradigms. This step is crucial to guar-
antee construct validity, i.e., to ensure that the search delivers the intended
results. These activities comprise Step 1 in Figure 2.4 and are presented
in detail in the next section.
Step 2 entails the actual keyword search in the selected databases and
publications. The results are evaluated through a three-step approach.
First, the title is evaluated to exclude literature on completely unrelated re-
search fields (e.g., medicine) or in languages other than English, German,
or Spanish. The second iteration is a qualitative abstract screening to ex-
clude thematically irrelevant contributions (e.g., literature on air traffic con-
trol optimization). The full-text evaluation is the last iteration to confirm or
refute the actual relevance of the contribution (e.g., non-content editorials
9
Own illustration following Vom Brocke et al. (2009).
Literature review 13
or book reviews). These steps need to be followed strictly for all individual
searches to ensure internal validity of the literature review.
The title of this study can be deconstructed as three major elements. “Big
data” describes the phenomenon that is the primary research subject. The
phenomenon is examined for a specific industry, namely “airlines,” and for
a specific business process, the “strategic network planning.” The two guid-
ing theories for this research are the resource-based view of the firm and
network theory10, which have also been included in the keywords. Syno-
nyms and adjacent search terms complement the extracted keyword in
each area (see Table 2.3).
Table 2.3 – Keyword extraction for the literature review11
10
Compare sub-chapter 3 for a detailed discussion on theory selection.
11
Source: Own illustration.
14 Methodology
For the identification of the relevant journals and databases, the primary
research domains for this study must be identified first. The research do-
mains of VHB-JOURQUAL 3 ranking (VHB, 2015), curated by the German
Academic Association for Business Research (VHB), are evaluated with a
scoring model based on the relative search results for each of the extracted
keywords (see Appendix 3 for the detailed scoring results).
12
Example: The complete Boolean Phrase for combination 4 would read as [“big data” OR
“data management” OR “business intelligence” OR “business analytics”] AND [“airline” OR
“aviation” OR “logistics” OR “transportation”] AND [“strategic network planning” OR “strategic
planning” OR “network planning”] AND [“resource-based view” OR “RBV” OR “network the-
ory”].
13
Based on VHB-JOURQUAL 3 ranking VHB (2015): Journals ranked A+ from all the relevant
research domains plus journals ranked A from logistics, operations research, strategic man-
agement and information systems. Appendix 4 provides a complete list of the 39 journals.
Literature review 15
During the initial search, the parameters were set to a full-text search and
limited to academic articles, conference proceedings, and books14. For the
initial search with one keyword only (combinations 1a to 1d in Appendix 2),
the search terms for object (big data), industry (airline), and theories (re-
source-based view OR network theory) resulted in at least 10,000 hits at
each access point, while the keyword chosen for the relevant process (stra-
tegic network planning) produced less than 1000 hits at all access points.
Therefore, the keyword selection for processes was extended with the two
decompositions “strategic planning” and “network planning.”
The initial round of the structured literature search followed the keyword
combination scheme presented in Appendix 2 and entailed 15 individual
searches. The search was commenced with the full combination of all rel-
evant keywords (4a) to identify the most relevant literature first. Subse-
quently, the search terms were relaxed by combining only three (3a–3d) or
two keywords (2a–2f). Finally, the single keyword search was repeated
(1a–1d). The titles and abstracts of the first 400 results of each search
were screened to identify the relevant works. Duplicates were only counted
in the first appearance. The detailed search results are listed in Appendix 6.
In total, 610 relevant studies were identified for the full-text evaluation, out
of which 207 passed the full-text review and were used for forward/back-
ward search. The main factor for rejection of a study was focus on different
research topics (158), such as air traffic control or communication network
planning. Other major reasons for rejection were lack of originality of re-
search content (86) and non-scientific research publications (71). The for-
ward/backward search was conducted iteratively during the full-text review
14
In Google Scholar, patents and citations were excluded.
16 Methodology
process and yielded 136 additional publications. The relevant body of liter-
ature for the theoretical aspect of this study consequently comprises 343
publications, as depicted in Figure 2.5.
Out of the 343 publications, 290 are articles from peer-reviewed academic
journals. Among these, 83 articles have been published in unranked pub-
lications, such as industry-specific journals and published conference pro-
ceedings. Appendix 7 summarizes the most frequent publications for each
VHB ranking category. Thus, most of the publications are from information
systems, operations research, and strategic management research do-
mains. The 53 remaining sources are mostly books (31), book chapters
(11), and research reports (4).
15
Source: Own illustration.
Literature review 17
16
The resulting literature is discussed in detail in section 3.7.3.
18 Methodology
Combination 3d, yields by far the most search results from all the third-
degree keyword searches17. A rich body of literature exists on airline net-
work planning, most of which originates from operations research (see sub-
chapter 3.5 for a detailed discussion). However, most of the studies focus
on optimization methods and not on used data.
Research that stem from the lower-degree keyword searches (1a–d and
2a–f) emphasize on specific aspects of big data, airlines, strategic network
planning, and the theoretical foundations. All of the analyzed studies lack
an integrated perspective of big data in strategic network planning for air-
lines, which presents a significant gap that this study aims to bridge.
The actual case study design follows the method proposed by Eisenhardt
(1989, p. 533), depicted in Table 2.4.
Table 2.4 – The case study method adopted from Eisenhardt (1989)
17
2’453 compared to less than 1’000 for all other 3rd-degree combinations (see Appendix 6).
18
Hypothesis on RQs developed within the respective chapter.
Comparative case study 19
The first step “Getting started” also includes the work required before de-
termining the case study design, which has been explained in the previous
sections of this chapter. This section starts with the selection of the case
study type (section 2.3.1), explains the theoretical case sampling (section
2.3.2), and presents the data collection and analysis techniques (section
2.3.3). Section 2.3.4 reviews the research quality criteria for the case study
research.
The first step in the actual case study design is determining the most ap-
propriate case study type. Eisenhardt and Graebner (2007, p. 27) have
proposed two generic case study types, namely single case studies and
multiple case studies. While single case studies are the most suited to de-
scribe the existence of a specific phenomenon, multiple case studies have
more explanatory power as they enable the researcher to compare the
same phenomenon within multiple settings. Baxter and Jack (2008, p. 550)
compared the design choice with the set-up of an experimental research
design. If the purpose is to discover a phenomenon, a single experiment
or case study is sufficient. If the properties of the phenomenon need to be
analyzed, multiple experiments – or multiple cases – are required. Yin
(2017, p. 47) has also stated that although multiple case studies have more
explanatory power, resource constraints can sometimes favor a single
case study setup.
For the airline cases, four relevant selection criteria were identified. Busi-
ness model is the most frequently used classification criterion in airline
management literature (Wittmer & Bieger, 2011, p. 31), and it will also be
used as the primary classification criterion in this study. However, other
influencing factors, such as fleet size, network type and route structure
have a considerable influence on the structure of the network process
(Goedeking, 2010, vii) and thus on the BDOs needed for its improvement.
For each selection criterion, three to four possible specifications were de-
fined (see Table 2.5).
Comparative case study 21
Criterion Specifications
Business model Full-service carrier (FSC)
Low-cost carriers (LCC)
Scheduled-charter airline
(SCA)
Cargo airline (CAR)
Fleet size Large (> 100 aircraft)
Mid-sized (30–100 aircraft)
Small (< 30 aircraft)
Network type Multi-hub
Single-hub
Point-to-point
Route structure Mostly long-haul
Mostly short-haul
Mixed
Four different business models are considered for the case sampling (Ster-
zenbach, Conrady, & Fichert, 2013, p. 225). Full-service or legacy carriers
(FSCs) are mostly developed from former state-owned airlines, and they
target a wide spectrum of passengers. They usually offer a premium prod-
uct and additional services, such as a frequent flyer program. Low-cost
carrier (LCCs) are the alternative draft to this model, which offers a product
with the bare minimum (“no frills”). Scheduled charter airlines (SCAs) or
“leisure carriers” focus on touristic routes and sell large seat contingents to
travel operators. Cargo airlines (CARs) focus on transporting freight only.
The business models are discussed in full details in Section 3.2.3.
Fleet size serves as a proxy for airline size, since that data is more easily
available than financial data or passenger data20. It is distinguished in three
levels, ranging from small (< 30 airplanes) to large (> 100 airplanes). Tra-
ditionally, network types are distinguished in hub-and-spoke (HS) and
19
Source: Own illustration.
20
Financial data for privately owned airlines is not published often; passenger data is availa-
ble consistently only for IATA member airlines.
22 Methodology
21
Source: Own illustration.
22
Assessed on October 4, 2016 with data from airfleets.net (2016).
Comparative case study 23
complement the data collected in the comparative case study. They are
not part of the actual case study group but serve as tri-angulation partners
who help define the feasibility of potential big data opportunities.
Figure 2.1: the first phase aims to assess the status quo of NP in airlines;
the second phase intends to discover and qualitatively evaluate potential
BDOs; and the last phase aims to develop an estimation methodology for
the financial impact of investment in big data.
23
Source: Own illustration. following Berg (2004).
Comparative case study 25
2.7 summarizes the data collection and analysis process of the status quo
assessment phase, whose results are presented in chapter 4.
There might be the case that an airline has not mentioned a data type in
response to the open question, even though it is being used. In order to
capture the complete picture of the data sources and IT systems used, the
first interview is complemented with a structured questionnaire that consti-
tutes all coded items, including the data types labelled as “new.” The struc-
tured questionnaire includes three questions per data source. The first
question entails a simple confirmation on whether a specific data type is
used (yes/no). The second and third questions evaluate the usefulness
and reliability of a data source. Berg (2004, pp. 105–108) recommends the
use of four-point rating scales for the measurement of usefulness. A clas-
sical Likert scale is not appropriate here, as the questionnaire does not
24
Source: Own illustration.
26 Methodology
capture personal attitudes towards an item but intends to rate them objec-
tively. The questionnaire for the status quo interviews is attached in Ap-
pendix 8 and the subsequent survey in Appendix 9.
Before the actual data collection, nine distinct information needs are de-
rived from the status quo assessment and the literature review on the NP
process. To satisfy these information needs, 23 potential BDOs are devel-
oped from three different sources, namely the initial status quo interviews,
a press review, and the analyzed body of literature.
These 23 BDOs are then presented and explained to the airline network
planners in the case study group. After the presentation, a standardized
online survey (see Appendix 10) asked for the perceived benefit potential
for the airline-specific NP. To increase granularity in the responses, a six-
point scale was chosen (Berg, 2004, pp. 105–108).
25
Source: Own illustration.
Comparative case study 27
gated for each BDO in order to calculate a total feasibility score. The aver-
age perceived benefit potential and the total feasibility score of each BDO
are then plotted on a two-dimensional matrix. The 12 BDOs among these
with the highest benefit potential and feasibility ratio are selected for further
analysis.
The last interview round was related to the key performance indicators
(KPIs), information on which was collected through the semi-standardized
interviews in the exploratory phase. The airline network planners were
asked to evaluate the impact of each of the 12 shortlisted BDOs on the
primary and secondary KPIs of their airline. Similar to the initial benefit
evaluation in the refinement phase, a six-point assessment scale was cho-
sen. The results of the structured interview were then used to inform the
development of the business case scheme to quantify the financial benefits
of the shortlisted BDOs.
In accordance with the data quality criteria derived from the literature in
sub-chapter 2, displayed in Figure 2.3, specific tactics have been deployed
to ensure the reliability and validity of the comparative case study.
A case study database, which includes the case study protocol, has been
created to ensure research reliability. This database contains all the inter-
view transcripts as well as the associated analyses26. The interview tran-
scripts and the report summaries were sent to the interviewees for verifi-
cation to ensure construct validity. Furthermore, specific sections of the
interviews were verified with system developers and data providers to tri-
angulate information.
26
Due to data confidentiality, the complete transcripts cannot be appended to this thesis, as
the doctoral thesis will be publicly available. The anonymized case reports can be requested
from the author.
28 Methodology
The theoretical sampling of the comparative case study has a positive in-
fluence on both internal and external validity. The comparison of different
cases enabled pattern matching, helping identify potential biases. Logical
models derived from the literature could also be tested with multiple cases,
since all sampling criteria were present in at least two cases. The research
design applied the same logic to all cases to ensure external validity. This
means that all interview protocols were standardized across cases as well
as across supporting sources.
3 Theoretical foundation
Figure 3.1 compares the search results of the combinations of the consid-
ered theories herein with research domains and research topics. It identi-
fies the most appropriate specific theories for this study. It is hereby as-
sumed that a large number of search results within each domain or topic
indicates whether a theory is widely used and well established.
The most widely cited theories, across research domains and topics, are
the resource-based view, game theory, and network theory. These theo-
ries are all well-established in the three research domains (Logistics, OR,
and IS) and two research topics (big data and airlines) being studied. How-
ever, the resource-based view (RBV) is the only theory which has a strong
influence on strategic management research, and network theory is the
only well-established theory for strategic NP.
Sub-chapters 3.2 and 3.6 define and describe the research objects “air-
lines” and “big data” respectively. After an introduction to network theory in
sub-chapter 3.3, the third research object “strategic network planning” is
27
Own illustration; the selection of candidate theories based on Defee et al. (2010); the circles
reflect a relative number of search results from EBSCOhost search combining each theory
with research area and research topics, e.g. (“resource-based view” AND “logistics”). Abso-
lute search results in title and abstract have been recorded and standardized for each re-
search domain/topic. A full circle represents >70% of the highest search result, an empty
circle <10% of the highest search result.
Airlines and their business models 31
This sub-chapter defines the research topic, airlines, for clarity in the fur-
ther course of this study. Section 3.2.1 delimits the airline industry within
the wider sphere of aviation systems. Developments and recent trends in
28
Source: Own illustration.
32 Theoretical foundation
the airline industry are summarized in section 3.2.2. The last sections pre-
sent different business models followed by airlines (3.2.3) and the most
important business processes (3.2.4).
Airlines are one group of actors in the broader aviation system. Wittmer
and Vespermann (2011, pp. 39–40) have described the aviation system as
a classic economic demand and supply system (see Figure 3.3).
The supply side of the system contains the aircraft manufacturers, airlines,
airports, and aviation service providers. Aircraft manufacturers produce the
movable transportation assets, whereas airports provide the immobile in-
frastructure assets. The broader industry of aircraft manufacturing is also
referred to as the aerospace industry, comprising end-manufacturers, such
29
Source: Own illustration. adapted from Wittmer and Vespermann (2011, p. 40).
Airlines and their business models 33
30
Examples: air traffic control, aviation system providers, aviation data providers, ground han-
dling, catering, maintenance providers, aircraft leasing companies.
34 Theoretical foundation
Historically, the development of the airline industry has been heavily influ-
enced by technological progress and changes in regulation. Most authors
such as Doganis (2010, pp. 25–63) and Gillen (2006, pp. 367–369) distin-
guish between a regulated and a deregulated era.
Until 1978, all major aviation markets were heavily regulated. All routes
needed governmental approval in specifying the permitted frequency, seat
capacity, and even the ticket price, which was calculated based on the dis-
tance and service components (Doganis, 2010, p. 36). This led to the cre-
ation of linear networks, which are economically inferior to the point-to-
point or hub-and-spoke networks (see section 3.4.2 for a detailed discus-
sion) in a free market.
The deregulation of the air transport system started with the Airline Dereg-
ulation Act in the United States in 1978 which was completely implemented
in 1985 (Gillen & Morrison, 2005, p. 162). This deregulation had a severe
impact on the airline business models and networks. The “open sky” poli-
cies31 allowed the establishment of more efficient network types, namely
the point-to-point and hub-and-spoke networks (see section 3.4.2). The
abolition of pre-defined fares led to a sharp decrease in airfare, hence in-
creasing cost pressure on the airlines (Doganis, 2010, p. 47). This sparked
both the consolidation of legacy carriers and the development of new busi-
ness models, most notably of the low-cost carriers (Gillen, 2006, pp. 370–
372).
31
Airlines can connect whichever airport pair they wish to, without regulatory approval; how-
ever, limited airport slots hinder the unlimited choice of connections.
Airlines and their business models 35
major alliances (Star Alliance, Oneworld, and Skyteam) control over 60%
of the total available passenger airline capacity (Statista, 2018).
Business models are used as the most common classification factors for
airlines, since they constitute the “fundamental differences […], i.e. the ser-
vice level offered, the regional reach, and the main function of an airline”
(Wittmer & Bieger, 2011, p. 31). Therefore, this study also adopts airline
business models as the main classification criterion.
After the de-regulation wave in the 1980s and 1990s, various new business
models emerged. The most popular new business model are low-cost car-
riers (LCCs) that serve point-to-point networks with a very low cost struc-
ture (Shaw, 2016, pp. 100–101). At the other end of the service spectrum,
super-differentiated airlines emerged to serve high-value business routes
with a premium product as a cheaper alternative for private jets (Bieger &
Wittmer, 2011, p. 98).
36 Theoretical foundation
32
Source: Own illustration. based on literature mentioned in the table.
Airlines and their business models 37
33
Airports with high traffic volumes assign landing time windows (i.e., slots) to airlines to avoid
congestion of the available infrastructure.
38 Theoretical foundation
34
Source: Own illustration. adopted from Sterzenbach et al. (2013, p. 225).
Airlines and their business models 39
Prior to the deregulation, SCAs were not permitted to market single seats;
rather, they could only sell seat allotments to tour operators (Bieger & Witt-
mer, 2011, p. 97). Post deregulation, many formerly independent airlines
were integrated into larger tourism conglomerates35 in order to reduce the
transaction cost and adjust capacity more flexibly. SCAs are characteristi-
cally smaller than FSCs, although they can serve both short-haul and long-
35
E.g., Hapag-Lloyd to Preussag (later TUI) in 1997, Condor to C&N Touristik (later Thomas
Cook) in 1997.
40 Theoretical foundation
Regional carriers
36
However, the revenue share of cargo in long-haul wide-body aircraft is up to 15% (Doganis,
2010, p. 306).
Airlines and their business models 41
and many ancillary services that can be assembled a la carte by the pas-
senger to best suit his preferences. Ancillary revenues can generate up to
30% of the total revenues (Doganis, 2010, p. 152). This led to a focus on
the load factor rather than average yield, since most ancillary revenues are
generated after the actual booking. At the same time, sophisticated yield
management systems, as used by FSCs, became obsolete.
Super-differentiated airlines
Hybrid carriers
With the market entry of the new LCCs, strategic reactions of the FSCs led
to innovations in their own business model and service offering. Simulta-
neously, LCCs added premium services to their own product set, so that
the lines between these business models became blurry (Daft & Albers,
2013, p. 47). The new breed of business models has often been called the
“hybrid” business model, since they incorporate elements of both FSCs
and LCCs in varying degrees (Lohmann & Koo, 2013, p. 7). Such business
model convergence can occur in two different ways: first, airlines can alter
their product portfolio. This entails either FSCs decomposing their bundled
full-service product and offering a la carte ancillaries similar to LCCs; con-
Airlines and their business models 43
versely, LCCs can start to bundle some of their a la carte products in some-
thing similar to full-service products (Daft & Albers, 2013, p. 50). Second-
arily, airlines can establish their own LCCs to compete directly with the new
market entrants and benefit from the low-cost structure of a newly founded
airline. This results in large airline groups operating both FSCs and LCCs
simultaneously but with clearly delimited scope and target markets (Pompl,
2007, p. 127). Airlines in North America have mostly chosen hybridization
through product portfolio adjustments, whereas European airlines have de-
ployed a mix of both hybridization approaches (Daft & Albers, 2015;
Klophaus, Conrady, & Fichert, 2012).
Operational processes are the responsibility of the classic cost centers and
thus strive to provide a satisfactory level of service at minimal cost. Hence,
efficiency is the most important innovation driver for operational processes
(Wensveen, 2015, p. 248). Commercial processes are usually revenue
centers, with the exception of NP, which can alternatively be a profit center
that coordinates commercial and operational processes (Goedeking, 2010,
p. 99). Strategic planning, including fleet planning, is usually considered an
administrative function reported directly to the CEO (Wensveen, 2015,
p. 241). Administrative processes are either cost or profit centers in the
organization, depending on the responsibilities and decision-making rights
assigned to them.
44 Theoretical foundation
Networks are omnipresent in almost all aspects of life and science. They
can take different appearances, sizes, and strengths. Networks are clearly
visible (e.g., a road network) or almost invisible (e.g., a social network).
Even in airlines, there exist numerous network structures: route networks,
airline alliance networks, computer intranet networks, and so on. In this
sub-chapter, section 3.3.1 first defines the term “network” in general and
discusses different network forms relevant for scientific research, with an
aim to clearly delimit the airline route networks from other network types.
Thereafter, section 3.3.2 presents the main scientific theories to describe
and analyze networks. The most important theories are introduced in de-
tail, namely the graph theory (section 3.3.3) and flow networks (sections
3.3.4 and 3.3.5).
37
Source: Own illustration. following Wensveen (2015, pp. 233–263).
Introduction to the network theory 45
the network. In social networks, the objects are usually human actors or
groups of actors and the connections are the relationship between these
actors (Brass et al., 2004, p. 795). In technical networks, the objects can
be computers, televisions, cities, or power plants as examples, and the
connection is usually a physical connection, such as a wire, an electromag-
netic signal, or a road.
38
Source: Own illustration. following Nakicenovic (1995), Bell and Iida (1997), and Frybourg
and Nijkamp (1998).
39
Let us consider social networks as an example. Network flows of affect are not quantifiable
(Granovetter, 1973), and information and knowledge are at least difficult to quantify (Brass et
al., 2004). The transmission of these flows in relationships is direct, i.e., there is no need for
a transition mode. The only capacity supply constraint present in social networks are thus the
network objects themselves – the amount of information, knowledge, and affect that social
actors are willing or able to transmit in a social network.
Introduction to the network theory 47
mode needs to return at some point to keep the network running. Counting
deadheads as bi-directional network flow, reciprocity is a necessary condi-
tion for transportation networks.
40
Retrieved via EBSCOhost on October 30, 2017 with Boolean search phrase “social net-
work” and “(“transportation network”) OR (“transport network”)” and filter for peer-reviewed
academic journals only.
Introduction to the network theory 49
Apart from theories describing the static and dynamic properties of net-
works, there also exist theories related to the impact of networks. From a
strategic management perspective, the expectation towards a network –
independent of its type – is to contribute towards building a competitive
advantage (Gulati, Nohria, & Zaheer, 2000, p. 204). When considering the
effect of networks in the management research domain, it is indispensable
to relate networks to the common theories on competitive advantage in
strategic management. For this purpose, section 3.7.5 discusses networks
in the context of the resource-based view.
The graph theory has mostly been developed by mathematicians and can
be traced back to Leonhard Euler’s Königsberg bridge theorem in 1736
(Casti, 1995; Newman, Barabasi, & Watts, 2011). Every graph consists of
41
Source: Own illustration.
Introduction to the network theory 51
Directed graphs allow one flow direction for each link. In contrast, undi-
rected graphs allow bi-directional flows between nodes (Newman, 2003,
p. 172). All graphs can also be represented in a matrix. Figure 3.6 presents
an undirected graph with six nodes, A to F, and five links. The properties
of graph connectivity can be best elucidated using the three matrices in
42
Source: Own illustration. following the authors mentioned in the table.
52 Theoretical foundation
Figure 3.6. The adjacency matrix summarizes whether two nodes are di-
rectly connected via a link in a binary system. The node pair AB is con-
nected by a direct link; hence, the adjacency matrix value is 1. Similarly,
node pair BC has no direct link, so the adjacency matrix value is 0.
Besides the strongly connected graph, there are other graph types that are
depicted in Figure 3.7. A graph where every node is connected with direct
links to all other nodes is called a hyper-connected graph (Casti, 1995). If
43
Source: Own illustration.
Introduction to the network theory 53
graphs contain unconnected nodes or two sub-graphs that are not con-
nected, these entities are called components (Newman, 2003, p. 173). The
graph in the middle displays a two-component graph, consisting on the
components DE and ABCF. Finally, there may be different types of links
as depicted on right side of Figure 3.7. These links may differ according to
the types of flows they transmit or in the flow capacity (Newman, 2003,
p. 181). Examples of these networks are road-railway cargo networks or
high- and low-voltage electricity networks.
Besides graph connectivity, Albert and Barabási (2002, p. 71) have men-
tioned three key network attributes: average path length, clustering coeffi-
cient, and degree distribution. Newman (2003, p. 189) has added network
resilience and betweenness to the attribute analysis.
The average path length describes the arithmetical average of all the geo-
desic paths of a network. This measure has received a large amount of
public and scientific attention, triggered by Milgram’s (1967) small-world
problem, in which he found that the average path length of the social net-
work of all humans on earth is between 6 and 7 links. In other words, most
44
Source: Own illustration.
54 Theoretical foundation
people are connected with at least half of the entire human population via
7 or fewer links. The implications of the small-world problem for graph the-
ory are discussed at the end of this section. The betweenness or centrality
measure indicates how many geodesic paths intersect at a specific node
(Newman, 2003, p. 194). This also is the theoretical base for influencer
modelling in social network analyses (Wasserman & Faust, 1994).
If we take the graph in Figure 3.6 as an example, we can identify one tri-
angle ABD, and ten triple relationships45. Therefore, the clustering coeffi-
∗
cient is = 0.3. As presented in Table 3.4, the degree of a node is the
number of links connecting to the node. The degree distribution indicates
the probability of a node i to have k degrees (Albert & Barabási, 2002,
p. 73). Figure 3.8 exemplifies a degree distribution for the graph introduced
earlier in this section.
45
Triple relationships in undirected networks are counted bi-directionally. For example, the
path ABE has its counterpart as EBA. The 10 triples in Figure 3.6 are ABE, AFC, BAF, CFA,
DBE, DAF, EBA, EBD, FAB, and FAD.
Introduction to the network theory 55
46
Source: Own illustration.
56 Theoretical foundation
Flow networks have some more properties than static networks. The start-
ing point of a specific flow is called the origin or source. The last node of a
specific flow is the destination or sink. Nodes with a negative net flow are
called demand nodes, since more flows originate than those that arrive. In
turn, supply nodes have more arriving flows than originating flows, creating
a positive net flow. If the net flow is zero, the node is called a transshipment
node (Bazargan, 2016). If a network is separated into multiple independent
networks by deleting one or more links, the combination of eliminated links
is called the cut (Casti, 1995).
Introduction to the network theory 57
In the following paragraphs, the most important network flow issues and
their practical implications for transportation networks are introduced.
The shortest path problem is the most fundamental network flow problem
and a basis for almost all network optimization theories. First formulated
by Czech mathematician Otakar Borůvka in the 1920s (Nešetřil, Milková,
& Nešetřilová, 2001), it aims to find the optimal paths for specific source
destinations in higher-order networks. In static network theory, using a link
comes with no cost, so the optimal path is always the geodesic path47. In
a flow network, links get assigned a distance or a cost (Ahuja, Magnanti, &
Orlin, 1993, p. 94). Hence, the problem is also known as the minimum cost
problem. Consider a network such as the one depicted by the directional
graph in Figure 3.9.
47
The geodesic path is the path connecting two nodes with the smallest number of links.
48
Own illustration following Bazargan (2016).
58 Theoretical foundation
Each link has an associated flow cost that can also be viewed as the travel
distance between two nodes. The geodesic paths connecting nodes A and
D are ABCD and AFED. However, if we consider the link distances (cost),
the shortest (cheapest) path from A to D is ABCED with a distance (cost)
of 10, whereas the geodesic paths have lengths of 12 (ABCD) and 14
(AFED).
Since the shortest path problem does not consider any capacity con-
straints, it is not wholly transferable to most transportation networks. How-
ever, it is very useful to understand the basic mechanism of geographic
routing that applies to all physical transportation networks (Goldman &
Nemhauser, 1967). The most common application in real transportation
networks is traffic planning on road networks, as it is applied by traditional
navigation systems without real-time traffic information for example (Zhan
& Noon, 1998, p. 66).
Consider that you want to drive from Frankfurt to Berlin at night with little
traffic, so the capacity constraint of the road system is negligible. Most nav-
igation systems provide you with two route options, the shortest and the
most economic route. When considering the shortest route, the navigation
system assigns distances to the links to calculate the shortest path. When
choosing the most economic option, the navigation system assigns travel
times (which equal cost) to the links to calculate the path with the shortest
travel time. This example demonstrates that one network – in this case the
German road network – can have varying allocations of distance and cost.
Nevertheless, the algorithm required to determine this remains the same.
Over the last 90 years, the classical shortest path problem has produced
a significant body of literature. Kruskal (1956) and Dijkstra (1959) proposed
the first solution algorithms with exact but computational heavy solution
algorithms that examined every network link. Subsequent research fo-
cused on reducing the computing time and not overly deteriorating the fol-
lowing empirical results obtained. Dreyfus (1969) and Ahuja, Magnanti,
Introduction to the network theory 59
According to Ahuja et al. (1993, p. 184), the most important theorem of the
maximum flow problem in the entire network optimization is the max-flow
min-cut theorem introduced by Ford and Fulkerson in 1956. Based on em-
pirical observations in the Soviet railway system, they developed a simple
rule to determine the maximum flow capacity of a network. The maximum
capacity of the network is the capacity of an individual cut with the smallest
capacity of all cuts. Consider the same network as in the minimum path
problem, depicted in Figure 3.10.
60 Theoretical foundation
The number value now indicates the maximum capacity for each link. The
dotted lines show all possible cuts of the network, which are summarized
in the table on the right. The maximum capacity of each cut is the aggre-
gated flows that can flow from the left part of the cut to the right part of the
cut under the consideration of flow directions. For instance, cut b separates
the left part AF from the right part BCDE. The links connecting the two parts
are AB, FB, FC, and FE with an aggregated capacity of 13 flow units. Cut
c relies only on the links BC and AF to connect the left part AB and the
right part CDEF of the network. Here, the maximum capacity is only 7 flow
units, which is also the maximum capacity of the entire network.
The generic problem with “bottlenecks” limiting the capacity of large trans-
portation networks are most commonly observed in road and rail transport,
but it is also present in airline networks and shipping networks. However,
most capacity restrictions for the latter are determined by node capacity
restrictions (airports and sea ports), although waterways, such as the Suez
canal or the Panama canal, have capacity restrictions for shipping routes.
49
Source: Own illustration. following Ahuja et al. (1993, p. 184).
Introduction to the network theory 61
The minimum cost flow problem is the combination of the maximum flow
and shortest path problems. In a directed network with one or more origins
and one or more destinations, each link has both capacity restriction and
associated flow cost. This is much closer to the reality of many transporta-
tion networks than the simpler problems presented earlier50 (Ahuja et al.,
1993, p. 295).
50
The minimum cost flow problem has undergone multiple specifications and generalizations
that lead to a variety of special cases and sub-problems. Ahuja et al. (1993, pp. 317–337)
provided a comprehensive collection of these modifications as well as developed solution
algorithms.
62 Theoretical foundation
Figure 3.11 – Exemplary network and solution for the minimum cost flow problem51
The minimum cost flow problem and its derivatives are of central im-
portance for transportation networks. First, it provides a sound set of solu-
tion algorithms for general routing problems. The network in Figure 3.11
could for example be understood as a simple network of a cargo airline
transporting a single commodity. The intermediate stop D has a short run-
way and thus can only be served by small aircrafts, carrying up to 50 tons
of cargo. The minimum cost flow problem here offers a simple routing al-
gorithm to optimize this network.
51
Source: Own illustration. following Bazargan (2016).
Introduction to the network theory 63
However, network design is in fact much more complex than network opti-
mization. Network optimization problems are characterized by solving one
well-specified problem, and more importantly, it is usually towards one sin-
gle optimization parameter. However, network design requires encom-
passing and weighing all individual optimization problems and usually has
multiple optimization parameters (Magnanti & Wong, 1984, p. 2). Consider
airline network design as an example. Besides defining the most profitable
network, you also want to design a feasible network, i.e., a network that
can be flown with reasonable schedules and that exhibits some level of
robustness towards external shocks52.
Bell and Iida (1997, p. 20) have distinguished two archetypical network de-
sign problems: the continuous and the discrete network design problems.
The continuous network design problem takes the network structure as
given and aims to change the network characteristics, such as capacities
and flow cost. Real-life examples of these design problems are the addition
of a lane to an existing road or the installation of a traffic light. In airline
networks, the determination of flight frequencies and the offered capacity
(fleet assignment) are continuous network design choices (Bell & Iida,
1997, p. 20). Discrete network design problems are concerned with the
determination of the network structure. In transportation networks, this can
include infrastructure investments, such as the construction of a new road
or railway, and the carrier route selection (Bell & Iida, 1997, p. 20). In an
52
Changes in the network structure usually entail adding and deleting destinations; shocks
such as political unrest or expired traffic rights can complement economic considerations for
network changes.
64 Theoretical foundation
airline network, the most common discrete network design problems are
network type and route selection.
53
For an extensive overview of the existing network design models, see Farahani et al.
(2013).
Airline networks 65
54
Source: Own illustration.
66 Theoretical foundation
If we look at the demand for airline network flows, the source of the demand
is called the origin, and the sink of the demand is called the destination.
The demand between a specific origin and destination is abbreviated as
O&D demand (Bazargan, 2016, p. 32). The demand for air travel is usually
not created at the airport itself but in the catchment area of an airport. Since
airports are connected to other ground or sea transportation systems, pas-
sengers and cargo start their individual itinerary not at the airport but at
individual homes, offices, or at manufacturing sites. Similarly, their journey
does usually not end at the destination airport but at some location in the
catchment area of those airports. These individual origins and destinations
are called the real O&D.
55
The terminology for links is inconsistent in literature. While some authors refer to links as
routes (Goedeking, 2010, p. 7), others use city pairs (Bazargan, 2016, p. 31) or flights
(Burghouwt, 2007, p. 22).
Airline networks 67
not optimized for it. A typical example of stations are airports that are crew
bases for low-cost airlines.
The geographic setup of airports connected via routes is the spatial dimen-
sion of an airline network (Burghouwt, 2007, p. 7). The analysis of the spa-
tial dimension is rooted in graph theory and has become increasingly pop-
ular since the 1980s (Chou, 1990; Teodorović & Guberinić, 1984). The de-
regulation of air transport in the United States in the 1970s and in Europe
in the 1990s has also attracted extensive research on the impact of dereg-
ulation on spatial structures of airline networks (Gillen, 2005; Gillen & Mor-
rison, 2005). The most important spatial trend has been the increased cen-
tralization of airline networks around the so-called hubs. Rooted in the gen-
eral network design theory (presented in the previous sub-chapter), schol-
ars have identified the hub location problem as a central design problem
for airline networks (O'Kelly & Miller, 1994).
56
Source: Own illustration. following Burghouwt (2007).
Airline networks 69
Chain networks
In the early days of aviation, aircraft ranges were small and refueling stops
common in connecting distant airports. Typical examples for chain net-
works were the intercontinental networks of European and North American
airlines before the suitable long-haul aircraft was introduced in the 1970s57.
Since chain networks are less efficient than other network types (see sec-
tion 3.4.3), most passenger airlines have developed different network
structures. Today, chain networks are still common in cargo airlines and
with some regional airlines connecting remote airports (Gillen, 2005,
p. 61).
57
Exemplary aircraft ranges: 1930s: below 2,500 km (Junkers Ju 52 or Douglas DC-3);
1950/60s: 6,000-8,000 km (Douglas DC-6, Lockheed Constellation, Boeing 707, Douglas DC-
8); 1970s: above 10,000 km (Boeing 747, Douglas DC-10).
70 Theoretical foundation
58
Source: Airportzentrale (2014).
Airline networks 71
Point-to-point networks
Hub-and-spoke networks
In the most extreme form of a HS network, all traffic flows are routed via a
single hub. The upper right graph in Figure 3.12 schematically pictures
such a “pure” HS network. The five spokes are only connected to one hub,
and any O&D demand between the spokes must change flights at the hub.
scale and scope, higher service levels, and an increased competitive po-
sition at the hub airports (Lederer & Nambimadom, 1998, p. 787).
The spatial design of the random radial (RR) network is similar to the HS
network but lacks the temporal coordination of the hub. This network type
was the typical network for the European legacy carriers prior to the liber-
alization of traffic rights. Before the spread of the so-called sixth-freedom
rights59 or beyond rights, airlines were restricted to carry passengers only
between two countries, one of them being the home country of the airline.
Lufthansa, for example, could only carry passengers from Buenos Aires to
Frankfurt (or vice versa) but not from Buenos Aires via Frankfurt to Mum-
bai. Connecting traffic flows thus had little relevance, and the schedule had
little need for temporal coordination. With the increase of bilateral beyond-
rights traffic agreements, European home base airports, such as Frankfurt,
59
The sixth freedom right is the right to carry passengers or cargo from a second country to
a third country by stopping in one's own country. Please refer to Annex 2 of Burghouwt (2007,
p. 258) for a complete overview of the traffic right freedoms.
Airline networks 73
Very few airlines actually serve a pure network archetype; rather, there are
varying shades of grey between these network types (Button & Stough,
2000, p. 26). The current networks are the result of a continuous network
development, which has been influenced by changes in regulation, com-
petition, aircraft technologies, and consumer preferences (Burghouwt,
2007, p. 2).
In the United States, for instance, networks first evolved from chain to PP
networks due to the availability of long-haul aircraft, and then after the de-
regulation of the 1970s, they developed into HS networks. All the large
United States (US) airlines have a multi-hub network, resulting from mer-
gers in a consolidating industry. In general, there has been a clear ten-
dency towards more spatial and temporal concentrations in the US airline
networks (Burghouwt, 2007; Button & Stough, 2000; Gillen, 2005).
In Europe, the situation has evolved differently. The legacy carriers moved
from chain to RR networks and then to HS networks. The spatial setup of
the legacy carrier networks has not changed substantially (Bel & Fageda,
2010), but the temporal coordination has increased significantly
(Burghouwt, 2007, pp. 3–4). The industry’s consolidation in Europe has
progressed slower than in the United States, but mergers in the past 20
years have produced more airlines with multiple hubs, such as Lufthansa,
Air France-KLM, and the International Airlines Group (IAG) 60 (Lordan,
60
Lufthansa Group currently operates with five hubs (Frankfurt, Munich, Zurich, Vienna and
Brussels), IAG with two hubs (London Heathrow and Madrid) and Air France-KLM with two
hubs (Paris Charles-de-Gaulle and Amsterdam).
74 Theoretical foundation
2014, p. 1191). The Open Skies agreement between the US and the Eu-
ropean Union (EU) has not led to further concentration but to an increased
importance of alliances and joint ventures (Pels, 2009, p. 88).
The size and the structure of networks have fundamental economic impli-
cations for airlines. Networks affect the cost structures, (de-)stimulate de-
mand, and influence the competitive situation of an airline (Burghouwt,
2007, pp. 2–4). The literature on airline network economics has long as-
sumed that in a deregulated market, the HS network type prevails, since
airlines can harvest economies of scale and scope, provide a broader ser-
vice offering, and increase the entry barriers to the airline industry (Goede-
king, 2010, p. 108). However, the expansion of low-cost-carriers focusing
on PP networks has led to a reconsideration of the traditional network eco-
nomics in competitive markets (Takebayashi, 2013, p. 93).
61
While Caves, Christensen, and Tretheway (1984) found a constant unit cost at the firm
level, Baltagi, Griffin, and Rich (1995) discovered economies of scale among North American
airlines.
Airline networks 75
Rietveld, 2000, pp. 431–432). This positive demand network effect is also
the primary driver for alliance formation, whereby airlines can virtually grow
their network by offering code-sharing agreements with their alliance part-
ners (Pels, 2008, p. 73). The temporal coordination of flights in an HS net-
work also enables shorter connection times, thereby increases the attrac-
tiveness of the offered product (Goedeking, 2010, pp. 28–29). Several air-
lines use the quality of service index (QSI) to quantify these network ef-
fects. The QSI measures customer preferences for O&Ds, flight time, ser-
vice quality, and price as well as the disutility of delays or connections.
Based on these estimates, airlines can not only optimize the network struc-
ture but also the temporal coordination and used equipment for specific
routes (Goedeking, 2010, p. 29). Serving a HS network provides more flex-
ibility for network and schedule adjustments, since all the services are
routed through the same hub.
Other than cost advantages and demand stimulation, there is a third factor
explaining the historic development of HS networks. Concentrating traffic
on one or more hubs increases the competitive position of an airline at
these hubs. Since they provide a large share of the total traffic at these
airports, they enjoy huge market power over the airports and the associ-
ated service providers (e.g., ground operations, catering, etc.). Besides the
potential financial benefits, they often have a say on traffic right decisions
and gate allocations. This market power increases the barriers on entry in
the home markets of the HS airlines, whereby they can achieve a monop-
oly-like market position (Brueckner & Spiller, 1991; Oum, Zhang, & Zhang,
1995).