Centrality in Strategic Transportation Network Design: Anne Lange

Edition KWV
Anne Lange
Centrality in Strategic
Transportation
Network Design
An application to less-than-truckload
networks
Edition KWV
Die „Edition KWV“ beinhaltet hochwertige Werke aus dem Bereich der Wirtschaftswissen
schaften. Alle Werke in der Reihe erschienen ursprünglich im Kölner Wissenschaftsverlag,
dessen Programm Springer Gabler 2018 übernommen hat.
Weitere Bände in der Reihe http://www.springer.com/series/16033

Anne Lange
Centrality in Strategic
Transportation
Network Design
An application to less-than-truckload
networks
Anne Lange
Wiesbaden, Germany
Bis 2018 erschien der Titel im Kölner Wissenschaftsverlag, Köln

Dissertation Universität zu Köln, 2010
Edition KWV
ISBN 978-3-658-24240-4 ISBN 978-3-658-24241-1 (eBook)
https://doi.org/10.1007/9783658242411
Library of Congress Control Number: 2019934949
Springer Gabler
© Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2011, Reprint 2019
Originally published by Kölner Wissenschaftsverlag, Köln, 2011
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, express 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
Foreword
Industrialized societies cannot exist without powerful transport networks. The qual-
ity of such networks can be measured along many dimensions: time and cost effi-
ciency, flexibility, and reliability, to name a few. However, in the scientific literature
the design of networks is primarily discussed under the objective of cost minimiza-
tion. Furthermore, it is regularly assumed that all necessary information for the opti-
mization is available, while in real life strategic network design decisions this is hardly
the case. Thus, the usual assumptions of strategic network design models do reflect
real life conditions only to a limited extent.
To overcome this discrepancy, strategic information about the relation between the
structural and procedural shape of transport networks and their performance is es-
sential. In this context the centrality of a network should play an important role as key
element of network design. Based on these thoughts, Anne Paul formulates two ob-
jectives for her dissertation: “firstly, to conceptually devise the relationship between
centrality and network performance in order to emphasize the outstanding impor-
tance of network centrality for network design, secondly, to suggest quantitative mea-
sures for transportation network centrality”.
These objectives are fully achieved. Based on the precise definition of network cen-
trality, Anne Paul links convincingly and for the first time the previously separate ways
of thinking of network topology and network concentration. The centrality of trans-
port networks is identified as overarching factor for network design. Established con-
cepts as the exploitation of consolidation potentials or the minimization of distance-
or volume-related costs can be linked by the concept of network centrality, integrating
service aspects. Hereby the concept takes on an extended perspective, complement-
ing classical, cost-oriented network design in an innovative way.
Furthermore, it is shown that network centrality has two dimensions. The quali-
tative dimension of network centrality and its relation to performance indicators of
transport networks can be used to structure the search for superior network alterna-
tives in the design process efficiently. This concept is of a conceptual nature and of
highest relevance for practical strategic network planning. The quantitative dimen-
sion of network centrality is structured by the development of indices of centrality for
the general cargo market. This must be understood as a fundamental methodological
contribution in the area of network analysis.
All together the present thesis provides a remarkable scientific contribution and
is pointing a new way for the strategic design of transport networks. By linking the
concept of network centrality with classical criteria of OR for network optimization
v
Anne Paul builds a bridge between previously separate research concepts. With a
wide range of applied analyses in her thesis she proves in a convincing way the fruit-
fulness and sustainability of her innovative approach and opens up a new, scientif-
ically ambitious and intriguing field of research with immediate utility potential for
the practical design of transport networks and far beyond.
I wish this thesis a very positive reception in academia and practice and hope it will
trigger an intensive discussion and further research.
Werner Delfmann
vi
Preface
Figuratively speaking, you are holding in your hands a travelogue. It describes the
course of a doctoral voyage through the fascinating land of transportation networks.
The land had steep hills and smooth plains, both traversed with equal care. I expe-
rienced hostile weather and very sunny days. And I realized there were a surprising
number of dead end streets along the way.
Many things are to be learned on such a trip. Traveling widens knowledge. Trav-
eling allows one to gain new perspectives regarding one’s own territory. Traveling
teaches one to take decisions. Traveling deepens the understanding of home. In-
evitably, traveling is about the people that have been there before the departure and
are awaiting the arrival, and about those one meets along the way.
A doctoral trip must, by definition, be a solitary one. Yet, it never truly is. So many
people have their share in it, and the trip would be doomed without them. Some laid
the foundation for the trip. They nurtured the wish to pursue it. They taught how to
walk, how to hike. They nourished the curiosity that generates the momentum nec-
essary to depart. Others aided in the preparations for the very trip. Provisions were
gathered. Maps were sketched and modified as obstacles appeared. Rest camps were
prepared, facilitating the accruement of new material on the way. Helpers awaited
me at those camps, sometimes unexpected, sometimes long planned.
Many walked alongside for some time, some even for the entirety. I do not believe
I spent a single day alone on this road. The faces changed. Some travelers took off on
crossroads to pursue their own journey. Others joined at some point and continued
all the way to the end. Travel companions kept up the spirit and helped to carry the
baggage. Some sat by the campfire at nights, listened to the stories of the day, told
their own stories, helped to modify and refine ideas. Advice was much needed and
freely given along the way: Which general direction to head, which turn to take, which
companion to seek. There were lively discussions that challenged and modified the
endeavor. I found myself defending the entire trip itself more than once during this
journey. Admittedly, I sometimes ignored advice and learned from that experience as
well.
In particular, I would like to thank my doctoral supervisor and mentor Professor Dr.
Dr. h.c. Werner Delfmann, for being the chief-cartographer for this trip. Not sparing
me the steeply rising paths, but always making sure there were a sufficient number of
bridges mapped to cross the crevasses and rivers along the way. Additional thanks are
due to Professor Dr. Dr. Ulrich Derigs and to Professor Ulrich W. Thonemann, Ph.D.
vii
who, as members of the doctoral committee, brought in different viewpoints, acted
as final gatekeepers as well as first receivers upon arrival at my destination.
Some companions are sadly missed upon the arrival, but most are here to see and
to celebrate the end of this journey - far and near. Thanks to all of you. I am grateful
for your preparation, your challenges, your support, and your presence. Pars pro toto:
Sascha Albers, Rowena Arzt, Lisa Brekalo, Trisha Conway, Jost Daft, Björn Götsch, Ralf
Günther, Caroline Heuermann, Vera Kimmeskamp, Jürgen Klenner, Kai Krause, Finn
Lange, Tobias Lukowitz, Antje Möckelmann, Ralph Müßig, Christoph Paul, Edda Paul,
Hans-Helmut Paul, Markus Reihlen, Ingo Reinhardt, Hilde Reuter, Jens Rühle, Heike
Schwegler-Kirch, Bastian Schweiger, and Pierre Semal.
A straightforward travelogue would not make a good story as such. I had to leave
out many parts of the voyage and elaborate more on certain aspects. All the same,
stories from the entirety of this trip are included and I am certain that each travel
companion will find familiar elements of the days spent together along the route - a
route that I have enjoyed every single day.
Anne Paul
viii
Ganz nebenbei
oder Das Derivat des Fortschritts
Indes sie forschten, röntgten, filmten, funkten,

entstand von selbst die köstlichste Erfindung:
der Umweg als die kürzeste Verbindung
zwischen zwei Punkten.
Erich Kästner
Kurz und bündig - Epigramme
(c) Atrium Verlag, Zürich 1950 and Thomas Kästner
xi
Contents
Figures xvii
Tables xxi
Acronyms xxiii
Variables xxvii
1. Introducing network centrality for strategic transportation network de-

sign 1
1.1. Emphasizing the role of network centrality in transportation network
design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2. Structure of the thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2. Transportation networks and their optimal design 9

2.1. Transportation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2. Transportation networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2.1. Network types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.2. Transporting cargo or passengers . . . . . . . . . . . . . . . . . . . . . . . 20
2.2.3. Creating efficiencies in transportation networks . . . . . . . . . . . . 22
2.3. Optimal transportation network design - an Operational Research per-
spective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3.1. Graph theoretic terminology for network design . . . . . . . . . . . . 29
2.3.2. Strategic decisions for network design . . . . . . . . . . . . . . . . . . . . 30
2.3.3. Tactical decisions for network design . . . . . . . . . . . . . . . . . . . . 34
2.3.4. Operational decisions for network design . . . . . . . . . . . . . . . . . 39
2.3.5. Service-oriented aspects of OR network design . . . . . . . . . . . . . 40
2.3.6. Performance indicators for transportation networks . . . . . . . . . 42
2.4. Trade-off identification as an alternative perspective on network design 44
2.4.1. Empirical challenges for optimal network design . . . . . . . . . . . . 44
2.4.2. Trade-offs in network design . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.5. Enriching optimal network design with qualitative aspects . . . . . . . . . . 50
3. Transportation network centrality 53

3.1. Network topology and concentration as dimensions of network cen-
trality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
xiii
Contents
3.2. Impact of network centrality on network performance . . . . . . . . . . . . . 59

3.2.1. Consolidation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.2.2. Link frequency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.2.3. Transportation distance and time . . . . . . . . . . . . . . . . . . . . . . . 63
3.2.4. Schedule reliability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.2.5. Network vulnerability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
3.2.6. System capacity flexibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.2.7. Cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.2.8. Summarizing network centrality and its relation to network per-
formance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
3.3. Centrality measures in networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.3.1. Freeman’s centrality framework . . . . . . . . . . . . . . . . . . . . . . . . 75
3.3.2. Measures of inequality in social groups . . . . . . . . . . . . . . . . . . . 78
3.3.3. Measures of market concentration . . . . . . . . . . . . . . . . . . . . . . 80
3.4. Measuring network centrality of passenger airline networks . . . . . . . . . 81
3.4.1. Assessing the outcome of deregulation . . . . . . . . . . . . . . . . . . . 82
3.4.2. Metrics for the centrality of passenger airline networks . . . . . . . 84
3.5. The role of network centrality in strategic network design . . . . . . . . . . . 93
4. Less-than-truckload network centrality 95

4.1. Logistics service providers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.1.1. Markets for logistics service providers . . . . . . . . . . . . . . . . . . . . 96
4.1.2. Categorizing logistics service providers . . . . . . . . . . . . . . . . . . . 97
4.1.3. Research topics related to logistics service providers . . . . . . . . . 100
4.2. Less-than-truckload transportation . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
4.2.1. Less-than-truckload characteristics . . . . . . . . . . . . . . . . . . . . . 102
4.2.2. Less-than-truckload operations . . . . . . . . . . . . . . . . . . . . . . . . 103
4.2.3. European less-than-truckload market . . . . . . . . . . . . . . . . . . . . 105
4.2.4. Services accompanying less-than-truckload transports . . . . . . . 105
4.3. Measuring less-than-truckload network concentration . . . . . . . . . . . . . 107
4.3.1. Similarities of airline and less-than-truckload networks . . . . . . . 108
4.3.2. Traffic flows in less-than-truckload networks . . . . . . . . . . . . . . . 109
4.3.3. McShan-Windle index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
4.3.4. Network concentration index . . . . . . . . . . . . . . . . . . . . . . . . . . 110
4.3.5. Hubbing concentration index . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.3.6. Comparing the indices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
4.4. Relevance of measuring less-than-truckload network concentration for
network design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
xiv
Contents
5. An algorithm for less-than-truckload network design 119

5.1. Algorithm specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
5.1.1. Stable overall transportation time in a less-than-truckload net-
work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
5.1.2. Centrality-driven network design . . . . . . . . . . . . . . . . . . . . . . . 122
5.1.3. High level specifications for network generation . . . . . . . . . . . . 124
5.2. Input and output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.2.1. Input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.2.2. Cost and transport time assumptions . . . . . . . . . . . . . . . . . . . . 130
5.2.3. Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
5.2.4. Type of generated output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.3. Assembling the algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.3.1. Trade-off direct vs. indirect: Separating direct and indirect traffic 136
5.3.2. Terminal allocation: Defining the number of hubs . . . . . . . . . . . 140
5.3.3. Number of transshipments: Ensuring full connectivity . . . . . . . . 142
5.3.4. Demand estimation: Determining the right number of vehicles . 143
5.3.5. Testing the generated network . . . . . . . . . . . . . . . . . . . . . . . . . 145
5.3.6. Possible enhancements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
5.4. The bigger picture: Contributions to strategic network design . . . . . . . . 149
6. Generated networks and their performance 153

6.1. Case study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
6.1.1. Input and parameters to the algorithm . . . . . . . . . . . . . . . . . . . 154
6.1.2. Generated results on stable transportation time . . . . . . . . . . . . 158
6.1.3. Generated results on cost . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
6.1.4. The role of network centrality for the case study . . . . . . . . . . . . 161
6.2. General Insights . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
6.2.1. Minimal frequency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
6.2.2. Maximal accumulating days . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.2.3. Distance-related transportation cost . . . . . . . . . . . . . . . . . . . . . 172
6.2.4. Percentage of sold empty capacity . . . . . . . . . . . . . . . . . . . . . . 172
6.2.5. Daily distance by vehicle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
6.2.6. The impact of the potential hubs in the network . . . . . . . . . . . . 176
6.2.7. Summarizing the most influential findings . . . . . . . . . . . . . . . . 182
6.3. Indicators of network centrality to compare networks . . . . . . . . . . . . . . 183
6.4. The benefit of including network centrality in the network design process 188
7. Closing remarks 193
A. Appendix 199
References 215
xv
Figures
1.1. Line of argument. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1. A typology of transportation networks. . . . . . . . . . . . . . . . . . . . . . . . . 16

2.2. Transshipment volume distribution. . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.3. Network types. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4. Economies of distance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.5. Economies of scale. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.6. Categorization of cost economies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.7. Situative examples of location problems. . . . . . . . . . . . . . . . . . . . . . . . 31
3.1. Topological centrality scale of networks. . . . . . . . . . . . . . . . . . . . . . . . 55

3.2. Two networks of identical topological centrality. . . . . . . . . . . . . . . . . . . 55
3.3. Topologically central networks differing in concentration. . . . . . . . . . . 56
3.4. Typology of network centrality. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
3.5. Network centrality. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
3.6. Consolidation strategies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.7. Average distances of networks with different degrees of topological cen-
trality. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.8. Taxonomy of transportation network flexibility. . . . . . . . . . . . . . . . . . . 70
4.1. Sizes of European LSP markets in bn Euro. . . . . . . . . . . . . . . . . . . . . . . 97

4.2. Relation of customization and functions for LSPs. . . . . . . . . . . . . . . . . 99
4.3. Routing schemes in LTL operations. . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
4.4. MW index example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
4.5. NC index example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.6. Example of NC index values in markets with concentrated demand. . . . 113
4.7. HC index example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.8. Examples of networks to illustrate centrality index values. . . . . . . . . . . 116
5.1. Impact of waiting time on overall transportation time. . . . . . . . . . . . . . 121

5.2. Algorithm: Construction process. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
5.3. Line of identical costs in direct and indirect routing. . . . . . . . . . . . . . . . 138
5.4. Algorithm: Separating direct and indirect traffic. . . . . . . . . . . . . . . . . . 139
5.5. Algorithm: Defining the number of hubs. . . . . . . . . . . . . . . . . . . . . . . . 141
xvii
Figures
5.6. Algorithm: Ensuring full connectivity in the network. . . . . . . . . . . . . . . 142

5.7. Algorithm: Determining the right number of vehicles. . . . . . . . . . . . . . 143
5.8. Overview of the developed algorithm. . . . . . . . . . . . . . . . . . . . . . . . . . 150
6.1. Shipment pay weight distribution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

6.2. Coefficient of variation of transportation time per OD-market in base
case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
6.3. Coefficients of variation of transportation time per OD-market in an
exemplary 1-hub scenario. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
6.4. Costs in the base case scenario. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
6.5. Comparison of least cost scenarios to base case. . . . . . . . . . . . . . . . . . . 161
6.6. Vehicle utilization against minimal frequency. . . . . . . . . . . . . . . . . . . . 167
6.7. Share of direct shipments against maximal accumulation time in 255
scenarios. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
6.8. Coefficient of variation of indirect transportation time against maximal
accumulation time in 255 scenarios. . . . . . . . . . . . . . . . . . . . . . . . . . . 169
6.9. Ratio of HC index / connecting share depending on the maximal accu-
mulation time in 255 scenarios. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
6.10. Handling cost at transshipment terminals against share of sold empty
capacity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
6.11. Centrality indices against share of sold empty capacity. . . . . . . . . . . . . 174
6.12. Number of transshipment terminals connected to each EoL terminal
against share of sold empty capacity. . . . . . . . . . . . . . . . . . . . . . . . . . . 175
6.13. Coefficient of variation of transportation times against daily distance. . . 176
6.14. Total cost against number of potential hubs. . . . . . . . . . . . . . . . . . . . . 177
6.15. Indirect transportation distance against number of potential hubs. . . . . 178
6.16. The impact of two distinct EoL terminals on the ratio of HC index /
connecting share. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
6.17. Centrality metrics against indirect vehicle utilization. . . . . . . . . . . . . . . 184
6.18. Centrality metrics against indirect transportation distance. . . . . . . . . . 186
6.19. Centrality metrics against number of arriving and departing vehicles at
a terminal per indirect connection. . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
6.20. Indirect transportation distance against cost. . . . . . . . . . . . . . . . . . . . . 188
6.21. Centrality metrics against total cost. . . . . . . . . . . . . . . . . . . . . . . . . . . 189
A.1. The Lorenz curve of a distribution. . . . . . . . . . . . . . . . . . . . . . . . . . . . 200

A.2. Coefficient of variation of transportation times in 255 scenarios. . . . . . . 203
A.3. Share of direct shipments against minimal frequency. . . . . . . . . . . . . . 204
A.4. Number of connected transportation terminals per terminal against
minimal frequency constraint. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
A.5. Coefficients of variation of transportation time against minimal
frequency constraint. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
xviii
Figures
A.6. Mean shipment distance against minimal frequency constraint. . . . . . . 205

A.7. Centrality metrics against minimal frequency constraint. . . . . . . . . . . . 206
A.8. Transportation cost in the network against share of empty capacity sold. 208
A.9. Indirect transportation time per scenario against maximal accumula-
tion time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
A.10.Centrality indices against maximal accumulation time. . . . . . . . . . . . . 210
A.11.Number of direct vehicles in the network against share of sold empty
capacity. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
A.12.Centrality indices against number of hubs. . . . . . . . . . . . . . . . . . . . . . . 212
A.13.Direct transportation against number of potential hubs. . . . . . . . . . . . . 213
A.14.Indirect transportation times against number of potential hubs. . . . . . . 214
xix
Tables
2.1. US GDP attributed to for-hire transportation services. . . . . . . . . . . . . . 11

2.2. Europe of 17 revenues in transportation. . . . . . . . . . . . . . . . . . . . . . . . 12
2.3. Global top 25 LSPs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.1. Relevance of cost categories associated with network types. . . . . . . . . . 72

3.2. Characteristic network performance depending on network centrality. . 73
3.3. Network topology indices. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.4. Scope of network concentration measurements. . . . . . . . . . . . . . . . . . 87
3.5. Markets and event periods for selected studies on aviation market
change. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.6. Compared networks in selected studies. . . . . . . . . . . . . . . . . . . . . . . . . 92
4.1. LSP literature contributions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

4.2. Top 10 LSPs in European LTL 2008. . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
4.3. MW index example. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
4.4. Values of concentration indices of networks depicted in figure 4.8. . . . . 117
5.1. Stability of transportation time: One departure on Thursday. . . . . . . . . 125

5.2. Stability of transportation time: Departures Tuesday, Thursday, Friday. . 126
5.3. Algorithm: The spread of departures over the week. . . . . . . . . . . . . . . . 145
6.1. Relevant shipment information. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

6.2. Shares of initial routing schemes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
6.3. Case study algorithm parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
6.4. Transport cost at different load factors. . . . . . . . . . . . . . . . . . . . . . . . . 157
6.5. Shipment shares by pay weight. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
6.6. Centrality indices per scenario. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
6.7. Connections and runs in the 5-hub example. . . . . . . . . . . . . . . . . . . . . 166
6.8. Centrality indices of 255 generated networks in relation to maximal ac-
cumulation time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
6.9. Centrality indices of generated networks against number of potential
hubs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
6.10. Indirect transportation time indicators as influenced by two explicit
EoL terminals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
xxi
Tables
A.1. European LTL: Coverage and delivery times. . . . . . . . . . . . . . . . . . . . . . 201

A.2. European LTL: Services. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
A.3. Value approximation for base case scenario. . . . . . . . . . . . . . . . . . . . . 203
A.4. KPIs against distance-related transportation cost. . . . . . . . . . . . . . . . . 207
A.5. Regression results. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
xxii
Acronyms
3PL third party logistics.

4PL fourth party logistics.
BB break-bulk.
BC before Christ.
bn billion = 1,000 million.
C4 4-firm concentration.
C8 8-firm concentration.
CEP courier, express, parcel.
CLSCM closed loop supply chain management.
CSCMP Council of Supply Chain Management
Professionals.
DSS decision support system.
EC European Community.
ECMT European Conference of Ministers of Transport.
EoL end-of-line.
EOQ economic order quantity.
EU European Union.
FAA Federal Aviation Administration.

FSC full service carrier.
FTL full-truckload.
GDP gross domestic product.
H&S hub-&-spoke.
HC hubbing concentration.
HH Hirschman-Herfindahl.
HH&S hybrid hub-&-spoke.
HMH&S hybrid multi-hub-&-spoke.
xxiii
Acronyms
ICT information communication technology.

IT information technology.
ITF International Transport Forum.
JIS just-in-sequence.
JIT just-in-time.
kg kilogram.
KPI key performance indicator.
LCC low cost carrier.

LLP lead logistics provider.
LSI logistics service intermediary.
LSP logistics service provider.
LTL less-than-truckload.
MH multi-hub.
MH&S multi-hub-&-spoke.
MW McShan-Windle.
NC network concentration.
OD origin-destination.
OR Operational Research.
P2P point-to-point.
P&D pick-up and delivery.
PTL part load.
SC supply chain.
SCM supply chain management.
SME small and medium size enterprise.
t metric tonne.
tkm tonne-kilometer.
TSP traveling salesman problem.
UK United Kingdom of Great Britain and Northern

Ireland.
US United States of America.
VRP vehicle routing problem.
xxiv
Acronyms
WLP warehouse location problem.
xxv
Variables
A Adjacency matrix.
α Detour incurred by going to a transshipment
terminal.
B Set of links to and from break-bulk terminals.
cap Capacity of a vehicle.

f ix
ci The fixed cost associated with a node a node v i .
ch Handling cost per kg.
chf ul l Handling cost per full vehicle.

χ Eigenvector.
Ci j Share of connecting passengers on the market for
transport from node v i to node v j .
ci j A value associated to an edge in a network from
node v i to v j , typically a cost.
ckm Cost per full vehicle km.
CV Coefficient of variation.
di The weight of a node v i , typically a demand at this

node.
dg Demand for traffic class g .
dij Demand for transport from node v i to node v j .
in
d hu bi Incoming demand for transport from hub i .
d i nd Demand for indirect transports, market may be
specified by index.
d i s ti j The distance from node v i to node v j .
ou t
d hu bi Outgoing demand for transport to hub i .
d ov e r Overflow demand from direct transports, market
my be specified by index.
E Set of edges or links in a graph.
Fm i n Minimal frequency.
Fπ Frequency of service for product π in a network.
xxvii
Variables
G Denominates a graph.
g Traffic class.
gij Number of geodesics (shortest paths) between
nodes v i and v j .
L Set of direct links.

l Number of available vehicles.
m A number of nodes in a graph or of terminals in a

network.
n Number of all nodes in a graph or network, may be

used for the number of airports or terminals in a
network as well.
ω Incidence relation mapping edges in a graph to

two nodes i and j .
p A path in a graph.
φ Entries in the adjacency matrix.
π A transportation product (or service).
pi j A path in a graph from node v i to node v j .
p i∗j Shortest path from node v i to node v j .
Q Overall traffic in a network.

qi j Traffic on a market for transport from node v i to
node v j .
2 2-dimensional real space.
S Any arbitrary subset of V .

si Share of traffic at node i .
σ Sample standard deviation.
s ii n c Income share of node v i .
pop
si Population share of node v i .
T Set of transshipment terminals.

θ Maximum eigenvalue.
u Average vehicle utilization.
xxviii
Variables
υ Number of vehicles.
V Set of vertices or nodes in a graph.

vi A node in a graph or network.
v ix The x-coordinate of node v i .
y
vi The y-coordinate of node v i .
x̄ Sample mean.
xxix
1. Introducing network centrality for
strategic transportation network design
Nothing is in the air.

What an idyll. People sunbathe right in front of major airports. Bar-
becues without the sound of aircraft flying above. Still, some do
not seem to enjoy it. Crowded trains. A never seen price jump for
rental cars. The British Navy picks up travelers on the European
continent. Car manufacturers run short on parts (BBC News, 2010).
Video-conferencing peaks. Perishable goods do not appear on store
shelves (Economist, 2010). Politicians are stuck like everyone else
(Erlanger et al., 2010). Those were some European days in April 2010.
Only too well, everything came back to normal quickly.
It all started with a short note: the Icelandic volcano Eyjafjallajökull
erupted on March 21, 2010; the largest fear of scientists being a likely
breakout of the nearby-located volcano Katla, having a violent history
of eruptions (Simons, 2010). About 500 people were temporarily
evacuated. Journalists quote a farmer who had to leave all her
animals behind (Spiegel Online, 2010a). Tourists are attracted to
the site (Spiegel Online, 2010b). It was spectacular, but local. And it
was petering out. Then came the second outbreak on April 14, 2010
(Devlin, 2010), causing the European airspace to be partially shut
down for several days. The news were filled with the danger of flying
through ash clouds. Companies started to reclaim their losses caused
by the closed airspace. The airline industry stated that the disruption
in flights resulted in costs of $400 million a day (Economist, 2010).
Eventually, people realized how much they depend on transportation
networks.
The Eyjafjallajökull outbreak is a perfect example of how strongly modern societies

rely on effective and efficient networks: to ensure the availability of products, to allow
for mobility, to permit data exchange, to provide energy to households and produc-
tion, to communicate worldwide, and so on. Transportation networks are interwoven
with modern societies.
The design of transportation networks has been of interest to researchers for
decades. Leonhard Euler’s work on the seven bridges of Königsberg in 1736 is an early
© Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2011

1
A. Lange, Centrality in Strategic Transportation Network Design,
Edition KWV, https://doi.org/10.1007/978-3-658-24241-1_1
1. Introducing network centrality for strategic transportation network design
and prominent example of research on transportation networks (West, 2001, p. 26).

Transportation network design is a relevant research field in Operational Research
(OR), where seminal work on optimal network design from different viewpoints is
available.
Consequently, it is legitimate to ask, what an optimal network is. What does the
term optimal refer to? In cost-oriented OR models, an optimal network is one with
minimal costs; for example resulting in minimal transportation cost. Having the
above example of Eyjafjallajökull in mind, an optimal transportation network would
have probably been a network without airplanes. When thinking of the Internet,
the optimal network is available at all times. An optimal spider-web is a network
that is structured to catch enough flies, yet not too difficult to construct, and strong
enough to withstand threats. Defining optimality depends on the context; no general
definition can be stated. Nature has developed networks that comply with many
restrictions and are optimized to meet multiple goals. Modern artificially constructed
networks must, equally, fulfill different aspects of desired performance, but we have
not achieved this objective. Thus far, we can only marvel upon the efficiency that
nature is able to construct suitable networks with, virtually without any central
coordination (e.g. Tero et al. (2010)). There is a long way to go for modern science
until this sophistication can be met. Overcoming the strong cost-orientation in
traditional network design and widening the perspective to more encompassing
objectives is a valuable first step.
Service and risk-management have become vital aspects of the performance of
modern transportation networks from a market perspective. Manifold dimensions
are used to describe network service. Unfortunately, there is no common under-
standing regarding which indicators best describe the service expectations for a trans-
portation network. Even so, it is understood that transportation time, the reliability of
schedules, and the frequency of departures are elementary aspects of it (Pfohl, 2004b,
pp. 94-100).
These are among the relevant indicators for network performance in general and
can be translated to network key performance indicators (KPIs). KPIs include cost
and service elements but also keep track of further indicators such as risk. The wide
spectrum of potential KPIs, and thereby their sheer number, creates a challenge for
decision-making as it is infeasible to consider them all simultaneously.
Many of the KPIs are strongly influenced by the underlying network’s appearance as
expressed by spatial network centrality.1 Network centrality describes to what degree
certain locations in the transportation network assume an outstanding role for the
network’s operations. A star-shaped hub-&-spoke (H&S) network with an important
1 Centrality may have an organizational connotation (in the sense of organizational centralization) apart
from its spatial dimension. This thesis does not address challenges in finding suitable organizational
structures to manage a transportation network. Whenever the term centrality is used in the following, it
is understood in its spatial sense.
2
1.1. Emphasizing the role of network centrality in transportation network design
hub in its middle is the prototype of a central network, whereas a point-to-point (P2P)
network is a decentral transportation network.2
Network centrality should be seen as a core KPI for transportation networks,
but is currently not perceived as such, neither in academic transportation network
design, nor in decision-making on transportation networks in the logistics sector.
Transportation network design will benefit from exploring the relation between
network KPIs and network centrality. Network centrality enriches cost-oriented
decision-making by allowing for a sound inclusion of service-orientation and further
aspects of network performance into network design.
1.1. Emphasizing the role of network centrality in

transportation network design
The exceptional relation between centrality in transportation networks and network
performance indicators is at the focus of this thesis. Network centrality influences
network KPIs applied in network design and plays an exceptional, yet currently ne-
glected role in transportation network design. Network centrality is the concept in-
terlinking network appearance and network performance.
Therefore, the objective of the thesis is twofold. Firstly, to conceptually devise the
relation between network centrality and network performance in order to empha-
size the outstanding importance of network centrality for network design. This as-
pect refers to the qualitative understanding of network centrality. Secondly, to sug-
gest quantitative measures for transportation network centrality. These measures
are necessary for analyzing and comparing in depth the centrality of transportation
networks.
The research field of network design is well-developed. However, its strategic appli-
cation to real life problems creates challenges for planners, including e.g. the require-
ment for detailed data on a long-term planning horizon. To bridge this gap between
research and application, the identification of trade-offs that impact network design
strategically is beneficial. The knowledge of strategic influence factors will not, most
likely, help to find optimal transportation networks, but rather guide the design pro-
cess. Transportation network centrality may serve as a very expressive influence fac-
tor for strategic network design. Thereby, it supports the identification of good design
alternatives.
The general objective to emphasize the outstanding role of network centrality for
transportation network design allows to formulate three subordinate goals set by this
thesis.
2 Comparing the transportation distance of different networks (as one possible KPI) is a tangible example
of the impact of centrality on network KPIs: a H&S network is typically characterized by longer trans-
portation distances than a P2P network.
3
Enriching OR by generating strategic insights for ill-defined decision situations.

The thesis strives to highlight that optimal network design in the way it has been
conducted by researchers in OR is not sufficient to support good decision-making
for strategic, long-term network design. Strategic decision situations are often char-
acterized by high uncertainties, general assumptions, and expectations about future
demand for the network being planned. At this stage, strategic insights into trade-offs
involved in planning are more valuable to decision-makers than theoretically sound,
optimal solutions on unreliable data. This aspect is not new in the literature, but
appears to have been of lesser importance over the last few years. It is, therefore,
necessary to reiterate and further enrich this line of thinking in the literature. A qual-
itative understanding of transportation network centrality and its relation to network
performance is a significant step in this direction.
Implementing an algorithm to simulate network scenarios. The theoretically de-

vised interdependencies between transportation network centrality and transporta-
tion network behavior will support the development of an algorithm for less-than-
truckload (LTL) network design. The insights from a qualitative understanding of
transportation network centrality allow to pre-structure the algorithm. Hence, the
algorithm serves as an example for the practical implementation of the knowledge of
the relation between network centrality and network performance.
Generating conclusions on non-measured network performance. It is challeng-

ing to assess all aspects of network performance simultaneously in a decision situa-
tion, even though simulations may be used to gather large amounts of data. It will be
shown how the quantitatively measured network centrality of a network may be inter-
preted in order to draw conclusions on aspects of network performance that were not
directly recorded. The interpretation is based on qualitative knowledge of network
centrality and its relation to network performance.
1.2. Structure of the thesis

The line of argument of the thesis is illustrated in figure 1.1. This first chapter outlined
the background and the objective of the thesis. The second chapter, Transportation
networks and their optimal design, will broadly introduce the concept of transporta-
tion network design with the goal of presenting the width of the research field, but
also of explaining the key deficiencies that exist for strategic decision-making. To do
so, it will give fundamental information on transportation network design before pre-
senting an overview of the manifold problems that were discussed in OR in order to
solve various aspects of network design. This overview is structured around the time
horizon for decision-making and the decisions to be made at each stage. Literature
reviews in the field of OR would oftentimes be structured according to the solution
4
techniques for existing problems.3 Orienting the presentation around the planning
time horizon appears more appropriate for this thesis, as it directly highlights one
main deficiency, namely the sequential planning approach. Most importantly, the
section will point out the strong cost-orientation of OR network design approaches.
This thesis can by no means claim to be the first work to identify the additional im-
portance of general, strategic insights in OR network design. Selected contributions
from this literature stream are, therefore, presented after the weaknesses of standard
transportation network design and its application in practice were highlighted. The
chapter closes with a claim for more general insights into strategic network design.
Network centrality will prove to be a helpful concept for this purpose.
The third chapter, Transportation network centrality, is devoted to presenting the
outstanding role of transportation network centrality in transportation network de-
sign. The first section defines terms and will provide the reader with a fundamental
understanding of network centrality. The second section, then, thoroughly discusses
the impact of network centrality on six network performance indicators of transporta-
tion networks. This qualitative understanding of transportation network centrality
will be exploited to structure network design later on. It is well known that several
prototypes of transportation networks exist which differ in their advantages and dis-
advantages and are, therefore, applied in different environmental settings. Their per-
formance is typically measured along different scales, including transport distance
and time, service frequency, or schedule reliability. By relating these KPIs to network
centrality, it is argued that it is the very concept of network centrality that determines
the other KPIs and, therefore, is at the root of transportation network performance.
In more concise terms, network centrality is the concept that discriminates networks
from each other. In the forefront of these considerations, the remainder of the sec-
tion is oriented to addressing the question of how the multifarious term of network
centrality is understood and measured in different contexts in the literature. Research
in social sciences is rich in network centrality indices and is consequently reviewed.
The last remaining issue to be examined in the chapter is how indices from the so-
cial sciences were adapted to transportation networks. Airline networks provide the
widest field of application and will serve as an example for transportation networks.
They are even more interesting, as researchers have started to develop customized
indices for the context of airline network analysis. Knowledge of existing measures
of (transportation) network centrality sets the stage for developing a quantitative un-
derstanding of network centrality for LTL metrics in the subsequent chapter.
The remainder of the thesis will demonstrate how transportation network central-
ity can be exploited for network design. To this point, the thesis will have been ori-
ented upon transportation networks in general. From here forward, the discussion
will be focused on one field of application, namely that of road-based LTL networks.
Notwithstanding, the general concepts can be easily applied to other transportation
3 Examples of this type are Laporte (2009), Wieberneit (2008), or Grünert and Irnich (2005).
5
networks that possess similar characteristics. The similarities between LTL networks
and passenger airline networks will explicitly be dealt with, but applications to e.g.
air cargo networks, structured intermodal transportation networks, and express mail
networks may also be considered with ease.
Chapter 4, Less-than-truckload network centrality, encompasses two important as-
pects. At first, logistics service providers (LSPs) in general are presented as these are
the main providers of LTL networks on the European market. The chapter then con-
tinues to illustrate the LTL business: the market, the offered services, the customer
demand, and the operations. It thirdly outlines suggestions on how network central-
ity - more precisely network concentration - can be measured expressively for LTL
networks. This is a necessary step to operationalize the conceptual claim from chap-
ter 3 for further application and discussion in chapter 6.
Chapter 5, An algorithm for less-than-truckload network design, presents an algo-
rithm in detail that was developed to generate LTL networks in order to compare
network scenarios. Qualitative transportation network centrality was a fundamen-
tal element for the development of the algorithm. It allows to identify the network
appearance that best suits the given context and leads to the desired network perfor-
mance. The algorithm creates and tests transportation networks of this appearance
and accounts for various KPIs that may serve to compare the networks in detail. It is a
practical application of the concept of network centrality and it is, therefore, valuable
to present the logic behind the algorithm. The reasoning behind the algorithm, in-
cluding the role of centrality, is discussed. This frames the more technical algorithm
details on input and output, as well as on the detailed procedure. The main goal here
is to present the way of thinking that structures the algorithm. The algorithm presen-
tation itself merely supports this aim and will clarify the results obtained using the
algorithm.
The sixth chapter, Generated networks and their performance, once again focuses
on two important aspects. It will firstly present, in the form of a case study, the results
generated for a project that the algorithm was applied in. The project was conducted
with a large LSP and aimed at identifying options for LTL network redesign on a Euro-
pean scale. Insights are obtained, allowing the generation of managerial implications
that are relevant for the field and, in their core, related to network centrality. These are
presented subsequently. Quantitatively measuring network centrality by the indices
suggested in chapter 4 supports conclusions regarding typical network performance
that was not in the set of measured KPIs. Lastly, the information acquired this far
serves to review the benefits generated by the underlying knowledge of transportation
network centrality. The conclusions stated here regarding network centrality bring us
full circle to the conceptual contribution of the thesis. Chapter 7 summarizes and
therein closes the thesis.
6
Demonstrate the relevance of efficient

and effective transportation networks.
2.2
Operational Research:
Empirical challenges for the application.
Optimal network design.
2.3 2.4
Chapter 2
Network centrality enriches strategic

decision-making concerning network
design.
Definition of transportation network

centrality.
3.1
Outline relation between network

Measure transportation
p network
performance
f andd network
t k centrality
t lit as 3.3
centrality.
a medium to structure network design. 3.2 3.4 Chapter 3
Introduction to the European LTL Suggest measures for LTL network

market
market. 44.11 centrality
centrality.
4.2
4.3 Chapter 4
Develop an algorithm to generate and

test networks based on qualitative
knowledge of network centrality.
centrality Chapter 5
Compare scenario KPIs.
Conclusions for non-simulated KPIs

provide additional perspectives for
decision-making.
Chapter 6
Figure 1.1.: Line of argument.
7
2. Transportation networks and their
optimal design
This chapter sets the stage for the discussion of transportation network centrality in
the chapters to follow. To do so, it firstly points out how transportation blends in with
logistics and supply chain management (SCM), and continues by introducing trans-
portation networks. Where do they play a role in practice? Why are they important for
modern societies? It will become clear that transportation networks must be designed
carefully, which leads to reviewing network design as a research field in Operational
Research (OR). As network design is an immensely wide field in OR, the presenta-
tion is intended to spotlight the breadth of research questions and developed models.
This suffices to illustrate shortcomings of OR approaches for real life network design.
Most importantly, an overly strong cost-orientation is identified. One possibility to
overcome these is to strive for general trade-offs in transportation network design.
The chapter ends with a statement to the need for strategic insights for the network
design process, which will be resumed by chapter 3.
2.1. Transportation
This first section introduces transportation and its relevance for economies and soci-
eties. It furthermore exhibits the role of transportation in logistics and SCM.
There is no uncontested terminology for logistics and SCM in academia or practice.
The understanding of the two terms is of minor importance here as transportation
takes an outstanding role in both concepts. In order to identify the assumed perspec-
tive, it should be pointed out that the following is based on a wide understanding
of logistics: customer and service orientation, the thinking in flows, and a systemic
perspective are fundamental building blocks.4 Transportation has always been con-
4 This thinking follows the unionist view by Halldórsson et al. (2008), or the European perspective as iden-
tified by Delfmann and Albers (2000, p. 7).
The United States of America (US)-based Council of Supply Chain Management Professionals (CSCMP)
defines logistics as: “the process of planning, implementing, and controlling procedures for the effi-
cient and effective transportation and storage of goods including services, and related information from
the point of origin to the point of consumption for the purpose of conforming to customer require-
ments. This definition includes inbound, outbound, internal, and external movements” (Council of Sup-
ply Chain Management Professionals, 2010).
The European perspective on logistics differs from this one. It has evolved from a narrow viewpoint to
a wide one nowadays. In the early 1980’s, logistics was limited to activities related to transporting goods
or passengers and to storing goods (Berens and Delfmann, 1984, p. 32). Today, a systemic perspective

9
2. Transportation networks and their optimal design
sidered to be one of the key sub-systems of logistics, in one line with transshipment
and warehousing / inventory management (Pfohl, 2004a, p. 8).5
Transportation is highly important for economic development. Efficient trans-
portation systems provide economical benefits, whereas deficient systems will hin-
der development. Transportation directly impacts development by increasing the size
of markets as more places are interconnected and, furthermore, to that it decreases
travel time and product cost. Transport indirectly influences economic development
by opening up competition in markets (Rodrigue et al., 2009, p. 83). It is even seen as
a factor of production (Rodrigue et al., 2009, p. 88).
The US generate 6% of their national gross domestic product (GDP) by transporta-
tion (Stock and Lambert, 2001, p. 312).6 For-hire transportation contributes about 3%
to the US GDP (table 2.1).7 The European perspective on the importance of trans-
portation over the years is shown in table 2.2.8 Both tables highlight that transporta-
is common for European logistics, emphasizing the interdependencies between different sub-systems,
players and entities in networks. It further includes an orientation to customers as well as to flows (Delf-
mann and Albers (2000, pp. 6-7), Delfmann (2000)). Furthermore, logistics is an applied science (Delf-
mann et al., 2010, p. 61). This idea closely relates to what is currently understood as SCM by the CSCMP:
“Supply Chain Management encompasses the planning and management of all activities involved in
sourcing and procurement, conversion, and all logistics management activities. Importantly, it also in-
cludes coordination and collaboration with channel partners, which can be suppliers, intermediaries,
third-party service providers, and customers. In essence, supply chain management integrates supply
and demand management within and across companies. Supply Chain Management is an integrating
function with primary responsibility for linking major business functions and business processes within
and across companies into a cohesive and high-performing business model. It includes all of the logis-
tics management activities noted above, as well as manufacturing operations, and it drives coordination
of processes and activities with and across marketing, sales, product design, finance and information
technology.” (Council of Supply Chain Management Professionals, 2010). This definition is coherent to
stating that SCM has an important philosophical aspect. Philosophical SCM characteristics are the above
mentioned systems approach that values the wholeness of the supply chain (SC), the strategic orienta-
tion towards cooperation in the supply chain and, finally, a strong customer focus (Mentzer et al., 2001,
p. 7). The definition identifies logistics as a part of SCM. This is not uncontested: other research groups
see SCM as a sub-area of logistics (e.g. Delfmann et al. (2010, p. 60)). Thus, the understandings of the
connection between the terms logistics and SCM are fuzzy in research and practice (Halldórsson et al.
(2008), Delfmann and Albers (2000, pp. 6-10)).
However, apart from the discussion on terminology, a systemic perspective on SCs and a strong customer
focus are agreed upon as constitutive elements amongst practitioners and researchers.
5 Logistics is said to create time and place utility. Time utility mainly arises out of warehousing and stor-
age. The value created by transportation lies in major shares in place utility. As long as a product is not
available to a customer, it has a limited value. Transportation bridges spatial distances between produc-
tion and consumption and, thereby, provides place utility. The speed of transport and the reliability to
scheduled arrival of goods may also create time utility for customers. If goods arrive with delay at their
destination, their value may be significantly decreased (Stock and Lambert, 2001, p. 313).
6 This includes derived production in non-logistics sectors such as the automotive production.
7 For-hire transportation is provided by common carriers, by contract carriers, and by exempt carriers.
All three carrier types are legally defined for the US. They offer their transport services publically on the
market for transportation, as opposed to private carriers (Stock and Lambert, 2001, pp. 321-322).
8 Table 2.2 provides data for the so-called Europe of 17: Austria, Belgium, Denmark, Germany, Finland,
France, Greece, United Kingdom, Ireland, Italy, Luxembourg, The Netherlands, Norway, Portugal, Swe-
den, Switzerland, Spain. Data based on internal and outbound cargo moved.
10
2.1. Transportation
Transportation service 2002 2005 2007

Air transportation 48.3 48.3 55.2
Rail transportation 26.2 33.5 40.5
Water transportation 7.0 10.0 10.7
Truck transportation 95.7 118.4 127.6
Transit and ground passenger
15.7 17.9 19.3
transportation
Pipeline transportation 11.5 9.5 12.0
Other transportation and
73.4 91.6 101.5
support activities
Warehousing and storage 26.8 35.6 40.3
Sum 304.6 364.7 407.2
GDP 10,469.6 12,421.9 13,807.5
% GDP 2.9% 2.9% 2.9%
Table 2.1.: US GDP attributed to for-hire transportation services in current bn US $

(U.S. Department of Transportation, 2010).
tion not only supports economic development but is, furthermore, an important in-
dustry by itself.
Transportation activities always involve several partners. Different classifications
exist in the literature, in the given context here, the most important transport par-
ticipants are the shipper, the consignee, and logistics service providers (LSPs). The
shipper (or consignor) is the party that is in possession of the goods before the trans-
port and, thereby, is the origin of transport. The destination of transports, or the re-
ceiver of goods is referred to as the consignee. These two parties generate demand for
transportation (Bowersox et al., 2010, pp. 194-195). The third group of transport par-
ticipants are LSPs, who carry the freight, organize the transport, or provide additional
activities related to the transportation. These three parties, shipper, consignee, and
LSP, form the so called logistics triad (Selviaridis and Spring, 2007, p. 137).9
The global top 25 largest LSPs 2008 (by revenues) are listed in table 2.3. The ta-
ble is restricted to providers of goods transportation, omitting passenger transport.
Not surprisingly, one finds many well known LSPs amongst the top 25, including di-
versified service providers such as Deutsche Post DHL, Deutsche Bahn, and Nippon
Express, as well as courier, express, parcel (CEP) providers such as UPS, FedEx, and
9 The logistics triad is usually discussed for cargo transportation, but the transition to passenger transport
can be done easily. Passengers assume the role of shippers and consignee in one party. Yet, this distinc-
tion contributes only marginally to the understanding of the situation. The challenges in cargo transport
arise out of the triangular relation formed by the three parties and do not apply as such for passenger
transport.
11
Mode 2002 2005 2008

Road 167.5 222.7 277.0
Rail 11.9 11.8 13.1
Inland waterways 3.8 3.5 3.4
Sea cargo 49.0 67.4 72.0
Pipeline 3.0 3.6 4.1
Air cargo 12.1 8.46 10.42
Sum 247.4 317.5 380.0
GDP 8,959.5 10,820.0 12,175.0
% GDP 2.8 2.9 3.1
Table 2.2.: Europe of 17 revenues in transportation in bn Euro in 2002 (Klaus, 2004,

pp. 51,54), in 2005 (Klaus and Kille, 2007, pp. 50,54), and in 2008 (Klaus
et al., 2009, pp. 50,54).
TNT. The important share of bulk transport service providers, rail, and ocean freight is
highly influenced by the large transportation volume that is moved by these providers.
Large airlines are not included in the given dataset. Even though e.g. Lufthansa re-
ports an annual revenue of 24.8 bn Euro in 2009, respectively 22.3 bn Euro in 2008,
most of these revenues are passenger-related and are therewith not relevant for table
2.3. Lufthansa Cargo with 2.0 bn Euro revenues in 2009 is too small to be included in
the list by itself (Lufthansa, 2010, pp. 2-3).10
Other players participate in transportation apart from those forming the logistics
triad. The government has a high interest in efficient transportation, as economies
rely on it. On the other side, the government is an important infrastructure provider
for many transportation modes, e.g. road. Governments, moreover, influence the
transport industry through regulation and law-setting (Bowersox et al., 2010, pp. 195-
196). Some operative impacts on transportation exist, such as slot coordination at
airports (Doganis, 1992, p. 103). Last but not least, the public impacts transporta-
tion. Firstly, people create demand for transportation indirectly by purchasing goods
or directly by using public transportation. Furthermore, the public is touched on by
the external effects of transportation (e.g. air pollution, noise) and will therefore ask
for regulation on environmental and safety aspects of transportation (Bowersox et al.,
2010, p. 196).
This section provided information about the role of transportation in various con-
texts and the involved players. It is now necessary to focus on transportation net-
works. Efficient and effective transportation typically needs transportation networks;
the reasons for this will be illustrated in the next section.
10 LSPs are introduced in more detail in chapter 4.
12
2.2. Transportation networks
Company Service (-s) Revenues

Deutsche Post diversified 44.1
United Parcel Service CEP, contract logistics 36.5
Maersk A/S sea freight 25.3
FedEx CEP 23.7
Deutsche Bahn railway, diversified 19.3
NYK Line sea freight 18.5
CMA-CGM sea freight 15.1
Kühne & Nagel diversified 13.6
Mitsui OSK sea freight 13.1
Chinese Railway railway 13.0
BNSF railway 12.8
Union Pacific railway 12.7
RZB railway 12.4
Nippon Express diversified 12.0
MSC sea freight 8.5
SNCF (incl. Géodis) diversified 8.0
CSX railway 8.0
Norfolk Southern railway 7.6
Indian Railway railway 7.0
TNT CEP 6.9
Sagawa Express CEP, diversified 6.9
Neptun Orient Lines Ltd. sea freight 6.6
Yamato Transport diversified 6.3
Panalpina forwarder, diversified 5.6
Hanjin Shipping sea freight 5.0
Table 2.3.: Global top 25 LSPs (Klaus et al., 2009, p. 193) by worldwide logistics rev-
enues 2008 in bn Euro.

Transportation networks are a special case of networks. All networks share common
elements. By its most neutral definition, a network is an abstract model of reality con-
sisting of nodes and links relating them. Usually the nodes in the network are actors
or locations such as people in social networks, airports in airline networks, or bus
stops in urban transportation networks. The links constitute a relation between the
actors or locations. Examples can be personal relationships in social networks, flights
in airline networks, or bus routes in the urban transportation case. In transportation
networks, the links are often permanent tracks e.g. rail tracks or channels, or refer to
scheduled services conducted between the nodes (Rodrigue et al., 2009, p. 18).
13
Various networks are discussed in the literature, including e.g. transportation

networks, telecommunication networks, distributed computer networks, centralized
teleprocessing networks, and energy networks (Minoux, 1989, p. 314). These net-
works have many similarities but differ in details. The following briefly highlights the
peculiarities of different networks before common features are presented.
Telecommunication and distributed computer networks are used to transport digi-
tal information packages. There is virtually no variable cost of transportation on these
networks. Furthermore, the information transmitted on the network is usually repli-
cable and it is possible to resend it if necessary. Current research questions include in-
ter alia reliable service in case of node or link faults (e.g. Abd-El-Barr (2009), Kim and
O’Kelly (2009)), quality-of-service routing (e.g. van den Berg et al. (2007), Pliakas et al.
(2008), Yao et al. (2008), Skianis (2008)), and the design of ad-hoc and wireless sensor
networks (e.g. the special issue in Telecommunication Systems (Woungang, 2010)).
Energy networks connect the points of harvesting the energy or the points of trans-
formation with the points of final utilization. Energy networks nowadays depend on
large power plants as the center of energy generation and transmission to many dis-
tributed points of energy consumption. New energy sources such as wind turbines
may lead to more decentral energy networks in the future (Bouffard and Kirschen,
2008).
Transportation networks exist for various modes of transport, especially the clas-
sical ones: road, rail, air, water (inland waterways and ocean), and pipelines (Ballou,
2004, pp. 171-176). The networks are used to transport cargo, passenger, or both.
Transportation networks describe the form of connecting terminals (e.g. railway sta-
tions or airports) by links operated with the transport vehicles.11
It could be drilled down much deeper on differences in these networks, but the
short descriptions already show that networks differ in details but that structural sim-
ilarities dominate. The design of all of these networks - including transportation net-
works - can initially be approached by the same toolkit mostly rooted in OR.12
It should be mentioned that the planning and design of transportation networks
is not a trend of industrialized societies but has been of importance all throughout
history. Their existence has been the backbone of societies, from ancient to mod-
ern times (Rodrigue et al., 2009, p. 5).13 Modern economies would not exist without
efficient transportation networks. The convergence of information and telecommu-
11 Westlund distinguishes single mode transportation networks and multi-mode transportation systems. A
transportation system is an elaborated transportation network in the sense that it consists “of different
types of transport networks which, taken together, form a system in which the principal task of the nodes
is to act as points of transshipment between the constituent primary networks” (Westlund, 1999, p. 100).
This distinction is mainly definitional in nature as most challenges arising in transportation network
design are not mode-specific.
12 OR approaches to network design are presented in section 2.3.
13 The ancient Greeks built agglomerations around the Mediterranean and the Black Sea between the
eighth and sixth century BC. They connected their agglomerations mainly by a shipping network which
hardly required any infrastructural investments (Westlund, 1999, p. 96). The Romans later realized that
governing an enormous empire requires mobility of goods and information, and started to invest heavily
14
nication networks and their potential for transportation networks is seen as the new
milestone in this evolution (Rodrigue et al., 2009, p. 86).
2.2.1. Network types

Daganzo (2005) defines four types of networks: one-to-one networks, one-to-many
networks, one-to-many networks with transshipments, and lastly many-to-many net-
works. One-to-one network types apply e.g. for full-truckload (FTL) transportation or
courier services. One-to-many networks with or without transshipment are common
for production networks dominated by one production site.14 Finally, many-to-many
networks are at the focus in the following; the less-than-truckload (LTL) market is a
typical example for this kind of network, other examples include e.g. CEP, airline, and
railway networks.
Rodrigue et al. (2009, pp. 19-21) approach their overview of transportation net-
works by laying out a typology of transportation networks (depicted in figure 2.1) that
illustrates twelve of the manifold ways to classify transportation networks; these are
briefly described in the following.
(a) Transportation networks can be formulated concretely (closely related to reality)

or abstractly (with a graph theoretic description).
(b) They can be classified according to their real life location as right by a coast or a
river.
into a widespread road network, whose impact is still visible in modern times. The road network per-
mitted trade, transport of goods from the provinces to Rome, and was needed for defending the empire
(Westlund, 1999, p. 96). Another well-known infrastructural network built in Roman times are the aque-
ducts constructed all over the empire. With the fall of the Roman Empire, the quality of the infrastructure
slowly degraded.
It is known that the Mongols had an early communication network in the late 13th century. Messengers
carried information through the empire, either walking or riding on horse-back depending on the mes-
sage priority. Stations existed where horses were exchanged or new messengers awaited those on foot
prompt at their arrival. These exchange stations and the smart interchanges conduced there allowed for
a significant decrease in travel time (Kidd and Stumm, 2005, pp. 1250-1251).
During the European medieval times, independent fortified castles, cities, and cloisters developed with
rather little interchange by transportation amongst them at first. With the increasing importance of
trade, transportation and its infrastructure also rose in importance. Italian commercial cities, the
Hanseatic League as trading enterprises (Westlund, 1999, p. 97), and the Princely House of Thurn and
Taxis as pioneers in postal services later on are only few examples for the rise in relevance of transporta-
tion networks.
The starting colonialism by the Portuguese and the Spanish in the late 15th century and the related
importance of high-sea sailing opened the door for early globalization (Westlund, 1999, p. 97). From
this time on, economic development was strongly coupled with innovations in transportation. The in-
dustrial revolution in the 18th century drove, but also depended on, the simultaneous development of
transportation modes with high transport capacity: inland waterways, steamships, and railways. Road
transportation gained its share with industrial mass production in the early 20th century. The idea of
coupling different transportation modes - including air transport - emerged around the mid-twentieth
century (Rodrigue et al., 2009, p. 86).
14 Many-to-one networks depict the mirrored situation for one-to-many networks.
15
Concrete Abstract
(a) Abstraction level (b) Relative location (c) Orientation and extent
1 Port
4 Railyard Highway Secondary road
3 Maritime
7 6 5
4 R il
Rail 100
2 1 Road 50
5 6
3 2 125 km
Depot 90 km
(d) Number of edges and (e) Modes and terminals (f) Distance, road type, and
nodes control of the vehicle
0.3
Contiuous
0.9
07
0.7
Divided 0.5 0.7
(g) Type of traffic (h) Volume and direction (i) Load and capacity
Hierarchical
Non-hierarchical Linear
R d
Random
Mesh
(j) Type of correspondence (k) Pattern (l) Change (dynamics)
Figure 2.1.: A typology of transportation networks following Rodrigue et al. (2009,

p. 21).
(c) The networks’ extent and orientation reflect their geographical coverage and the
market they serve.
(d) The number of edges and nodes of the network is important for the formal
model: this information is necessary for the graph theoretic translation.
(e) Networks can further be described by the modes and terminals involved in them:
the terminals are represented by nodes and the modal routes by links in the
model.
(f) The information about modes and terminals can be refined for a distinct mode
such as road transportation, showing for example roads, speed limits, and dis-
16
tances.
(g),(h) The type of traffic as well as volume and direction of traffic flows allow to
further cluster transportation networks.
(i) The load of a network is defined as the ratio of transported volume to available
capacity and is another characteristic to describe a transportation network.
(j),(k) Networks differ in node hierarchy as well as in the pattern (or topology) they
form.
(l) Finally, Rodrigue et al. (2009) suggest the network dynamics related to new cir-
cumstances as the final criterion for their network typology.
This typology is beneficial in its descriptive way to show different network types. The
literature on network design addresses abstract networks and expresses the infras-
tructural and environmental contingencies therein.
Two main types of network patterns for transportation research on many-to-many
networks are typically identified: hub-&-spoke (H&S) networks and point-to-point
(P2P) networks. The following presents and compares these two network patterns.
Some terminology must be introduced to do so.
The term (transshipment) terminal describes any node in a transportation network
that is used for logistical activities. These activities include in particular transship-
ment operations, consolidation, or bulk-breaking. The term hub is oftentimes used
to describe a terminal that acts as a “major sorting or switching center in a many-to-
many distribution system” (O’Kelly and Miller, 1994, p. 32).15 The concrete definition
of the term is subject to debate and naturally depends on a specific context. Most
authors would agree that the outstanding share of sorting and switching activities, or
transshipment operations, is what distinguishes a hub from a simple transshipment
terminal.16 Accordingly, when the term hub is applied in the following, this is done
15 Crainic (2003, p. 456) similarly defines the term hub as a major consolidation center.
16 Taaffe et al. (1996, p. 17) restrict their usage of the term hub to airline networks and put gateway forth as
their all-embracing term. Their interpretation is not applied in the following.
Alternatively, a hub may also be a terminal that is highly connected to others (e.g. Reggiani et al. (2009,
p. 260)). This characteristic by itself appears less expressive for many-to-many networks than defin-
ing the term as related to sorting or switching activities. Notwithstanding, high connectivity and major
switching activities will not be independent from each other. Yet, the mere existence of highly connected
nodes does not imply high transshipment volumes.
Button (2002) presents many perspectives of interpreting the term hub only in the context of air trans-
portation.
The McShan-Windle (MW) index will be presented in section 3.4.2 as a measure of network concen-
tration. It measures the traffic share of the 3% largest transshipment terminals in the network. When
interpreting this as a metric of hubness, it can be stated that McShan and Windle (1989) perceive as hub
the 3% most important transshipment terminals in the network.
17
100%
%
transshipment 50%
volume
0%
1% 33% 64% 95%
Transshipment terminals
Figure 2.2.: Transshipment volume distribution.
to point out that the very transshipment terminal conducts significantly more trans-
shipment operations than most of the other terminals in the network.17
It is common for transportation networks to have some transshipment terminals
that conduct significantly more transshipment operations than others. Figure 2.2
depicts the distribution of transshipped volume in an exemplary LTL network. The
example shows that few terminals contribute strongly to the overall transshipment
volumes in the network, e.g. in this case only 4% of the terminals transship more than
30% of the entire transshipment volume in the network.
Connections in transportation networks refer to links in the network that connect
terminals. In airline networks, the term (flight) leg is common (e.g. Barnhart et al.
(2003a, p. 370) or Derigs et al. (2009, p. 370)), in LTL networks one would rather use the
term (trade or traffic) lane for the connections defined by direct services (e.g. Keaton
(1993, p. 345)). Connection is suitable for all network types. A path is a sequence of
connections through the network that relates an origin to a destination. Again, the
airline expression for a path is a route flown by an aircraft or traveled by a passenger
(Ball et al., 2007, p. 42). Path is the embracing term used for all modes.
The Oxford English Dictionary defines vehicle as “any means of carriage, con-
veyance, or transport; a receptacle in which anything is placed in order to be moved”
(Oxford English Dictionary, 2010). It is an umbrella term for all transportation
equipment. Commonly used for road and rail transportation, it will also denote any
other equipment such as aircraft and vessels that serve to hold and transport goods
or passengers in discrete loads.
The two prototypes of networks that are contrasted in most literature contributions
on transportation networks are P2P networks (also referred to as fully-meshed or grid-
17 The term hub appears in hub-&-spoke (H&S) network. Originally, this stems from the idea, that all H&S
networks have one or very few such hubs. Transportation networks with many transshipment terminals
are commonly referred to as multi-hub-&-spoke (MH&S) networks as they are seen as an extension of
H&S networks, but here, the outstanding role of some transshipment terminals may be contested. Even
so, the terms H&S network and MH&S network are used to describe these networks in the literature and
in practice, whether the transshipment terminals actually have an outstanding amount of traffic or not.
18
networks) and H&S networks (e.g. Hall (1987b), Bryan and O’Kelly (1999), Lumsden
and Dallari (1999), Mayer (2001), Jara-Díaz and Basso (2003), Alderighi et al. (2007),
Derudder and Witlox (2009)). Figures 2.3(a) and 2.3(b) respectively depict these two.
In a P2P network, the nodes are interconnected with direct links. No transshipment
is necessary as direct transports between the nodes are possible.
In a single-H&S network, one of the nodes assumes the central role of a major trans-
shipment terminal, a hub.18 When all nodes are connected only to the hub (the star-
type network), it is sometimes also referred to as a 1-hub-system (Mayer, 2001, p. 11).
No additional links exist in this most pure H&S network. All traffic flows between ter-
minals (apart from flows originating or terminating at the hub) are channeled through
the hub.
One distinction must be pointed out here as it will become important in the fol-
lowing. From a topological point of view, the network depicted in figure 2.3(b) is a
star-shaped network. The dotted line around the central node intends to indicate it
as a hub. A star-shaped network with a hub at its center is a H&S network. How-
ever, not all star-shaped networks are H&S networks: if no transshipment operations
take place at the central node, the network is not a H&S network. This difference is
sometimes overlooked in the literature.19 Burghouwt (2007, p. 257) lists definitions of
H&S networks and by far most accept the notion of transshipment operations at the
central node a constitutive element in the definition. In practice, most star-shaped
networks are H&S networks.20 Even though this is a lookahead on chapter 3 and will
be lined out in detail there, it is worth noting that the definition of a H&S network
has two components: the topological star-shaped network and outstandingly strong
traffic flows at the central node.
The cost elements of H&S networks and P2P networks differ as the operations are
not the same. Jara-Díaz and Basso highlight that the costs assessed in a study of trans-
portation network costs should always include a flow-oriented term21 to assess the
handling operations at the terminal as well as a flow-distance-oriented term22 to ac-
count for costs incurred in the transport (Jara-Díaz and Basso, 2003, p. 286). Based
on this, the key cost differences between the two networks can be identified. The net-
works differ in the importance of flow- or flow-distance-oriented costs. H&S networks
require transshipment operations, whereas P2P networks do not. The transported
18 It is also possible to introduce an additional hub node into the network. The difference between the two
cases is marginal. For the context here, however, it is more convenient to assume that the hub is one of
the nodes in the network and not explicitly added for the transshipment purpose.
19 For the case of telecommunication networks, Kim and O’Kelly (2009, p. 285) point out that hubs are
often reduced to an important node out of a number of nodes, neglecting the importance of switching
operations.
20 Then again, it is known that some airlines, especially regional airlines or low cost carriers (LCCs) (e.g.
Hawaiian Airlines (Hawaiian Airlines, 2010), Wizz Air (Wizz Air, 2010), or SmartWings (SmartWings,
2010)) operate star shaped networks with only very few connecting passengers at the central node. These
networks are no H&S networks.
21 The flow-oriented term is measured e.g. in weight such as metric tonnes (t).
22 The flow-distance-oriented term is measured commonly in tonne-kilometers (tkm).
19
(a) Point-to-point-network. (b) Hub-&-spoke-network. (c) Hybrid multi-hub-&-spoke-

network.
Figure 2.3.: Network types.
distances will be longer in H&S networks than in P2P networks as channeling the traf-
fic through the hub implies that a detour can be expected for passing there, which in-
creases distance-related costs. Whether flow or flow-distance costs are higher in any
network type depends on the given situation.
Many different network types are derived out of the two network prototypes. Well
known are multi-hub (MH) networks and hybrid hub-&-spoke (HH&S) networks. MH
networks are an extension of H&S networks in that they possess several hubs. The
hubs are interconnected in a fully-meshed network, but the other terminals will be
allocated to one of the defined hubs. In other words, MH networks are several distinct
H&S networks that are connected by a fully-meshed network only containing the hubs
(Mayer, 2001, pp. 12-13).
Another mixed form of networks are HH&S networks. They are based on a H&S net-
work but include “short cuts” (Lumsden and Dallari, 1999, p. 54) between terminals so
that it becomes possible to bypass the hub if necessary. It seems rather uncommon
to set up a HH&S network with a 1-hub-system at its base, so hybrid multi-hub-&-
spoke (HMH&S) networks are commonly found in practice, e.g. airline networks and
LTL networks are typically structured accordingly. Figure 2.3(c) sketches a HMH&S
network.
2.2.2. Transporting cargo or passengers

It is important to be aware of what is transported in a network when it is designed:
cargo or passengers. There are differences in expectations and feasibility for trans-
portation operations.
Transported objects is the umbrella term that covers all goods and passengers that
are transported in transportation networks. In the context of transportation, goods
are also referred to as cargo or freight which is used synonymously here.23 A shipment
23 The literature as well as practice seem to use the term cargo more frequently for air and rail transporta-
tion whereas freight is more common for road based transports. As this distinction is not generally agreed
upon, it is not uncommon to use the two terms synonymously.
20
describes a defined bundle of goods, possibly several pieces that belong together, and
that is transported on behalf of one shipper.
By definition, transportation deals with the two categories of transported objects,
cargo and passengers. By far most modes of transport are able to handle both, either
by actually transporting them on the same vehicle (such as transporting air cargo in
the belly of passenger aircraft (Doganis, 2010, p. 21)) or with distinct vehicles (such
as cargo trains and passenger trains). The general concepts are the same for both
categories of transported goods. The demand by the two transport objects is, how-
ever, different and will influence the actual operations. It might be seen as a trend
that for certain subsets of cargo and passenger, the demand converges almost to the
same. For example, express letter delivery shows certain characteristics that are usu-
ally related to passenger transport. Similarly, the demand fulfilled by low cost airlines
resembles that of cargo transportation in some aspects.
Nonetheless, typical demand patterns for passenger and cargo transportation may
be distinguished (Rodrigue et al., 2009, pp. 151-152).
Speed of transport. In general, passengers value the speed of travel highly, as trav-
eling by itself does not fulfill a purpose; the service offered is to arrive at the
destination.24 Apart from express shipments and mail, cargo is less sensitive to
the speed of transport.25
Frequency of service. Passengers have a higher demand concerning the frequency

of service than cargo does. Especially in urban transport, frequency by itself is a
goal for network design. In certain contexts, frequency is of relevance for cargo
as well, yet, the time intervals between two consecutive services will typically
not be as short for cargo as for passenger transport.26
Schedule reliability. If schedules are set up, passengers expect transports to meet
those schedules. Delays are unacceptable. Cargo is less sensitive to that. It
must, nevertheless, be mentioned that in certain supply chain contexts, sched-
ule reliability will be crucial as well. For instance, just-in-time (JIT) and just-
in-sequence (JIS) is infeasible without a high reliability of the defined arrival of
goods.
Peaks in demand. Passenger transport demand has high peaks during the day, es-
pecially at commuter times in the morning and the late afternoon. Seasonal
peaks around holidays are perceived as well. Cargo demand is more evenly
24 Exceptions do certainly apply, especially when thinking about cruises or scenic drives in tourism.
25 Various time components are commonly included within the planning objectives for urban passenger
transportation (e.g. Wirasinghe and Vandebona (2010)).
26 Continuous transportation modes exist for passengers and for cargo for specific applications: e.g. esca-
lators for people, conveyor belts in production processes for goods, or pipeline transports for cargo in
terms of liquids and gas. Thus, for specific contexts, the desired frequency of service for passenger and
cargo converges.
21
split throughout the day, mainly depending on production cycles and opening
hours of businesses. Nonetheless, exceptional high demand is experienced on
a seasonal level (e.g. highs before Christmas and lows over the summer), on
a repetitive basis (e.g. at the end of the month, related to so called MRP-jitters
(Lee et al., 1997, p. 554)), and on a weekly schedule (peaks towards the weekend
(Humphrey et al., 2007)).27
Traffic balances. Passenger traffic flows are usually assumed to be balanced, as most
passengers return home after their trip. This is a major difference when com-
pared to cargo, as such flows are characterized by high imbalances of traffic.28
Routing options. Passengers are free to choose any routing through a transportation
network and will have many personal reasons to prefer one routing over an-
other. Cargo does not have these individual preferences and will usually only
demand transportation from origin to destination without specifying the rout-
ing (Derigs et al., 2009, p. 371). Hence, it can be routed following a planning by
the transport provider.
2.2.3. Creating efficiencies in transportation networks

Efficient many-to-many transport operations rely on transportation networks. Sev-
eral triggers allow to create cost efficiencies in transportation networks. It may not
be possible to exploit all of them simultaneously in a given situation. Still, it is ben-
eficial to be aware of the typical efficiencies that may be generated in transportation
networks.
Economic research has long since discussed the cost impacts of production pro-
cess characteristics. The focus lies on identifying the relation of input and output
to a production process. Efficiencies exist if the output increases more than the re-
quired input or if the same output can be achieved with lower input. The same can be
done for the special production process of transporting goods through a transporta-
tion network.
The input to produce transportation on a network is the network itself (Jara-Díaz
and Basso, 2003, p. 274). This includes inter alia the number and location of nodes in
the network, the routes between the nodes, the route frequency, the route structure,
and the transportation distance. This input induces costs for the network operator.
The output of the process are the transports that are conducted. A product in this
context is an origin-destination (OD)-connection of a certain service class. A service
class is e.g. a type of transported object. Since distinct OD-transports are different
27 For the case of time-sensitive cargo transports, daily peaks may occur. For instance, (express) mail trans-
port experiences high traffic demand at night-time when sorting operations must be carried out during
a short interval of time to allow for fast transport to the destinations afterward (Crew et al., 1990). In that
sense, specific cargo segments may experience similar constraints as for passenger transport.
28 Doganis (2010, p. 298) stresses this aspect for the comparison of air passenger and air cargo traffic.
22
cost
distance
Figure 2.4.: Economies of distance.
products, any many-to-many network by itself offers multiple products. Disaggregat-

ing all products in a transportation network is unfortunately rarely feasible in practice
(Pels and Rietveld, 2000, p. 322).29 It is nevertheless beneficial to disaggregate outputs
to highlight different efficiencies that can be created in transportation in the follow-
ing.
An often stated research question in economic research is whether or not certain
economies exist in certain production or transportation situations. Furthermore, cost
function analysis has been looked at broadly.30 The following sections are intended
to highlight the main categories of economies that are at the root of efficiencies in
transportation networks and give some high-level reasoning in what situations they
are likely to prevail.
Transport-related economies
Certain sources for cost economies are particular to transportation processes. First of
all, transportation cost is highly impacted by characteristics of the transported object
such as stowability and easy handling (Bowersox et al., 2010, pp. 219-221). Economies
of distance and of vehicle size are, furthermore, specific for the context of transporta-
tion.
Economies of distance Economies of distance exist, if the transportation cost per

unit of distance decreases as the transport distance increases. The reasoning for their
existence is straightforward: the fixed cost associated with a transport is distributed
over a longer distance and the cost per unit of distance decreases. Figure 2.4 shows the
generalized relation between distance and cost. The term tapering principle is usu-
ally associated with economies of distance in transportation (Bowersox et al., 2010,
p. 219).
29 The most popular aggregate measurement of the output is tkm, even though its significance is contested
in the literature (Jara-Díaz and Basso, 2003, p. 273). It is, however, most commonly used in the industry
and will be easy to understand for practitioners.
30 See Pels and Rietveld (2000) for an overview of cost functions in transportation.
23
cost
shipment weight
Figure 2.5.: Economies of scale.
Economies of vehicle size If the cost for increasing vehicle size is proportionally
less than the gain in transport capacity, the situation exhibits economies of vehicle
size. Morrison and Winston (1985, p. 59) note that most transport industries may
exploit economies of vehicle size. The special case of economies of aircraft size cer-
tainly is the most prominent one in the literature,31 even though the same arguments
apply for any other transport vehicle. Economies of vehicle size are “crucial for the
existence of economies of traffic density” (Basso and Jara-Díaz, 2006b, p. 158). Only
because an increase in capacity of a vehicle will not induce the same increase neither
in purchasing nor in operating costs will transport providers have the possibility to
exploit economies of traffic density.
Economies of scale
Economies of scale exist in many contexts and are well known in production research.
They refer to the long run development of average total cost. Economies of scale
in production exist, if “long run average total cost falls as the quantity of output in-
creases” (Mankiw, 2001, p. 284). The very concept has been translated to transporta-
tion networks and is “the concept used in applied research to study issues such as
ownership, regulatory reform, the scope for competition, or postderegulation market
structures” (Basso and Jara-Díaz, 2006a, p. 260). Economies of scale in transportation
are generally defined as the decrease of cost per unit as the size (or volume, weight) of
the shipment increases (e.g. Bowersox et al. (2010, p. 194)), as depicted in figure 2.5.
The input dimension is the weight of the shipments, but may as well be in terms of
volume. The increase in efficiency is mostly related to the fixed cost of transportation
being distributed to a larger weight of shipments. The fixed cost includes e.g. time to
position a vehicle to load and unload, invoicing, and cost of equipment.
Economies of scale can be looked at in more detail. The size of a transportation net-
work, being the input to the system, is understood as the number of nodes n in the
network (Oum and Waters, 1996, p. 430). The OD-dimension of the product of a trans-
31 The interested reader is referred to Wei and Hansen (2003) for an overview.
24
Route structure
fix variable
Economies of scale
Network ssize
Economies of
fi
fix with
ith fixed
fi d network
t k
density
size
Economies of (spatial) scope
variable
N
(or economies of network size)
Figure 2.6.: Categorization of cost economies following Basso and Jara-Díaz (2006b,
p. 152).
portation network is not independent of the network size. Accordingly, economies of

scale in transportation networks are often seen as a mixture of the impact on cost of
traffic density and network size (Oum and Waters, 1996, p. 429). Moreover, it is highly
probable that with an increase in traffic, the routing through the network will tend
more to direct routings bypassing central transshipment nodes. For the context of
transportation networks, the concept of economies of scale usually assumes a vari-
able network size and, thus, “studies both product and network growth” (Basso and
Jara-Díaz, 2006a, p. 259). Basso and Jara-Díaz (2006b) split the concept of economies
of scale in transportation networks based on the observation that route structure and
network size impact the output of transportation processes. The authors distinguish
between (i) economies of density, (ii) economies of scale with fixed network size, and
(iii) economies of scope as sketched in figure 2.6.
Economies of density Economies of traffic density refer to “the impact on aver-

age cost of expanding all traffic, holding network size constant” (Oum and Waters,
1996, p. 429), which is the same as saying that they show “the unit cost implications
of a change in the volumes transported in a given network” (Pels and Rietveld, 2000,
p. 323). Especially economies of vehicle size are important for their existence in trans-
portation networks (Basso and Jara-Díaz, 2006b, p. 153). Basso and Jara-Díaz (2006b,
p. 151) point out that in order to be able to interpret returns on traffic density, the
routing of the network has to be kept at a fixed level.32
From a strategic point of view, economies of traffic density can be expected to exist
in most transportation networks. More traffic on existing network arcs allows inter
alia to achieve higher fill rates on the transport vehicles used. Additional shipments
can be transported with hardly any additional cost.
32 The changes in traffic routing must not be included into the assessment of economies of traffic density.
If the possible modification in the traffic routing is to be taken into account, the authors suggest the
measure of economies of scale with fixed network size (Basso and Jara-Díaz, 2006b, p. 160).
25
Economies of scale with fixed network size Economies of scale with fixed net-
work size are used to assess returns to scale when the network size is kept constant
but route structure may change. They “are related with the convenience or inconve-
nience of expanding proportionally the flows in all OD-pairs”(Jara-Díaz and Basso,
2003, p. 283). A proportional increase of flows in a transportation network will induce
a shift from a H&S network towards a P2P network since the need to consolidate flows
decreases. This situation is typical for the transportation context: “The need to make
a decision on a route structure is, finally, a consequence of the spatial dimension of
[the] product” (Jara-Díaz and Basso, 2003, p. 274).
Economies of (spatial) scope Economies of scope are “the impact on cost of

adding new outputs to the line of production” (Jara-Díaz and Basso, 2003, p. 283).
They “exist when it is cheaper to produce two or more outputs jointly by a single
firm than producing each of them separately by an independent firm” (Oum and
Waters, 1996, p. 435). As mentioned above, in transportation networks the products
typically are OD-connections and different types of transported objects. And so,
when economies of scope are discussed in the literature on transportation networks,
one usually refers to situations where cargo and passengers are transported on a
single vehicle or to situations where new OD-connections are offered. Whereas the
former may be treated as a special case of economies of density, the latter cannot be
discussed without modifying network size. The notion of economies of spatial scope
explicitly refers to adding new OD-connections. Making this distinction allows to
analyze correctly the advantageousness of network expansion and, thereby, supports
economic research (Jara-Díaz and Basso, 2003, p. 286). Moreover, being aware of the
distinction between the different cost economies allows a better understanding of
the different elements of potential cost reduction.
Elements of cost reduction in transportation networks

The presentation of different cost economies above sheds light on some key elements
that may or may not lead to cost reductions as transportation networks grow. In sum-
mary, the key elements discussed above that may influence the cost efficiencies of
transportation networks are:
• the demand for transportation on certain network links,
• the nodes in the network, which relate to the OD-connections offered in the
network,
• the routing through the network,
• bundling different products on one transport vehicle,
• the transportation distances, and finally
26
2.3. Optimal transportation network design - an Operational Research perspective
• the transportation vehicles.
One way or the other, these elements will influence the costs in transportation net-
works and should therefore be kept in mind when designing these.
The following section will present OR models that find optimal - usually cost-
optimal - transportation networks. The above mentioned elements are of importance
for OR research as well.
2.3. Optimal transportation network design - an Operational

Research perspective
“OR is the discipline of applying appropriate analytical methods to
help make better decisions.” (INFORMS, 2010)
The term Operational Research (OR) has been known since around the Second
World War. It is rooted in two areas: the military area focusing on applied research
concerning, at first, radar systems as well as the private sector that was eager to ap-
ply methodology from sciences to management questions. From early on, OR was
an applied science with a special interdisciplinary focus. Research continued to de-
velop new models, solution algorithms, and heuristics to support real-life decision-
making. It benefited strongly from advances in computer science that provide re-
searchers nowadays with computational support beyond thought decades ago (Zim-
mermann, 2008, pp. 6-9).33
Many different terms are used to describe the application of analytical methods for
managerial decision-making. Operational Research is the term the INFORMS Society
decided to put forward in 2004. Amongst others, the prominent wording of Man-
agement Science basically covers the same aspects; Operations Research is more fre-
quently used than Operational Research in Germany but describes the same research
field. Thus, Operational Research shall be the umbrella term for the entire field (IN-
FORMS, 2010) and, therefore, also the term used in the following.
Network design creates networks explicitly, e.g. by defining the nodes and the links
that should be included in the networks.34 The problems arising in network design
can often be solved to optimality with the toolkit provided by OR. Even a short glance
at the literature on network design will highlight the importance of optimization
models in this context. Nevertheless, optimization is only one method established in
OR: scenario analysis and simulation will also support the decision-making process.
33 When taking a look at research streams in OR, an orientation to methods and less to applications is
discovered. This might be a misleading picture due to the fact that application topics are more difficult
to publish in academic journals.
34 The term network design is sometimes used in a restricted understanding in the field of urban planning:
“The network design problem involves the optimal decision on the expansion of a street and highway
system in response to a growing demand for travel” (Yang and Bell, 1998, p. 257). As opposed to this, a
broad understanding - independent of a specific type of network - will be applied in the following.
27
Furthermore, good heuristics play a major role in network design. Any method fulfills
a different purpose. Scenario analysis suggests decision alternatives, simulation
helps to evaluate alternatives, heuristics find good solutions for difficult problems,
whereas optimization provides optimal solutions for a given model (Grünert and
Irnich, 2005, pp. 8-9).
Abstractly, network design will always address the same questions for all networks:
Number of nodes. Some nodes in the network will be externally given, such as cus-
tomer or supplier locations. It remains an open question how many additional
nodes are to be added to the network in order to achieve the expected perfor-
mance.
Location of nodes. If nodes are added or should be relocated for the network, one
will have to decide where to locate these. One example of this is the decision
where to locate a new production site, considering location information of sup-
pliers and customers.
Connecting nodes with links. For most networks, the planner will have to decide
which links are to be added or operated in the network. Energy providers can
think about actually constructing new cables while airlines will rather have to
decide which airports to connect by a scheduled aircraft.
Operating the links and nodes. Finally, companies will have to set policies con-
cerning how to operate the constructed or opened links and nodes. This
includes setting schedules for public transport but also figuring out how
the landing / take-off and ground operations at an airport are conducted
efficiently.
Networks usually consist of several layers that are linked to and embedded in each
other. The self-similarity of the network model (Delfmann et al., 2010, p. 59) implies
that the above illustrated questions in network design may arise on all layers. For
instance, intralogistics transport flows within a production facility form one layer of a
network that is linked to the external transportation network for final products.
Transportation network design is an important field of application for OR methods.
Crainic and Laporte state that “transportation planning is undoubtly one of the great
success stories” of OR (Crainic and Laporte, 1997, p. 435).
This section presents important streams of research to design optimal transporta-
tion networks. Certain overlaps in content exist with neighboring fields such as SCM,
SC design or supply network design.35
Different time horizons are relevant when designing networks. Decisions are taken
on a strategic, long-term planning horizon; others on tactical, mid-term planning
35 Chopra (2003) illustrates different options for designing distribution networks. Klibi et al. (2010), Melo
et al. (2009), Meixell and Gargeya (2005), and Lapierre et al. (2004) provide overviews of distribution
network design from an optimization point of view.
28
horizon and eventually on the operational, short-term horizon. Crainic and Laporte
(1997) draw an overview of the different planning issues and have influenced the fol-
lowing sections.36
The next section will very briefly introduce some graph theoretic terminology that
is necessary for the presentation of the models in the following. Subsequently, OR
formulations for strategic and tactical decisions in transportation network design are
presented. Further, operational decisions are sketched to complete the picture. The
goal of the presentation is twofold. Firstly, the type of decisions to be taken at each
planning horizon is characterized by exemplary model formulations. This will pro-
vide a clear picture of the manifold challenges to be solved in transportation net-
work design. Secondly, the presentation allows to identify a strong cost-orientation in
the OR models for transportation network design. Yet, the service-oriented design of
networks is also present in the pertinent literature; the addressed aspects of service-
orientation are briefly outlined following the model orientation. Based on the pre-
sentation of the models, some key performance indicators (KPIs) for transportation
network design are identified for further discussion in chapter 3.
2.3.1. Graph theoretic terminology for network design

Networks have a graph representation.37 A brief introduction to the graph theoretic
terminology that will be applied in the following will simplify the understanding of
the OR models in this section.38
The mathematical field of graph theory provides researchers with a wide terminol-
ogy and toolkit to support the analysis and design of networks. A graph G consists of
a set of nodes (or vertices) V , a set of edges (or links) E as well as a relation ω (inci-
dence relation) that maps each element of E onto two elements i and j in V . Thus,
E contains 2-tuples of nodes. If the pairs in E are ordered, the graph is directed; if
the pairs are unordered, the graph is undirected (Domschke and Drexl, 2005, p. 65).
Edges in directed graphs are also referred to as arcs or arrows. Directed graphs have
paths: a sequence of arcs (a 1 , ..., a t ) is a path p 0t in G if a sequence of nodes (v 0 , ..., v t )
with a h = (v h−1 , v h ) for all h = 1, ..., t exists (Domschke and Drexl, 2005, p. 67).
A graph (directed or undirected) is a weighted graph if some value is associated with
the edges of the graph. The value can have many interpretations, common are the
36 Barnhart et al. (2002) suggest a time line of four planning activities: strategic planning, tactical planning,
operations (or market) planning, and contingency planning. The tasks that Crainic and Laporte (1997)
allocate to the tactical and operational horizons are allocated to the tactical, operations, and contingency
planning by Barnhart et al. (2002).
37 Figure 2.1 touched on this aspect before.
38 The formal definitions express the same ideas as presented above in section 2.2.1, where the terms were
already introduced without making reference to the graph-theoretic interpretation of it. The brief pre-
sentation of fundamental graph theoretic terminology here, is aimed to support the understanding of
the next sections and not intended to be a comprehensive introduction to graph theory. Many textbooks
exist to this goal, e.g. West (2001).
29
cost to use that edge, the length of the edge or the capacity of the edge. The notation
of c i j is used for the value (especially the cost) associated to edge (v i , v j ).
If p 0t= (v 0 , ..., v t ) is a path in G , the sum of all associated values with edges in p ,
t
c (p ) = h=1 c v h−1 v h is defined as the length of the path p . A path p i∗j is a shortest path
from i to j if no other path p i j in G exists with c (p i j ) < c (p i∗j ). In that case, c (p i∗j ) is
the shortest distance from i to j (Domschke and Drexl, 2005, p. 68).
2.3.2. Strategic decisions for network design

Strategic issues in transportation network design are usually related to infrastructural
decisions, as these must be taken on a long-term planning horizon. A typical example
is the location of terminals in the network and their capacity planning. In contexts
where the transport connections depend on infrastructure networks such as tracks,
decisions on their location and capacity are just as important.39 Similarly, for the
planning process of freight carriers, Crainic (2003, p. 469) characterizes strategic de-
cisions to “determine general development policies and broadly shape the operating
strategies of the system over relatively long-term horizons”.
Optimal terminal location

Many different location models have been developed in OR. Generally speaking, all
location models find optimal geographic positions (e.g. for constructing a production
plant or a transshipment terminal). Figure 2.7 depicts the different location decisions
and classifies the examples given by ReVelle and Eiselt (2005, pp. 2-3). Location prob-
lems differ in the topology of potential positions: either in the plane or on a network.
Planar problems are usually formulated in 2-dimensional real space, but e.g. clus-
ter analysis maps locations in n-dimensional space by making use of planar location
problems. Location models on networks limit the potential positions to a given net-
work which is common in transport-related problems. Both types of problems may
further be divided into continuous and discrete models. Continuous problems allow
the location decisions to be taken anywhere in the plane or network whereas discrete
models limit the potential locations to choose from.
Even though one finds relevant applications for all four possible combinations of
problem classes, by far most of the problems discussed are either continuous planar
or discrete network models (ReVelle and Eiselt, 2005, Klose and Drexl, 2005).
Continuous location models in the plane In the context of location decisions, the
models usually refer to 2 but the methodology is easily adapted to n . The following
39 Song et al. (2008, p. 1265) identify two different perspectives in strategic network design: the shipper and
the carrier perspective. The shipper will usually include inventory costs into the models while the car-
rier is more interested in optimally utilizing its assets. The following presentation will provide a general
overview of strategic decision in network design and not differentiate between the two perspectives.
30
Set of ppotential loccations

Positioning of Locating retail
discrete transmitter stations facilities on lots
at some that are zoned for
permissible points them
Landing a Placing a tow truck

continuous hhelicopter
li t for f along
l a stretch
t t h off
trauma pickup highway
planar network
Topology of potential locations
Figure 2.7.: Situative examples of location problems.
assumptions are at the core of continuous planar location models (Domschke and
Drexl, 1996, p. 162):
1. Customer locations are situated on a homogeneous plane.
2. Every point on the plane is a potential location.
3. The distance between two points on the plane is measured according to a cer-
tain metric.
Domschke and Drexl (1996, p. 164) present common metrics to measure the dis-
y y
tance between two points v i (v ix ,v i ) and v j (v jx , v j ):
• the rectangular distance metric, often used in intralogistics contexts

x x y y
d i s t i j := v i − v j + v i − v j
• the euclidean distance metric , suitable for many situations

y y
d i s t i j := (v ix − v jx )2 + (v i − v j )2
• the squared euclidean distance metric, especially applicable when long dis-
tances are strongly unfavorable, such as in defining locations for emergency
buildings (firefighters, hospitals, etc.)
y y
d i s t i j := (v ix − v jx )2 + (v i − v j )2
The Steiner-Weber model is a formulation for a continuous location model with

euclidean distance metric (Delfmann, 1987). The n known customer locations are in
y
the form (v ix , v i ), each having a weight of d i . The coordinates of the potential location
31
is given in the form (x , y ). The model is stated as follows (Domschke and Drexl, 1996,
p. 167):

n
y
min F (x , y ) = di (x − v ix )2 + (y − v i )2
i =1
The Steiner-Weber model cannot be solved analytically, but iterative solution pro-
cedures exist, especially Miehle’s algorithm converges to accurate solutions after only
a few iterations (Miehle, 1958, p. 239). It is apparent that the Steiner-Weber model
targets at minimizing the distance-weight-related costs in the plane.
Discrete location models in networks The warehouse location problem (WLP) is

the most prominent discrete network model. The simple WLP describes the following
situation. A company delivers to n customers and wants to decrease its distribution
cost. It plans to open one or several warehouses. m possible locations are prese-
f ix
lected for these. There are fixed costs c i associated with opening a warehouse at a
location i . Transports from warehouse i to customer j cost c i j monetary units. The
objective of the model is to minimize the overall costs of the network, by opening the
best number of warehouses and define the allocation of customers and warehouses
(Domschke and Drexl, 1996, pp. 51-52).

m
n
m
f ix
min F (x , y ) = ci j xi j + ci yi (2.1)
i =1 j =1 i =1
s .t .
x i j ≤ y i for all i = 1, ..., m and j = 1, ..., n (2.2)

m
x i j = 1 for j = 1, ..., n (2.3)
i =1
y i ∈ 0, 1 for i = 1, ..., m
x i j ≥ 0 for all i and j
The decision variables in the WLP are understood as:

1 if customer j is fully served by warehouse i
xi j =
0 if j does not deliver to i
0 ≤ x i j ≤ 1 if i delivers d j x i j units to j

1 open warehouse at location i
yi =
0 else
32
Extensions of the simple WLP include capacitated warehouses, non-linear costs,

and the introductions of additional echelons (which leads to transshipment prob-
lems).40 It can already be seen from the simple WLP that the minimization in this
case includes not only transport-related costs but also additional cost elements such
as fixed warehousing costs. Alumur and Kara (2008) review in detail network location
problems.
Optimal network links

The other strategic perspective - apart from finding the optimal terminal location -
is to decide on the links in the network. Formulations start with a given demand for
transport services from origins to destinations and will then find the best set of links
to produce the related transportation service (Crainic and Laporte, 1997, p. 413).41
Finding optimal links is of varying importance for different modes of transport. In
rail transport, finding optimal links often relates to the construction of tracks which
is a strategic decision for companies. In road transportation, finding optimal links
supports the mid-term purchase of truck capacity,42 since roads as the infrastructure
is mostly provided by the states and part of the planning of companies.
Many model formulations for the identification of network links stem from graph
theory. Two basic versions of the problem are minimum spanning trees and shortest
path algorithms for a graph. Both allow to identify links that should be included in
a network. They may also be used for tactical decisions. On a strategic level, both
formulations may support the strategic infrastructure planning concerning network
links. A minimum spanning tree is a group of arcs connecting all nodes in a net-
work with minimal cost (Winston (2004, p. 456), West (2001, p. 95)).43 Shortest paths
problems identify the shortest path from an origin to all or a predefined destination.
Algorithms to find shortest paths are applied in network design (Bazaraa et al., 1990,
p. 575).
Summing up, strategic decisions in transportation network design identify optimal
network nodes and - in certain contexts - network links. The solutions depend on
cost and demand assumptions. The strategic decisions are usually taken as input for
tactical decision-making.
40 For a capacitated model, each customer has a demand of d j units. An additional constraint to force the
model to meet the entire demand is then included.
41 The WLP already includes the variable x i j that hints at which links are to be used in the network design.
From this perspective, the WLP finds optimal nodes and links simultaneously.
42 Opening a link in a road transport networks usually means to schedule a truck on that link.
43 Ahuja et al. (1993) give a comprehensive overview of elaborated algorithms to find minimum spanning
trees.
33
2.3.3. Tactical decisions for network design

Once strategic decisions about infrastructure in a network were taken, tactical issues
arise in network design with a mid-term planning horizon. Strategic decisions of-
ten act as input to tactical decision-making. The tactical planning is concerned with
“services on an existing network configuration” (Song et al., 2008, p. 1266). The aris-
ing challenges differ by network purpose. For networks that require consolidation,
such as LTL networks, finding the optimal service network is of crucial importance.
The routing of freight on this network is referred to as the load plan and its generation
constitutes an important challenge in this context. As opposed to this, in cases where
several pick-up and delivery operations are performed, such as in the courier busi-
ness, the vehicle routing problem (VRP) is highly relevant. Moreover, crew schedul-
ing is important for all transportation networks based on tight schedules and relying
on a skilled workforce. The following sections will shed light on these challenges and
related models.
Further elements are of importance in the tactical planning but will not be dis-
cussed as typical and prominent examples for tactical network design. Amongst them
are terminal policies. These include the definition of tasks to be performed within the
terminals, e.g. how shipments are to be consolidated. O’Kelly (2010) draws atten-
tion to the planning of routing in the transshipment terminals. He points out that
this questions has hitherto been neglected in the literature on transportation net-
work design. The efficient task distribution between terminals is also seen as part of
the terminal policies to be set (Crainic and Laporte, 1997, p. 421). In MH networks
this touches the task of allocating terminals to hubs. The literature provides generic
trade-offs in this decision.44
Another element of tactical planning is related to the balancing of empties. The
question to be solved includes how to reposition empties (vehicles, containers, boxes,
etc.) for future (such as next-day) operations (Crainic and Laporte, 1997, p. 421). The
literature on repositioning empty vehicles has many contributions directed to the rail
transport industry as empty rail car distribution is an important planning issue in
this industry (e.g. Holmberg et al., 1998) but is not limited to this field (Turnquist,
1985, pp. 361-362). The management of returnable packaging such as containers of
any kind (e.g. Kroon and Vrijens, 1995, Mevissen, 1996) is another element but less
directly connected to transportation network design.45
44 Section 2.4.2 explicitly addresses these.

45 Closed loop supply chain management (CLSCM) has developed more recently. Reverse flows and (re-
) distribution circuits for returnable containers are explicitly treated together with forward flow SCs with
the goal to create models and therewith solutions that can describe and provide answers for the flows
in the entire networks. Dekker et al. (2004) and Dyckhoff et al. (2004) provide valuable overviews of the
variety of aspects discussed in reverse logistics and CLSCM.
34
Optimal service network

The goal of service network design models is to “plan services and operations to an-
swer demand and ensure the profitability of the firm” (Crainic, 2000, p. 280). The term
service in this context expresses the frequencies of transport. Coming from the ter-
minals or facilities located during the strategic planning phase, the service network
design now aims at selecting the connections between the network’s nodes and espe-
cially their frequencies (Crainic and Laporte, 1997, p. 421).
Crainic (2000) gives an overview of different models for service network design.
These can be distinguished into two categories: models treating frequencies as
decision variables as well as models deriving service frequencies as their output. The
former use decision variables that express how often the service is offered in the
planning period. The latter have binary “operate or not” variables (Crainic, 2000,
p. 281) and define the service frequency based on these and a minimum service
constraint. Both formulations lead to a load plan, i.e. the plan on “how a shipment
will be routed through the network, in the form of the sequence of [break-bulk (BB)
terminals] through which the shipment must pass before reaching the destination
[end-of-line (EoL) terminal]” (Powell and Sheffi, 1983, p. 472).
A short example of a model with frequencies as decision variables gives an idea
about the types of formulations found in the literature (Crainic and Laporte, 1997,
p. 423).
min Ψ(X pg , Fπ ) (2.4)

s.t.

X pg = d g for all g (2.5)
k
X pg ≥ 0 for all p, g
Fπ ≥ 0 and integer for all π
Fπ is the decision variable representing the frequency of service for each OD-
connection π in the network. Hence, Fπ stands for the fixed cost of the transportation
g
system. The variable cost of the system is expressed in X p being the routing of freight.
It is the amount of flow of traffic class g on the itinerary p .46 An itinerary may cover
several services. The transportation demand per traffic class is d g . The objective
g
function Ψ(X p , Fπ ) accumulates the total cost of the transportation system by incor-
porating fixed and variable elements. Crainic and Laporte (1997, p. 423) provide an
example of the explicit form such an objective function may take. Their formulation
further highlights that delay costs or penalty terms may be included directly into
46 A traffic class represents the transport demand for a certain commodity, e.g. passenger or cargo, from an
origin to a destination.
35
the objective function. As frequencies in this model are decision variables Fπ , the
optimal frequency per service can be read from the solution of the model.
An often referenced example of a model that derives service frequencies as its out-
put is the model provided by Powell and Sheffi (1983). In this case, the frequency F (x )
depends on the traffic flow x in vehicle loads on a route. A minimal service frequency
Fm i n is enforced with the help of a minimal service restriction. This situation reflects
nicely how LSPs often operate their networks. The service frequency then is defined
as ⎧
0 x =0
⎨
F (x ) = Fm i n 0 < x ≤ Fm i n
⎩
x x > Fm i n
The primary decision variable y isj stands for the developed load plan

1 if freight at node i , destined for s should go next to terminal j , i j ∈ L, s ∈ T
y isj =
0 otherwise
The formulation further requires the binary variable δir js

1 if link i j is on the path defined by y from r to s
δir js =
0 otherwise
Let x be the vector of flows as F is the vector of frequencies. Several sets are of im-
portance: L includes all direct links joining terminals, the set of all nodes V , the set of
all transshipment terminals T and finally the set of all links representing movements
through break-bulks B . The costs taken into account are the transportation cost per
vehicle from i to j c i j , and the respective cost for handling operations of a full vehicle:
c h f u l l . D r s is the total flow from origin terminal r to destination terminal s . t i j is the
time spent on a link and ςr s is the maximum delivery time constraint.
With these variables, the full model can be stated:

x , F ) =
hf ul l
min Z ( c i j Fi j (x i j ) + ci j xi j (2.6)
i j ∈L i j ∈B
s.t.

y isj = 1, for all s ∈ T, i ∈ V (2.7)
J

xi j = D r s δir js (2.8)
r ∈T s ∈T

t i j (x i j ) · δir js ≤ ςr s , for all r s (2.9)
ij
36
This formulation also relates to the frequencies but needs the binary variable δir js
to match if a service is operated at all. Powell and Sheffi (1983) suggest a solution
algorithm to solve the problem.
Looking at both of these optimization formulations for service network design it
is again worth mentioning that their objective functions are purely cost-oriented
and service considerations are incorporated either as constraints or by using the
workaround of penalty costs.
The service network design problems stated above are relevant in the long distance
transportation where terminal consolidation plays an important role. The generated
load plan represents the routing of freight. For settings with shorter distances such
as distribution tours in urban areas, the desired information is less oriented towards
efficiencies by consolidating but rather by reducing transportation distance. OR treats
these questions with the help of VRPs.
Optimal routing
Basic vehicle routing problems (VRPs) are used to find cost optimal tours in a specific
situation. The number and location of customers as well as their demand is known.
A service provider aims to fulfill demand from its depot with a limited number of
transportation vehicles. The service provider has some restrictions as to the capacity
of its trucks and as to time. It strives to find the most cost efficient tour (as a sequence
of customers to visit one after another) by respecting the constraints (Domschke and
Drexl, 1990, p. 131).
The single depot VRP reads as follows (Laporte, 2009, pp. 409-410):

min ci j xi j (2.10)
(i ,j )∈E
s.t.

n
x 0j = 2l (2.11)
j =1

xik + x k j = 2, for all k ∈ V \ {0} (2.12)
i <k k <j

x i j ≤ |S| − v (S), for all S ⊆ V \ {0} (2.13)
i ,j ∈S
x 0j = 0, 1, or 2, for all j ∈ V \ {0}

x i j = 0 or 1, for all i , j ∈ V \ {0}
The special node 0 typically denominates the depot from where all l trucks start. S
is any arbitrary subset of V , v (S) is the lower bound on the number of vehicles needed
to produce the service. In a capacitated case where each truck has a capacity of c a p
37
di
units and each customer i has a demand of d i , v (S) = i ∈S c a p
. x i j is the decision
variable of the model. It can only take the values 0, 1, or 2 and equals the number of
times a vehicle travels on link i j .
The model will find a tour per vehicle always starting and returning to the depot.
Any customer is visited exactly once by one truck in the planning horizon. Equation
2.13 guarantees connectivity of the solution in the sense that sub-tours are excluded.
The VRP is a highly relevant but NP-hard problem; many good heuristics to solve
the VRP have been developed. Amongst the most prominent ones are the sweep- and
the savings-heuristic. The two heuristics solve the VRP by dividing it in two smaller
problems: allocation and routing. The sweep-heuristic allocates customers to tours
(Gillett and Miller, 1974) whereas the savings-heuristic creates the routing of one tour
(Clarke and Wright, 1964). The VRP with only one vehicle - omitting the allocation
issues related to several vehicle - is also known as the traveling salesman problem
(TSP). Thereby, the savings-heuristic basically provides a solution for a TSP.47
Optimal crew scheduling

“Crew scheduling can be defined as the problem of assigning a group of workers (a
crew) to a set of tasks” (Barnhart et al., 2003b, p. 517). Crew scheduling challenges
arise in any transportation network that involves scheduling personnel and are of
major importance in airline networks (Gopalakrishnan and Johnson, 2005, p. 306), in
mass transit systems (Guihaire and Hao, 2008), and in rail operations (Caprara et al.,
1997, p. 125).48 Barnhart et al. (2003b) explain that crew scheduling for airlines is
especially interesting to discuss as the problem is important for the airline industry
having relatively high wages for their crews and the provided transportation service
relies on strict schedules to function properly. Furthermore, from an OR perspective,
the airline crew scheduling problem provides a context for reflecting many problems
arising in any crew scheduling situation as the industry must meet an outstandingly
large number of regulations by legislation, unions, and agencies. Crew scheduling for
other transportation networks usually has lighter constraints and might have a lower
impact on the total operating costs of the network.
47 Milk run is a term that is often used in connections to transportation networks. “A milk run is a route
in which a truck either delivers products from a single supplier to multiple retailers or goes from multi-
ple suppliers to a single retailer” (Chopra and Meindl, 2004, p. 421). It is a tour where goods are either
dropped of and / or picked up as the vehicle drives along. Mathematically, the optimization of such a
tour refers to the TSP. It has long since been established and saw a revival lately in industry (e.g. Kohagen
(2010), Toth and Wagenitz (2010)). Peddling is sometimes used as a synonym for milk-run (e.g. Burns
et al., 1985).
48 Caprara et al. (1997) see the crew scheduling problem as a sub-problem of what they label crew manage-
ment.
Crainic (2003, p. 495) groups crew management as an operational problem. Nonetheless, for many ser-
vice providers, the related questions already arise on a mid-term planning horizon; operational modifi-
cation to the tactical plan are common.
38
Barnhart et al. (2003b, p. 519) present crew scheduling with four components as
its building blocks: flights, duties, pairings and monthly schedules. Single flights are
grouped into duties that represent a crew’s workday. So called pairings are subse-
quent duties for a crew separated by longer rest periods. Finally, a monthly schedule
is the sequence of pairings and rest times for each crew during one month.
It is relatively easy to model and interpret the crew scheduling problem (Gopalakr-
ishnan and Johnson, 2005, p. 306). However, due to the sheer complexity of the
problem it is usually solved in various steps by airlines. Mathematically, the crew
scheduling sub-problems are set partitioning and set covering problems (Barnhart
et al., 2003b, 529). A wide spectrum of optimal and heuristic solution procedures exist
in the literature. Barnhart et al. (2003b), Kohl and Karisch (2004), Gopalakrishnan
and Johnson (2005) provide valuable overviews for the interested reader.
2.3.4. Operational decisions for network design

Transportation networks are planned on the long and medium term but whether or
not the network will operate efficiently and create value for the network’s operator
is highly influenced by the quality of the operational decisions taken in daily busi-
ness. Accordingly, Crainic (2003, p. 494) formulates “strategic and tactical plans can
be drawn up to guide operations, but the operational capabilities of the firm will ul-
timately determine its performance”. A transportation network is strongly influenced
by external events and will require many ad hoc decisions by the personnel involved
in the operations to keep the network running. These events may have very different
causes, some examples might show the bandwidth of events:
Customer demand. Parcel services usually receive their pick-up orders from cus-
tomers during the day while they are on their delivery tours. What is the best
way to modify the routing for a driver in order to pick up the new parcels on the
tour?
Traffic congestion. A delivery truck is stuck in traffic and will not be able to return
to the terminal before the cut-off time. Is it worth delaying the terminal opera-
tions in order to wait for the truck?
Terminal break-down. A terminal cannot operate as planned because it was hit by a

lightning strike. How should the shipments be re-routed through the network?
Natural causes. A volcano eruption may be at the root of the closure of the entire
West-European airspace. How can air shipments be delivered anyway?
Shipment increase. Due to an early winter, the demand for transportation of snow
tires peaks unexpectedly. The tires are distributed through the network and
this was not foreseeable. What should be done to transport the additional ship-
ments?
39
Decisions often need to be taken on the spot and they will influence future opera-
tions. However, the side-effects on other operations cannot be judged at the moment
of decision-making: If a transport vehicle is used to transport excess demand, this
implies that the vehicle might not be available for its assigned tour later on. Delaying
long-haul trucks in the network might be possible but might as well result in spread-
ing the delay across the entire network.
These ideas hint at how important high quality decisions are for the final value cre-
ated with the transportation network. Two research streams exist. The first approach
is to find acceptable ways to resolve operational challenges. The second approach is
to plan for the inherent stochasticity. Decision support systems (DSSs) to improve the
quality of real time decisions have been of rising in importance over the last years.
2.3.5. Service-oriented aspects of OR network design

The previous sections summarized fundamental OR models to find optimal solutions
on different planning horizons. The goal was to show the manifold aspects of trans-
portation network design that have been addressed in the literature. Cost-orientation
is an important element in the models. In the brevity of the presentation, this can
merely serve as examples to hint at the breadth of the field. Service-orientation has
started to influence transportation and supply network design during the last years.
The following will point out some of the prominent aspects of service-orientation in
order to illustrate alternatives to the above mentioned cost-oriented models.
Various aspects of risk

It can be stated that the perception of risk in transportation networks has become
more common lately. Limiting risk may take different forms, discussing network flex-
ibility and reliability (see below) are merely two facets of tackling the challenge of
network risk. The perception that future networks must be more robust and that this
aspect must be incorporated into network design is supported in the literature review
by Klibi et al. (2010). The authors strongly advocate the future development of robust
supply network models. They claim that only robust networks are able to ensure sus-
tainable value creation. Meixell and Gargeya (2005) identify only few contributions
integrating flexibility and robustness into SC design. More recently, Melo et al. (2009)
indicate the rising importance of risk management for global SCs design. Pfohl et al.
(2010) review existing literature on risk in SCs and develop a vision for future research
in the field; another indicator of the rising importance of this topic in network design
and neighboring fields.
Verma and Verter (2010) actually formulate an optimization model that is, inter
alia, to limit the exposure of the population to the risk of transporting hazardous ma-
terials.
40
Flexibility
Researchers have begun to conceptualize flexibility for transportation networks. Flex-
ibility in transportation networks is directed to different aspects, e.g. flexibility in
adaption to changes in the demand structure for the network (Feitelson and Salomon,
2000, Morlok and Chang, 2004). Ukkusuri and Patil (2009) develop a framework of
network flexibility covering different aspects of flexibility for transportation networks.
Hsu and Wen (2002, p. 197) highlight that network design (in their application airline
network design) needs to incorporate flexibility to better cope with only rough esti-
mates of future travel demand.
System reliability & robustness

Shippers expect transportation networks to provide reliable service. Network relia-
bility is not a new topic in the literature concerning telecommunication networks;
Kim and O’Kelly (2009) provide a brief review of the pertinent contributions. Even so,
transportation system reliability has been looked at only sparsely in the past. Yet, re-
cent contributions address this topic more often. Hsu and Wen (2002) briefly review
the existing research approaches prior to 2002 and suggest a network design aimed at
finding a reliable airline network.
Scott et al. (2006) bring together the aspects of capacity flexibility and network re-
liability to suggest their network robustness index for highway transportation net-
works. The optimization of the robustness of a logistical network while minimizing
cost is discussed by Meepetchdee and Shah (2007). The authors furthermore link the
concepts of robustness and complexity.
Delays
Fowkes et al. (2004) present the results of a study on the value of timeliness in differ-
ent transportation contexts. Amongst others, the researchers identify that delays are
very unfavorable for JIT transports. Whilst this is not astonishing by itself, it is im-
portant to note that the authors explicitly targeted the research goal of transportation
delays: evidence for the perception of this issue being important for transportation
research. Fosgerau and Karlström (2010) discuss the value of reliability of trip dura-
tion with an abstract model. In general, delays seem to be more prominent in pas-
senger transportation research (e.g. Lederer and Nambimadom (1998)) than in cargo
transportation.
Green logistics & SCM

Greening the supply chain and providing environmentally friendly logistics has risen
in importance over the last years. The International Journal of Physical Distribution
and Logistics Management had a double special issue on The sustainable agenda and
41
energy efficiency: logistics solutions and supply chains in times of climate change in
January 2010; taking a closer look at the extensive research being done in the field
of green logistics that has seen a rise of attention during the last years. Mollenkopf
et al. (2010) review, inter alia, literature contributions in the field of green supply
chain strategies. The authors also address the different motives of greening the supply
chain, cost reduction being only one of them.
Technologies
It was mentioned above that the great breakthrough in transportation networks is
currently expected at the interface of information technology (IT) and transportation
(cf. section 2.1). New IT systems are available at reasonable prices nowadays. This de-
velopment leads to many potential applications in transportation. Different aspects
on this have been discussed by researchers. Mageean and Nelson (2003) evaluate
the potential of telematic-based demand responsive transport for urban passenger
transport. Crainic et al. (2009) review the fast developing field of intelligent freight
transportation systems. The authors point out that even though technology is widely
available, the upcoming challenge is to transform the accumulated data in useful in-
formation and exploit it properly.
2.3.6. Performance indicators for transportation networks

The above presentation of different aspects of optimal network design was oriented
to give an overview of the field. It mentioned performance indicators used in the
models of network design. Network KPIs are applied to measure the performance
of a transportation network from manifold perspectives. These are oftentimes cost-
oriented measures, reflecting the strong cost-orientation in the models for optimal
transportation network design. Additionally, the service-oriented aspects that have
been illustrated, are also accounted for in different KPIs. This section explains which
network KPIs will be of interest in the following chapters.
Performance measurement systems for SCs exist in the literature. Gunasekaran
et al. (2004) summarize performance indicators for SCs and suggest a framework for
an entire SC. For the sub-aspect of product delivery, the authors suggest strategic,
tactical, and operational metrics.49 The metrics reflect that the research orientation is
49 Strategic metrics are flexibility of service system to meet customer needs and effectiveness of enterprise
distribution planning schedule. On a tactical level, the authors put forth flexibility of service system to
meet customer needs, effectiveness of enterprise distribution planning schedule, effectiveness of deliv-
ery invoice methods, percentage of finished goods in transit, and delivery reliability performance. Fi-
nally, quality of delivered goods, on time delivery of goods, effectiveness of delivery invoice methods,
number of faultless delivery notes invoiced, percentage of urgent deliveries, information richness in car-
rying out delivery, and delivery reliability performance serve as operational metrics (Gunasekaran et al.,
2004, p. 345).
42
on production based SCs as opposed to measuring the performance of a transporta-

tion network. Nevertheless, two aspects are worth pointing out: the authors include
a notion of flexibility into their metrics and also account for reliable performance.
Chopra (2003, p. 124) highlights that the performance of a distribution network
must be measured based on the customer needs that are met and on the related cost.
The author compares different types of distribution networks amongst each other.
To measure customer needs, he focuses on response time, product variety, product
availability, customer experience, order visibility, and returnability. The cost perspec-
tive includes cost for inventory, transportation, facilities, handling, and information.
These performance measures cannot be directly applied for transportation networks.
However, it is interesting to note that the author actually puts service aspects before
cost aspects. This strengthens the claim for the importance of service aspects in net-
work design.
The classical models of optimal transportation network design strive for cost-
minimal networks; not surprisingly distance- and flow-oriented costs are important
here. Already mentioned above were the Steiner-Weber model, the WLP, and the
VRP. Melo et al. (2009) review facility location problems representing one aspect of
strategic network design. Inter alia, the authors come to the conclusion that 91% of
the papers in their review utilize cost minization or profit maximization objective
functions for their network design (Melo et al., 2009, p. 408); expressivly highlighting
that cost-oriented measurements persist as the major goal in network design. The
same holds true for SC design. Meixell and Gargeya (2005) review literature on
global SC design and find a strong cost focus well. “Even tough SC performance has
broadenend in scope, the research community in global SC modeling has not yet
given due attention to alternative objectives” (Meixell and Gargeya, 2005, p. 546).
Different aspects of how frequencies are applied in the models were also indicated.
Transportation time is another well-known network KPI (e.g. Powell and Sheffi
(1983), Jeong et al. (2007), and Gelareh et al. (2010)).
Other service or performance measures were incorporated into the mindset of net-
work design more recently. Network risk, tardiness, robustness, and flexibility were
amongst the more recent aspects of network performance to be measured by net-
work KPIs. Discussing the role of complexity and network robustness, Meepetchdee
and Shah (2007, p. 203) note that “linking logistical network complexity measurement
to a trade-off between efficiency and robustness is certainly of great potential to un-
veil insights in logistical network performance that have not been made so far”.
Many of the KPIs are related to the transportation network’s centrality. Six of the
above identified KPIs were chosen for the detailed discussion of their relations to net-
work centrality in chapter 3.50 The selection of the KPIs is based on two main ways
of thinking. Firstly, those KPIs were picked that are important for OR models as ele-
50 Additionally, the potential for consolidation is included as the seventh KPI for the analysis in chapter 3
due to its close connection to the other KPIs.
43
ments of objective functions or well-acknowledged constraints: cost, transportation

time and distance, and service frequency. Secondly, reliability, flexibility, and vulner-
ability were chosen for their prominent position as desirable performance indicators
of networks, especially in the more recent literature.51
2.4. Trade-off identification as an alternative perspective on

network design
The scope of methods for network design from the OR discipline was outlined above.
As shown, current literature contributions continue to refine existing methods, es-
pecially taking into account new business challenges and changes of perspectives in
network planning such as incorporating risk or flexibility.
This section points out fundamental points that hinder the application of OR so-
lutions in practice. Based on these challenges, it continues to illustrate an alternative
approach to network design: the identification of general trade-offs. This research
stream devises general guidelines on how certain decisions in transportation net-
works should be taken, depending on a given situation. Yet, the guidelines do not rely
on cost assumptions as much as the OR models presented in the previous section.
On the contrary, they are based on general insights in network design, highlighting
basic relations between network characteristics. This approach has proved to sup-
port empirical decision-making on transportation networks while overcoming some
of the empirical challenges of optimal network design. Chapter 3 addresses the role
of network centrality in the sense of providing comparable, but more general insights
for transportation network design.
2.4.1. Empirical challenges for optimal network design

The optimal OR models for transportation network design show drawbacks concern-
ing their application in empirical situations. OR models are applied in practice and
provide sound decision support. Nonetheless, certain challenges to the application
persist and will be outlined in the following.
Sequential process
Typically, the process of designing transportation networks is a sequential one. As
shown above, certain decisions arise at different time horizons for network planning.
These decisions typically involve financial investments for different time spans as
well. The planning process follows the structure as outlined: strategic decisions are
51 Service frequency could be part of either of the two groups. Then again, the categorization is not of
importance for the following, all selected KPIs are discussed in a similar manner.
44
2.4. Trade-off identification as an alternative perspective on network design
taken, tactical ones accept the strategic decisions as their input, and finally opera-
tional challenges will be solved.52
This approach is necessary due to the complexity of the problem to be solved. Fur-
thermore, information that is required for the operational planning will not be avail-
able on a long-term scale, thus, it cannot be used for the strategic planning. Neverthe-
less, by redrawing to a stepwise approach, one has to be aware that strategic decisions
will always limit the potential solutions for other decisions to be taken later. Even if
the question may be raised what an “optimal” solution for the entire system can be,
it is obvious that the sequential approach will most likely lead to suboptimal overall
decisions.
To find a trade-off between the infeasible overall perspective and the strive for very
narrow optimal sub-solutions, researchers are solving jointly some problems that are
likely to have an especially high interdependence. Inventory and transportation de-
cisions are often coupled. The economic order quantity (EOQ) model might be the
most prominent example that takes both aspects into account for identifying an op-
timal order quantity. Berman and Wang (2006) suggest formulations for choosing the
appropriate distribution strategy considering transportation and inventory cost. The
authors also review the pertinent literature in this field. Aykin (1995) is an exam-
ple for another type of studies, jointly solving hub location and routing problems.
Moreover, Nagy and Salhi (2007) review the field of location-routing problems that
has developed recently. Inventory-routing problems discuss the coordination of in-
ventory management and vehicle routing (Andersson et al., 2010). Ambrosino and
Grazia Scutellà (2005) suggest two models for distribution network planning by inte-
grating decisions on facility location, transportation, and inventory management.
Framing the problem

It has just been pointed out that the classical network design process will most likely
never lead to optimal decisions, no matter how sophisticated the underlying model
is. One obstacle is the complexity of the problem if one was to formulate an all-
embracing model. The complexity will lead to a large number of constraints to con-
sider. How far would the model need to be stretched? It will be necessary to limit
the scope of any model somehow, so even an overwhelmingly large model will always
52 An exemplary network planning process is described by Keaton (1993). He conducts a study on LTL
transportation network and presents the solution procedure that is employed to find networks. The pro-
cedure starts by determining the demand for transportation from a given dataset. Afterwards, terminal
decisions are fixed before load plans are generated (Keaton, 1993, p. 348).
Jansen et al. (2004) suggest a planning system that identifies a cost-efficient transportation plan. This
planning system is another example of a sequential planning process.
Barnhart et al. (2003a) address the specific case of aircraft and crew schedule planning faced by airlines
at a mid-term planning horizon. Even this sub-problem of network design must be split into smaller
problems due to its size and complexity in order to obtain “suboptimal, yet feasible aircraft and crew
plans” (Barnhart et al., 2003a, p. 369).
45
have limitations.53 Moreover, some input is always missing, the example that opera-
tional information is unavailable for long-term planning has just been stated above.
Manager acceptance
One highly relevant point mentioned in the literature is that highly sophisticated
modeling approaches and solution techniques may construct an artificial barrier
between analyst and decision-maker. Daganzo (2005, p. 8) refers to it as the “commu-
nication bridge”. It is furthermore stated that “practitioners do not seem to use the
research findings because practitioners consider most [OR] research to be irrelevant,
incomplete, narrow, trivial, and unrealistic” (Swamidass, 1991, p. 798). Similarly,
“managers are [...] unwilling to accept the results of the models, since they would
rather live with a problem which they understand than accept a solution which
they do not understand” (Meredith, 1993, p. 3). These approaches apply to network
design as they do apply to the entire field of OR. Some of this has changed over the
last decades.54
Gathering data
Large networks in practice typically show problems in data gathering accuracy. Even
though information communication technology (ICT) has evolved a long way over
the last decades, the gathering of data relies on the interplay of technology and hu-
man workers. Consider the example of gathering shipment data for transshipment
operations of LTL transports. The shipments are typically labeled with barcodes, so
the information gathering is a mere scanning process that will not be subject to errors
in data entry. Yet, it is not absolutely secured that a worker does not miss to scan a
single shipment. Furthermore, as the scanning takes place locally at terminals as the
shipments move through the network, synchronization between the scanners and the
local database is an issue, as well as between local databases and the central one. Sit-
uations with more manual operations are more prone to human error, no matter how
extensive the training.
Forecasting demand
Forecasting demand for transportation networks is difficult. Firstly, demand in trans-
portation is a derived demand (Rodrigue et al., 2009, p. 2), i.e. it is strongly depen-
53 The above mentioned location-routing problems (e.g. Nagy and Salhi (2007)) integrate two types of de-
cisions on networks. Ambrosino and Grazia Scutellà (2005) are an example of adding a third type of
decision. Manzini and Bindi (2009) go even beyond that by including decisions on facility location, pro-
duction planning, inventory management, and routing. Further extensions are imaginable. Yet, it should
be considered if this approach will eventually lead to improved decision-making.
54 Initiatives such as the one started by the INFORMS community (see www.scienceofbetter.org) are a visi-
ble sign that there still is a line between OR research and an overwhelming acceptance of it in the entire
industry.
46
dent upon environmental influences and seasonal trends. Secondly, even though
forecasts for the aggregated demand may be made up to a certain precision, distinct
OD-market forecasts are more difficult to make.55 Lastly, any forecast that is based
on data that is known to be prone to accuracy deficits is only reliable up to a certain
point. As forecasts tend to get worse the longer the forecasting period, good data for
strategic decisions is hard to obtain.
Assuming that current data is not available in absolute accuracy and that fore-
casts will at most provide probable pictures of reality the quest for optimal solutions
based on this data seems to be artificial. “Conventional modeling exercises are only
as convincing and effective as the assumptions on which they are based” (Eilon, 1999,
p. 155). The current literature supports this stream of thinking. Sensitivity analyses
are state-of-the-art in decision support and are taught to students in fundamental
business administration courses (e.g. Domschke and Scholl, 2000, p. 251).
2.4.2. Trade-offs in network design

Even though several shortcomings of the traditional OR modeling approach to strate-
gic network design exist, “the need for advice and assistance” (Eilon, 1999, p. 160) for
strategic decision-making remains. The broad scope of expectations towards strate-
gic decision on network design makes it “unrealistic to believe that a single formu-
lation, mathematical or otherwise, or a single procedure may encompass all relevant
elements, address all important issues, and fulfill all goals” (Crainic, 2003, p. 463). All
the same, some general guidelines may be expressed, covering the most relevant ele-
ments impacting the final solution and purposefully steering the network design pro-
cess. The identification of fundamental trade-offs in transportation network design
is one approach to evolve in that direction. It allows to support network design with-
out many detailed assumptions; hence it is worthwhile introducing the fundamental
contributions in the stream.
The notion that optimal planning is not the best approach in all situations, is not
new in the literature. Identifying general relations or basic trade-offs for decision-
making has been a relevant research focus in general and for transportation networks
in particular. These contributions are related to the strategic management in logistics
and tend to be of a lower mathematical sophistication than the above presented work
on network design. The obtained results are by no means of lesser importance.
Trade-offs between production and transportation were of interest to researchers
from early on. Karnani (1983) is a very good example for this; he strives for trade-offs
between production and transportation cost to find optimal plant sizes. Interestingly,
he is able to show that this decision is mainly driven by the degree to which economies
of scale, either in production or in distribution, can be achieved. With the dramatic
55 Ballou (1995) gives examples for the aggregation and disaggregation of information for inventory man-
agement; structurally similar arguments apply for the separating aggregated information to distinct OD-
markets.
47
rise in crude oil prices, the production-distribution-cost perspective regained atten-

tion lately.56
The trade-offs between transportation cost and inventory cost is the underlying
principle behind the EOQ model used in inventory modeling to find the optimal or-
der quantity (Winston, 2004, ch. 15). Transportation-related insights may as well
be gained; e.g. Burns et al. (1985) find that shipment lot size is at the root of the
inventory-transportation-cost trade-off for direct shipments in a one-to-many trans-
portation network. The authors furthermore compare direct and milk-run strategies
in one-to-many networks based on it. Daganzo and Newell (1985) take inventory
in transit explicitly into account for their analysis; these inventories inside travel-
ing vehicles are often neglected in the decision-making process. Blumenfeld et al.
(1985) determine optimal shipping strategies in many-to-many networks by consid-
ering transportation, inventory, and production set-up costs. Moreover, the authors
suggest an approach to decide when to transport through a transshipment terminal
and when to transport directly in a many-to-many transportation network.
Clarens and Hurdle (1975) and Banks et al. (1982) both aim at finding zones and
especially the size of zones where a transportation service takes place. Both contri-
butions aim at finding a good solution but, on the way to reach their goal, use the
more general and comprehensive parameter of the zone size and discuss its strate-
gic impact. Daganzo and Newell (1986) give guidelines for distribution network de-
sign based on influence areas.57 The authors find properties of good, or even near-
optimal, solutions that are able to steer the development of transportation networks
in a beneficial direction.
Another well-known example of contributions concerning strategic insights in net-
work design is Fisher’s fundamental contribution pondering the question What is the
right supply chain for your product? (Fisher, 1997).
The following paragraphs present insights from the literature concerning trans-
portation network design on aspects that were already touched on as optimal OR
models. It will be shown that general trade-offs are developed concerning the very
decisions. These do not rely on cost-assumptions and allow to structure the situation
clearly for decision-makers. This provides fundamental support for network design
decisions in practice.
Allocation strategies In any transportation network, the question arises how to

connect EoL terminals with transshipment terminals. In essence, this revisits the
WLP. Several strategies exist to achieve this goal. One option is to connect all EoL
terminals with all transshipment terminals in the network. As this is generally not fea-
sible at a reasonable cost, Hall (1987c) defines two general types of allocation strate-
gies, namely (1- and 2-stop) closest and best-nearby allocation. The closest allocation
56 See e.g. Simchi-Levi et al. (2008) or Butz et al. (2010) on this topic.
57 An influence area is the region served by one terminal. The idea is closely related to catchment areas.
48
will connect each EoL terminal to the closest transshipment terminal that is available
and route all shipments to and from that EoL terminal via this transshipment termi-
nal. Best-nearby routing transports each shipment “through the terminal that offers
the minimum travel distance” (Hall, 1987c, p. 421). Hall points out that this typically
implies making use of the three or four closest transshipment terminals per EoL ter-
minal. This number of terminals to choose from will allow to find the best-nearby
terminal for most transports.
The allocation strategies differ in their application. Best-nearby routing will lead
to the shortest transportation distances, as it always chooses to route traffic flows
through the terminal with the least detour. It will oftentimes result in more transport
connections in the network. Hall (1987c, p. 427) concludes that “best-nearby routing
is desirable when it is important to keep travel distance small, such as when origins
and destinations are spread far apart or when the flow of material between origins
and destinations is large”.58
Role of break-bulk terminals Daganzo (1987) touches on a similar topic when

he discusses the role of BB terminals in distribution networks. As opposed to Hall
(1987c), who assumes transshipment terminals to exist in the network, he questions
the issue of making use of transshipment terminals. He concludes that their usage
for consolidation purposes is only justified in many-to-many networks.59 The author
furthermore agrees with Hall (1987c) when he states that whenever transshipment
“terminals are used, items [...] should be directed to a terminal in their general direc-
tion of travel” (Daganzo, 1987, p. 553). This, firstly, implies that each EoL terminal is
connected to several transshipment terminals and, secondly, touches on the aspect
of best-nearby routing as described above. Finally, it is shown that only two parame-
ters are relevant for the decision whether to go directly, with one transshipment (later
referred to as 1-stop routings), or two transshipments (2-stop routings): namely the
number of EoL terminals60 and the potential for consolidation in the network. Hall
(1987c) and Daganzo (1987) agree that multiple allocation (several transshipment ter-
minals per EoL terminal) is superior to single allocation (one transshipment terminal
per EoL terminal) for typical many-to-many distribution networks.
58 This condition holds for the European LTL context. Interestingly, Hall (1987c, p. 427) for his part, con-
cludes that LTL carriers should make use of 2-stop closest routings as the number of origins and desti-
nations is large for these networks. He further explains that 2-stop routings are especially beneficial if
shipments are standardized and easily handled by automated equipment. This feature certainly does
not apply for LTL transports.
59 The implementation of transshipment terminals in many-to-one or one-to-many networks is driven by
service aspects but is not be related to consolidation effects.
60 Daganzo (1987) refers to nodes in the network instead of EoL terminals, yet, using the terminology of EoL
terminals is consistent with the remaining presented literature.
49
2.5. Enriching optimal network design with qualitative

aspects
In brief, this chapter showed that sophisticated network design is of high importance
for various industries, transportation being one of them. The network is one of the
main resources for transport providers to create a competitive advantage. In the fore-
front of the strategic importance of efficient transportation networks, the OR disci-
pline has developed and continues to develop models to find optimal transportation
networks. It is beyond the scope here to present the wide variety of models developed
for network design in various contexts. Still, the models may be broadly classified by
their planning time horizon. Presenting some fundamental models gave an impres-
sion of the bandwidth of research foci that are relevant for optimal transportation net-
work design. The models define optimality often based on a cost-perspective. Least
cost solutions are perceived as optimal networks.
Apart from the almost philosophical question of “what does optimality mean for
real life applications”, these models generate optimal or almost-optimal, good solu-
tions based on data, whose accuracy can often be questioned. Therefore, even the
best models cannot guarantee optimal solutions in practice. Service-oriented net-
work design aspects exist in the literature, yet they are smaller in numbers. The rele-
vant service-aspects for transportation networks were pointed out: various perspec-
tive of risks, such as reliability, robustness, and vulnerability are important. Network
flexibility is a desired network characteristic. Delays are to be limited. Green logistics
and green SCM approaches are evolving rapidly. Finally, new technologies and their
implementation for logistics and in particular transportation offers new opportuni-
ties for network design.
However, the practical application of optimal network design models is hindered,
creating a gap between research and its application. One approach to overcome
the research-practice gap is to search for strategic insights that can be taken from
the optimality network design models. General guidelines can often be developed
from the analytically described models that might generate more valuable insights
for decision-makers in practice than any optimal solution might be able to give. It
is essential for sound decision-making on transportation network design in practice
to be aware of the fundamental relations and trade-offs in network design. These
allow to incorporate different aspects of desired network performance into the
design process and structure the obtained solutions based on these. This permits
to enrich optimal network design. Supporting this point, Daganzo (2005, pp. 8-11)
emphasizes the surprising accuracy of results obtained by simple, strategic models of
transportation networks.
Interestingly, many of the strategic insights that have been indicated by researchers
in the literature are related to the centrality of transportation networks, even though
this very term is not used. Reference is rather made to transshipment or BB termi-
50
2.5. Enriching optimal network design with qualitative aspects
nals.61 When taking a closer look, the existence of a transshipment terminal does al-
ready hint at a certain degree of network centrality. The following chapter will address
the question how network centrality conceptually relates to transportation network
performance and may provide strategic guidance to network design.
61 The introduction of general trade-offs in network design in section 2.4.2 never used the term network
centrality. Even so, the next chapter will highlight that there is a clear relation to network centrality.
51
3. Transportation network centrality
Most people have talked to call center agents before.

Computers need a central processing unit.
The center usually is the tallest player on a basketball team.
The Central Intelligence Agency is headquartered in Langley, Virginia.
The term centrality is broadly used in every-day life. Network centrality is at the fo-
cus of this thesis, and this chapter introduces the concept. It firstly presents node and
network centrality and provides a definition of network centrality. Node centrality is
an important foundation for network centrality and, thus, supports further discus-
sion on it. The next section then derives the relation between transportation network
centrality and network performance (respectively the related key performance indi-
cators (KPIs)). The subsequent reasoning concerning the application of the concept
of network centrality for transportation network design relies on the derived relations
between network centrality and the performance of transportation networks. Hav-
ing this in mind, the chapter continues to present how network centrality has been
measured in the literature so far; main contributions from the field of social sciences
are presented to provide the relevant background. Furthermore, it is shown in how
far transportation research has applied metrics of network centrality. Air transport
is a relevant example for measuring the centrality of transportation networks. The
methods applied and the results obtained for this context are presented.
3.1. Network topology and concentration as dimensions of

network centrality
The concept of network centrality may be understood in an organizational sense as
well as in the meaning of the spatial appearance of the transportation network. The
following refers to network centrality as the description of network appearance. The
challenges in finding a suitable organizational structure to manage the transportation
network is not approached here.
Centrality refers to different elements of a network or the entire network itself. On
the one hand, the question might be to identify the most central node or the most
central edge in a network. “The centrality of a node in a network is a measure of
the structural importance of the node” (Alderighi et al., 2007, p. 534). On the other
hand, one can not only look at the centrality of nodes or edges in the network, but
also on the centrality of entire networks and compare them to each other. Networks

53
with few central nodes tend to be central networks. In that sense, network centrality
depicts the appearance of a network. Eventually, network centrality is the perspective
of interest here. To reach this point of view, the next section will exhibit some node
centrality measures that lead to metrics for network centrality.
Centrality is especially relevant in many-to-many networks as these can assume all
degrees of centrality. One-to-many or many-to-one networks can never be totally de-
central; one highly central node will always be in the network.62 The following presen-
tation and discussion of centrality will be conducted for the context of many-to-many
networks.
Defining a common understanding of centrality is difficult; and most researchers
will agree that there is no such common definition (e.g. Freeman (1979, p. 217),
Koschützki et al. (2005a, p. 17), or Borgatti and Everett (2006, p. 467)). Borgatti and
Everett (2006) classify centrality measures and conclude that all measures point at
the involvement of a node in the network. The outstanding role of certain nodes will
also be reflected in the discussion throughout this chapter.
Transportation networks are made up of nodes that represent important locations
(e.g. airports, cities, harbors, etc.). The nodes are interconnected by network links.
Consequently, the understanding of network centrality in the following is also ori-
ented to nodes.
Two main aspects of network centrality are discussed in the literature: network
topology and network concentration. The former describes how the transportation
network is set up in terms of links and nodes; it focuses on the “spatial layout of the
network” (Scott et al., 2006, 216). “Topological measures exploit static network struc-
ture” (Wojahn, 2001, p. 38). The latter - measurements for the concentration of the
network - are used to assess how strongly the network’s operations are focused on
single or few hubs. These measurements rely on traffic flows in the network.
From a topological point of view, a network shall be defined as being central, if
it takes the form of a star-shaped network. A topologically central network is a net-
work where one node is connected to all other nodes, while each other node does
not have any connection to any other node but to the one in the center. A network
is topologically decentral if it forms a circuit where all nodes are connected to two
other nodes. Figure 3.1 illustrates schematically the scale of network topology from
decentral to central networks. Different degrees of topological centrality in-between
the examples in figure 3.1 exist. This graph-theoretic understanding of centrality has
been present in the academic literature from early on. Bavelas (1948, 1950), Leavitt
(1951), Sabidussi (1966), and Freeman (1979) are early advocates of this point of view.
Circuit-type networks are rare for the context of transportation networks. Fully con-
62 One-to-many and many-to-many networks were introduced in section 2.2.1. As the name describes,
one-to-many networks describe the situation where one origin sends shipments to many destinations
(vice versa for many-to-one networks). The origin (respectively the destination) will assume a central
position in the network.
54
3.1. Network topology and concentration as dimensions of network centrality
circuit fully-meshed wheel star
decentral Network topology central
Figure 3.1.: Topological centrality scale of networks.
(a) (b)
Figure 3.2.: Two networks of identical topological centrality.
nected (or fully-meshed) point-to-point (P2P) networks are often seen as the most
topologically decentral transportation network.
Figure 3.2 depicts two exemplary networks of identical topological centrality. Even
though these two networks’ nodes differ in their geographic positions, the nodes are
connected in the same manner and both networks are, in their layout, similar to a star.
Thus, for a fundamental understanding, topological centrality describes how the links
in a network interconnect the nodes. Nonetheless, a direct link to the geographical
position of the nodes and their connections does not necessarily exist.
The second aspect concerning network centrality is spatial network concentra-
tion.63 Network concentration describes how strongly traffic flows in the network are
gathered at certain network nodes. The networks in figure 3.1 do not show actual
63 The concept of network concentration has a spatial and a temporal dimension. The spatial dimension
of network concentration applies in the following. The temporal dimension of network concentration
refers to that transportation networks organize the arrivals and departures at a terminal in a certain way
to allow for transshipment. Airline networks operate this system at the hubs. Burghouwt and de Wit
(2005) define what they refer to as the temporal configuration of an airline network as “the organization
of the airline flight schedule at an airline’s station resulting in a given number and quality of indirect con-
nections offered through that airline’s station. The number and quality of indirect connections through
the station can be enhanced by concentrating the flight schedule in time; i.e. by adopting a wave-system
structure in the airline flight schedule” (Burghouwt and de Wit, 2005, p. 186). Similar coordination is
common for other transportation networks that rely on transshipment operations, such as less-than-
truckload (LTL) networks. Hence, a transportation network is temporally concentrated when arrivals
and departures take place during short time intervals and provide short transition times. The spatial
dimension of network concentration describes the appearance of a transportation network. Whenever
55
(a) Concentrated (b) Mixed (c) Dispersed
Figure 3.3.: Topologically central networks differing in concentration.
traffic flows; but when taking these into account, different situations may be distin-
guished; three cases are sketched in figure 3.3. In case 3.3(a) all shipments are sent
from the middle node to the outlying nodes. Case 3.3(b) depicts that the central node
acts as a transshipment point for all the shipments in the sample; yet, no shipments
are consumed in the central node. The third case 3.3(c) illustrates the situation that
all outlying nodes send shipments to the middle node where they are consumed.
Let traffic be measured by the number of departing shipments for the example in
figure 3.3. Under this assumption, network 3.3(a) is the most concentrated network,
as all the traffic occurs at the central node; it is the only node in the network with
departing shipments. Network 3.3(c) shows the lowest degree of concentration, it is
the most dispersed network: all outlying nodes have an equal number of departing
shipments while the central node has no departing traffic. Finally, 3.3(b) is an inter-
mediate situation. All nodes, including the one in the center have a share in the overall
traffic of the network: the outer nodes each send off a shipment, the central node acts
as a transshipment node and sends off all shipments after the transshipment.
From the flow-oriented understanding of centrality - network concentration - the
utmost concentrated network is a network where all flow exists only at one single
node. This definition relies on measuring traffic flows. How flow is accounted for will
differ by network and by the given context.64 This flow-oriented perspective com-
pares traffic shares in the network. It is closely related to measuring the concentra-
tion in other contexts, e.g. income shares (poverty distribution in a society) or market
shares in an industry (Reynolds-Feighan, 1998, p. 241).
Together, topology and concentration form a broad picture of network centrality.
Any single perspective will not give an overall picture of the network appearance.65
the term concentration is used without specifying the dimension, it is understood in the sense of spa-
tial network concentration. There are some interfaces between temporal and spatial concentration in
transportation networks.
64 Answering the question which network in figure 3.3 is more concentrated will depend on the measure-
ment of traffic, too. If traffic in the context was measured differently, for example in terms of arriving
shipments, the outcome of the example would vary accordingly, it would be the opposite compared to
the described example.
65 The distinction between network topology on one hand and network concentration on the other is not
consistently used in the airline literature, e.g. Reggiani et al. (2009, p. 258) state the goal of their paper
as “... to examine how concentration measures can point to the different network topologies ...”. The
56
3.1. Network topology and concentration as dimensions of network centrality
central
Regional LTL Letter mail

network network
Nettwork topology
y
Boulevard
l d
Périphérique, Paris German road
/ Circle Line, network
London
decentral
low Network concentration high
Figure 3.4.: Typology of network centrality.
In sum, the observation by Borgatti and Everett (2006) suits the above indicated un-
derstanding of centrality well: from a topological as well as from a flow-oriented per-
spective, a central network is one where one node is outstandingly involved in the
network.
A transportation network is central if it is both, topologically central and concen-
trated. As already mentioned in section 2.2.1, a hub-&-spoke (H&S) network is a cen-
tral network because it forms a star and has outstandingly high traffic shares at the
node in the middle. A decentral transportation network is topologically decentral and
lowly concentrated. Many different degrees of centrality are situated in-between the
two extremes.
Figure 3.4 contrasts topological centrality and concentration to firstly illustrate the
understanding of network centrality as constructed from concentration and network
topology and secondly to classify some examples of networks.
Illustrating examples for all combinations of topological centrality and concentra-
tions can be found and are mentioned in figure 3.4. Transportation networks with a
low concentration and a low topological centrality form a circuit and have similar traf-
fic shares at all nodes on the ring. The freeway ring surrounding Paris, the Boulevard
Péripérique, or the Circle Line as part of the London Underground network are exam-
ples for this type of transportation network when they are considered independently
from the remaining network. A topologically decentral network with a high network
concentration is the German road network. Traffic is concentrated around the large
authors draw a direct link between the two terms. Alternatively, Scholz (2009) states that his work applies
two groups of indices, “concentration and centrality measures” (Scholz, 2009, p. 4). He sees the two con-
cepts as independent. The applied centrality metric by Scholz (2009) is Freeman’s network betweenness
centrality, which will be classified as a metric for network topology in the following.
57
central
Nettwork topology
y
decentral
low Network concentration high
Figure 3.5.: Network centrality.
cities, and the appearance of the network is highly interconnected (admittedly not cir-
cular) and far from forming a star-shaped network.66 A regional LTL network is often
structured as a P2P network, interconnecting many LTL end-of-line (EoL) terminals
directly. The networks usually have one transshipment terminal where all overflow
shipments are sent to. This terminal receives relatively small shares of shipment traf-
fic, but it forms the center of a star-shaped network. Lastly, a central network with
high topological centrality and concentrated traffic flows is e.g. a letter mail network,
commonly operated by courier, express, parcel (CEP) providers. All letters are brought
to a central sorting facility where they are sorted and sent out to the delivery regions;
an example of a H&S network.
Summing up, network centrality is a concept that allows to describe the appear-
ance of a network. This is achieved by two subordinate aspects: topological network
centrality and network concentration. Figure 3.5 summarizes the elements of the
concept of network centrality. Network topology refers to the layout of the network in
terms of nodes and links. Network concentration focuses on grasping the way traffic
flows through the network. Together, topological centrality and concentration form
a consistent picture of network centrality. A network is central if it is topologically
central and concentrated.
The subsequent presentation of the relevance of centrality for network design will
refer to network centrality in its two subordinate aspects. It will eventually be shown
that network concentration is of particular importance for the analysis of transporta-
tion networks. The relevance of hubbing in a transportation network is important for
66 The French road or rail network however, would actually form a star-shaped network with Paris as its
center.
58
3.2. Impact of network centrality on network performance
its analysis. This thinking is closely coupled with network concentration as it is im-
possible to assess hubbing without the notion of flows. Aggregate information in the
sense of one single value per network is necessary for network comparisons.

The following section is devoted to presenting how network centrality is conceptu-
ally related to typical network KPIs.67 It will be shown that centrality is at the root of
network performance measured by the KPIs. Therefore, it is possible to distinguish
networks by their centrality. This, further, highlights that some facets of performance
of a many-to-many network are mutually exclusive. These well-known trade-offs in
transportation networks are actually interlinked by network centrality. Changing the
centrality of the network will impact the KPIs and result in a transition to a different
network.
Chapter 2 extensively introduced different Operational Research (OR) models on
transportation network design. Section 2.3.6 summarized the KPIs that were identi-
fied as important for network design because they are either used as relevant vari-
ables in the objective functions of the models (especially the cost-oriented measures)
or are used as palpable descriptions for network performance (oftentimes the service-
oriented measures). Many of these KPIs make use of different consolidation strategies
as drivers of efficiency. Therefore, the potential for consolidation as related to differ-
ent network appearances is included in the discussion even though it was not identi-
fied as a KPI previously. Nevertheless, the discussion of the relations in the following
will benefit from its inclusion. Thus, the following will discuss the relations between
network centrality and seven relevant indicators of network performance:
• Potential for consolidation
• Link frequency
• Transportation distance and time
• Schedule reliability (focusing on congestion and late arrivals)
• Network vulnerability
• System capacity flexibility
• Cost
67 The idea that the centrality of a network influences its characteristic performance in the context of social
groups was the target of the seminal research conducted by Bavelas (1950) and Leavitt (1951). Admit-
tedly far stretched, the intention to relate transportation network centrality to transportation network
performance is the same thinking.
59
The basic insight that each single indicator is related to network performance is for
most cases not fundamentally new knowledge. All the same, the embracing charac-
teristic of network centrality, covering all of the performance indicators simultane-
ously, is what will pre-structure network design decisions.
It is always assumed in the following that the described networks operate fairly
efficient for their context. That is to say that e.g. the discussed networks will not use
obviously inefficient network links, nor will they be planned for surprisingly low load
factors of vehicles. Which network type is efficient depends on the environmental
setting, as well as the requirements, and constraints of the situation.
The presented comparisons are always understood in the sense of comparing two
transportation networks in the same environmental setting that operate to serve com-
parable levels of demand. It is certainly true that the below discussed indicators of
network performance will also react e.g. to changes in demand.68 Yet, this is explic-
itly not the focus of the discussion below. For the coherence of the arguments, the
only difference between the networks is their degree of centrality, all other things - in
particular the demand for transportation - being equal.
3.2.1. Consolidation
Hall (1987a) is one of the early authors to explicitly address different options to con-
solidate shipments to a full vehicle load. He defines consolidation to be “the process
of combining different items, produced and used at different locations and different
times, into single vehicle loads” (Hall, 1987a, p. 57). His paper presents three consoli-
dation strategies that are depicted in figure 3.6 and briefly introduced in the following.
Inventory consolidation (a) might be the most intuitive form of consolidation that
one can think about: items are accumulated in storage until the necessary vol-
ume is gathered to fill up the vehicle load. This approach is a consolidation over
time. A real-life example is an airport terminal shuttle bus that waits to have a
certain number of passengers on board before departing.
The basic trade-off in inventory consolidation is between transportation cost
and inventory holding cost. The more items may be accumulated to be loaded
onto a single vehicle, the less costly the transport per item will be.69 Then again,
the longer the items are stored to be loaded onto a vehicle, the higher the in-
ventory holding cost for these items will be (Hall, 1987a, p. 61). This trade-off is
an economic order quantity (EOQ) model.
Vehicle consolidation (b) is a milk-run type of consolidation. Transport vehicles

will stop at several origins and destinations to either drop off or collect items
68 For example, comparing a network with a high demand for transportation with a low demand network
will typically reveal that frequencies are higher in the high demand network, no matter what the central-
ity.
69 It is implicitly assumed that the vehicle has not yet reached its limiting loading capacity.
60
(a) Inventory consolidation
(b) Vehicle consolidation
(c) Terminal consolidation
Figure 3.6.: Consolidation strategies (Hall, 1987a, p. 59)
to fill up the vehicle load. In that sense, vehicle consolidation is a consolidation

strategy in space (Hall, 1987a, p. 58). Krause (2007, pp. 85-92) demonstrates
that vehicle consolidation is of ongoing importance and raises strategic chal-
lenges for logistics service providers (LSPs) especially for optimizing part load
(PTL) operations.
The key triggers for vehicle consolidation are the time elapsed between two
consecutive vehicles and the number of stops on a route. Hall (1987a, p. 63)
points out how these effects can be modeled in a modified EOQ formulation.
The basic idea is that the longer the time between two consecutive vehicle de-
partures and the more stops are on one route, the higher the potential to find
additional items to consolidate on the vehicle. However, long times between
departures result in longer inventory holding times and more stops on a route
increase the time goods spend in transit and might further lead to an increased
transportation distance. Stopping itself is typically related to additional cost as
well.
Terminal consolidation (c) relies on efficient transshipment operations. Items

from many origins are brought in full vehicle loads to a terminal where they
are sorted and exchanged to leave the terminal to different destinations again
in full vehicle loads. Hall (1987a, p. 66) distinguishes between one-terminal
(1-stop) and two-terminal (2-stop) routing. In one-terminal routing the
61
shipments are transshipped only once on their way from origin to destination,
as for example in a pure H&S network. Two-terminal routings have shipments
transshipped twice on their way. This yields the additional benefit that on the
leg between the two transshipment terminals extremely high utilization rates
can be achieved.
Consolidation strategies are closely related to centrality. Inventory consolidation

fills the vehicle capacity by waiting. This strategy does not require any other nodes in
the network but origin and destination. If milk-runs (vehicle consolidation) are con-
ducted, the vehicle stops at several nodes to pick up or drop off shipments. These two
consolidation strategies can be implemented efficiently in a topologically decentral
network and may be the strongest lever to create high vehicle utilizations in decentral
networks.70 Terminal consolidation relies on transshipment terminals. Depending
on the routing and the network appearance, the number of transshipment terminals
in the network may vary but terminal consolidation cannot be done in a topologically
decentral network. It will usually lead to an increase in network concentration, too.
It is not uncommon to find literature contributions on network design that will
favor central networks due to the efficiencies generated by terminal consolidation.
But this solution is usually obtained under the assumption that inventory consolida-
tion can hardly be exploited. For example, Barnhart et al. (2002) discuss a network
design for express shipment delivery. Under the assumption of a “single day plan-
ning horizon” (Barnhart et al., 2002, p. 244), the authors conclude that “a single hub
model of operation is more cost-effective [...] because a maximum degree of demand
consolidation is attained” (Barnhart et al., 2002, p. 259). Widening the perspective,
inventory consolidation is just as much a possibility to generate efficiencies given a
setting that tolerates waiting times to accumulated transportation volume. As signif-
icant parts of research on transportation network design originate in time-constraint
environments such as passenger air transport or the above mentioned express ship-
ment delivery, the contribution of inventory consolidation to create good transporta-
tion networks appears to be underrepresented. There is no advantageousness per se
of central transportation networks over decentral ones that is related to their consol-
idation potential. The relevant issue is to identify the correct source of consolidation
potential (inventory, terminal, or vehicle) for a given environmental setting and then
choose a suitable transportation network design.
3.2.2. Link frequency

Central networks tend to have a higher schedule frequency than decentral networks
(Basso and Jara-Díaz (2006b, p. 158), Mayer (2001, p. 23)). This is easily demonstrated
by taking a look at a topologically decentral P2P network and a central H&S network.
70 In fact, vehicle and inventory consolidation are feasible in all network types; their application in decen-
tral networks is especially beneficial.
62
Let both networks have the same number of nodes n. Then, it is directly seen that
the P2P network has n(n − 1) directed arcs in the network whereas the H&S network
only consists of 2(n −1) directed arcs. Let the demand for transportation between any
two nodes in the network be of x items per time unit. The transport vehicles have an
identical transport capacity of c a p units (integer c a p ). The traffic in the P2P network
will flow on any arc; for the sake of simplicity it is assumed that items are sent on the
direct arc to the destination. In the P2P network case, any arc will have a demand of
cap
x items per time unit, resulting in an accumulation time for a full vehicle of x time
units. In the H&S network case, each arc into the hub is used to serve the demand
from an origin to all other destinations in the network, thus x ∗ (n − 1) items per time
unit. The same holds true for all routes leaving the hub: the transport vehicle is loaded
with all the items sent from all other nodes to the destination which will be x ∗ (n − 1)
cap
as well. Therefore, a transport vehicle is filled up every x ∗(n −1) time units; more often
than in the P2P network case. Making the safe assumption that a transport vehicle
leaves when its capacity is reached, this yields a higher frequency for the central H&S
network.71 The increase in frequency for central networks is related to the interplay
of network topology and network concentration.
Brueckner (2004) presents a formal model to compare H&S networks and P2P net-
works in the airline sector analytically regarding their frequency. His model supports
the above presented argument and confirms that H&S networks in the airline indus-
try have a higher service frequency than P2P networks. Lederer and Nambimadom
(1998) study the advantages of different network appearances and come to the con-
clusion that “the optimal service frequency for direct service is lower than for [H&S
networks]” (Lederer and Nambimadom, 1998, p. 796). This, again, confirms the above
stated argument from a different perspective.
3.2.3. Transportation distance and time

Topologically central many-to-many networks directly imply the usage of one or sev-
eral transshipment terminals. As the routing to a transshipment terminal typically
induces a detour, the circuity in these networks is higher than in networks that offer
direct connections between the nodes. A fully-meshed network provides the lowest
circuity.72 This leads to a longer average transportation distance than in central net-
works.
71 The main point in the argument is derived from the higher demand that is accumulated in the H&S net-
work. The argument assumes a fixed capacity per transport vehicles. Admittedly, there are transportation
contexts where various vehicle sizes with different capacities are used, e.g. in passenger air transporta-
tion. In those contexts, an alternative to increasing the frequency of transport is also to use a vehicle of
larger capacity for the transport.
72 The case of a circular network, by definition even more topologically decentral than a fully-meshed net-
work, may lead to longer transportation distances than the fully-meshed network. The circuit lacks direct
connections between the nodes.
63
2a
a
(a) Distance 1.6a (b) Distance 1.4a
Figure 3.7.: Average distances of networks with different degrees of topological

centrality.
Figure 3.7 sketches a small example for the transportation distances in a star-
shaped and a fully-meshed network. Let the distance between the central node and

the outlying ones be a , the distance between outlying nodes is 2 a .73 It can then be
seen that the two networks differ in the average distance from one node to another:
the average transportation distance. In case (a), all transports are operated via the
central node, the average distance is of 1.6a . In contrast, in case (b), the trips may
additionally bypass the central node and thereby save detour and travel distance.
The average distance reduces to 1.4a .
Transportation time is correlated with the transportation distance (Hall, 1987a,
p. 60). It can be expected that topologically central networks will have longer
transportation times as well. It should be noted that the above mentioned frequency
of transportation also impacts the overall transportation time. Overall transportation
time denotes the time between a shipment is ready for departure and its final arrival
at the destination.74 It therefore includes the waiting time before the initial departure
as well as waiting times at transshipment terminals in addition to the transportation
time that is directly related to the transportation operations and the time spent on
transshipment activities. Low link frequency is likely to lead to higher waiting times
than high frequencies of service. Therefore, central networks will have rather low
waiting times as part of the overall transportation. The proportion of the waiting time
element to the operation-related elements of overall transportation time will finally
decide if central or decentral networks actually have a shorter overall transportation
time. A general indication is not feasible for this aspect.
73 The diagonal links in case (b) have a length of 2a , even though the drawing suggests a longer distance for
its illustrative purpose.
74 For passenger transportation, it is shown that passengers will adapt their arrival at a station to the
planned schedule if the intervals between consecutive departures exceed a certain threshold (Bowman
and Turnquist, 1981). The passengers try to minimize the waiting time that is included in the overall
transportation.
64
3.2.4. Schedule reliability

It is important to assess how well the preplanned schedules are met in regular op-
erations for scheduled transportation processes. Schedule reliability is therefore an
important measurement in transportation network design. Different terms exist in
the literature for schedule reliability, even though the pertinent “literature is a small
and evolving one” (Batley et al., 2011, p. 61). Batley et al. (2011) identify four major
terms related to it, namely reliability, punctuality, lateness, and delay. Lederer and
Nambimadom (1998, p. 798) use the term schedule reliability to describe the absence
of real delay for the passenger in an airline network. Hence, schedule reliability de-
notes the degree to which a transportation network is able to meet its schedule. Net-
work operators often include a certain amount of slack into the schedules to buffer
for uncertainty. In order to compare different network appearances concerning their
schedule reliability, it must be assumed that the included slack in the networks is of
the same level.75 In general, decentral networks are known to have higher schedule
reliability than central networks due to their low network concentration.
Concentrated networks rely on few important hubs for their operations. The mal-
functioning of a hub will impact large parts of the network, if not even its entirety.
For the case of airlines, Ball et al. (2007, p. 2) note that “any significant disturbance
at a hub, rapidly leads to disruptions of extensive parts of the carrier’s schedules”.
The hubs are the most critical nodes in central networks when considering schedule
reliability.76 Moreover, central networks have fewer but very important links than de-
central networks where the importance of links is spread more evenly. The analog
argument applies as for the hubs: if one of few links in the network is not operable, it
is far more expensive and complex to avoid that link; compared to the decentral case,
stronger impact on the schedules in terms of long delays is expected.
Late arrivals and congestion are two reasons for perturbations in transportation
networks. The following two sections will review their impact in order to support the
comparison of schedule reliability for central and decentral networks.
Impact of late arrivals

A central many-to-many network is efficient amongst others because it makes use of
transshipment operations, e.g. through terminal consolidation. The interchange of
75 In their analysis, Lederer and Nambimadom (1998, p. 792) find that direct networks include the lowest
slack in their schedules to buffer against uncertainty to limit the delay in the network to an acceptable
level. In the following, it is assumed that slack is comparable between the networks. This will lead to dif-
ferences in schedule reliability. The information that slack of different magnitude is included in different
networks actually is a sign for their difference in schedule reliability.
76 Some passenger airlines rely heavily on highly central transportation networks where hubs play a major
role in the operations. Perturbations at the hub will lead to a decrease in productivity and punctuality. To
overcome this, some airlines start “de-peak” their operations and carry out “random hubbing” (Franke,
2004, p. 20). The airlines reduce temporal traffic peaks that result from desired connectivity and high
traffic concentration at the hubs in the network which indicates how serious the challenge for these
transportation networks is.
65
goods or passengers is a crucial part of the production process.77 The transshipment

process itself creates an additional factor of risk in the case of late arrivals. Transship-
ment operations are organized so that arriving items are unloaded at the terminal,
resorted and quickly loaded to the transport vehicles so that they will leave as soon
as possible.78 More arriving transport vehicles are beneficial for the network’s oper-
ations as long as the terminal does not become congested. It is the latest vehicle to
arrive that will impact all the departing vehicles. In the worst case, items on that lat-
est vehicle will be transshipped to the vehicle that is scheduled to leave the earliest.
There is a trade-off between a long time window of arriving vehicles (more transport
objects to be transshipped) and the aim for departing vehicles to leave the terminal
as early as possible (to speed up the transport).79
The situation becomes difficult as soon as an incoming vehicle is delayed for some
reason. The operational planner will have to decide on the spot whether or not to
delay all departing vehicles in order to wait for the late one to arrive or whether to
have them leave on schedule and hinder the transshipment operations of the items on
the late vehicle. When the departing vehicles are held back, the delay of the arriving
vehicle is propagated into the network. In the freight transport business, this is solved
with the help of cut-off times. These are general rules and leave no space for ad hoc
decisions. Only if vehicles arrive before a certain cut-off time, the transshipment will
be guaranteed. This is a pragmatic rule to guarantee the viability of the network.
Lowly concentrated networks do not have as many transshipment operations at
terminals as concentrated networks do. The impact of late arrivals in terms of propa-
gation to the network is much smaller in decentral than in central networks.
Thus, late arrivals are a more relevant issue in central than in decentral networks.
It may manifest itself in different ways:
1. If the planner adds additional slack into the network, it will be less challenging
for the vehicles to arrive on schedule.80 Additional slack will, however, increase
the time items or passengers spend in the network and is usually valued nega-
tively.
2. Ad hoc recovery plans for late arrival situations are sometimes executed. Anec-
dotal evidence for this situation are taxis that transport train passengers once
they missed the last connecting train at night. Along the same line, LSPs report
on situations where delayed shipments were loaded on a van to catch up with
a truck that already left the terminal. These plans will serve their purpose, but
are obviously costly for the network operator.
77 Section 2.3 gives an idea of the complex planning processes in interchanging goods or people at termi-
nals.
78 The same concept is valid for passenger transport: passengers disembark at the terminal, walk to their
connecting vehicle and board it.
79 This trade-off has been approached mathematically with the goal to find an optimal solution (Kara and
Tansel, 2001, Wagner, 2004).
80 Lederer and Nambimadom (1998) note that this practice is common for airline network planning.
66
3. It is even an option to allow late arrivals to occur and propagate the delay onto
the departing vehicles. There will be situations where this is operationally ben-
eficial. The drawback in this situation will be the impact on schedule reliability
which is usually valued negatively by shippers and passengers (e.g. Batley et al.
(2011), Fowkes et al. (2004)).
Congestion
“Congestion occurs when transport demand exceeds transport supply in a specific
section of the transport system. Under such circumstances, each vehicle impairs
the mobility of others” (Rodrigue et al., 2009, p. 248). Congestion may arise on net-
work links (e.g. traffic jams on highways) as well as in nodes (e.g. congested airports).
Congestion is yet another reason for network perturbation. As Flores-Fillol (2010,
p. 358) puts it “local problems may have a general impact because they are trans-
mitted throughout the network, so that problems affecting hubs become especially
serious”. Congestion is a challenge for today’s airline networks. The concentration
on certain airports that results from using H&S network in passenger air transport
leads to capacity constraints. Unconstrained major airports are an exception (Wo-
jahn, 2001, p. 68).
Grove and O’Kelly (1986) address the question how network appearance and
schedule delays are related in airline networks. The authors conclude that the more
connecting flights are served at a terminal, the more likely congestion will occur and
the longer the delay of the passengers. The more a network is dispersed, the less
delay occurs. The authors confirm by simulation that concentrated networks run a
higher risk of congestion at the hubs then lowly concentrated networks.
Central networks are more exposed to transmitting local problems at topologically
central nodes or arcs through the network, just as it was pointed out for propagating
delays in the case of late arrivals. Again, if the local problem congestion appears in a
network, its impact will be much higher in central networks as the problem is spread
faster and more difficult to overcome than in decentral networks.
What is important when considering congestion is that the probability of conges-
tion to appear in the network is influenced by the centrality of the network itself, more
explicitly by the network concentration. Whilst late arrivals are a challenge for all net-
works and only the propagation from central nodes is critical for the network, con-
gestion itself is more likely in highly concentrated networks. The additional challenge
of propagating the perturbation only complicates the situation.
3.2.5. Network vulnerability

Node and network vulnerability and reliability are two different but connected con-
cepts. Transportation network reliability has been studied intensively. One com-
monly distinguishes connectivity reliability, travel time reliability, and capacity reli-
67
ability. Connectivity reliability is the probability that a pair of nodes remains con-
nected if links in the network break down whereas travel time reliability refers to the
probability to keep a certain travel time from origin to destination. This was referred
to as schedule reliability above. Capacity reliability describes the probability that a
network can provide a certain capacity (Taylor and D’Este, 2007, pp. 11-12). In all
cases, reliability stands for the probability that a system can provide a somehow ex-
pected service. It usually expresses the percentage of time that the network is able to
do so. High network reliability is a hygiene factor for any transportation network and
is not impacted by its degree of centrality.
Network vulnerability as opposed to reliability is especially important for “large-
scale, sparse regional or national networks” (Taylor and D’Este, 2007, p. 13). Vulner-
ability describes the consequences of a node or link failure, and not the probability
of its occurrence. It expresses the weakness of a network (Taylor and D’Este, 2007,
p. 13).81 Taylor and D’Este (2007) define node and link vulnerability based on how
strongly the accessibility of a link or node depends on the expected performance of
few other links in the network. Their research focus is on national road infrastructure
vulnerability. In order to extend this thinking from the background of physical road
networks to transportation service networks, the relevant role of terminals needs to
be incorporated. When looking at national road infrastructure researchers usually
identify cities as the nodes of the network and the goal of the network is to connect
the cities. Transportation networks, on the contrary, have nodes that are required for
the viability of the network as they connect the links in the network. In this light, a
transportation network is vulnerable if the (quasi) loss of a few nodes or links in the
network diminishes the viability of the network substantially.82
Central networks are more vulnerable than decentral networks; the argument is
structurally similar to the one for congestion of central networks. This is directly re-
lated to the prominent position of the central nodes. Central nodes are important in
terms of connecting the network due their topological centrality. Moreover, the net-
work relies on them for transshipment operations. If one of them breaks down or is
congested, the entire network’s viability is diminished substantially. The impact is
81 The concept of network robustness is related to that of network vulnerability. “Robustness signifies that
the [network] will retain its [...] structure intact when exposed to perturbations” (Holmgren, 2007, pp. 33-
34). A vulnerable network is not robust. However, a network may be not vulnerable even if it is not robust
due to the fact that it can find a new stable position when facing perturbations. The network does not
retain its structure but will nevertheless remain viable (the network is resilient); see also Grubesic et al.
(2008). Extending this view, Klibi et al. (2010) review different understandings of robustness.
82 The idea that infrastructure can be vulnerable moved into focus more and more over the last years. The
United States of America (US) White House defined a list of eight sectors that are identified as critical
infrastructure, transportation being amongst them. The challenge to secure infrastructure is increased
as modern societies not only depend on these eight sectors of infrastructure but that these sectors are
closely interwoven: if telecommunication networks break down, banking and finance networks will be
impacted. Just the same, transportation networks rely on electrical power systems or gas and oil stor-
age to receive the required infrastructure. Moreover, if transportation networks break down, emergency
services will be disrupted and gas and oil storage is likely to be impacted as well (Murray and Grubesic,
2007).
68
larger than in decentral networks. Furthermore, as not only the nodes but also the
links that connect central nodes with other locations are highly utilized and impor-
tant for the viability of the network, they are another source for network vulnerability.
Links as important as these with as large an impact on the network do not exist in
lowly concentrated networks.
3.2.6. System capacity flexibility

Network flexibility can take many forms. Feitelson and Salomon (2000) see flexibil-
ity as mainly connected to the adaptation to changes in infrastructure of a network.
Ukkusuri and Patil (2009) refer to flexibility as the potential of a network to adapt
to new situations. The authors provide a taxonomy of transportation network flex-
ibility which is depicted in figure 3.8. They state that any network has an intrinsic
flexibility which relates to the “inherent characteristics of the system” (Ukkusuri and
Patil, 2009, p. 627) and an extrinsic flexibility that builds on the intrinsic one by pro-
viding resources and strategy for improving flexibility. Following their taxonomy, the
topology-related aspects of intrinsic and extrinsic flexibility are interesting. They refer
to the “flexibility offered by the structural arrangement of nodes and links” (Ukkusuri
and Patil, 2009, p. 627). The key question in the context here is to asses if certain net-
work appearances are more or less flexible to adapt to changes in demand as this is
a day to day challenge in transportation networks. This is known as system capac-
ity flexibility, defined as “the ability of a transport system to accommodate variations
or changes in traffic demand while maintaining a satisfactory level of performance”
(Morlok and Chang, 2004, p. 406).
Do central or decentral networks have a higher system capacity flexibility? The
special case of how airlines will structure their networks when facing demand un-
certainty was studied econometrically (Barla and Constantatos, 2000, 2005). The au-
thors show that central H&S networks provide “increased flexibility in adjusting their
capacity allocation across markets as new information about demand conditions be-
comes available” (Barla and Constantatos, 2000, p. 180).
This conclusion is valid for other transportation networks as well and the under-
lying reasoning supports that. Let a topologically central H&S network and a topo-
logically decentral P2P network be trimmed to the same origin-destination (OD)-
demand, they both operate their vehicles at reasonable load factors for each situa-
tion. Vehicles in the H&S network case operate at higher load factors than in the P2P
network case.83 This also implies that the H&S network is able to consolidate flows
and operates higher frequencies on the links to and from the hub as presented above.
The P2P network, in comparison, will operate more direct OD-connections at rather
83 See section 5.3.1 for details on this reasoning.
69
Transportation Network
Fl ibilit
Flexibility
Intrinsic Flexibility Extrinsic Flexibility
Operations Resource
User Planner Strategy Dominant
Manager Dominant
Access Link Node degree User Planner Operations Manager

Temporal
p Node Topology
p gy
Mode Temporal
Day-to-day
Divisibility Survivability
Usability Technology functionality
Recourse
Route Topology
Frequency Traffic
Management Disjoined
Access Technology paths
Route
Reserve
Capacity
C it
Figure 3.8.: Taxonomy of transportation network flexibility (Ukkusuri and Patil, 2009,
p. 627).
low frequencies.84 The H&S network will be more concentrated than the P2P network.
Three different types of changes in demand can be distinguished:
General increase in demand. When demand rises, both networks will start to fill
up the already existing transport capacities. In both network types, load fac-
tors will rise. However, the P2P network was operating with lower load factors
from the start. So, there is more potential for filling up remaining empty ca-
pacity. Up to a certain level of increase, the P2P network swallows the increase
more easily. As this border is crossed, it will be easier for the H&S network to
buffer the increase in demand as consolidation helps here. A further increase in
OD-demand requires the provision of new transport vehicles in both networks.
More of these vehicles will be needed for the P2P network as (in worst case) one
vehicle per link has to be added. This is not only an increase in cost for trans-
port capacity but also in administrative complexity to find and purchase that
transport capacity. The consolidation in the H&S network allows to add fewer
vehicles and fill the added capacity more easily. Thus, for higher increases in
demand, the central H&S network arguably has a higher system capacity flexi-
bility.
General decrease in demand. The opposite case is a decrease in demand on all OD-
markets. The decrease of demand in the P2P network case will directly result in
84 It is a safe assumption that changes in demand will rather quickly generate changes in the network pat-
tern. The discussed reasoning does not incorporate this change in demand pattern as it would be a
second step adaptation to the network triggered by potentially identified decreases in efficiency.
70
lower load factors, but no reasonable adaptation in transport capacity is possi-

ble as all OD-links need to be served (assuming some demand remains for the
market). In the H&S network, there will also be a decrease in required capacity.
This can be done rather easily by decreasing the frequency on the links to and
from the hub. This might not be possible in all situations but the potential is
higher in a H&S network than in a P2P network.
Shift in demand structure. The third characteristic situation is a shift in demand

structure, i.e. the increase of demand on some OD-markets and the decrease
on others. This situation will result in the challenges mentioned above for the
P2P network as the OD-connections are independent in this network and the
decrease and increase in capacity confront the network with the same issues as
in the general increase and decrease situations. In comparison, the H&S net-
work is much better off. In the worst case it will find itself in one of the situa-
tions above, but more likely, the decrease and increase happens on markets that
are consolidated on the same link to or from the hub. In this case, the network
would not even have to deal with any significant changes in required capacity
as the changes level out.
Summing up, the systems capacity flexibility is related to the fixed capacity of trans-
portation vehicles and the therein originating jump-fixed cost. The increase in cost
and capacity is the same for the vehicles on the direct OD-connections and indirect
connections to the transshipment terminals. Accordingly, whenever an increase or
decrease in capacity results in a jump in terms of capacity, the H&S network is better
off.
Based on the reasoning just expressed, in general, a central transportation network
is more likely to have a higher system capacity flexibility because in all cases it is able
to accommodate the changes in demand more easily without harming the perfor-
mance of the system.
3.2.7. Cost
It is impossible to draw general insights into what type of network is more costly than
another. Finding cost optimal networks is an important aspect of OR modeling ap-
proaches that were discussed in section 2.3. If it were as easy as to say this type of net-
work is always cheaper than the other ones, OR network models would hardly have a
justification for existence. It is however possible to identify the cost categories that are
more or less important in central respectively decentral network operations. Typical
cost categories include fixed costs related to infrastructure and variable cost related
to operations (Rodrigue et al., 2009, p. 97).85 Table 3.1 summarizes some of the ex-
pressions of these categories in transportation network design. The general idea is
85 Daganzo (2005, ch. 2) discusses extensively different costs arising in logistics operations.
71
cost central decentral

fixed cost elements
administrative terminal costs - +
company-related administrative cost = =
sales-related cost = =
variable cost elements
distance-related transportation cost + -
transshipment cost + -
unused capacity cost - +
Table 3.1.: Relevance of cost categories associated with network types.
that central networks are able to transport full vehicles to central nodes. This implies
a detour that increases the transportation distance. At the same time, the transship-
ment operations at the topologically central nodes induce a cost. Decentral networks
will have less detour due to more direct connections between origins and destina-
tions, but in general cannot achieve load factors as high as the feeder lines to the
central nodes. Moreover, as the central nodes of the network are rather important, it
is likely to have more administrative and overhead costs associated with these.
The appearance of transportation networks depending on transportation cost has
also been studied econometrically. Flores-Fillol (2009) finds that airlines tend to
structure their networks as H&S networks when transport cost (in forms of marginal
cost per seat) is high, while low transport cost will lead to P2P networks. This finding
unfortunately does not give an information on the general (dis-) advantage of one
network type in terms of overall cost.
3.2.8. Summarizing network centrality and its relation to network

performance
Six elements of network performance in transportation operations were shown to be
related to network centrality. Table 3.2 summarizes these relations. The network per-
formance indicators above are of different kinds: the more cost-oriented, classical
measures such as transportation cost, distance, and time as opposed to the rather
service-oriented measures for e.g. flexibility and vulnerability. Both kinds are linked
to the concept of network centrality. The discussion of optimal network design in
chapter 2 has already concluded that elements of network performance such as reli-
ability and flexibility gain in importance in current network design research.
Even when having in mind that the relation between network performance and
network centrality is always only a tendency, the exceptional role of network centrality
that allows to connect the different elements of network performance is outstanding.
72
Performance Central Decentral

Potential for consolidation
inventory consolidation low high
vehicle consolidation low high
terminal consolidation high infeasible
Link frequency high low
Transportation distance long short
Schedule reliability
Impact of late arrivals high relevance low relevance
Impact of congestion high low
Vulnerability high low
System capacity flexibility high low
Table 3.2.: Characteristic network performance depending on network centrality.
Firstly, it pinpoints known trade-offs in transportation network design.86 Secondly -

and probably most importantly - the qualitative understanding of network central-
ity allows to draw conclusions about the network performance. It goes beyond the
single elements of network performance by encompassing and structurally aggregat-
ing them on a higher level of perception. This knowledge can be used to successfully
guide the network design. From early on, some possible network configurations can
be excluded from further investigation if the early centrality analysis reveals that these
networks will not be acceptable later on, due to a certain performance (e.g. network
vulnerability being to high). Knowledge of the importance of centrality can influence
the network design procedure. If the network is to have certain characteristics, then
the planner will know early on whether to strive for more or less central networks in
the design phase.87 In that sense, the knowledge of network centrality and its rela-
tion to network performance may pre-structure network design decisions. It allows
to eliminate network appearances that will not fulfill the expressed demand for the
network design. Potentially beneficial design options may then be granted more time
and resources for their in-depth analysis and comparison.
In addition to the qualitative understanding of transportation network centrality
that has been presented this far, a quantitative approach towards network centrality
exists and is of importance for network design. It supports the comparison of dif-
ferent networks regarding their centrality. Measuring network centrality is common
in network analysis, but may also support network design when different network
86 For example, it becomes directly apparent why high frequencies usually imply a detour in a network. To
achieve both, high frequencies of departures and low distances of transport at the same time, is usually
infeasible by modifying the network appearance.
87 The algorithm to be presented in chapter 5 was highly influenced by the degree of centrality that the
generated networks should have in order to fulfill the demand for the specific transportation network.
73
scenarios are compared based on various performance indicators. This is always an

ex-post analysis; measures of network centrality stand alongside other performance
indicators. Yet, network centrality has one important feature that the six above pre-
sented elements of performance already hint at: network centrality is conceptually
related to network performance. Hence, analyzing network centrality of scenarios
goes beyond the comparison of KPIs as it allows conclusions to elements of network
performance that were not assessed in the scenarios. It might be that at the time
of the scenario generation, the performance indicator did not appear to be relevant
or that the scenario generation did not allow for an element of performance to be
tested. Moreover, assessing certain aspects of network performance will require ad-
ditional assumptions about the network and its future development. For instance,
measuring expected congestion for a transportation network will require detailed as-
sumptions about the capacity of links and nodes. These assumptions are error-prone.
However, assessing the centrality of a network allows to conclude, on a general basis,
about the risk of congestion in a network. If this aspect turns out to be important for
the subsequent planning process, detailed analysis - potentially based on the neces-
sary assumptions - may follow. As network centrality allows many such conclusions,
it enriches the set of KPIs applied in network design valuably.
Measuring network centrality on a numeric scale reduces the concept in its expres-
siveness. A homomorphic function of a qualitative concept such as centrality to a
quantitative scale will reduce its versatility. Many such functions exist in the literature
in forms of centrality indices. Any index of centrality always reflects the researchers
understanding of centrality. The next section reviews existing indices and presents
some applications of these.
3.3. Centrality measures in networks

The outstanding role of transportation network centrality was pointed out in the pre-
vious section. It stands to reason, now, to find expressive metrics for measuring net-
work centrality. Some contributions on measuring the centrality in transportation
networks exist; especially for passenger airline networks. The there applied indices
mainly come from centrality metrics used in social sciences.88 Thus, fundamental
88 One might expect the measuring of centrality in networks to arise out of graph theoretic research. Quite
counterintuitively, social sciences have been driving the early development of centrality measures (or
synonymously centrality indices) in graphs, starting in the late 1940’s. The two probably most fundamen-
tal papers on network centrality by Sabidussi (1966) and Freeman (1979) are from this field. The former
appeared in Psychometrika which states that it “is devoted to the development of psychology as a quan-
titative rational science, including the advancement of theory and methodology for behavioral data anal-
ysis in psychology, education, and the social and behavioral sciences generally” (Psychometrika, 2010).
The latter was published in Social Networks, a journal dedicated to research on the “empirical struc-
ture of social relations and associations that may be expressed in network form” (Social Networks, 2010).
Thus both fundamental publications on network centrality are published in highly ranked journals that
address research communities related to social networks, but can certainly not be seen as typical outlets
for graph theory advances. Even though the suggested metrics are widely used, their role as a proper
74
concepts of measuring network centrality in social sciences will be reviewed in the

following. Subsequently, it will be presented how these metrics were applied to air-
line networks: which modifications were necessary and which results were obtained.
This will support the development of metrics for LTL networks in chapter 4.
“Centrality measures can be regarded as generating expected values
for certain kinds of node outcomes (such as speed and frequency of re-
ception) given implicit models of how traffic flows, and [...] this provides
a new and useful way of thinking about centrality” (Borgatti, 2005, p. 55).
The following section will shed light on the metrics developed in social sciences
on how to measure the centrality of networks. Freeman (1979) structured the basic
centrality metrics developed in social sciences and introduced indices that are fun-
damental for network analysis. The presentation will closely follow his structure. The
developed indices are mainly related to the layout of networks and, therein, repre-
sent the aspect of topological network centrality. Measures for the concentration of
networks come from different backgrounds, namely that of measuring inequality in
social groups and metrics to asses market concentration. The main metrics from this
field to be applied in the following are introduced subsequently. The literature on
all three fields is, most obviously, much wider than what can be presented here; the
selection of metrics is based on their fundamental influence and the further applica-
bility to transportation networks.
3.3.1. Freeman’s centrality framework

Freeman states that by the late 1970s much research had already been undertaken
in the field of network centrality in human groups. Yet, he stresses that the commu-
nity has “no unanimity on exactly what centrality is or on its conceptual foundations
[and that] there is very little agreement on the proper procedure for its measurement”
(Freeman, 1979, p. 217). He extracts three measures of point centrality in a network
from the existing literature, that will eventually lead to three metrics for measuring
the centrality of a graph. Knowledge of point centrality is important to define net-
work centrality later on.
Freeman’s point centrality

Point centrality in social sciences is intuitively related to the idea, that a person in the
middle of a star-shaped network has a central position in that network. That intuition
is easy to follow but allows for different mathematical interpretations of that position
(Freeman, 1979, pp. 218-219).
methodology is contested. Just as one example, Friedkin (1991, p. 1479) reproaches that “the proposed
hypotheses are not derived from any broader theory but are ad hoc formalizations of plausible ideas”.
Then again, the large amount of contributions and the applications described herein, the importance
for research, and the applicability of the metrics hold a strong position to their relevance.
75
Degree of a node From a graph theoretic point of view, the degree of a node is
defined as the number of links connected to that node (Domschke and Drexl, 2005,
p. 66). It can be argued that a person is central in the network if it has many connec-
tions to others. That is true for the person in the middle of a star-shaped network as
well. For a network with n nodes, degree centrality can be calculated as (Freeman,
1979, pp. 220-221)89 : n
a (v i , v k )
C D (v k ) = i =1

(3.1)
n −1
where

1 iff there is an arc in the network between v i and v k
a (v i , v k ) =
0 otherwise
Degree centrality is the number of incoming arcs into a node divided by the total
number of nodes (minus one) in the network. Freeman (1979, p. 221) sees the degree
centrality of a node as “an index of its potential communication activity”, as having
many connections to other members makes it likely that communication takes place
with them.
Betweenness of a node A second point of view on node centrality is expressed

with the notion of betweenness. If a person is positioned so that it will be on the
“communication paths linking pairs of others” (Freeman, 1979, p. 221), that person
is a central one in the network. Again, this is true for the example of the middle of
a star-shaped network. The definition of betweenness point centrality relies on the
existence of geodesics.90 The number of shortest paths (geodesics) relating two points
v i and v j is expressed as g i j . Then, g i j (v k ) is the number of geodesics between v i and
v j that contain v k . Finally, the probability that a randomly selected geodesic from v i
g i j (v k )
to v j contains v k is b i j = gij
. Based on these definitions, Freeman (1979, p. 224)
defines the betweenness point centrality as:
n n
i =1 j =1
b i j (v k )
C B (v k ) = (3.2)
n2 − 3n + 2
Betweenness point centrality is the normalized measure for a node on how likely traf-
fic will pass through it. Freeman (1979, p. 224) expresses that “betweenness is useful
as an index of the potential of a point for control of communication “; i.e. the node
89 Refer to Freeman (1979) for the rigorous development of the presented formulae.
90 A geodesic is the shortest path in a graph, however it is measured. Graph theoretic network models apply
different definitions of a shortest path as introduced in section 2.3.1. The special terminology applied by
Freeman (1979) is due to the social science application. The length of the shortest path is irrelevant for
betweenness centrality, only the number of shortest paths is of interest; see Freeman (1979, p. 222) for an
illustrative example.
76
would be transmitter of information in the network and is, therefore, able to control
the communication by deciding which information to pass on or which to reject.
Closeness of a node The third perspective on node centrality is expressed by the

closeness centrality of a node. Authors see the centrality of a node based on its dis-
tance to the others. Again, the middle of the star-shaped network has the shortest
distance to all other nodes in the network and holds the most central position in this
understanding of closeness centrality as well. Let d i s t (v i , v k ) be the number of edges
in the geodesic that connects v i and v k :
n −1
C C (v k ) = n (3.3)
i =1
d i s t (v i , v k )
Node closeness is the inverse of the average distance from one node to all other nodes
in the network. The closeness node centrality may be used “when measures based
upon independence or efficiency are desired” (Freeman, 1979, p. 226). Knowing that
a point is not close to others can be interpreted as its independence from these.
Eigenvector centrality
Eigenvector is not one of Freeman’s initial centrality measures, but it is usually
grouped in with these (e.g. Kiss and Bichler (2008)). This centrality measure was
introduced by Bonacich (1972), refined by Bonacich (1987), and is continuously de-
veloped further (Bonacich, 2007, p. 556). The underlying idea is to measure indirect
connections and not only directly related nodes. If person A in a network is closely
connected to a central person B, in many settings, A will also be central. Formally,
this relation is obtained by measuring the “eigenvector of the largest eigenvalue of
an adjacency matrix” (Bonacich, 2007, p. 555). Let the adjacency matrix A contain
is the eigenvector with the maximum eigenvalue θ .
entries in the form of φv j v i . χ
The eigenvector point centrality of a node p i is computed as (Kiss and Bichler, 2008,
p. 236):
1
n
C E (v i ) = χv i = φv v χv (3.4)
θ j =1 j i j
An extension to this basic eigenvector centrality is beta-centrality where the indirect

centrality gained by a person is weighted with a factor (Bonacich, 1987).
Freeman’s network centrality

Freeman (1979, p. 227) points out that there are two streams for measuring the cen-
trality of a network, i.e. in the context here for measuring the topology of the network.
He firstly identifies authors who seem to understand the graph theoretic compact-
ness of a network as network centrality. In their understanding a network is central if
77
it is compact, i.e. if the distances between the different nodes are short. This thinking
is closely related to node centrality as presented above. The second literature stream
values a network to be central if the differences in node centrality are high. This ba-
sically expresses the thought that in a central network some nodes are very central
whereas others are not. Freeman (1979) stresses the point that he is convinced the
second literature stream to be relevant for the study of centrality of social networks.
Freeman (1979, p. 228) suggests a generic index structure for graph centrality in-
dices that respects two properties:
1. It assesses how strongly the most central node in the network differs from the
other nodes, and
2. It is normalized with the maximal value of graph centrality that is possible in a

network with the given number of nodes.
C X (v i ) is any node centralitymeasure, and C X (v ∗ ) denominates the largest value of

n
C X (v i ) in the network. m a x i =1 [C X (v ∗ ) − C X (v i )] then describes the maximum pos-
sible sum of point centrality differences in a network with n nodes. The generic struc-
ture for a graph centrality index therefore reads (Freeman, 1979, p. 288):
n
i =1
[C X (v ∗ ) − C X (v i )]
CX = n (3.5)
m a x i =1 [C X (v ∗ ) − C X (v i )]
Based on the three node centrality measures presented above, Freeman (1979) de-
velops three different network centrality measures from the generic formula by sub-
stituting the different point centralities in the generic structure in equation 3.5. Not
surprisingly, these three network centrality indices are known as network degree cen-
trality, network betweenness centrality, and network closeness centrality.
3.3.2. Measures of inequality in social groups

Metrics for network concentration come mainly from two literature streams: inequal-
ity in social groups and market concentration. The metrics from these two streams
that were applied to measure transportation network centrality are introduced in the
following two sections.
Social scientists have a long tradition of measuring the inequality within a given
population.91 Two main metrics have evolved in this field. The Gini index or related
concepts that are drawn from Lorenz curves make up the first group while the second
group comprises indicators of dispersion, e.g. an entropy index. Axiomatic deriva-
tions of inequality indices as well as normative measures stemming from social wel-
91 The idea is interestingly closely related to Freeman’s suggestion that the difference between the most
central and the other nodes is important for measuring graph centrality (Freeman, 1979, p. 228).
78
fare functions exist further, but are of minor importance (Kaplow, 2005, p. 65).92 The
following paragraphs will briefly introduce the Gini index as well as give an example
of an entropy-based index.
Gini index The Gini index is the most widely used index to assess inequality in so-
cial groups (e.g. Atkinson and Brandolini (2010, p. 5), Ogryczak (2009, p. 62), Kaplow
(2005, p. 71), Atkinson (1970, p. 244)). The Gini index is defined as “half the mean
difference divided by the mean” (Atkinson and Brandolini, 2010, p. 5). Theil (1967,
p. 121) formally defines the Gini index to measure the income inequality in a social
group of n individuals as:

s ji n c
1 pop pop s ii n c
n n
GI = s s j pop − pop (3.6)
2 i =1 j =1 i si sj
pop
where s i is the population share of the i t h group in the population and s ii n c is its
income share. The minimal income inequality is achieved if all groups in the popu-
lation have an income share directly equivalent to their population share with a Gini
index of 0. The Gini index approaches 1 if the income is all with one group in the
population and this very group “becomes smaller and smaller” (Theil, 1967, p. 121).
The Gini index is the comparison of the Lorenz curve of a distribution with a line of
perfect equality (e.g. Theil (1967, pp. 121-122), Rodrigue et al. (2009, pp. 184-185)).93
Entropy index Different entropy-based indices have been developed, but the en-
tropy index usually refers to the so called Theil index presented by Theil (1967, p. 91):

n
1
T (i n c ) = s ii n c l o g (3.7)
i =1
s ii n c
where n is again the number of individuals and s ii n c the income share of the i t h indi-
vidual. The Theil index will be maximal (l o g n ) for an equal distribution of income
and amount to 0 when the entire income is attributed to a singe individual (Theil,
1967, p. 91). As this is somewhat counterintuitive, studies sometimes use the inverse
Theil index to measure (in-) equality (e.g. Wojahn (2001, p. 29)).
Conçeicão and Ferreira (2000) highlight that the Theil entropy measure of inequal-
ity has one particular strength: the Theil index is superior to the Gini index in that it
provides a much better understanding of the concentration that exists within a cluster
of grouped individuals, which is a result from its decomposability.
92 Kaplow (2005, p. 65) argues that normative indices derived from social welfare functions are of limited
usefulness to measure inequality in a social group.
93 Figure A.1 in the appendix graphically shows the relation between Gini index and the Lorenz curve. The
Gini index is twice the area A between the Lorenz curve of the distribution and the 45◦ line in relation to
the entire area: G = A/(A + B ) (Rodrigue et al., 2009, p. 185).
79
3.3.3. Measures of market concentration

The context of measuring market concentration for economic purposes is the last sit-
uation presented here where measuring network centrality originates. Economists
search for single value indices to measure the distribution of market shares within an
industry for applications such as antitrust analysis. The m-firm concentration ratio
and the Hirschman-Herfindahl (HH) index are commonly used to assess market con-
centration and will be presented in the following. An entropy-type index is sometimes
also used for measuring market concentration (Tirole, 1988, p. 222) such as the Theil
index that was already introduced above.
Tirole (1988, p. 222) highlights that formal definitions of the concentration mea-
sures and well thought axioms supporting them do not necessarily “tell us what use to
make of concentration indices”. For the context of concentration of markets, the au-
thor suggests a potential relation of concentration and profitability to be constructed.
m-firm concentration ratio The m-firm concentration ratio adds up the highest m
market shares in the industry. It describes the importance of the top m players in the
market. It is formally defined as (Tirole, 1988, p. 221):

m
C= si (3.8)
i =1
where n is the number of firms in the industry and s i is the market share of the i t h
firm. The n firms in the sample must be ordered by decreasing market share.
Many studies include either 4-firm concentration (C4) ratios or 8-firm concentra-
tion (C8) ratios for their analysis (e.g. Borenstein, 1992, Saunders and Shepherd, 1993,
Lee, 2003).
Hirschman-Herfindahl index The HH index is defined as (Tirole, 1988, p. 221):

n
HH = s i2 (3.9)
i =1
where the symbols have the same meaning as defined above.94 The HH index sums
the squared market shares per firm on the market. Notice that the market shares of all
n firms are taken into account for the HH index, but not for the m-firm concentration
ratio.
Summing up, metrics for measuring either the topological centrality of social net-
works, the inequality in social groups, or the concentration of markets have been
94 The HH index appears in the literature also as the Herfindahl-Hirschman index (e.g. Lijesen (2004)) or
simply the Herfindahl index (e.g. Tirole (1988, p. 221), Wojahn (2001, p. 28)), whereas the formal defini-
tion is undebated.
80
3.4. Measuring network centrality of passenger airline networks
introduced in the literature. All of the above presented metrics can be seen as fun-
damental for their field of application and have been applied extensively. When re-
searchers started to measure the centrality of transportation networks, they redrew to
existing metrics, namely those introduced above. Some of the metrics were adapted
to the context of transportation networks, some could be applied directly. Further-
more, the existing metrics and their adaptations were supplemented by new metric
developments for the specific context of transportation networks. The next section
will present how the different metrics were applied to measure the centrality of trans-
portation networks; passenger airline networks will serve as an example.
3.4. Measuring network centrality of passenger airline

networks
The context modeled in a network is important when interpreting its centrality.
Whether or not a centrality metric is appropriate is dependent on the situation
(Koschützki et al. (2005b, p. 84), Sabidussi (1966, p. 582), Freeman (1979, p. 217),
Borgatti (2005, p. 56)). Borgatti (2005) points out that the measures commonly used
in social network analysis make implicit assumptions on the flow of elements in
the network. However, the path that objects travel along through a network may be
very different. Borgatti identifies seven exemplary objects that have different travel
behaviors, e.g. e-mails are forwarded by duplication and may exist in multiple copies
in a network. Packages are also amongst the examples he lists. They are characterized
by the special feature that they have a defined origin and destination and it can
be further assumed that they always travel along the shortest path (a geodesic) in
the network (Borgatti, 2005, p. 58). This section will take a look at measuring the
centrality of transportation networks in one specific context.
Transportation networks in industrialized countries have seen a history of regula-
tion and deregulation in the 20th century. Related to this development, researchers
have been eager to quantify the centrality of transportation networks to point out
the (expected) performance of regulated and deregulated transportation networks.
Passenger airline networks have obtained an especially large share of attention from
policy makers and researchers.
This section will highlight how some of the above introduced centrality indices
were adapted to transportation networks. The area of passenger airline network cen-
trality literature is a very interesting example to review in more depth. Firstly, it pro-
vides direct adaptations of centrality indices from other fields as well as new devel-
opments for the specific research context. Furthermore, it will serve as a valuable
starting point when striving for LTL network centrality indices in chapter 4.
81
3.4.1. Assessing the outcome of deregulation

Transportation industries which are characterized by networks underwent important
changes in the regulatory environment during the last forty years.95 Before, through-
95 Measuring centrality in transportation networks has been of importance since the beginning of deregu-
lation. Traditionally, transportation markets were heavily regulated by national governments. Network
industries with natural monopolies were to be restricted and destructive competition to be limited (Win-
dle, 2005, p. 50). By today, most of the regulation of Western transportation markets has been reduced
to operational and safety regulations (Rodrigue et al., 2009, p. 141). The following will briefly intro-
duce (transportation) network regulation, provide background information on deregulation activities,
and continue to explain the connection of deregulation and centrality measures.
Transportation networks have been regulated by governments since their economic importance has
been realized. The ocean shipping industry has had cartels since the 19th century, these were granted US
antitrust immunity as early as 1916. The shipping industry still remains special within the transport in-
dustry in that international maritime institutions play a major role in terms of safety regulation that is not
comparable to any other transportation mode. An interesting overview is provided by Talley (2005). The
first modern time transport industry to be regulated in the US were railroads (since 1887), later followed
by the trucking industry (1935), the entire airline industry following in 1938. Other network industries
such as telecommunication, cable TV, and electric utilities have seen similar regulation (Windle, 2005,
pp. 50-51).
European states have had similar national regulations on transportation. The Treaty of Rome created a
single internal market for the European Community (EC) in 1957 and a common transport policy was
necessary to accomplish this (Van Reeven, 2005, p. 710). Many regulative issues have been dealt with on
a European-wide level since the mid 1950s. The following will briefly outline the deregulation of the US
and European Union (EU) airline and the trucking industries as two interesting examples.
The mindset of governments as well as public opinion changed over time and deregulating transporta-
tion networks became an option. The air transport markets have gone through a phase of deregulation
that was closely followed by the public. First criticism on the regulation of the airline industry arose in
the 1950s / 60s (Morrison (2005, p. 406), Doganis (2010, p. 43)). The United Kingdom of Great Britain and
Northern Ireland (UK) started to liberalize the market from 1975 on (Doganis, 2010, p. 44); the US Air
Deregulation Act of 1978 opened the US American market to competition (Rodrigue et al., 2009, p. 141),
and the US started to work on bilateral agreements in the 1980s (Doganis, 2010, pp. 45-46). 1983 brought
the first EC directive on liberalizing regional air services between EC member states (Doganis, 2010,
p. 44), which also fell in the phase for European countries to renegotiate their bilateral agreements with
other countries (Doganis, 2010, pp. 46-47). From 1992 on, the US and the EC (since 1993 EU) began to de-
velop towards open skies agreements (Doganis, 2010, pp. 49-53). The first phase of the open skies EU-US
Air Transport Agreement took effect on March 30, 2008, allowing inter alia European airlines to operate
flights from anywhere in the EU to anywhere in the US, and vice versa for American airlines (European
Union, 2008). Furthermore, cooperative agreements on both sides are less restricted (Ezard, 2008). The
second phase of the EU-US open skies agreement was signed in June 2010 (Reals, 2010); the lawfulness of
cabotage being one of its new elements (Ezard, 2009). Additional states, e.g. Iceland, Norway, and Japan,
have also accepted the open skies agreement; negotiations with other states are ongoing (Govindasamy
and Ranson, 2010, Knibb, 2010).
The US trucking regulation was contested throughout the 1970s until the US Motor Carrier Act was signed
in 1980 that formalized the liberalization of the trucking industry. Entry barriers were reduced and free
price-setting was introduced. Whether or not this liberalization was beneficial for the entire trucking
industry remains an open point for discussion (Rakowski, 1988, Giordano, 2008).
The European market evolved later and slower. As mentioned above, the 1957 Treaty of Rome stated
that a common transport policy was to be introduced for the EC member states. Yet, it was not until
1985 when a ruling by the European Court of Justice forced member states to start passing legislation ac-
cordingly. Before that, international road transport within the EC was possible under one of four cases:
(a) Bilateral agreements between states allowed for open transports between the states or set quotas.
Additional permits for transit states could be necessary. (b) Each country was granted a certain num-
ber of community permits that it could attribute to its carriers. These community permits allowed free
operations within the EC. (c) Similarly, European Conference of Ministers of Transport (ECMT) permits
82
out, and still ongoing, the literature contains many contributions on the expected or
perceived changes in the concerned transportation networks as well as on discussing
various outcomes of network changes (e.g. Ghobrial and Kanafani (1995), Oum et al.
(1995), Saunders and Shepherd (1993), Borenstein (1992)). Some of these studies pre-
pared deregulation, others validated or contradicted the expected network changes.
Apart from these type of studies, there are - of course - also further research foci to
measure the development of networks. Inter alia, the rise of low cost carriers (LCCs)
as new players in the passenger airline industry led to studies that compare the dif-
ferent network types operated by LCCs and full service carriers (FSCs) as well as their
development over time.
To assess these changes, researchers reviewed metrics developed in other sciences
that were originally destined to give information about networks and graphs. As
transportation networks are abstract networks just as any other, the context tran-
sition can be done without major difficulties. Moreover, researchers developed
additional indices for transportation networks that actually found their way to other
disciplines.96
The next section will illustrate how transportation network centrality is measured
for the specific context of air transport networks.
allowed for free operations in the ECMT states (ECMT states included in the 1980s the entire EC plus
Switzerland, Finland, Norway, Sweden, Austria, Yugoslavia and Turkey. Since 2006, the ECMT developed
into the International Transport Forum (ITF), today having more than 50 member states (International
Transport Forum, 2010)). (d) Transports that were not subject to quotas were always allowed to travel
through the EC and some neighboring states. Typically either transport prices or the number of trans-
port licenses or both were regulated in the EC member states. Cabotage was entirely prohibited and time
consuming controls at borders within the community were performed. Own accord transports never fell
under most of the regulation; but even until today, these carriers are not allowed to transport third-party
freight. The market liberalization in the EC started in 1987 by increasing the number of community per-
mits; in 1990 a limited number of cabotage licenses were issued and the market is finally perceived as
liberalized since 1998 (Lafontaine and Valeri, 2009, pp. 22-25).
96 An early example for such independent measures are those proposed by Taaffe and Gauthier (1973) to
measure connectivity in a transportation network. Connectivity is seen as a highly relevant measure of a
network: “Since the expansion or intensification of transport linkages between nodes is directly related
to increases in demand for transportation facilities to move goods and people, the degree of connectivity
of a transport network is indicative of the complexity of the spatial order that it imposes on the region
it serves” (Taaffe and Gauthier, 1973, p. 101). To assess the connectivity of a network, Taaffe and Gau-
thier (1973) suggests the Gamma index as well as the less frequently used Alpha index. Furthermore, the
authors apply the metric by Shimbel (1953) that was originally designed to measure the dispersion in
communication networks to the context of transportation. The ideas expressed by Shimbel (1953) with
his network measures are in the same stream of thought as Freeman’s centrality measures and Bonacich’s
eigenvector centrality. This work is often seen as one of the earliest contributions in transportation net-
work measurements. It stems directly from graph theory and is not yet related to the other disciplines. As
measuring transportation network centrality developed further, more allusions to the above introduced
fields in social sciences, inequality metrics, and market concentration were made. Even so, the indices
provided by Taaffe and Gauthier (1973) continue to influence current literature contributions.
83
3.4.2. Metrics for the centrality of passenger airline networks

One important field of application for network metrics are airline networks. Three key
reasons have already been touched upon. Firstly, airline networks have been an in-
teresting field of research for a long time and the optimal planning of their networks
has been studied intensively in the past. Thus, it stands to reason to keep track of
their development. Secondly, the airline industry has lived through major regulative
changes in the past, as just presented, and their impact on the network appearance
is interesting (from both, an academic and a policy making perspective). Thirdly, the
industry was shaken by the entrance of LCCs onto the market. Their network strategy
is different from the traditional FSCs.97 This clash of strategies has been an interest-
ing field of research. The following section will therefore shed light on this specific
context.
This section will present the application of a selection of above introduced central-
ity metrics to the specific context of passenger air transport networks. Many of the
indices were applied in their general form as it was introduced above. However, the
McShan-Windle (MW) index is a development of the m-firm concentration ratio spe-
cific to airline networks, similarly, the network concentration (NC) index enhanced
the Gini index for this context. Moreover, the hubbing concentration (HC) index was
devised explicitly for airline networks, inter alia based on the HH index. These metrics
are introduced in the following. The presentation closes with an overview of results
obtained by the studies analyzing transportation networks.
Airline network topology

The study of airline network topology is related to assessing the network’s spatial ap-
pearance. The “measures of network configuration do indeed depend upon the shape
of the network and its centrality” (Alderighi et al., 2007, p. 530). The goal of one of
these studies is to “typify and map out complex airline network configurations” (Reg-
giani et al., 2009, p. 258), this is a good example for the entire group of topology mea-
sures. Research is directed to show similarities to certain network topologies, such as
a star-shaped or a fully-meshed network (Alderighi et al., 2007, p. 534). In this con-
text, researchers define a hub as a node that is highly connected to others (Reggiani
et al., 2009, p. 260).98 Therefore, measures to highlight if a network possesses hubs
will point out networks that have highly connected nodes.
An early example of topological measures is the Gamma index by Taaffe and Gau-
thier (1973). It is a “useful measure of the relative connectivity of the entire network”
(Scott et al., 2006, p. 219). Wojahn (2001, pp. 38-39) provides a normalization of the
Gamma index and presents index values for large airlines.
97 Pels (2008, p. 71) points out that most LCCs use P2P networks while FSCs operate H&S networks.
98 When considering only the typology of a network, a traffic-based definition for the term hub cannot be
applied.
84
Toh and Higgins (1985) create a hub index that is calculated “by dividing the num-
ber of outlying cities served by the hub by the number of spokes radiating from it”
(Toh and Higgins, 1985, p. 19).99 The index is not normalized and always greater than
1. A centralized network with a strong hub that directly connects to many outlying
cities will have a low index value close to 1. This index is an interesting example of in-
tuitive measures for airline network centrality. As one drawback, this measure needs
to know beforehand not only what the airports in the network are, but also which
of these airports is a hub. It will serve to compare networks only when the hubs are
predefined.
An example of the applicability of the Shimbel index to the airline context is given
by Bowen (2002). He makes use of Shimbel’s trip length between airline hubs in the
network, relates these to the minimal trip length to connect the airports in a hypo-
thetical P2P network and finally normalizes the value (Bowen, 2002, p. 429).
Alderighi et al. (2007) suggests to use a graph centrality index based on Freeman’s
betweenness centrality (equations 3.2 and 3.5) to assess the similarity of a network to
a perfect star which the authors directly relate to an airline H&S network (Alderighi
et al., 2007, p. 536). The authors state that Freeman’s betweenness graph centrality
represents economic behavior of passengers as it identifies whether a network has
hubs that lie on geodesics through the airline network and are, therefore, preferen-
tial for the passengers as connection points (Alderighi et al., 2007, p. 538). Their re-
search, furthermore, applies Bonacich’s eigenvector centrality measure to identify the
appearance of different airline networks, but cannot contribute to comparing differ-
ent networks amongst each other (Alderighi et al., 2007, p. 539).
A general review of different network measurements with a special focus on topo-
logical aspects is provided by Reggiani et al. (2009). Table 3.3 gives an overview of the
metrics the authors suggest to assess network topology.100 The used indices can be
traced back to two categories: graph theoretic elements (degree, diameter, clustering)
or to the indices from social sciences (closeness, betweenness and the betweenness
centrality index).
Airline network concentration

Measuring airline concentration seems to create more research opportunities than
measuring network topology when judging from the number of literature contribu-
tions. Table 3.4 gives a non-comprehensive overview of the wide spectrum of metrics
used to measure airline network concentration. Measuring airline network concen-
tration - especially in the early contributions - is strongly connected to measuring the
concentration of airports, which is seen as a special case of measuring industry con-
99 Multi-hub (MH) networks attribute weights to the hubs.

100 The authors see the Freeman betweenness centrality index as a measurement for concentration, how-
ever, in the light of our discussion it is rather perceived as a measure for topology in the spirit it was
introduced by Alderighi et al. (2007).
85
Index or measure Description

The degree of a node is given by the
Degree
number of its links.
It indicates a node’s proximity to the
Closeness
other nodes.
It indicates a node’s ability to stand
Betweenness between the others, and therefore, to
control the flows among them.
It measures the maximum value of the
Diameter
geodesic distances between all nodes.
Clustering coefficient It measures the cliquishness of a node.
It is a measure of similarity to a perfect
Freeman betweenness centrality index
star network.
Table 3.3.: Network topology indices (Alderighi et al., 2007, pp. 263 and 265).
centration. The amount of enplanements at an airport compared to the entire en-

planements in the airline network gives the traffic share at this airport.101 It is not sur-
prising that metrics from the industry concentration literature were quickly adopted
to airline networks. Other indices also found their way into airline research, e.g. the
Gini index. Moreover, airline specific adaptations of existing indices were presented
(e.g. the NC index) and airline specific index development is also represented in the
literature (e.g. the HC index). The following paragraphs will exemplarily introduce
some of the measurements used in the conducted studies.
What is known in the literature as the MW index actually is a special case of the
m-firm concentration ratio applied to airports (McShan and Windle, 1989, Wojahn,
2001).102 It is independent of the size of the players which gives it an advantage over
the m-firm concentration ratio (McShan and Windle, 1989). It measures “the propor-
tion of an airline’s total departures [...] leaving from the three percent most utilized
airports [...] in that airline’s network” (McShan and Windle, 1989, p. 214). Wojahn
(2001) modified the MW index slightly by adding a fractional term to the sum of the
top 3% of airports in the network. Is is then calculated as:

m
MW = s i + (m − m )s m +1 (3.10)
i =1
101 The Federal Aviation Administration (FAA) uses the term enplanement in the meaning of “revenue pas-
senger boarding” (FAA, 2009). Most studies use outgoing traffic flows for their purpose (e.g. Debbage
and Delk, 2001, Toh and Higgins, 1985, Reynolds-Feighan, 2001, Martín and Voltes-Dorta, 2008). Rare
examples (e.g. Hensher, 2002) consider incoming and outgoing movements to assess traffic. Borenstein
(1991, p. 1262) highlights that some statistics on airport traffic report only enplanements whereas others
distinctly report en- and deplanements at airports.
102 The calculation of the m-firm concentration index is described in equation 3.8.
86
Contribution Main measurement

McShan and Windle (1989) McShan-Windle index
Borenstein (1992) Hubness, Airport Herfindahl index
Saunders and Shepherd (1993) C8 ratio, HH index
Ghobrial and Kanafani (1995) HH index
Reynolds-Feighan (2001) Gini index
Wojahn (2001) Broad literature overview
Burghouwt et al. (2003) NC index
Lee (2003) C4 ratio, C8 ratio, HH index
Martín and Voltes-Dorta (2008) HC index
Derudder and Witlox (2009) Entropy-based measurements
Huber (2009a,b) Gini-based analysis
Martín and Voltes-Dorta (2009) HC index
Papatheodorou and Arvanitis (2009) Gini-based analysis
Reggiani et al. (2009) Normalized Gini index, entropy index
Costa et al. (2010) Modified HH index
Table 3.4.: Scope of network concentration measurements.
where m = 0.03n , n being the number of airports in the network and the distinct air-
ports i are ordered in decreasing magnitude of traffic. s i is the relative share of traffic
at the airport i . Even though the MW index has received little attention over the last
years, it is an expressive and yet computationally simple measurement of concentra-
tion, highlighting the importance of the top 3% airports (in terms of traffic) in the
network. The MW index rises as the concentration in the network increases; it ranges
between 0.03 (not concentrated network, all nodes have equal traffic) to 1 (concen-
trated network).103 McShan and Windle (1989) use the MW index to measure in how
far the routing in the network resembles a typical routing in a H&S network in airline
networks and relate it to cost and competitiveness of airlines. Their reasoning is that
if the importance of the top 3% airports increases, the network becomes more similar
to a H&S network.
Studies using HH index type measures are common in the literature. As mentioned
above, some authors use the HH index to measure the concentration in the airline
industry by comparing the traffic shares of airports (e.g. Borenstein, 1992, Lee, 2003,
Papatheodorou and Arvanitis, 2009, Costa et al., 2010). Costa et al. (2010) provide a
HH index normalization that is specific for the airline context. It is worth mentioning
that Wojahn (2001, p. 32) shows that the interpretation of the HH index for an entire
network is scale dependent and, hence, of limited use in interpreting it as a measure
for network concentration. The other possibility to make use of HH indices is to com-
103 Formally, the MW index may assume the value of 0 when there is no traffic in the network at all. All the
same, this is merely a theoretical case.
87
pare city-pair specific routes, as e.g. Borenstein (1992) and Saunders and Shepherd
(1993) do. This measure allows to compare the concentration or vice versa the com-
petition on these routes. Martín and Voltes-Dorta (2008, 2009) make use of city-pair
specific HH indices for the development of their HC index (see below).
Studies with entropy-based measurements for airline networks are limited. Xie and
Levinson (2007) study road networks with an entropy based approach. Only recently,
Derudder and Witlox (2009) suggested a framework for evaluating the “overall evo-
lution of the spatiality of airline networks” (Derudder and Witlox, 2009, p. 276). By
stating that hierarchical differentiation can be interpreted in terms of concentration
or dispersal of an airline network, they develop entropy-based measures to asses this
differentiation.
The Gini index was firstly devised to measure the concentration of airline networks
by Reynolds-Feighan (1998) and prominently applied to compare the networks of
LCCs and FSCs by Reynolds-Feighan (2001). She adapts the formula for the Gini index
as presented in equation 3.6 to the specific context of airline networks:
1
GI = si − s j (3.11)
2n i j
Since then, the Gini index has become very popular in the field, and is commonly
used by researchers (e.g. Wojahn (2001), Huber (2009b), Papatheodorou and Arvanitis
(2009)). The value of the maximum of the Gini index (G I m a x ) depends on the number
of airports in the network. This situation calls for a normalization, which is what the
NC index provides (Burghouwt et al., 2003). The Gini index reaches its maximum in
an airline network where all traffic is consolidated on one route (Burghouwt et al.,
2003).
2
G Imax = 1 − (3.12)
n
The interpretation of the Gini index is therefore difficult when comparing networks of
different sizes. Burghouwt et al. (2003) propose the NC index, which corrects the Gini
index for network size.
GI
NC = (3.13)
G Imax
In contrast to the Gini index, the NC index varies in the interval [0; 1], no matter what
the network size. The NC index increases if traffic becomes less evenly split across the
network (Burghouwt et al., 2003). The lower the NC index, the less concentrated the
network.
An index that was specifically developed for the airline context is the HC index. Air-
line hubbing strategies may lead to a distorted picture when considering traffic at an
airport. Passengers at the airport for initial departure are mixed with those switching
planes to continue onward in their travels. Accordingly, traffic at an airport repre-
sents an ambiguous role of this airport as an origin and as a hub in the original sense
88
(O’Connor, 2003, Alderighi et al., 2007). To be able to explain hubbing expressively,

Martín and Voltes-Dorta (2008) argue that one needs to distinguish between a hub
and an important origin or destination in an airline network. The authors suggest the
HC index as a new concentration metric that takes into account this difference. It is
an index that relates concentration to two major aspects: the importance of hubbing
in the network and the concentration of hubbed traffic on each OD-market. As a first
step, the hubbing behavior of the connecting passengers on a market i j needs to be
calculated104 :

n
Hi j = C i j s k2 ,i j (3.14)
k =1
where C i j is the share of connecting passengers on the market i j and s k ,i j equals the
traffic share of airport k on the market i j . Taking a closer look at this formula, H i j is
a variant of the HH index of the city pair i j and represents the routing concentration
on this market. The HC index is further calculated by weighting the hubbing behavior
according to the relevance of this route for the airline. The relevance is expressed by
the traffic on this route qi j in relation to the overall traffic Q:
qi j
HC = Hi j . (3.15)
i j
Q
The HC index falls in the interval [0; 1]. A single H&S network yields a HC index of 1
(Martín and Voltes-Dorta, 2008). A HC index equal to 0 represents inter alia a network
where passengers do not connect at all to ongoing flights. A high HC index indicates
that hubbing is important in the network and that it is done on a large share of the
traffic. Martín and Voltes-Dorta (2009) demonstrate how the HC index is superior to
the NC index to indicate which airline networks actually make use of hubbing for a
relevant share of their passengers. The HC index can depict this information whereas
the NC index might falsely label airports with high origin traffic as hubs even though
no passengers switch planes there.
Results obtained for the airline sector by measuring network centrality

The literature reviewed above applied various measures of network topology and con-
centration to the research field of air transportation. Two research streams can be
identified. Firstly, authors discuss the changes in transportation networks, the indus-
try, or the airport landscape over time. This approach is usually related to the effects
of deregulation in the market or to LCCs entering the market and thereby challenging
the FSCs. Selected studies of this type are shown in table 3.5. Secondly, the network
centrality indices were used to compare networks of different scope with each other.
Table 3.6 shows exemplary studies of this type. The following will briefly comment on
104 A market is the set of all possible flights (including direct flights and flights via a hub) to travel from an
origin i to a destination j .
89
the obtained results by studies (a) measuring airline market change and (b) compaing
different networks.
Airline market change Borenstein (1992) shows with the help of the C4 ratio, the
C8 ratio, and the HH index that US airline networks have become more concentrated
between 1982 and 1990. Furthermore, he identifies a growth in the appearance of
networks in form of H&S networks. Reynolds-Feighan (2001) confirms this result for
the period between 1969 and 1999. She employs the Gini index and can thereby high-
light that the importance of hubs for US FSCs increased in parallel with deregulation.
Burghouwt et al. (2003) find similar tendencies for European networks that, in sum,
show more concentration from 1990 to 1999.
Saunders and Shepherd (1993), O’Connor (2003), and Papatheodorou and Arvani-
tis (2009) orient their research to airports in the networks. As concentrated airline
networks typically imply high traffic shares at few airports and vice versa, this can
be seen as a shifted perspective to a similar phenomenon. Saunders and Shepherd
(1993) take the classical economical viewpoint on deregulation and can, based on the
C8 ratio and the HH index, suggest that limiting hub dominance may increase total
efficiency. O’Connor (2003) is interested in different size airports and concludes that
“long term evolution of air travel lies in the mid-sized and smaller aircraft capable of
providing links between those cities in the middle and lower levels of the top 100 air-
ports” (O’Connor, 2003, pp. 89-90). Hence, he expects those smaller airports to gain in
importance in the future. Finally, Papatheodorou and Arvanitis (2009) examine Greek
airports with the help of a Gini index analysis. Interestingly, the authors show only low
reactions to market liberalization in Greece; an evolution that is very different from
central Europe.
Airline network comparison Reynolds-Feighan (2001), Burghouwt et al. (2003), and

Huber (2009a,b) compare networks of different carrier types with each other. Based
on the Gini index, Reynolds-Feighan (2001) shows that the European LCCs have a
high variety of degrees of network concentration. She identifies two groups of LCCs;
the first group operates P2P networks with a low concentration. The second group has
much higher degrees of concentration (Gini index above 0.6) and organizes the oper-
ations around a small number of key airports.105 The author furthermore concludes
that, overall, LCCs show lower degrees of concentration than FSCs. Burghouwt et al.
105 In this context, Reynolds-Feighan (2001, p. 274) notes that “this group organizes their traffic flows in a
manner similar to the full-service carriers”. It appears as if the author herself addresses a shortcoming
of the Gini index in that it cannot express the degree of hubbing operations in the network. Dobruszkes
(2006, p. 255) reviews this claim and explains in a more detailed way that “the nodes of the low-cost
networks are in fact traditional points of access or exit of the network and not transfer platforms. The
star-shaped networks of certain airlines do not correspond to the hub-and-spoke model”. Pels (2008,
p. 71) comes to the same conclusion; even though from the network layout, some LCCs appear to have
H&S networks as these are star-shaped; the fact that no connecting travel is planned for is a clear sign of
P2P network operations.
90
Contribution Region Time period Research focus
Borenstein (1992) US 1982-1990 Airline networks
Saunders and Shepherd (1993) US 1978-1984 Hub dominance
Reynolds-Feighan (2001) US 1969-1999 Airline networks
Burghouwt et al. (2003) Europe 1990-1999 Airline networks
O’Connor (2003) World 1990-2000 Airports
Papatheodorou and Arvanitis (2009) Greece 1978-2006 Airports
Table 3.5.: Markets and event periods for selected studies on aviation market change.
91
Contribution Networks
Reynolds-Feighan (2001) FSC, LCC
Burghouwt et al. (2003) Various carrier types
Derudder and Witlox (2009) Europe
Huber (2009a,b) US, EU, carrier types
Reggiani et al. (2009) Lufhansa vs. StarAlliance
Table 3.6.: Compared networks in selected studies.
(2003) compare the networks of national airlines, regional carriers, low-cost carriers,
and extra-EU airlines. The authors derive detailed results mainly based on applying
the NC index. National airlines (“flag carriers”) operate radial networks based at na-
tional airports. Regional carriers, too, operate radial networks, yet at a smaller scale.
There are hardly any examples of linear networks in Europe. LCCs in Europe operate
focused on a limited number of airports; but the authors see early signs of more lin-
ear networks by LCCs on the European market. Lastly, extra-EU airlines, i.e. airlines
operating under the fifth air freedom right106 , operate linear networks with low NC
index values. Huber (2009a) compares US and European airline networks and finds
many differences. Two findings are interesting here. Firstly, US airlines operate more
H&S networks than European airlines. Secondly, the author points out that spatial
concentration (measured with a Gini index) in European air networks is highest for
intercontinental, long-distance flights. Almost identical findings are documented in
Huber (2009b).
Derudder and Witlox (2009) and Reggiani et al. (2009) provide network compar-
isons that take unusual perspectives, but are certainly worth mentioning. Reggiani
et al. (2009) compare four networks with very different scopes regarding network con-
centration and network topology: European Lufthansa, worldwide Lufthansa, Euro-
pean StarAlliance, and worldwide StarAlliance networks. By applying a wide spec-
trum of measures, the authors conclude that the European Lufthansa network ap-
pears to be a H&S network and is the most concentrated in the sample.
Derudder and Witlox (2009) take a different approach. The authors develop an
entropy-based measurement that is not directly related to the widely applied mea-
sures as presented above. Furthermore, the authors compare national networks for
traffic flows that are separated into business and economy traffic. Amongst others,
it is shown that business traffic flows are more concentrated than economy flows.
Furthermore, the authors conclude that the domestic flights within Germany are less
concentrated than domestic flights in France and the UK.
106 The fifth air freedom allows airlines to carry traffic between two foreign countries if the flight either orig-
inates or terminates in the home country. Rodrigue et al. (2009, pp. 138-141) provide an illustrative in-
troduction to air freedom rights.
92
3.5. The role of network centrality in strategic network design
This excursus into measuring airline centrality served several purposes. It firstly
presented centrality metrics that will, in the following, form the basis for developing
measurements for LTL networks. It secondly showed the manifold theoretical ap-
proaches that researchers in this field have been pursuing. So far, no agreement has
been reached concerning how network centrality in the limited field of air transport
should best be measured. Finally, the section allowed to present some exemplary re-
sults obtained by the conducted research in the field.
3.5. The role of network centrality in strategic network design

This chapter demonstrated the strategic relevance of network centrality for network
design. Firstly, it gave a definition of network centrality. Network centrality was di-
vided into the two aspects of topological network centrality and network concentra-
tion. While the former refers to the layout of the network, the latter incorporates the
notion of traffic flows.
Secondly, the relation of centrality and relevant network performance indicators
was demonstrated, forming a qualitative understanding of network centrality.
In brief, central networks have a high potential for terminal consolidation while
inventory and vehicle consolidation play minor roles, operate high frequency
connections, are characterized by long transportation distances, are prone to low
schedule reliability due to high risks of congestion and strong impacts of late arrivals,
are comparably vulnerable, and show a high system capacity flexibility. The opposite
is true for decentral networks. The demonstration depicted that commonly accepted
trade-offs in transportation networks can be directly linked to the concept of network
centrality. This qualitative understanding may be used to pre-structure network
design as it allows to frame potentially suitable network appearances once desired
network performance is expressed.
Thirdly, a wide toolkit of measurements for network centrality, more concretely of
network topology and network concentration, was presented, supporting the quanti-
tative understanding of network centrality. It provided an overview on the one hand
of the variety of understandings of network centrality and the related measurements
for it. On the other hand it presented a foundation of network centrality metrics. The
literature in disciplines that - at first sight - seem rather far away from transporta-
tion research provides much methodological background for centrality metrics. Es-
pecially social sciences provide well developed foundations. An example of transfer-
ring the measurements to transportation networks has been given by the spotlight on
measuring the centrality of airline networks. Many tools can be fruitfully applied and
some indices were developed (further) to suit the context of air transportation. Net-
work topology is mostly assessed with directly applied metrics from other fields while
new developments of indices apply mainly for network concentration.
93
Significant parts of the literature use the concentration metrics for airline networks
in order to depict if airline networks show characteristics of H&S networks or P2P net-
works.107 Even though no final agreement about the expressiveness of certain metrics
has been reached in the literature, the general concept of using indices to depict net-
work appearance is uncontested. It allows to compare networks in detail as to their
network appearance. This may be exploited for different purposes. In the context of
network design, it may be used to compare different scenarios of potential networks.
It is hardly worth mentioning that the logical connection between centrality index
value and network performance is only indicative and that one always needs to be
aware of what aspects of centrality are depicted by the applied index. Nevertheless,
centrality indices of networks are related to network performance. In this function,
they can steer the search for the development of transportation networks that fulfill
the decision-maker’s expectations regarding the performance of the final network.
The following chapter will continue by presenting the LTL industry and then sug-
gest indices to measure the concentration of LTL networks.
107 These research questions arise from industry events such as the deregulation (that supposedly led to
more H&S networks) or the market entry of LCCs (who supposedly operate more P2P networks than
FSCs).
94
4. Less-than-truckload network centrality
The content of the previous chapters was generally applicable to all transportation
networks. The perspective is now drawn to less-than-truckload (LTL) networks oper-
ated by logistics service providers (LSPs). This concretization allows for the consec-
utive chapters to develop an algorithm for LTL network design based on the general
relations between network key performance indicators (KPIs) and network centrality
and discuss results obtained with its help.
This chapter firstly introduces LSPs as companies offering a wide variety of logistics
services, among them LTL transports. An overview over the operations carried out by
LSPs is given and different types of LSPs are distinguished. The chapter then contin-
ues by providing detailed background on the market for LTL transports. LTL trans-
portation refers to a many-to-many road-based transportation market for palletized
shipments between 30 kg and 2.5 metric tonnes (t). The shipments are individually
labeled. Operations involve the consolidated transport of shipments of various ship-
pers on efficient transportation networks.108
With the information on LSPs and LTL operations in mind, the chapter revisits the
different measurements for transportation networks from the previous chapter. Ex-
pressive measurements of LTL networks are necessary to analyze the centrality of LTL
networks for network design. Measures of topology need not be fitted to the LTL con-
text. However, modifications for concentration metrics are necessary. Many similari-
ties exist between the LTL and the airline context. Therefore, the concentration mea-
surements can be derived from the existing metrics for airline networks, but some
peculiarities of LTL networks must be considered. The suggested metrics are used
further to analyze LTL network scenarios in chapter 6.
4.1. Logistics service providers

“Generally speaking, logistics service providers are companies which
perform logistics activities on behalf of others” (Delfmann et al., 2002,
pp. 203-204).
The following section presents background information on LSPs. Before doing so,
the markets for LSPs are briefly introduced with the aim to give a first idea how differ-
ent the markets are that LSPs operate on. The presentation of LSPs is intended to be
a basis for pointing out their role in LTL transport in the subsequent section.
108 The term groupage is sometimes used as a synonym for LTL transports.

95
4.1.1. Markets for logistics service providers

Klaus et al. (2009, pp. 74-77) suggest a segmentation of the entire LSP market into nine
segments; plus the additional segment of mail transport due to its rising importance
in public perception:
Bulk logistics is concerned with the transport of large shipments of over 25 t. Trans-
ported goods often comprise raw materials or waste in gas, liquid, or granular
form.
Full-truckload (FTL) & part load (PTL) transportation encompasses shipments

between 2.5 t and 25 t. Goods can be carried and stored in standard equip-
ment such as ISO-containers or swap bodies. This segment is restricted to land-
based transports.
General LTL networks transport (usually palletized) shipments between 30 kg and

2.5 t.
Specialized transportation encompasses all shipments that do not fit into any of

the other categories as they require special care. Heavy transports belong also
to this group as they require special equipment. Moreover, normal LTL cargo
requiring special conditions or unusual transport are in this segment, e.g. per-
ishable goods, hanging garments, and live animals.
Courier, express, parcel (CEP) covers shipments too small to be in the LTL seg-
ment. Hence, it refers in particular to small and time-critical shipments.
Contract logistics is the market for all customer-targeted, individual service offer-
ings by LSPs. It usually spans standard LSP operations in combination with
value-added services.
General warehousing and terminal operations refer to operating terminal infras-

tructure (such as sea- and airports) as well as conducting warehousing func-
tions independent from the contract logistics segment.
General ocean freight is the analog to the FTL segment above, but focused on
worldwide water-based transports and forwarding activities.
Air freight similarly encompasses worldwide air transports.
Mail & Postal services for printed materials and letter deliveries are not part of
the CEP segment and focus on the transport of letters below 2 kg and related
operations.
96
400
350
300
250
200
150
100 M k t size
Market i
50
0 thereof outsourced
Figure 4.1.: Sizes of European LSP markets in bn Euro following Klaus et al. (2009,
pp. 79-115).
Figure 4.1 exhibits the market sizes for the segments as listed above.109 The impor-
tance of the contract logistics segment, especially when considering the potential for
outsourcing in this segment, directly catches the eye. The LTL segment is small in
comparison, yet amongst those segments with a very high degree of outsourcing to
LSPs.
4.1.2. Categorizing logistics service providers

A wide scope of different service providers exists. Not surprisingly, the providers do
not compete with each other on all of the markets listed above. Specialists for certain
activities have developed. LSPs work together with various other partners who will
always influence the services provided by LSPs. Most importantly, LSPs are part of the
logistics triad, together with shippers and consignees.110 The following will introduce
three main criteria that serve to categorize different LSPs; namely the service offer,
the degree of customer orientation, and the role of assets.
Scope of service
LSPs differ in the services they offer.111 The service offer will, therefore, provide a
discriminating characteristic to distinguish LSPs.
Niebuer (1996, pp. 46-49) differentiates between the offering of isolated logistics
services, bundles of logistics services, and customized services. Isolated logistics ser-
109 The entire LSP market is estimated to 930 bn Euro, excluding the mail segment (Klaus et al., 2009, p. 77).
110 See section 2.1 for details.
111 Table 2.3 listed the top 25 European LSPs with their main services offered. These are distinct services
from the segmentation above or combinations of these.
97
vices are independent service offers such as only providing transport or warehousing.
Bundles of logistics services are the combination of independent services in order to
offer a broader service to a customer without actually having to adapt to every single
customer. Customized services are provided when the individual services are joined
to suit the demand of one specific customer.
Isolated services offered by LSPs comprise essentially transportation, transship-
ment, and warehousing. Very few additional services must be provided to complete
the service. For example, transportation administration is necessary for pure trans-
portation activities. Transshipment will most likely include some short-time ware-
housing. Operating a warehouse must include the administration as well as (un-
) loading activities.
Bundles of logistics services are an intermediate service offer between isolated and
customized services. Customized services involve at least one of the isolated services
mentioned above but go far beyond by offering a range of value added activities ac-
companying it. Examples are the inventory management at spare parts warehouses,
light assembly in warehouses, just-in-time (JIT) deliveries, and manufacturing activ-
ities.
Customer orientation
One major criterion to distinguish different LSPs from each other is the degree of ori-
entation to the needs of a specific customer. There is a direct relation between cus-
tomer orientation and the offered services. Some services explicitly address a large
number of customers and, thereby, cannot be customized for each of them. On the
contrary, very customized services are targeted to suit a single customer and will most
likely not be applicable to any other customer.
The distinction of services by Niebuer (1996) is revisited by Delfmann et al. (2002)
to identify three types of LSPs: standardizing LSPs offering isolated logistics services,
bundling LSPs offering bundles of logistics services, and customizing LSPs that pro-
vide customized services to their clients. The authors point out an implicit relation
between the services offered by an LSP and its degree of customer orientation, e.g. of-
fering customized services must encompass the provision of core logistics processes.
Figure 4.2 depicts the authors’ approach to classify LSPs based on these two dimen-
sions.
The role of assets

LSPs may be distinguished based on whether or not they own the assets needed for
their operations. Asset-based LSPs have the ownership over their truck fleet or ware-
houses. Usually, operating these assets at a high utilization is a major part of the
business of asset-based LSPs. Non asset-based providers (so called asset-free LSPs)
do not own the assets needed for the operations but rather subcontract the physical
98
Management support
Financial services
Customizing
LSP
nctions
Fun
Value added services
Bundling LSP
Core processes Standardizing

LSP
Degree of customization
Figure 4.2.: Relation of customization and functions for LSPs (Delfmann et al., 2002,
p. 207).
activities or set up joint-ventures for the operations. Asset-free service providers typ-
ically focus on management-related activities for their customers (Stefansson, 2006,
p. 132). A carrier is an LSP that offers transportation services (Bowersox et al., 2010,
p. 195). Carriers are asset-based providers and aim at efficiently utilizing their assets.
Freight forwarders and brokers are logistics agents that will provide services to their
customers but subcontract the actual doing to other partners, such as carriers (Bow-
ersox et al., 2010, p. 195). All degrees of asset ownership exist between asset-based
and asset-free providers. Freight forwarders often own only few assets themselves
and operate any additional demand with the help of subcontractors. This situation is
referred to as asset-light. A purely asset-free LSP is known as a fourth party logistics
(4PL) provider. It provides the management of all other LSPs necessary to run the en-
tire logistics services for the customer. A 4PL provider is a knowledge-based service
provider (Delfmann and Nikolova, 2002).
Types of LSPs
Different classification frameworks exist in the literature (e.g. Baumgarten and Thoms
(2002), Delfmann et al. (2002), Hertz and Alfredsson (2003), Stefansson (2006)). The
frameworks differ in details by the dimensions they apply to categorize LSPs. Never-
theless, they identify three large groups of LSPs that exist in the market. Even though
these groups may be split further into sub-groups to account for research specificities,
academics agree on the three main groups:
99
Carriers along with warehousing firms form the group of LSPs with a narrow focus
of isolated services (Delfmann et al., 2002, Hertz and Alfredsson, 2003) and di-
rectly related accompanying products (Stefansson, 2006). The customization
of products to specific customers is limited. The LSPs in this group are char-
acterized by asset ownership. Carriers usually offer origin-destination (OD)-
transports.
Third party logistics (3PL/TPL) providers provide bundles of services (Niebuer,

1996, p. 48) including value added services to customers. This group is some-
times referred to as LSPs (e.g. Stefansson (2006)), yet this viewpoint is not ap-
plied here. 3PL providers are often asset-light, subcontracting excess trans-
portation and warehousing demand to carriers and warehousing firms.
4PL providers - also known as lead logistics providers (LLPs) (Selviaridis and Spring,
2007) or logistics service intermediaries (LSIs) (Stefansson, 2006) - form a group
of mostly asset-free service providers whose core competences are in knowl-
edge, consulting, and coordination. It is often stated that their service is to co-
ordinate the 3PL providers involved in the operations (Delfmann and Nikolova,
2002, p. 424).
4.1.3. Research topics related to logistics service providers

Apart from literature reviews and classifications, three distinct aspects of LSPs have
undergone major research: relationships, strategy, and operations. Table 4.1 summa-
rizes selected literature contributions on LSPs. Firstly, LSPs have a very special role as
an intermediary partner between shippers and their customers or in an entire supply
chain. This position leads to research projects concerning the relationships between
the different partners and arising challenges and benefits. Horizontal cooperations
between different LSPs are also found in this category.
Secondly, the logistics industry is a growing industry that has seen public aware-
ness on this fact in the recent past.112 Researchers are interested in management
strategies of this developing and fast growing service industry.
Thirdly, LSPs provide efficient and customer-targeted operations. Therefore, qual-
itative and quantitative aspects of the services offered by LSPs creates many research
opportunities.
112 For instance, The Economist had A survey of logistics: The physical Internet on its front page on June
16, 2006. As similar examples, the German weekly Wirtschaftswoche featured Zukunftsbranche Logis-
tik - Die neue deutsche Jobmaschine (“Sunrise industry logistics - The new German job machine”) in
September 2005, McK Wissen 16 headlined Logistik (“logistics”) in November 2005 and brandeins titled
Die Problemlöser - Schwerpunkt Logistik (“The problem solvers - special feature on logistics”) in March
2010. None of these publications is scientific, but all are reputable for their well researched management
articles.
100
Operations
Relational
Strategy
Contribution
Berglund et al. (1999) x
Bruns et al. (2000) x
Murphy and Poist (2000) x
van Hoek (2000) x
Bolumole (2001) x
Larson and Gammelgaard (2001) x
Delfmann et al. (2002) x
Delfmann and Nikolova (2002) x
Zäpfel and Wasner (2002) x
Hertz and Alfredsson (2003) x
Lemoine and Danæs (2003) x
Halldórsson and Skjøtt-Larsen (2004) x
Makukha and Gray (2004) x
Panayides (2004) x
Carbone and Stone (2005) x
Gopalakrishnan and Johnson (2005) x
Knemeyer and Murphy (2005) x
Panayides and So (2005) x
Halldórsson and Skjøtt-Larsen (2006) x
Hoi Yan Yeung et al. (2006) x
Markides and Holweg (2006) x
Stefansson (2006) x
Cruijssen et al. (2007) x
Davis et al. (2008) x
Juga et al. (2008) x
Mortensen and Lemoine (2008) x
Song et al. (2008) x
Erera et al. (2008) x
Andersen et al. (2009) x
Derigs et al. (2009) x
Taylor and Whicker (2010) x
Klaas-Wissing and Albers (2010) x
Özener et al. (forthcoming) x
Zhou et al. (2011) x
Table 4.1.: LSP literature contributions.
101
Extensive literature reviews are available that give a complete picture of the re-
search concerning LSPs. Marasco (2008) provides an overview of LSP research, but fo-
cuses on strategic and relational aspects. The review by Selviaridis and Spring (2007)
develops a taxonomy of LSP research and a future research agenda for the field.
4.2. Less-than-truckload transportation

Based on the general introduction of LSPs, this section introduces their operations on
the European LTL market, the market in terms of numbers and players, the provided
services to customers, and recent developments concerning the LTL demand on the
market.
4.2.1. Less-than-truckload characteristics

The LTL business can be described based on several characteristics. Klaas-Wissing
and Albers (2010) list seven characteristics to distinguish the LTL business from other
transport sectors.
• Heterogeneous shipments between about 30 kg and 2.5 t,
• Multi-user dedication by offering open systems for many customers,
• Efficient and easy to compare service offers (see below),
• Many-to-many network,
• Complete coverage of a geographical area,
• Hub-&-spoke (H&S) network,
• Standardized products and operations (see below).
These characteristics illustrate that the transport in an LTL network is in-between

isolated services and bundles of logistics services (cf. figure 4.2). The product is more
than the mere OD-transport as it involves coordinating and efficiently operating an
entire network. The LTL network must combine shipments from many customers
to efficiently operate the many-to-many network. Hence, customized services to all
shippers are infeasible. Some examples exist where the network is customized to
one specific customer,113 and some more intense partnerships, including additional
value-added services, may be based on the LTL network operated by an LSP.
LTL networks allow for the efficient and effective transport of shipments. The focus
of the following presentation is on system-based LTL networks that operate on regular
113 For instance, DHL provides extensive services to BMW in Central Europe since 2005 and actually modi-
fied parts of the LTL network for this partnership (Ernst, 2005).
102
schedules. These networks are common in Europe. Some more spread-out countries,
especially the United States of America (US) operate different networks to reflect e.g.
the different geographical situation. These will be briefly touched on where applica-
ble. Erera et al. (2008, pp. 3398-3399) give a comprehensive introduction to the US
LTL system.
4.2.2. Less-than-truckload operations

An LTL network usually has two echelons. Firstly, the end-of-line (EoL) terminals for
a city or region that are the bases for local pick-up and delivery (P&D) trucks and,
secondly, transshipment facilities (or break-bulk (BB) terminals) where shipments
are consolidated and transshipped (Hall and Zhong, 2002, Cheung and Muralidha-
ran, 2000, Powell, 1986). It may be the case that this distinction is merely organiza-
tional and some EoL terminals act as transshipment terminals in addition to their
role as EoL terminal. Several operational routing schemes between EoL terminals are
used in an LTL network, as seen in figure 4.3.114 A direct transport link, as depicted
in scheme (i), is implemented if the volume of shipments between origin and des-
tination is sufficiently high (Cheung and Muralidharan, 2000). Schemes (ii) and (iii)
make use of one transshipment facility either near the origin or the destination ter-
minal. It is also possible to route shipments through two transshipment facilities (iv)
to increase the loads on the trucks by consolidating shipments further. This routing
is referred to as 2-stop routing, as it involves two stops at transshipment facilities.
The volume or weight of shipments from an origin to a destination will determine the
routing scheme that is used. All of these routing schemes are routinely implemented
in European LTL networks.115
It does happen in practice that shipments are injected into the network directly
(direct injection) or delivered out of the network directly (direct delivery). A direct in-
jection is a transport directly from the sender’s site to a transshipment terminal or
even a destination EoL terminal. The origin EoL terminal is bypassed. This makes
sense if one customer or very few, closely located customers send off enough ship-
ment volume to the transshipment or EoL terminal to justify an entire vehicle. The
114 Local P&D of shipments is omitted from figure 4.3 for the sake of simplicity; all shipments are depicted
as if they had their origin respectively destination at an EoL terminal, which will be the viewpoint for the
case study in chapter 6 as well.
115 Some variations to this scheme exist in practice to solve operational challenges. These are additions
to the general procedure outlined above, but of minor importance for the general operations. Exam-
ples are stops at intermediate EoL terminals to fill empty capacity, vehicles from opposite directions
meeting each other to exchange trailers so that every driver will take the other trailer back home (truck-
meets-truck traffic), various kinds of additional emergency transports in case of important late arrivals
at terminals, etc.
103
(i)
Transshipment terminal
(ii)
EoL terminal
(iii)
((iv))
Origin Destination
Figure 4.3.: Routing schemes in LTL operations.
handling operations at the EoL terminal can be spared. A direct delivery is the analog
procedure bypassing the destination EoL terminal.116
The networks operate on standardized schedules, typically highly repetitive. The
schedules define when vehicles have to leave a terminal. Cut-off times exist: trans-
shipment and onward transport of shipments is only possible if the arriving vehicle
with the shipments is at the terminal before a certain deadline.
The timing in LTL operations is standardized as well. Shippers inform the LSP of
shipments during the day. Local pick-up operations bring shipments to be trans-
ported to an EoL terminal in the afternoon or early evening. Vehicles leave the termi-
nals at different times during the evening. Early vehicles might stop at an additional
EoL terminal on their trip and pick up shipments there. All vehicles are scheduled to
arrive at their destination early the next morning. The local delivery takes place in the
morning. P&D may be combined, so delivering shipments to a customer might at the
same time be the pick-up of new shipments for the evening.
European LTL networks tend to be denser than the US American operations. The
routing schemes depicted in figure 4.3 will hold for typical European operations, as 1-
and 2-stop routings suffice to increase fill rates between the terminals to an accept-
able level. In contrast, US LTL transportation has to bridge longer distances between
the major commercial areas of the states. The setting is different, so operations will
differ.117
116 Erera et al. (2008, p. 3399) use direct load to describe the same idea. Cheung and Muralidharan (2000)
introduce the term opportunistic directs for transports that were not planned but make sense when the
operations take place. This is related to the idea of direct injection / delivery, however, these are usually
planned operations.
117 It is interesting to see that Cheung and Muralidharan (2000) use as a small example of an LTL load plan a
sketch of a network with three BB terminals. Giordano (2008) highlights that warehouses for (un-) load-
ing are necessary for US LTL networks. BB terminals in European networks tend to carry very little in-
ventory as schedules are set tightly to transport shipments off quickly.
104
4.2.3. European less-than-truckload market

The European LTL market size in 2008 is estimated to 38 bn Euro.118 90% of the market
is outsourced to LSPs. The market share of the top 10 LSPs in the outsourced part of
the market is only of 42%. Furthermore, there is no dominant player on the market.
These numbers hint at the high competition on the market (Klaus et al., 2009, p. 87).
The European market for LTL transports is not homogeneous. The German market
is known to have dense transportation networks and a strong international orienta-
tion in comparison to other countries, such as France (Kloss, 2010). South- and East-
European markets are overcoming their deficiencies in density, delivery times, and
technology (Klaus and Kille, 2007, p. 90). The market is comparably small in some
countries (for instance United Kingdom of Great Britain and Northern Ireland (UK),
Italy, and Spain), as providers from the CEP and the PTL / FTL markets enter and
provide services in the typical LTL range (Klaus et al., 2009, p. 88).
The German LTL market is characterized by three main groups of providers: large
corporations, cooperations, and small and medium size enterprises (SMEs) (Koha-
gen, 2010). Individual SMEs face the challenge to provide geographical coverage for
their customers. Thus, many of them started to form LTL cooperative networks. While
large corporations are able to control their geographically scattered terminals from an
organizational central unit, the network’s organization is more difficult for coopera-
tive networks (Klaas-Wissing and Albers, 2010, Albers, 2009).
Table 4.2 lists the top 10 LSPs on the European LTL market. All top 10 LSPs in the
LTL market are 3PL providers.119 The major cooperations on the market are included
for comparison with their accumulated revenues. DSV as well as Kühne & Nagel are
partners of the cooperation IDS Logistik GmbH for the German market but have fur-
ther LTL business as well.
One finds international, integrated LSPs among the leaders in the LTL segment.
Furthermore, international freight forwarders and cooperations of SMEs (all from
Germany) are able to generate high revenues in the segment.
The players operate fully integrated, distinct networks. About 12 of these networks
exist in Germany alone (Klaus et al., 2009, p. 87). A new LTL network that targets at air
transport providers by operating airport-to-airport connections has been announced
in 2010 (Cremer, 2010, Lauenroth, 2010b). An additional network with a strong fo-
cus on concentrated transports via one transshipment terminal is expected to start
operations in Germany in early 2011 (Lauenroth, 2010a).
4.2.4. Services accompanying less-than-truckload transports

The service offered by LSPs on the LTL market is, for most part, standardized and
interchangeable between the providers. This is a very difficult situation for the LSPs
118 The LTL market makes up 4% of the entire market for logistics services, as shown in figure 4.1.
119 The LTL subsidiaries of corporations are listed in addition where applicable.
105
Rank LSP Revenue

1 Deutsche Bahn Group 3,200
DB Schenker Logistics 3,200
2 Dachser 2,413
3 Deutsche Post DHL 2,200
4 SCNF 1,800
Geodis 1,800
5 Kühne & Nagel 1,417
6 DSV 1,300
7 Hellmann Worldwide 705
8 Mory Group SA (F) 610
IDS Logistik GmbH (Cooperation) 572
System Alliance (Cooperation) 560
CargoLine GmbH (Cooperation) 530
24plus Systemverkehre (Cooperation) 455
9 Heppner SA (F) 428
10 Ziegler Group 400
Table 4.2.: Top 10 LSPs in European LTL by LTL revenue 2008 in Europe (in m. Euro)
(Klaus et al., 2009, p. 89).
as differentiation from competitors is difficult.

Klaas-Wissing and Albers (2010) summarize that, in order to be successful on the
LTL market, an LSP has to operate a cost efficient network on a national and mostly
international scale, provide high density geographical coverage, and respond to cus-
tomer demand in terms of reliability, responsiveness, timeliness, speed, and techno-
logical support. Industry specialists continuously point out that customer satisfac-
tion must be reached for successful LTL operations as the competition in the market
is high. This can be achieved by the aspects mentioned above, especially the interna-
tional coverage and competitive pricing (Kohagen, 2010, Jüngst, 2008). Crainic (2000)
highlights that high quality, high speed, and low prices are important factors for suc-
cess in the LTL market. It is often mentioned that the relationship between shipper
and LSP is important in the LTL business, even though the operations are standard-
ized in many ways. For example, SME shippers typically prefer SME LSPs as their
partners (Lauenroth, 2006). Klaus et al. (2009, pp. 89-90) indicate that geographical
coverage, reliable delivery times, and track & trace abilities are the most important
success factors on the market from their point of view.
All major European LTL providers provide a full geographical coverage. Networks
tend to be denser in central Europe than in remote areas, as network density is con-
nected to demand to or from a region. Corporate network providers usually operate
their own networks on a European scale, while the German cooperations operate their
106
4.3. Measuring less-than-truckload network concentration
own networks with the partners’ terminals in Germany and have international part-
ners to cover Europe. Extra-European deliveries remain exceptional as of now (Bollig,
2009).
Standard delivery times in the LTL market are 24 or 48 hours nationally or to neigh-
boring countries. In general these are promised delivery times that the LSPs meet with
high service levels. In contrast, guaranteed delivery times with penalties are special
services offered by virtually all providers.
Information technology (IT) services such as track & trace are provided by most
LSPs. Electronic data exchange is no longer a challenge.120
Many of the above mentioned services were valuable to differentiate an LSP from
its competitors in the past. For example, track & trace was innovative some years
back, while nowadays customers expect this service as a basic requirement. Low cost,
fast delivery times, IT infrastructure, etc. have become hygiene factors in the market.
LSPs must strive to identify other sources of differentiation in this highly competitive
market. The scope of services offered is constantly widened. LSPs offer more specific
value-added services with their LTL products.
Another trend currently perceived by LSPs is that customers do not ask for faster
transports, but rate reliability, flexibility, and stable transportation times higher.121 It
is a clear sign that LSPs on the LTL market strive to adapt to new customer demand in
order to obtain a competitive advantage.

This section revisits indices presented in sections 3.3 and 3.4 and continues to de-
velop expressive indices to measure the centrality of LTL networks. Centrality indices
for airline networks were presented in depth in section 3.4.2. These will serve as the
foundation for developing metrics for the LTL context.
The presentation on airline network centrality revealed that metrics for network
topology can be directly applied for transportation networks due to their abstraction
from the underlying network to a graph theoretic representation. The development of
concentration metrics is not independent from the network since the routing through
the network and the resulting traffic flows are analyzed. Therefore, concentration in-
dices must be adapted to fit the LTL context. This is pursued based on airline network
concentration metrics. Similarities between airline and LTL networks will firstly be
pointed out. One key difference between the networks is that passenger airline traffic
is usually assumed to be balanced whereas traffic flows are imbalanced in cargo trans-
port in general, and in LTL transportation in particular. To account for this, measures
120 Tables A.1 and A.2 in the appendix highlight service offers by the main players on the European LTL
market.
121 One specific form of transportation time reliability was the goal for the case study to be presented in
chapter 6 and drove the algorithm development in chapter 5. Detailed explanation on the understanding
of stable transportation times as reliable delivery times is presented in chapter 5.
107
for traffic and traffic shares must be modified to suit the LTL context. Based on these
preliminary considerations, the section will suggest modifications to the McShan-
Windle (MW) index, the network concentration (NC) index, and the hubbing concen-
tration (HC) index to measure LTL network concentration. These indices were chosen
for different reasons. The MW index is one of the fundamental indices that provides
information without manipulating the underlying data strongly. It is of particular in-
terest since it is very intuitive in its interpretation. The NC index is the normalization
of the Gini index. It was selected due to its high influence in current literature con-
tributions. Lastly, the HC index was chosen because it is a recently developed index
that considers the relevant role of hubbing. To the best of my knowledge, no other
concentration index exists that depicts hubbing as expressively as the HC index. LTL
networks rely strongly on transshipment operations and the HC index promises to
provide meaningful information.
4.3.1. Similarities of airline and less-than-truckload networks

Airline networks are constructed from airports and flight routes. From a structural
point of view, airline networks can be placed in a continuum between point-to-point
(P2P) networks and H&S networks (Wojahn, 2001, p. 23). Derudder and Witlox (2009)
emphasize that many airlines do not rely on one central hub, but have several hubs
in their network. Furthermore, the authors identify the current trend towards more
P2P networks, which is on one hand driven by the recent rise of low cost carriers
(LCCs), and on the other hand supported by the perception within the industry that
customers prefer direct flights. Hybrid multi-hub-&-spoke (HMH&S) networks are a
likely outcome of these trends.
An LTL network consists of terminals and truck routes. The transported shipment
volume or weight can be interpreted as the intensity on a route. The network spanned
by the EoL terminals and the transshipment terminals is set up similarly to an airline
network and often takes the form of a HMH&S network.
Airline and LTL networks have many characteristics in common. Their structural
similarities have just been outlined. Furthermore, both are characterized as many-
to-many networks and operate in highly competitive market environments. Efficient
and reliable transports at a competitive price are important factors for success in the
LTL market just as in the aviation sector (Doganis, 2010, p. 23). Airline and LTL net-
works both need to be designed for efficiency and customer orientation.
That said, there are key differences between these networks as well. In general,
LTL networks tend to have far more direct routes between EoL terminals than a full
service carrier (FSC) airline network. This is due to the different types of goods trans-
ported through the network: passengers and shipments.122 As airline passengers are
sensitive to waiting time and expect high service frequencies, airlines are forced to
122 The differences between transporting passengers and cargo were discussed in detail in section 2.2.1.
Some relevant points for comparing air passenger and LTL will be brought to attention again.
108
make use of transshipment (hub) airports. In contrast, shipments can often be con-
solidated over time so that higher shipment volumes are accumulated, which allows
for direct transport links in LTL networks.
Another important difference that stems from transporting passengers rather than
shipments is that passengers have individual preferences, whereas shipments do not.
Passengers may choose certain routings on their air journey for a variety of personal
reasons. Shipments in an LTL network should always be routed according to prede-
fined planning.
Furthermore, passenger traffic flows are usually assumed to be balanced, as most
passengers return home after their trip. Hence, incoming and outgoing passenger
traffic flows are equal (Burghouwt et al., 2003). This is a major difference when com-
pared to air cargo (Doganis, 2010, p. 298), as well as cargo in general (Rodrigue et al.,
2009, p. 152), and LTL shipments in particular. Such traffic flows are characterized by
high imbalances of traffic.
Airline and LTL networks share many properties, which suggests to use similar in-
dices to measure their concentration. However, as the networks differ in some as-
pects, modifications to the afore presented indices are necessary. The (im-) balance
of traffic flows will prove to be a major point of modification. To account for this, the
definition of traffic flows must be adapted from airline networks to the context of LTL
transportation. Secondly, 2-stop routings for LTL transports must be considered. Af-
ter highlighting these two aspects, the section will continue to suggest the MW index,
the NC index, the HC index, and a variant of the HC index for measuring LTL network
concentration.
4.3.2. Traffic flows in less-than-truckload networks

As pointed out earlier, airline traffic is usually measured in terms of enplanements
(see section 3.4.2). The idea to measure only outgoing flows is valid as passenger air
transport flows are usually assumed to be balanced. The LTL context shows strong
imbalances of traffic. Traffic at an LTL terminal shall be defined as the sum of incom-
ing and outgoing transportation volume.123
2-stop routings (figure 4.3 (iv)) are often used in the airline context, as most travel-
ers will agree.124 Nevertheless, they are not explicitly mentioned in the literature on
measuring airline network concentration. This may be related to the datasets used by
the authors: it would be surprising if they held the necessary information. Productive
datasets available to LSPs operating LTL networks, on the contrary, contain detailed
routing information. Therefore, modeling 2-stop routing for the centrality indices is
feasible. The necessary modification concerning 2-stop routings will differ between
the indices and is discussed in the following when applicable.
123 More operatively, shipments in the case study in chapter 6 are measured in terms of pay weight.
124 Woxenius (2007, p. 738) presents the example of air travel between two minor European cities that will
be connected through two hubs.
109
EoL Traffic share

EoL1 0.019
EoL2 0.018
EoL3 0.018
MW index value 0.055
Table 4.3.: MW index example.
4.3.3. McShan-Windle index

There is no need to modify the procedure for calculating the MW index once it was
defined how the traffic and traffic shares in an LTL network are to be understood.
Equation 3.10 is directly applicable.
The MW index will be the sum of the traffic shares of the top 3% terminals (in terms
traffic) in the entire network. Even though highly probable, it needs not be the case
that all top 3% terminals in the network are transshipment terminals. As outlined
above, traffic at a terminal is the sum of incoming and outgoing transportation vol-
ume. Thus, a terminal with very high origin or destination traffic may well be within
the top 3% terminals.
Figure 4.4(a) is a randomly generated distribution of traffic shares for 100 EoL ter-
minals. It serves as an example to demonstrate the calculation of the MW index value.
In a network of 100 nodes, m = 0.03 ∗ 100 = 3. The MW index equals the sum of traffic
of the 3% EoL terminals with the highest traffic shares. Table 4.3 lists the values. The
MW index value for the example is of 0.055.125 The network in the example has a low
degree of concentration: the traffic shares graphically depicted in figure 4.4(a) can be
interpreted similarly.
It is easy and quick to calculate the MW index of a transportation network, which
might be its strongest advantage. Then again, the information obtained by the MW
index is limited since it neglects all information about the remaining 97% of termi-
nals in the network. For example, the network with the traffic shares as shown in
figure 4.4(b) has the same MW index value as the one in figure 4.4(a), even though the
distribution of traffic is different. Nevertheless, the MW index gives a first impression
of network concentration.
4.3.4. Network concentration index

The maximum of the Gini index (equation 3.12) for airline networks is limited due
to the balanced traffic flows. No airport commands more than half the traffic in the
125 Section 3.4.2 presented that the MW index varies from 0.03 (not concentrated network, all nodes have
equal traffic) to 1 (concentrated network).
110
0.020
0.018
0.016
0.014
traffic share
0.012
0.010
0.008
0.006
0.004
0.002
0.000
EoL terminals in decreasing order of traffic share
(a)
0.020
0.018
0.016
0.014
traffic share
0.012
0.010
0.008
0.006
0.004
0.002
0.000
EoL terminals in decreasing order of traffic share
(b)
Figure 4.4.: MW index example.
network (Burghouwt et al., 2003); otherwise, incoming and outgoing flows could not
equal out. This argument is not valid for the LTL case. Nonetheless, since traffic in
the LTL case is counted at the outbound as well as at the inbound side, there are al-
ways two terminals that share traffic. Therefore, any single terminal can accumulate
at most half of the total traffic in the network. This leads to the same conclusion
as derived for the airline case, namely, that the Gini index will never reach its theo-
retical maximum and the NC index should be used instead in order to simplify the
interpretation. To measure network concentration, the NC index will be applied as
introduced.126 The necessary modification is the definition of traffic that was given
above.
Figure 4.5 shows nine networks that will serve to illustrate some values of the NC
index. Nodes A - F represent EoL terminals. They send and receive shipments; node
A acts as a transshipment terminal. Flow is counted in both directions. The arcs rep-
resent routes between the terminals. The shipments sent on each link are displayed
126 The NC index for airline transportation was introduced in section 3.4.2.
111
F 1 B F B F 1 B
1/6 1/6 1/10 1/10 1/6 1/6
1 1 1 1
A 1 C A 1 C A 1 C
1/6 1/6 1/2 1/10 5/12 1/12
1 1 1 1
E 1 D E D E D
1/6 1/6 1/10 1/10 1/12 1/12
(a) NC index = 0 (b) NC index = 1/2 (c) NC index = 1/2
F 1 B F 10 B F 11 B
1/2 1/2 11/30 11/30 1/6 1/6
1 1 1 1
A C A 1 C A 10 C
0 0 5/30 1/30 7/36 5/36
1 1 1 1
E D E D E 11 D
0 0 1/30 1/30 1/6 1/6
(d) NC index = 1 (e) NC index = 7/10 (f) NC index = 5/108
F 1 B F 1 B F B
2/14 2/14 1/9 1/6 1/86 10/43
1 1 1 1 1 1 10 10
A 1 C A 1 C A C
5/14 1/14 5/18 1/6 11/43 10/43
1 1 1 1 1 1 10 10
E 1 D E 1 D E 1 D
2/14 2/14 1/9 1/6 1/43 21/86
(g) NC index = 5/14 (h) NC index = 1/4 (i) NC index = 81/172
Figure 4.5.: NC index example.
next to the links.127 The traffic share related to each terminal is noted in the circle
representing the terminal. The NC index value for each network is below each graph.
Cases (a) - (d) show some extreme NC index values for the given network. The NC
index is equal to 0 if all nodes have the same traffic in the network; this is the case
of the lowly concentrated network in case (a). An NC index value of 0.5 is related
to a perfect star network (as case (b) displays), but other situations may lead to the
same NC index value as well; case (c) is such an example. A network where all traffic
is shared between two terminals, while all other terminals have no traffic at all (case
(d)), will lead to an NC index value of 1: a concentrated network. Cases (e) and (f)
underline how some aspects in a network may outweigh others. Case (e) is a star with
one strong arc and is therefore characterized by an NC index value between these two
types. Further, case (f ) possesses three strong routes with some additional star-type
traffic, therefore the NC index value of this network is low.
127 The direction of flow is neglectable as both, incoming and outgoing flows, are accounted for.
112
A A
20/42 22/44
10 10 11 11
B 1 C B C
11/42 11/42 11/44 11/44
(a) NC index = 0.429 (b) NC index = 0.500
Figure 4.6.: Example of NC index values in markets with concentrated demand.
Cases (g) - (i) are three networks where node C may assume the role of an addi-
tional transshipment terminal. This does not have an impact on the NC index, as for
instance case (i) highlights. The depicted network does not provide information con-
cerning the flow from B to C and from C to D. Whether the transports are independent
of each other or whether they are actually only one transport from B to D that is trans-
shipped at C does not make a difference for the NC index. It does not need the infor-
mation on which node in the network acts as a transshipment terminal. No structural
difference exists between single H&S networks or multi-hub-&-spoke (MH&S) net-
works.
To take 2-stop routings into account for the NC index, the additional transshipment
is treated as another unloading and loading operation at the second transshipment
terminal; it therefore increases the total traffic in the network.
The suggested modification in traffic measurement for the NC index is straightfor-
ward. The examples above show that the calculation and the interpretation of the
index remains as easy as for the case of balanced flows.
The modified NC index values are strongly related to the demand concentration
in the networks. Figure 4.6 shows two NC index values for network examples. Case
(a) and case (b) have 10 units of demand from B to A and from A to C. There is a
demand of 1 unit from B to C. The cases differ in the routing of the flow from node B to
node C. The difference in the NC index value between the networks is relatively small,
even though the networks differ in that case (a) does not carry out transshipment
operations, whereas case (b) does. Still, the transshipment traffic directed to node A
is limited compared to the already existing direct flows. It appears as though the NC
index is impacted by the demand concentration for the network and only partially
by the routing of flows. The OD-demand is truly exogenous to any LSP (Jara-Díaz
and Basso, 2003, p. 285). In contrast, the routing of flows is at the disposition of the
planner. Whether or not this is a general deficiency of the NC index when applied to
markets with imbalanced traffic flows cannot be answered in the scope of this thesis,
but should be in the focus of further research in cargo transportation networks.
113
hub1
hub1a hub1b
s1 = 0.75
origin destination
s2 = 0.25
hub2
(a) HC index 2-stop routing.
hub1
s1 = 0.75
origin destination
s2 = 0.25
hub2
(b) HC index 2-stop routing, double in-

volvement.
Figure 4.7.: HC index example.
4.3.5. Hubbing concentration index

Measuring traffic is easier for the calculation of the HC index than for the NC index
since the focus of the HC index is always market-oriented. It is sufficient to identify
for each OD-market the share of traffic that flows on any possible routing option. To
do so, it is important to know the initial origin of any shipment and its final destina-
tion. In passenger airline networks it is often difficult to keep track of overall flight
itineraries, but for the LTL context this data is usually available to the LSPs.
The key modification to the HC index is related to 2-stop routings, which are im-
portant in LTL networks. Martín and Voltes-Dorta (2008, p. 179) give meaningful ex-
amples on the intuition behind and the calculation of the HC index for 1-stop cases,
but do not explicitly refer to 2-stop routings. 2-stop routings are incorporated into the
modified HC index with the help of virtual hubs.
Figure 4.7(a) shows an example of how 2-stop routings are accounted for. When-
ever there is a shipment that uses two consolidation terminals on its way, these two
terminals are treated as one virtual hub. Thus, hubs hu b 1a and hu b 1b together make
up the virtual hub hu b 1 for the path from origin to destination and account for a traf-
fic share of 75%. Further calculation of the HC index then strictly follows the proce-
dure provided by Martín and Voltes-Dorta (2008). The HC index value of this example
is of 0.625 (H = 1∗(0.752 +0.252 ) = 0.625). The idea of virtual hubs is easily extended to
x-stop cases; one virtual hub represents one possible path that shipments take from
origin to destination.
114
One needs to be aware that the suggested modification has one major drawback:
it is not possible to assess the importance of a single hub. Figure 4.7(b) presents a
slightly modified version of the situation shown in figure 4.7(a). Hu b 1 and hu b 2 now
both are virtual hubs and share the common gray terminal in the upper left corner.
Hu b 1 and hu b 2 have a traffic share of 0.75 and 0.25 respectively. The hubbing be-
havior H for this case is of 0.625 as well. Still, all the traffic from origin to destination
is channeled through the gray hub. Hence, one could argue that there are reasons to
expect H to be equal to 1. Therefore, the modification with virtual hubs could also be
interpreted as misleading.
Nevertheless, it is believed that this modification extends the HC index in an ac-
ceptable manner. The HC index was not formulated to compare the relevance of cer-
tain hubs in a given network; it was designed to provide a measure of the relevance of
hubbing in a network and the modification follows that concept.
Martín and Voltes-Dorta (2009) point out that the share of connecting passengers is
highly correlated with the HC index. The authors compare HC index and connecting
share. The ratio of the HC index over the connecting share is, therefore, interesting to
take an additional look at. The ratio is defined as:
HC
r a t io = C ∗q (4.1)
ij ij
i j Q
The ratio of HC index over connecting share for the entire network measures the
degree to which only the actually transshipped shipments in the network have differ-
ent routing options through the network. In other words, the ratio shows on average
how strongly the transshipped shipments on one OD-market use the same routing
through the network.
4.3.6. Comparing the indices

Figure 4.8 depicts networks with different degrees of topological centrality as well as
variations in their traffic flows. Table 4.4 summarizes and contrasts the values for
the four above introduced concentration indices. It allows to illustrate several points
about them.128
• All MW index values are low due to the small number of nodes in the networks.
• The fully-meshed networks in cases (a), (b), and (c) have different MW index
values since the traffic share of node E (as the node with the highest traffic
share) varies in the networks.
128 In addition to the information given in figure 4.8 an assumption is made to allow to calculate the HC
index. The indirect traffic flow from node A to node D as well as from nod B to node C is of 1 unit in
networks (a), (b), (d), and (e). It is transshipped at the central node. Moreover, in network (f), the indirect
flow from A to D is again of 1 unit, while the indirect flow from B to C is of 100 units. Once more, node E
acts as a transshipment terminal.
115
1 100 0
A B A B A B
1 1 1 100 0 0
1 1 E 1 1 1 E 100 0 0 E 0
1 1 1 1 100 1 0 0 0
C D C D C D
1 1 0
(a) (b) (c)
A B A B A B
1 1 1 100 1 100
E E E
1 1 1 1 100 1
C D C D C D
(d) (e) (f)
1 1 100
A B A B
A B 1 1 1 100
C D C D
1 1 1
(g) (h) (i)
Figure 4.8.: Examples of networks to illustrate centrality index values.
• The star-shaped networks in cases (d), (e), and (f) have only one option for
a routing path for transshipped flows. Therefore, the ratio of HC index over
connecting share is equal to 1.
• The networks without any transshipment operation (network (g), (h), and (i))
have HC index values of 0 and a connecting share of 0 as well. This prohibits to
calculate the ratio of the two.
• The NC index can only be computed for networks with more than two nodes.
The Gini index of network (g) is of 0. This is true for all 2-node networks where
traffic is accounted for at the origin and the destination of the traffic flow. All
flow takes place between the two nodes; they must have the identical share of
traffic in the network.
• None of the nine networks has the same index values for all four indices as all
indices assume a different perspective on network centrality.
116
HC index
Case MW index NC index HC index connecting share
4.8(a) 0.030 0.000 0.125 0.500
4.8(b) 0.074 0.488 0.004 0.750
4.8(c) 0.000 0.000 - -
4.8(d) 0.075 0.500 1.000 1.000
4.8(e) 0.075 0.981 0.020 1.000
4.8(f ) 0.075 0.827 1.000 1.000
4.8(g) 0.030 - 0.000 -
4.8(h) 0.030 0.000 0.000 -
4.8(i) 0.059 0.735 0.000 -
Table 4.4.: Values of concentration indices of networks depicted in figure 4.8.
The MW index is very easy to understand, even for the unexperienced, and it gives
quick information about the prominence of few major terminals in the network. The
NC index highlights the concentration in demand as well as the additional concen-
tration generated by routing shipments in the network. By doing so, it hints at how
strongly an LSP depends on the market demand that it can only partially influence.
Neither of the two indices, MW index nor NC index, requires additional informa-
tion about the networks but the share of traffic at each node. The information if a
node acts as a transshipment terminal is not relevant for the indices.
The HC index is clearly aimed at showing how important hubbing is in the network.
The ratio of HC index over connecting share provides the additional information on
how strongly the OD-market specific flows depend on few routing options. This in-
formation can be interpreted as a form of risk in terms of dependence on a transship-
ment terminal. It is important to have the entire routing of all shipments at hand in
order to calculate the HC index. Information about single links in the network is not
sufficient for this index.
These four perspectives give relevant insights on an aggregated level into the net-
work’s concentration. This can be related to likely network performance. The ob-
tained information serves as a first comparison between networks and may guide
further evaluation.
When interpreting the values obtained from the indices, one must have in mind
that each index assumes its specific viewpoint on network concentration. Further-
more, none of the indices can be seen as the superior index for measuring network
concentration, as for each of them drawbacks were pointed out. Nevertheless, taken
together, the breadth of information about the network concentration allows to pic-
ture the situation.
117
4.4. Relevance of measuring less-than-truckload network

concentration for network design
This chapter started by introducing LSPs and the LTL market. LTL transports are con-
ducted in a competitive environment where efficient and effective many-to-many
transportation networks provide a competitive advantage. Hence, network design for
LTL transports is of importance for LSPs operating on the market. Offering specific,
demand-oriented services allows an LSP to differentiate itself from its competitors.
Knowledge of network centrality supports transportation network design. In order to
measure the concentration of LTL networks, context-specific modifications are nec-
essary. Accordingly, the chapter continued to suggest indices to measure the central-
ity - or more precisely the concentration - of LTL networks, based on the existing liter-
ature on centrality metrics presented in chapter 3: the MW index, the NC index, and
the HC index. The ratio of HC index over connecting share is the additional, fourth,
metric. Their application to LTL networks will be shown in the following chapters.
As the modification of indices was based on the background information for the
LTL context, they are applicable to the simulated networks. Other applications to
networks with similar characteristics can be easily pictured.
Chapter 5 will continue by presenting an algorithm for LTL network design that
generates and tests different network scenarios. Chapter 6 illustrates the applica-
tion of the algorithm in a project on redesigning an LTL network as well as outline
extensive simulations on network performance in general. For this purpose, several
network scenarios were simulated and data was accumulated concerning cost and
delivery times, network utilization, and network centrality.
The suggested metrics for network centrality as presented in this chapter will be
used to operationalize network centrality with the goal to compare the simulated
networks. This opens a new facet to the evaluation of (simulated) transportation net-
works as it allows for two things: firstly, service-oriented aspects gain in importance as
compared to the often used cost-oriented metrics. Secondly, centrality is structurally
related to many different aspects of network performance and can be connected to
various network KPIs. Therefore, measuring network centrality provides additional
information on many of these.
118
5. An algorithm for less-than-truckload
network design
This chapter describes an algorithm that generates potential less-than-truckload

(LTL) networks on a heuristic basis. The algorithm creates LTL networks based on
end-of-line (EoL) terminal locations, potential hubs, and given demand forecasts. It
then simulates the transport of shipments on the generated networks while keeping
track of key performance indicators (KPIs). Specifying different settings of potential
hubs allows to simulate scenarios for comparison.
The algorithm is specific to LTL operations in that it generates networks that fulfill
desired performance for the European LTL market. The market environment as pre-
sented in section 4.2 is reflected in the routing decisions modeled by the algorithm.
Moreover, the algorithm aims at respecting the fact that both, cost and service, need
to be balanced for an LTL network so that it will be efficient and effective at the same
time.
Knowledge of transportation network centrality and how it relates to network per-
formance was fundamental for the development of the algorithm. Assessing the LTL
market situation and LTL transport characteristics led to the desired network perfor-
mance. Based on the conceptual relations devised in chapter 3, it was decided what
kind of network appearance will most likely fulfill this network performance. The al-
gorithm was then developed to generate these networks and compare scenarios.
The presentation of the algorithm is structured in three steps. First of all, the gen-
eral reasoning behind the algorithm is exhibited; the concept of network centrality
being a crucial part of it. This supports the understanding of why the algorithm was
designed in the presented way. Secondly, the input available to the algorithm as well
as the output expected from it, are presented. The algorithm is designed so that the
input should be easily available for logistics service providers (LSPs) that operate net-
works on the LTL market. The output consists of generated networks, shipment rout-
ings, and network KPIs gathered by simulating shipment transportation on the net-
works. The algorithm supports long-term decision-making on the appearance of the
transportation network by depicting the impact of modifications within the network
on its KPIs. The chapter then continues to show in-depth the single steps of the algo-
rithm; this includes all decisions taken by the algorithm concerning how to construct
the network and how to route shipments through it. The background on how exactly
the algorithm is designed allows to discuss and evaluate the obtained results for the
case study presented in chapter 6.

119
5. An algorithm for less-than-truckload network design
5.1. Algorithm specifications

This section presents the considerations that led to the design of the algorithm as it
will be shown subsequently.129 Three key aspects need to be pointed out. It is firstly
introduced how the demand for stable transportation times manifests itself on the
European LTL market. It is shown what stable transportation time refers to, what
may hinder it in LTL operations, and why it is important for shippers. Then, it is
exhibited how knowledge of transportation network centrality and related network
performance were exploited for structuring the algorithm from early on. Finally, high
level specifications for the algorithm is outlined. The specifications are based on the
desired network performance and the network centrality related to it, as well as the
demand for stable transportation time.
5.1.1. Stable overall transportation time in a less-than-truckload

network
LTL transportation was introduced in section 4.2. The following highlights one aspect
of service orientation on the European LTL market: stable transportation time. The
demand for LTL transportation by industrial customers has developed from a purely
transportation time focused demand that required fast delivery to a demand that val-
ues schedule reliability on the one hand and consistently stable transportation time
on the other hand. If an LSP promises delivery for a certain date, the customer expects
the arrival of the shipment for that date, neither earlier nor later. Punctual, reliable
arrival of shipments is of fundamental importance for the production planning of in-
dustrial customers. In the LTL business, it is typical to promise a certain delivery time
e.g. 24 hours within Germany; special products also guarantee a punctual delivery in
terms of penalties for late delivery. This guaranteed service is - in general - an addi-
tional service that is rather new in the LTL market. There seems to be a tendency of
some market segments in LTL to develop towards the express market where very fast
delivery to a predefined delivery time with strict guarantees is common (Klaus et al.,
2009, p. 88). Delivery promises that bind the LSP rather by reputation and customer
relationship than by penalties remain common in the LTL sector.
The second element of the new demand in LTL networks is that for consistently
stable transportation time. Producing companies oftentimes plan their production
on a repetitive schedule, without respect to certain weekdays. The LSP’s planning is
129 The term algorithm is understood in the following in the sense of a high-level “problem-solving method
suitable for implementation as a computer program” (Sedgewick, 1998, p. 4). The algorithm may also
be perceived as (part of) a decision support system (DSS). A DSS is defined as an “interactive computer
based system, which helps decision makers utilize data and models to solve unstructured problems”
(Sprague and Watson, 1989, p. 4). The algorithm is used to support decision-making by providing infor-
mation about generated network scenarios and their performance.
120
pure transportation time due to

operations
shipment A shipment B
available available
vehicle vehicle
departure arrival at
at origin destination
Mon Tue Wed Thu Fri Mon Tue
transportation time
shipment B
transportation
p time
shipment A
Figure 5.1.: Impact of waiting time on overall transportation time.
usually not taken into account, but should comply with the producers planning.130
Therefore, it is important for producing companies to have a consistent overall trans-
portation time for a transport from origin to destination independent of other exter-
nal circumstances; in particular independent of the day that the shipment is available
for transport. On first sight, it appears counterintuitive for the overall transportation
time to vary by day. The following example, shown in figure 5.1, will explain the situ-
ation.
Shipment A is available for transport some days before the departure. The ship-
ment will wait until the vehicle departure and is then transported directly to the fi-
nal destination. Its overall transportation time is the waiting time from Monday to
Thursday plus the pure transportation time due to operations from Thursday to the
following Tuesday; thus six workdays of overall transportation time. Shipment B is
available for transport on Wednesday and is loaded onto the same transport vehicle
as shipment A. Its overall transportation time is only from Wednesday to the follow-
ing Tuesday; four workdays in total. In this case, a customer will not be able to plan
a repetitive schedule with a stable transportation time; overall transportation times
with high changes are experienced by the customer. Shipment transportation time
depends on the day the shipment is available. The customer cannot know the general
transportation time of its shipments which is needed for a soundly repetitive sched-
ule.131
The key trigger for this problem is the variable waiting time before the transporta-
tion. The longer the intervals between consecutive vehicle departures, the higher the
130 Transportation schedules from the side of the LSP do typically not influence production schedules at the
producing company that is the LSP’s customer.
131 The following will always assume workdays when transportation time is summarized in days.
121
potential for variations in the waiting time before the transportation. Multiple trans-
shipment operations will most likely worsen the situation as additional waiting time
at the transshipment terminal is possible.
Transportation time in the following will always be understood including the wait-
ing time prior to the departure. This may also be referred to as the lead time of trans-
portation; those two terms could be used synonymously. It is the overall transporta-
tion time as it was just introduced.
5.1.2. Centrality-driven network design

Special attention was given in the LTL presentation in section 4.2 to the fact that the
market is very competitive and that only few factors allow for the differentiation in
service. It was furthermore indicated that the networks operated by the LSPs are de-
signed to be efficient and effective. As of now, network congestion and network vul-
nerability are minor challenges and well controlled by LSPs.
The stability of origin-destination (OD)-specific transportation time is an impor-
tant aspect of service, that is not yet offered on the European LTL market. The goal
of the algorithm is to generate networks with stable transportation time and simu-
late their performance regarding various KPIs. Many OD-markets in central Europe
already have a stable transportation time. The main focus of the algorithm is, there-
fore, on those OD-markets currently having an instable transportation time. The in-
stability can be overcome by increasing the frequency of departures from terminals.
Accordingly, the derived goal for the algorithm is to increase the frequencies in the
transportation network for the OD-markets with low transportation time stability.
The general idea is the following. The European LTL market has certain OD-
markets with high demand, that are already served with high frequencies. The goal
for the highly utilized OD-markets is to provide fast, cost efficient, and reliable
transport. Based on the results from chapter 3, this situation will best be served with
a decentral network, connecting origins and destinations directly. The benefits are
short transportation distances, few handling operations (both leading to limiting
costs), and few perturbations by propagating delays through central transshipment
terminals. Sufficiently high frequencies are generated by the high demand and need
not be increased by terminal consolidation.
Other OD-markets are characterized by low demand. The challenge for these mar-
kets is to increase the frequencies of departure while maintaining the transportation
cost within an acceptable bandwidth.132 The high service frequencies will lead to a
stable overall transportation time. Chapter 3 depicted that high frequencies are re-
lated to central transportation networks. This knowledge will be applied in the sense
that a topologically central network with a high concentration of transshipment activ-
132 It might be necessary to increase the transportation and transshipment costs in order to offer this service.
This may be an option as long as the increase in cost is justified.
122
ities to few potential hubs is striven for. This allows to create efficiencies by terminal
consolidation. Moreover, it is beneficial in terms of systems capacity flexibility.
The entire network, having both central and decentral elements, will form a hybrid
hub-&-spoke (HH&S) network. The literature already provides examples of beneficial
implementations of HH&S networks as intermediary solutions for situations where
different aspects of transportation network performance are required (e.g. Lumsden
and Dallari (1999) or Groothedde et al. (2005)). The first priority in routing will be
to send shipments through the direct network for a short transportation time. The
second priority will be to send shipments through the central network to increase
the frequencies. In that sense, the central network is the back-up network for the
decentral one.133
The following aspects have to be considered as well:
• Transportation distance may become relevant due to the central element.

Hence, the impact of distance-related transportation costs must be kept track
of in the network design.
• Schedule reliability is likely to decrease in central networks. This element has

not been critical for the LTL market so far. Therewith, it does not contradict a
strive for a central back-up network.
• The high centrality of the back-up network leads to network vulnerability. Then
again, the vulnerability of road-based LTL transports is limited compared to
other more infrastructure-related transportation modes (e.g. air transportation
relying on airports or rail transportation relying on tracks). It seems appropriate
to pay minor attention to vulnerability in LTL transportation network design.
• System capacity flexibility increases in central transportation networks which

is very beneficial for an LTL network. It was illustrated earlier that forecasting
OD-specific flows is difficult even for experienced planners. This aspects is a
further benefit of a central transportation network. The challenge for the direct
connections in the decentral network must not be overlooked.
• A central network will explicitly increase network concentration at very few po-
tential hubs in order to increase the frequency. This will also increase the risk
of congestion for the back-up network. This aspect is not directly modeled in
the algorithm, yet, it must be kept in mind for further investigation.
133 Groothedde et al. (2005) suggest an intermodal hybrid transportation network design. In their case, the
routing through the indirect hub-&-spoke (H&S) network is the first priority. The direct network merely
serves as a buffer for excess demand. The difference between their network and LTL networks discussed
here is that demand is too low to achieve an acceptable frequency of transport on the direct connections.
The general considerations are comparable, yet the given setting differs and so does the approach. The
authors come to this conclusion without explicitly addressing network centrality. Even so, the under-
standing of network centrality as an embracing concept might have been helpful in their research.
123
These considerations gave rise to the task to develop an algorithm for generating
HH&S networks for application in the European LTL market. The qualitative knowl-
edge of network centrality allowed to pre-structure this decision as outlined. This al-
lows to focus the network generation to these types of two-layered networks. Thereby,
the algorithm has a specification of what appearance the generated networks should
be. This provides the framework for the networks. The decisions taken by the algo-
rithm within the framework will not directly be based on the knowledge of centrality.
Here, well-established cost considerations come into play, as the generated networks
must be able to operate at a competitive level.
5.1.3. High level specifications for network generation

Eventually, the aim of the algorithm is to suggest network design guidelines that lead
to a consistently stable transportation time in an LTL network. The algorithm is to
generate networks at reasonable cost based on various input parameters concerning
network appearance.134 The parameters may be varied in order to create different
network scenarios, simulate LTL transport, and compare their KPIs.
The generated networks consist of EoL terminals and the operated connections
between these, as well as an aggregated, day-specific, weekly repetitive schedule that
specifies if a connection is operated on a day or not.135 For the strategic planning
purpose, a more detailed schedule is not necessary.
The algorithm relies on four design specifications, that will be outlined in the fol-
lowing. Firstly, a central back-up network for the lowly utilized OD-markets allows
to create high frequencies of departure and arrival for them. Secondly, cost consid-
erations are respected by making use of different consolidation strategies to increase
vehicle utilization. Thirdly, the number of transshipments is limited as additional
transshipment operations will induce the risk of process perturbations and additional
waiting time. This section closes by pointing out the fourth specification, namely that
the algorithm is focused to the network spanned by EoL terminals and omits local
pick-up and delivery (P&D) operations.
Central networks create high frequencies and stable transportation time

To generate networks with low variations in shipment waiting times, high departure
frequencies are necessary. As central networks create higher service frequencies than
decentral networks, the network design will work towards a high concentration of
flows in the back-up network.
134 Hence, the algorithm will respect cost considerations, but does not intend to find minimal cost trans-
portation networks.
135 It is important for the application that the schedule repeats weekly as this allows for repetitive operations
for the LSP (cf. Berman and Wang (2006, p. 298)).
124
Available on ... Waiting Pure transport Transport

Monday 3 3 6
Tuesday 2 3 5
Wednesday 1 3 4
Thursday 0 3 3
Friday 4 3 7
x̄ transport time 5.000
σ transport time 1.581
C V transport time 0.316
Table 5.1.: Stability of transportation time: One departure on Thursday.
The following small example highlights the connection between departure fre-
quencies and the stability in the waiting time, respectively the transportation time
of shipments. Tables 5.1 and 5.2 depict two cases to compare. Let an OD-market
have a pure transportation time of three workdays. Consider one case with only one
departure for that OD-market, table 5.1. The departure takes place on Thursdays. So,
all shipments that are available for transport on other days will have to wait a defined
amount of workdays before their transport. Thus, their transport time varies even
though the pure transportation time is the same. Table 5.1 lists the mean x̄ , the stan-
dard deviation σ, and the coefficient of variation C V of the transportation time.136
The coefficient of variation will serve to compare the stability of the transportation
time. The second table 5.2 shows the same OD-market, but departures occur three
times a week, on Tuesday, Thursday, and Friday. Again, table 5.2 lists the waiting and
transportation times for five shipments arriving on different days. The statistics are
listed below the tables. The difference between the two examples lies in the waiting
time prior to departure. This has an impact on the transportation time of shipments.
The more often departures occur, the lower the mean transportation time as well as
the standard deviation of transportation times. More interestingly, the coefficient
of variation of the transportation times depicts that the instability of transportation
time decreases as the frequency of departures at a terminal increases. Accordingly,
planning for more frequent departures in the network leads to an increase in stability
of market-specific transportation time.
Terminals with little arriving and departing traffic will be connected to few, even-
tually only one transshipment terminal.
If low demand terminals cannot even generate demand high enough to fill fre-
quent vehicles to a single transshipment terminal, there is no possibility to generate
more services with reasonable load factors from a network planning point of view.
136 The coefficient of variation is a normalization of the standard deviation by the mean of the sample (C V =
σ
x̄
). It is used to compare the dispersion of data with varying means (Bolch (1998, p. 7), Mosler and
Schmid (2006, p. 94)).
125
Available on ... Waiting Pure transport Transport

Monday 1 3 4
Tuesday 0 3 3
Wednesday 1 3 4
Thursday 0 3 3
Friday 0 3 3
x̄ transport time 3.400
σ transport time 0.548
C V transport time 0.161
Table 5.2.: Stability of transportation time: Departures Tuesday, Thursday, Friday.
For the situation of having too little demand even though central operations are im-
plemented, the network operator will have to define concrete steps. The four general
options are:
Stop operations at this terminal. If an EoL terminal cannot generate demand high
enough to justify its existence, management will have to consider closing that
terminal.
Exclude terminal from waiting time goal. It is an option to exclude the low de-
mand terminal from the goal to reduce the waiting time intervals, especially if
it is not foreseeable that customers in that geographical area will be impacted
heavily by such a step.137
Identify local solutions. It is an option to find local solutions that allow to provide
the service of stable transportation time for a terminal with low demand. De-
cisions could involve cooperating with external partners, in order to increase
the demand, or sourcing transportation capacity flexibly. However, this option
requires additional coordination and it might rather be included in a next, mid-
term oriented step after the strategic design process.
Enforce minimal frequencies. If an LSP wants to ensure low waiting time variations
even for lowly utilized EoL terminals, it can enforce minimal frequencies. Ser-
vice will be conducted at the predefined frequency, even if the vehicle load is
rather low due to the low demand. The cost of operating lowly used vehicles
can be seen as the price of the service to reduce waiting times and, thereby,
lower the variation of transportation time.
The option that is applied in the algorithm is the fourth one: enforcing minimal
frequencies. It will generate the most information as it allows to simulate how costly
137 This option is also conceivable if a network provider feels that customers in a certain region are not
interested in a stable transportation time.
126
the decrease in waiting time variation will be. Thus, it gives valuable information to
an LSP for evaluating which option to choose in a distinct case.
Consolidation decreases transportation cost

Consolidation may decrease transportation cost by generating high load factors on
vehicles. The algorithm consolidates shipments as much as possible. In the planning
phase of the algorithm, three elements are employed:
Go when full. If it is forecast that a vehicle is filled to capacity, its departure is sched-
uled as early as possible. Several vehicles can be scheduled for a single day.
Inventory consolidation up to a certain threshold. Shipments can be consoli-

dated in inventory, i.e. at the EoL terminal, for the time interval between
two consecutive departures. The time interval is limited by a maximal
accumulation time for direct shipments between EoL terminals.
Terminal consolidation. The next consolidation logic that is applied by the algo-
rithm is terminal consolidation. If inventory consolidation will generate load
factors too low for the direct transport between EoL terminals, shipments with
different destinations will be transported to a transshipment terminal where
they are resorted to be transported further.138
Vehicle consolidation is not implemented in the algorithm. Terminal consolidation

is an important element, especially for EoL terminals with low demand. As termi-
nal consolidation is directly related to central networks (there are no hubs to use for
transshipment operations in totally decentral networks), the goal to reduce costs by
increasing vehicle load factors will lead to more central networks the more terminals
with low initial demand are in the network.
Few transshipments reduce waiting time and complexity

The network generating algorithm will, if transshipments are necessary to increase
the vehicle load factors, plan for 1-stop routings whenever possible. Multiple trans-
shipments are avoided. This logic reduces waiting time and its variation as well as
planning complexity.
Waiting time reduction. Figure 5.1 depicts a situation where two shipments have
different waiting times before the departure at the origin EoL terminal. The
same situation will appear at the transshipment terminals again. Shipments
arrive on different vehicles and wait for the vehicle that continues further to
their destination. Arrivals and departures are coordinated. Still, some kind of
138 Detailed reasoning on whether shipments are transported directly or through a hub is discussed in sec-
tion 5.3.1.
127
time gap will appear. The less often such a time gap occurs in the transport, the
better in terms of waiting time. It may be advantageous to increase the time as-
sociated with pure transportation operations to go to a transshipment terminal
that is farther away if less waiting time can be expected for this routing.
Planning complexity reduction. Any coordination of arriving and departing vehi-

cles at transshipment terminals creates additional planning complexity. Even
though this aspect is not in the scope of the network design algorithm, the
more transshipment operations are necessary the higher that complexity will
be. Therefore, limiting the number of transshipments on the path of the ship-
ments will facilitate the planning process.
2-stop routings will be necessary on exception, detailed information is given in sec-

tion 5.3.2.
Restricting the algorithm to the EoL network

The algorithm is restricted to the network spanned by EoL terminals. Local P&D op-
erations are blinded out from the network for several reasons:
• The operations in P&D and long-distance transport differ significantly and are
usually planned independently.139
• In real life P&D operations are mainly performed by independent subcontrac-

tors that are responsible for their own operations.
• P&D is typically done in different, smaller vehicles than the long-haul transport
in LTL transportation (Crainic, 2003, p. 458).
• The algorithm runtime is diminished strongly if the number of locations in-

cluded is limited to the EoL terminals instead of all end-customer locations.
The approach to model only EoL terminals requires data manipulation to properly
model direct injections140 . If shipments that are in practice not channeled through an
EoL terminal are to be modeled, they will be treated as if they were handled at the EoL
terminal that is responsible for the area that the end-customer is located in. It would
not be a good option to exclude these shipments from the modeling as the load is
typically rather high and omitting it would distort the demand in the model.
139 For example, P&D mainly generate efficiency by creating optimal tours, mathematically related to the
vehicle routing problem (VRP) presented in section 2.3.3. Long-distance transports are rather strongly
influenced by location decisions and service network design as introduced in sections 2.3.2 and 2.3.3.
140 A direct injection is a transport that bypasses the origin EoL terminal, see section 4.2.2
128
5.2. Input and output

The algorithm requires various elements of input and will generate networks, ship-
ment routing, and network KPIs as output. The goal during the algorithm’s develop-
ment was to require only data that an LSP will be able to provide without major data
collection efforts.
The following subsections will briefly present the input available to the algorithm,
give an overview of parameters relevant for trimming the algorithm, specify the un-
derlying assumptions on cost and transport times in the algorithm, and finally de-
scribe what kind of output is generated.
5.2.1. Input
The algorithm requires data about the network and shipments.
Network data
The network data that is necessary for the algorithm are the existing EoL terminals, a
set of potential hubs, and a distance matrix. The existing EoL terminals are used as
the fixed nodes for the newly generated networks.
The set of potential hubs must be specified for the algorithm. The algorithm will
choose the best transshipment terminals possible to route shipments through from
this set, but limiting the number of potential hubs is necessary for structuring the
problem. The potential hubs are chosen out of the set of EoL terminals. Visual in-
spection of a map of the existing EoL terminals will quickly identify candidates for
potential hubs. Furthermore, an LSP should be able to specify some EoL terminals
where large transshipment operations are conceivable and others where this would
not be feasible without major investments in infrastructure. These potential hubs
are treated differently by the algorithm to model transshipment operations. As the
potential hubs are chosen from the set of EoL terminals, they have origin and desti-
nation traffic just as any other EoL terminal. Different sets of potential hubs can be
specified and compared. This is the most powerful tool for generating different net-
work scenarios. The number of potential hubs in a network is used to denominate the
scenario; e.g. a 2-hub scenario is a scenario with two potential hubs.
A distance matrix with traveling distances between all EoL terminals is required.
This gives the algorithm the possibility to calculate costs for all links potentially used
in the network. The distance matrix can be obtained in different ways. Software tools
calculate the distances based on road maps. If this is not possible, one rather sim-
ple option is to calculate the distances with a geographic distance function based on
the locations’ coordinates and to multiply them by a circuity factor to account for
differences between beeline connections and physical road layouts. Berens and Kör-
ling (1985) point out that it is necessary to choose correct, region-specific parameters
129
when estimating road distances. Ballou et al. (2002) give a list of circuity factors for
various countries.
Shipment data
Two types of shipment data are necessary for the algorithm: traffic flow forecasts that
the network is planned for as well as distinct shipments that allow to simulate the
performance of a generated network when shipments are transported.
Both types of shipment data must have the following information:
• The origin EoL terminal of the shipment (or in case of direct injection the EoL
terminal that is responsible for this consignor).
• The destination EoL terminal of the shipment (or in case of direct delivery the
EoL terminal that is responsible for this consignee).
• The volume or weight of the shipment.
• The start date when the shipment is ready to be transported.
The data should be easy to obtain for an LSP. Large production databases exist
where each transported shipment is entered. The historic data can serve as test-set
for the distinct shipment case. The forecasts can be generated on a different time
frame from the database. In the most simple case, average values from another time
period than the test-case can be used as forecasts.
5.2.2. Cost and transport time assumptions

Two types of costs are modeled in the algorithm: distance-volume-related cost, i.e.
transportation cost, and volume-related cost, i.e. transshipment cost. The distance-
volume-related cost is a linear function of the distance run by a transport vehicle.
It may or may not be influenced by the load of the transport vehicle. The distance
matrix mentioned above supports to calculate this measure.
The transshipment cost is calculated for operations at a transshipment terminal.
Each loading and unloading of a shipment at a transshipment terminal induces a cost
that is related to its weight or volume. Usually, an LSP will be able to specify this cost
as it oftentimes exists in the organization in form of a transfer price. To take a more
realistic perspective, the loading and unloading operations at the EoL terminals for
each shipment are also accounted for. This measure will not discriminate the scenar-
ios as it only depends on the shipments in the dataset and not on specifying hubs or
directing flows in the network.
Other cost elements as presented in table 3.1 are not included as they either do not
differ amongst the solutions or are accounted for indirectly (e.g. unused capacity cost
is included in the distance-volume cost).
130
The transport time is roughly estimated by the algorithm. It is assumed that a ve-
hicle has a maximal travel distance per day. If that distance is exceeded, even by a
small amount, the vehicle will arrive the following day. It was mentioned above, that
the algorithm only generates a day-specific schedule. Therefore, this approximation
is acceptable.
It is furthermore assumed that all demand for transportation that arises during a
day are ready to be transported with a vehicle that leaves that night. All arrivals at
a transshipment terminal on one day are available for the consecutive transport the
next day.
5.2.3. Parameters
Modeling parameters are used to trim the algorithm to various settings. The most
relevant ones are described in the following. The above described cost assumptions
are also included as parameters.
Minimal number of trips per week. The minimal number of trips per week en-
forces the minimal frequency constraint. It applies for connections to, from,
and between the potential hubs. Thus, it guarantees that each terminal is
connected at least with a minimal frequency to the entire back-up network.
Maximal accumulating days. For connections that do not fall under the minimal
frequency constraint, there is a maximal timespan for accumulating shipments
before they are loaded to a vehicle. This applies in explicit for direct connec-
tions between EoL terminals. As no transshipment with additional waiting time
is necessary for this case, the timespan can be longer than the minimal fre-
quency without strongly impacting the stable transportation time. If a ship-
ment cannot be loaded within the accumulation time, it must be transported
by the back-up network.
Transport costs per km per full truck. Transportation cost is measured per km
that is run by the vehicle.
Handling cost at transshipment terminals. (Un-) loading operations at the trans-

shipment terminals create a cost that depends on the weight or volume of the
shipment. That cost may be a transfer price and includes indirect costs for op-
erating the transshipment terminal.
Handling cost at EoL terminals. Similarly, (un-) loading operations at the EoL ter-
minal also induce costs. These are typically lower than the cost at the trans-
shipment terminal.
Percentage of sold empty capacity. It is assumed that some of the created empty
capacity in the network can be sold to other players on the LTL market. This
131
helps to decrease the cost of empty capacity. To incorporate this aspect, the
algorithm can assume a certain share of the existing empty capacity to be sold
to reduce the costs per truck.
Capacity per truck. The maximal load of a truck is a parameter of the algorithm.
Kilometers per day. The kilometers a truck runs per day are related to how long the
transportation operations take. This measure cannot support the implementa-
tion of details such as cut-off times that are necessary for more detailed plan-
ning.
Further implementation parameters exist in the simulation but these are not of
interest for adjusting the algorithm to a situation or to verify the impact of environ-
mental changes.
5.2.4. Type of generated output

The algorithm generates three types of outputs. Firstly, it creates networks and day-
specific schedules. Secondly, it keeps track of the KPIs of a network when it is tested.
These KPIs serve to compare the different networks and highlight differences between
them. Finally, the distinct shipment routings are extracted and can be interpreted.
Networks and day-specific schedules

The networks that are created by the algorithm include the EoL terminals that were
provided as input as well as the connections between them. The decision which con-
nections are to be included in the final network depends mainly on the demand for
transport on that connection. Intuitively, if transport demand is high, direct con-
nections will be included in the network. If demand is low, indirect connections are
included.141
Once a connection is part of the network, the related day-specific schedules are
also included in the output. It will be defined whether a vehicle leaves or does not
leave on a certain weekday.142 The schedules represent the frequency of transport on
a connection. The frequency is derived based on the minimal frequency restriction
mentioned above as well as the actual demand for transport. If demand is high, more
transports will be scheduled than for low demand.
As the scenarios are to reflect the operations conducted by an LSP, much attention
was given to the schedules being repetitive over the weeks. Most LSPs will agree that
transshipment and loading operations at a terminal benefit in efficiency if the general
task remains the same over time.
141 Section 5.3.1 will treat this decision in detail.
142 Detailed planning with definite departures and related cut-off times is not taken into account. For the
overview purpose of the scenario analysis this appears to be an acceptable approximation for the situa-
tion.
132
Key performance indicators

Several KPIs are tracked for each network scenario that is generated and explored by
the algorithm. The KPIs are in many cases related to the data typically applied in
network design; measures of network concentration are part of the KPIs.143 The main
KPIs for the algorithm are listed in the following.
Cost. The cost elements that are recorded are the induced cost at the EoL terminals,
transshipment cost at transshipment terminals as well as volume-distance-
related cost for the truck connections.
Utilization. The average utilization of the truck capacity is another performance in-
dicator that is generated as output.
Percentage of direct volume. The algorithm keeps track of the shares of shipments
that are transported directly and indirectly.
Network topology. The static network layout is tracked by several parameters. No

indices of network topology are used here, yet the appearance is described by
values that suffice to describe the networks:144
• number of departing and arriving trucks from EoL terminals
• number of hubs an EoL terminal is connected to
• number of EoL terminals without direct lines
Detour. The detour generated by trips to the transshipment terminals in comparison

to direct connections is calculated, simply by distance and also relating to the
fill rate of vehicles.
Transportation time. Detailed mean and average values of shipment transportation

time are calculated. Transportation time is the time from the day the shipment
is ready for transport until the delivery to the destination EoL terminal. Hence,
it includes waiting times at the departure EoL terminal and at potentially used
transshipment terminals. The standard deviation of transport time is relevant
to assess the quality of the algorithm, as the goal is to limit the variation in
transportation time.
It is assumed that all shipments arriving on one day are ready for the sub-
sequent transport (if necessary) the following day. Therefore, the calculated
transportation time forms an upper bound of the transportation time to be ex-
pected. It will usually be possible in real life to transship much quicker and
actually have shipments available for departure the very night of arrival.
143 These KPIs are the concrete variants of the KPIs discussed in chapters 2 and 3. Some general KPIs are not
among the ones specified here as they cannot be measured by the algorithm.
144 The values could be used to calculate network topology indices, but their expressive power in its simple
form serves its purpose.
133
Distance. The average and standard deviation of transported distance are further-
more assessed.
Concentration measures. Based on the shipment routing that is generated by the

algorithm, the McShan-Windle (MW) index, the network concentration (NC)
index and finally the hubbing concentration (HC) index of each scenario is part
of the output. The ratio of HC index over connecting share is, moreover, in-
cluded as a KPI.
Most KPIs are divided into those elements related to direct connections and those
related to hub connections.
The algorithm is not aimed to review the network performance if perturbations
arise. Variations in transportation demand are not modeled. Furthermore, none of
the EoL terminals has a capacity restriction for the algorithm. This allows to follow the
transportation volumes to be handled at a terminal in an unconstrained network. For
these reasons, schedule reliability, congestion and system capacity flexibility are not
among the output KPIs. Several reasons underline the exclusion of these indicators:
• The algorithm is oriented to support long-term decision-making. The key in-

formation to be obtained is of general, indicative network performance.
• Modeling perturbative effects in the network requires additional data and as-
sumptions. It creates additional effort to obtain the input information and de-
creases the expressiveness of the generated data.
• Knowledge of the overall centrality of the networks combined with all detailed
information concerning utilization and handling volumes at hand, renders the
additional information about perturbative effects is unnecessary. The central-
ity of the networks provides indications to the performance of the network
when facing perturbations. Therefore, the general information is available. Any
more detailed information can be accumulated in a next step if necessary.
Thereby, the generated data, combined with the argumentation of chapter 3 in mind,
will provide sufficient information to review the results obtained by the algorithm for
long-term decision-making.
Shipment routing
As a final part of the output, the shipment routing needs to be mentioned briefly. The
algorithm tests the network with real or artificially generated shipments. For any of
these shipments, the entire routing through the network is recorded and can be ex-
tracted from the algorithm. This information allows detailed analysis of the situation.
However, most of the KPIs that are of interest to be taken from the shipment routing
are already implemented in the KPI output as outlined above.
134
5.3. Assembling the algorithm

The network construction process is depicted in figure 5.2. The following sections will
illustrate each step in detail, but a first brief overview provides a better understanding
of the structure of the algorithm.145 Firstly, the algorithm will separate direct and indi-
rect transport connections. The algorithm decides per OD-market whether the rout-
ing will primarily be direct or whether the shipments are planned to be transshipped
at a transshipment terminal. Secondly, it will be defined how many transshipment
terminals are connected to each EoL terminal. This decision is based on the incom-
ing and outgoing transport volume of each EoL terminal and is directly impacted by
the minimal frequency in the network. Thirdly, the connectivity of the network must
be ensured. As the fourth and, therewith, final step of the network design, the al-
gorithm will calculate the number of discrete vehicles that are necessary to serve all
demand.
Split direct and hub 5.3.1 Trade-off detour vs. load factor
connections
ti
Identify right number of hubs 5.3.2 Minimal frequency constraint

per terminal
t i l
5.3.3 2-stop connections

Ensure
su e full
u coconnectivity
ec v y
Calculate necessary number 5.3.4 Transition from demand to

of trucks vehicles
Figure 5.2.: Algorithm: Construction process.
Subsequently, the algorithm continues by testing the generated networks with a

dataset of distinct shipments. For all days in the planning horizon, the transport of all
available shipments is simulated by virtually loading them on the scheduled trucks
and routing the shipments on the trucks through the network. The simulated rout-
ing may differ from the initially planned routing since the forecast average shipment
demand may vary from the shipments in the testset. The detailed routing logic is
presented following the network construction steps.
145 It is important to note that the algorithm generates networks based on planned demand for transporta-
tion. During the testing phase, the routing may differ from the initial planning.
135
5.3.1. Trade-off direct vs. indirect: Separating direct and indirect

traffic
It has been mentioned above that a major element of the algorithm is to decide
whether shipments are transported directly or are transshipped on their way. This
section will illustrate the criteria for this decision.146 It will further be demonstrated
why the approach can only be heuristic.147
Load factor vs. detour: The α-rule

The key trade-off between direct routing and indirect routing is best depicted when
considering the costs related to both routing types. The goal is to define, for each OD-
market whether to route shipments directly or indirectly. This decision is taken once
in the planning horizon and is by definition OD-specific. The relevant cost elements
and decision parameters are:
c k m Cost per full truck km.
c h Cost per weight or volume unit of handling (loading and unloading) shipments
at a transshipment terminal.
d i s t i j The distance of the direct connection from location i to location j (or other
locations that are specified. In particular, d i s t OD is the distance of the direct
connection from origin O to destination D.).
d A demand to be specified further.
d i j The demand from location i to location j .
d i n d The indirect demand from location to location.
d ov e r The overflow of direct demand from location to location.
α The detour incurred by going to the transshipment terminal.148
u The assumed utilization of a vehicle to and from the transshipment terminal.149

146 The challenge to decide on whether or not to transship freight in a transportation network has also been
discussed by Blumenfeld et al. (1985). The authors break down the entire many-to-many networks into
sub-networks with only one origin. The optimality of this approach is proven, under the assumption of
known traffic flows to the transshipment terminals. The idea of facilitating the decision to sub-networks
is tempting and will be applied in the following. However, the assumption of known traffic flows will not
hold true for the application here.
147 The considerations are shown for 1-stop routings only. Section 5.1.3 explained that 1-stop routings are
necessary as they allow for terminal consolidation. Additional transshipment operations as required by
2-stop routings shall - as far as possible - be avoided since they may lead to additional waiting time and
raise the risk of perturbations in the network.
d i s tO−hu b +d i s t hu b −D −d i s tOD
148 α= d i s tOD
.
149 Utilization refers to the ratio of used capacity over available capacity.
136
c a p The capacity of a transport vehicle in weight or volume units.

Based on these costs and parameters, the cost for direct (C d i r e c t ) and indirect
(C i nd i r e c t ) routing can be derived. The direct cost in equation 5.1 assumes that the
cost for the entire truck is paid by the LSP for any direct connection. The induced cost
is independent of the actual utilization of the transport vehicle and only comprises
distance-related costs.
C d i r e c t = d i s t OD ∗ c k m (5.1)
As opposed to this, the indirect costs in equation 5.2 are calculated differently to
approximate the terminal consolidation. The distance-related, indirect costs are at-
tributed to the shipments based on their contribution to the total load on the trans-
port vehicle. It is furthermore assumed that the vehicle is filled to a certain percent-
age. Additionally, the handling cost related to the shipments on the OD-market needs
to be paid. In sum, the distance-related costs are only attributed partially but addi-
tional handling cost is incurred.
dij
C i n d i r e c t = (1 + α) ∗ d i s t OD ∗ c k m ∗ + d i j ∗ ch (5.2)
cap ∗u
The algorithm takes into account the location of the best-nearby (resulting in the
least detour) transshipment terminal for the OD-transport and calculates both, direct
and indirect transportation cost. It chooses the option with lower costs.150 Figure 5.3
shows the general relation between the load factor on the direct line (d i j /c a p ) and the
detour factor α. The longer the detour to the next transshipment facility, the lower
the utilization of the direct truck will be to generate indifferent costs. As detour and
direct load are deterministic for the algorithm, the decision whether to transship or
to route directly depends on the location above or below the line of identical costs. As
this decision is amongst others impacted by the detour factor α, it is referred to in the
following as the α-rule.
To summarize, the trade-off evaluated by the algorithm is the one between high
load factors and low detour. If demand on a market does not lead to high load fac-
tors on a direct line, the decision has to be taken whether to accept medium to low
load factors for a direct line or whether to route to a transshipment facility which will
lead to detours but most likely increase vehicle utilization due to consolidation. The
algorithm makes this decision based on simplified cost estimates for both routing
schemes.151
In particular, one strong assumption that may influence the decision-making is
the assumed utilization of vehicles to transshipment terminals (u ). If considered in
150 The decision of the α-rule reflects the general idea that “consolidation at a break-bulk (BB) terminal be-
comes attractive when the savings in line-haul transportation costs exceed the terminal costs” (Daganzo
and Newell, 1986, p. 122).
151 If some empty capacity is assumed to be sold on the market, this will be included in equation 5.2. This
modifies some decisions as direct connections tend to become more favorable; yet, the general trade-off
remains the same.
137
90%
80%
70%
direct
60%
looad factor directt

50%
40%
30%
20% transship
10%
0%
0%
15%
30%
45%
60%
75%
90%
105%
120%
detour factor (alpha)
Figure 5.3.: Line of identical costs in direct and indirect routing.
detail, this value can only be obtained after the routing is defined and the network
has been tested with data. However, it is necessary for the decision-making up front.
It must be kept in mind that this value is an externally chosen parameter.
Applying the α-rule in the algorithm

The application of the α-rule in the algorithm (sketched in figure 5.4) is based on the
EoL terminals, the potential hubs, the expected transportation demand (in terms of
shipment forecast), and the distances between all terminals. The daily demand per
OD-market is kept track of; the variable d stands for the sum of demand in the ac-
cumulation period for one specific market. Then, the potential hub with the least
detour is identified and the detour factor α is calculated per OD-market. Splitting
traffic between direct and indirect flows is the following step. The function labeled
f (d , α) in figure 5.4 expresses the α-rule: if the demand d for the OD-market is so
low to justify the detour α to go to the best-nearby hub, the shipments on this market
are routed indirectly through the hub. Vice versa, if demand d is high, the shipments
are transported directly. One final modification is necessary due to overflows of di-
rectly served markets. The overflow d ov e r is the partial vehicle load that remains if the
directly transported shipments were loaded to fill up entire vehicles. It may also in-
clude shipments which exceed the maximal accumulation time. The remainder may
be only a fraction of a vehicle. Here again, the algorithm will evaluate if direct or indi-
rect transport for the remainder is appropriate. It may well be the case that the over-
flow of a directly served OD-market is transported indirectly. This implies, that the
indirect demand is to be increased by this amount. After completing this procedure,
the entire demand d is split into demand that will be transported on direct connec-
138
EoL Potential Shipment

Distances
terminals hubs forecast
For all arcs in the network

Calculate forecast traffic
fl d
flow
Find best (least detour) hub

among the potential hubs and
detour Į
hub direct For overflow:

f(d, Į)
find dover & Į
Hub hub direct Direct

f(dover, Į)
connections connections
Figure 5.4.: Algorithm: Separating direct and indirect traffic.
tions and demand that is transported by making use of a transshipment terminal. The
subsequent steps will use this information as their input.
Decision rules for special cases

The α-rule implicitly assumes that there is only 1-stop routing and that the trans-
shipment terminal that creates the least detour will be available for transshipment
operations. These two assumptions do not always hold true and decision rules for
other situations were set up.
Once the decision has been taken to primarily route shipments for an OD-market
through a transshipment terminal, this decision is not reconsidered. If it becomes
apparent that 1-stop routing is infeasible for the OD-market, it will not be checked if
it actually might be cheaper to route shipments directly than use the 2-stop case. Vice
versa, for OD-markets that are defined to be routed directly, the general possibility to
route on a 2-stop path is not considered, either.152 2-stop routings are seen as an
exceptional routing scheme as the number of transshipments should be reduced.153
152 Some authors argue that in 2-stop routings on the connection between the two transshipment terminals
extremely high load factors can be achieved which they represent by a discount factor in their analysis
(e.g. O’Kelly et al. (1996)). This thinking is not taken into account for the algorithm as high load factors
can be generated differently in LTL transports.
153 See section 5.3.3 for details.
139
In the same line of argument, the decision taken by the α-rule is not modified if a
transshipment terminal is not available for that OD-market. An OD-market will be
re-routed to another transshipment terminal but not switched to direct transports.
This approach was taken due to two key understandings. Firstly, if all decisions
were to be revised as soon as something changes, it would significantly impact the
runtime of the algorithm. Secondly, and more importantly, reviewing the decision
would most likely never lead to an end. All decisions taken in network design are
highly interdependent. It was shown in chapter 2 that Operational Research (OR)
provides various approaches to deal with these interdependencies and find optimal
solutions. If the algorithm was to revise the decisions that were just taken, it would
probably result in changes, which then in turn lead to new revisions and so on. In the
context of the algorithm, this is not desirable.
5.3.2. Terminal allocation: Defining the number of hubs

The next step in the algorithm allocates EoL terminals to transshipment terminals.
These transshipment terminals are necessary for indirect shipments. Hence, indirect
demand is the basis for the following considerations.
One fundamental element of the algorithm is to enforce minimal frequencies on
connections between EoL terminals and transshipment terminals.154 This may result
in unused but available vehicle capacity in the case that demand of transport on a
given link will be lower than the vehicle capacity c a p and the enforced minimal fre-
quency Fm i n in a time period.
There are two drivers of this unfavorable situation: low demand and a network
topology that does not suit the situation. If incoming or outgoing demand at an EoL
terminal is so low that operations will inevitably lead to unused capacity, the decision-
maker must consider the situation and decide how to proceed with this EoL terminal.
The algorithm chooses to enforce a minimal frequency for all EoL terminals.
To this point, all EoL terminals are allowed to send shipments to all potential hubs
in the network. The decision which transshipment terminal to use is based on the in-
curred detour to the hub. At this point, the question arises whether it might be better
to connect each EoL terminal only to a limited number of potential hubs; in the ex-
treme case even only to one single transshipment terminal. This allows to consolidate
shipments.
The sum of all flows from an EoL terminal j to the potential hub i in a given time
ou t
period is denoted by d hu bi :

ou t
d hu bi
= d ji nk d |hu b i is best-nearby hub (5.3)
k
in
The analog applies for d hu bi .
154 Section 5.1.3 presented the reasoning for this approach.
140
Potential Shipment
hubs forecast
For all EoL terminals in the network
Calculate sum of forecast

traffic flows dhubi per hub i
Suitable number of hubs

(in / out may differ):
¦d hubi
i
tm
cap * Fmin
Choose the m “most wanted”

hubs as used hubs
Redirect all OD-flows to the

used hubs if necessary
Figure 5.5.: Algorithm: Defining the number of hubs.
An EoL terminal is given only so many routing choices as it can justify by its trans-
port demand. This may differ between the outgoing and the incoming side of trans-
ports. Then the number of connected hubs m ou t to this EoL terminal is limited to:
ou t
d
i hu b i
m ou t ≤ (5.4)
c a p ∗ Fm i n
The connection to at least one hub must be satisfied and may overrule equation
5.4. This ensures the connectivity of the entire network. The same rule applies for the
in
incoming flows denoted by d hu bi .
In the forefront of these considerations, figure 5.5 shows the algorithmic steps to
define the number of connected hubs per EoL terminal. The incoming and outgoing
expected flows per EoL terminal that result from section 5.3.1 are firstly accumulated
per potential hub to find the demand for transportation on the links between the EoL
terminal and the potential hub. It allows to calculate the number of connected hubs
following equation 5.4. Knowing that m i n and m ou t transshipment terminals will be
connected to the EoL terminal, these have to be selected in the third step. Based
in ou t
on the d hu b i and d hu b i calculated initially, the distinct transshipment terminals are
chosen by selecting the ones with the highest demand for transportation; all other
transshipment terminals become unavailable for this EoL terminal. Lastly, all traffic
flows on OD-markets that were directed through transshipment terminals that are
now unavailable, are re-routed to the suitable best-nearby transshipment terminal
that remains available after the reduction just carried out.
A situation may arise, where full connectivity is no longer given and must be en-
sured by the algorithm.
141
For all hub legs
Best hub for

yes connection is no
connected to
OT and DT
Do OT and
yes DT share a no
common
hub?
Redirect that OD connection Create a 2-stop routing

to the common hub through the two hubs
Routing
complete
l t
Figure 5.6.: Algorithm: Ensuring full connectivity in the network.
5.3.3. Number of transshipments: Ensuring full connectivity

The algorithm strives to limit the number of transshipments. No more than two trans-
shipments can be planned between EoL terminals. In general cases, the routing deci-
sion is only between direct and 1-stop routing. This decision is based on the α-rule.
Even though there is a strong point for 1-stop routings, 2-stop routings need to be
planned. Consequently, all potential hubs must be interconnnected with a point-to-
point (P2P) network.
Some EoL terminals will not be connected to all transshipment terminals in the
network. This may lead to a situation where origin and destination EoL terminal
are no longer connected to the initially defined best-nearby transshipment terminal.
In that case, the algorithm validates if the EoL terminals share any other transship-
ment terminals and picks the best-nearby out of the potential hubs in common to
be the one that shipments for this OD-market are routed through. If this approach
fails as well, 2-stop routings are necessary. In that case, the shipments have to be
transshipped twice and routed between the transshipment terminals. The described
procedure is depicted in figure 5.6.
2-stop routings will only be necessary for very few OD-markets with little demand.
The concerned shipments are therefore limited in volume and number. Furthermore,
the above outlined terminal allocation is only possible if 2-stop routings may be car-
ried out. This is necessary for the algorithm to limit the number of allocated trans-
shipment terminals per EoL terminal, leading to the beneficial reduction of unused
capacity. 2-stop routings guarantee the connectivity of the entire network. The ben-
efit obtained by allocating EoL terminals smartly to potential hubs outweighs the im-
pact of 2-stop routings.
142
Shipment For all connections in the network

forecast
Sum up forecast transport

d
demandd on each
h li
linkk d
direct Leg hub

ttype
Set number of trucks per Set number of trucks per

week (ȣ) for each link week (ȣ) for each link
ª d º ª d º
°« » if d in accu. period ! cap max( « » , Fmin ) X
X
° « cap » « cap »
®
° ª days per week º else
°̄°¯ «« accu.
accu days »»
Spread trucks over weekdays
Figure 5.7.: Algorithm: Determining the right number of vehicles.
5.3.4. Demand estimation: Determining the right number of

vehicles
By completing the three afore explained steps, the algorithm determined the links
that will be part of the designed network. It firstly identified the direct and indirect
connections, it then continued by limiting the hub connections for lowly utilized EoL
terminals and ensured full connectivity by interconnecting potential hubs and cre-
ating 2-stop routings. The final step in the network design is to define and plan the
right number of vehicles υ on each existing connection. As depicted in figure 5.7, the
planning depends on whether or not a connection is a direct one.
For both situations, the demand d i per link is the input for the decision-making.
Two decisions need to be taken: firstly, how many vehicles are necessary for the given
demand and secondly, in how far should excess vehicle capacity be used to buffer
against uncertainty in demand. The way the algorithm solves this problem for the
case of direct and indirect connections is discussed below.
Direct connections
Direct connections must respect the maximal accumulation time. It ensures that
shipments will not exceed a certain waiting time. There are two cases to discriminate
for direct connections.
143
Demand in the accumulation period is greater than or equal to than the vehicle capacity.
Here, the truck capacity simply is the number of vehicles necessary to transport
the demand:
d
υ= (5.5)
cap
Demand in the accumulation period is smaller than the vehicle capacity. The
number of vehicles in this case is defined as the number of accumulating
periods within one week:155

days per week
υ= (5.6)
accumulating days
The number of vehicles that is calculated as described above includes some unused
capacity due to rounding up fractional results. Additional capacity can be included in
the planning but is more powerful for indirect connections, as these exist for each
OD-market and may act as back-up solutions when extra vehicle capacity is needed.
Indirect connections
The number of vehicles necessary for connections between EoL terminals and trans-
shipment facilities, as well as between two transshipment facilities is constraint by
the implied minimal frequency. It is therefore calculated as:

d
m = max( , f req) (5.7)
cap
If the demand on a link justifies more vehicles than the minimal frequency, that is
- of course - the number of vehicles that is planned for it. If even the consolidated
demand is too low, the minimal frequency of departures will be ensured by placing
the necessary number of vehicles, even though they might in the end have a lower
utilization than desired.
Additional buffer capacity can be included at this point to buffer against uncer-
tainty. Artificially increasing the demand is one option to do so, even small increases
such as 105% of demand are possible. This adds an additional vehicle if demand just
about fills an entire vehicle but will not impact those links where enough excess ca-
pacity is available in any case.
155 This planning of direct capacity is conducted for all links that were defined as directs by the α-rule. That
does not necessarily imply that the consolidated load will fill up an entire vehicle. Nevertheless, the
planning is set in a way that the same schedule repeats weekly.
144
υ Weekdays
1 Thursday
2 Wednesday, Friday
3 Tuesday, Thursday, Friday
4 Monday, Wednesday, Thursday, Friday
5 Monday, Tuesday, Wednesday, Thursday, Friday
>5 departures = 5 + (υ-5)
Table 5.3.: Algorithm: The spread of departures over the week.
Scheduling departures
Once the two decisions on the number of vehicles υ were taken, the truck departure
is planned for certain days. The idea is to spread the departures over the weekdays,
and in tendency, create more departures towards the end of the week. Table 5.3 shows
how the algorithm allocates departures to days.156 If more than 5 departures during a
week are necessary, the algorithm will allocate one departure per day and recursively
increase the number of departures per weekday following the same scheme.
With that, the algorithm has planned not only the links to be used in the network
but also the available vehicle capacities per day. As shown in figure 5.8, the perfor-
mance of the network will then be tested by simulating the transport of different ship-
ments on the network.
5.3.5. Testing the generated network

Testing the generated network is done by virtually loading distinct shipments on the
scheduled vehicles and simulate their transport. This allows to gather the above de-
scribed output, in particular the KPIs to compare the performance of different net-
works. The testing starts with the generated network, i.e. with the links and the sched-
ules for the vehicles. The simulation of operations covers several weeks.
Each shipment is initially located at its origin EoL terminal. Firstly, it is checked if
there is a direct vehicle to the destination EoL terminal available that provides suffi-
cient empty capacity and departs within the maximal accumulation time. If so, the
shipment is allocated to this vehicle and the transport may take place. Shipment-
specific KPIs are recorded and the operations for this shipment are terminated.
In the cases where no direct vehicle is available, either because no direct service
exists or because no empty capacity remains, an indirect vehicle is identified. The
algorithm checks which transshipment terminal has been defined previously for the
OD-market the shipment is part of. The shipment is then allocated to the next vehicle
156 In its details, this planning is arbitrarily set and can be easily modified to suit any situation.
145
that leaves for the transshipment terminal and provides sufficient capacity.157 The
shipment arrives after a certain time at the transshipment terminal. It is added to the
transshipment terminal’s queue of departing shipments for the day after its arrival.
First KPIs are recorded as the shipment awaits its subsequent transport.
The subsequent transport steps follow the same logic as outlined above. From the
transshipment terminal, the shipment may be routed directly to the destination EoL
terminal. In rare cases where 2-stop routings are necessary, no such direct vehicle
exists and the shipment is channeled to a second transshipment terminal from where
the routing to the destination EoL terminal will then take place.
Contrasting the routing in the testing phase with the planned routing in the net-
work generation phase, several points are important to mention. Firstly, the defini-
tion in the network generation phase whether an OD-market is served directly or not,
is used to schedule the vehicles. The routing of the shipments through the network
is independent: it checks if a direct connection is available and if not, it sends ship-
ments to the transshipment terminal. The testing phase takes the generated network
for granted and tries to find suitable routings through it. Secondly, shipments on one
single OD-market may take either the direct or the indirect routing. The routing de-
cision depends on available empty capacity, not on defined routes to follow. Thirdly,
the testing phase respects the network generation in that it may only transship ship-
ments at the predefined potential hubs. Even if there might be capacity available go-
ing to and from an EoL terminal that is not a potential hub, transshipment operations
will not be carried out.158
5.3.6. Possible enhancements

The algorithm is able to provide interesting and helpful insights for strategic decision-
making. It gives high level information about the performance of simulated LTL trans-
portation networks in different settings. Two paths for further development were
identified and can be followed for future work. Firstly, to improve the capacity plan-
ning by making use of the interplay between direct and indirect connections for an
OD-market, and secondly, to consider in detail the impact of different shipment pri-
orities in the network.
Capacity Planning
Additional excess capacity to buffer for uncertainty in the network is implemented
roughly but rather effectively in the algorithm. It does not consider in detail the inter-
play between the direct transports in the network and the transshipped connections.
157 It is not necessary to check for the maximal accumulation time in this case as this routing is the only
remaining option. However, the minimal frequency constraint that was applied in the planning of the
networks ensures that sufficiently enough departures occur.
158 From a practical perspective, space and equipment for transshipment operations are available at the
potential hubs, but may not at the other EoL terminals.
146
Direct connections are included in the network whenever demand is high enough
to justify a direct vehicle. Reliable network operations are difficult to plan for low
demand markets. All links between EoL terminals and transshipment terminals op-
erate frequently. This sub-network is available and backs up the direct connections:
in times of high demand on the direct connections it may be used as an alternative
path for routing shipments.
This creates a difficulty for planning. The OD-specific demand can be transported
directly or indirectly. The capacity planning for indirect transports is difficult. De-
mand and, therewith, capacity requirements come from three independent sources:
Low volume OD-markets are operated by indirect transport. Their demand can be
forecast just as the direct demand. Aggregation effects occur as different OD-
markets are consolidated on the connections to and from the transshipment
terminals.
Planned overflow from direct connections is channeled through the transship-

ment terminal to deal with low volume excesses that may occur.
Unplanned excess demand on direct connections must be dealt with on indirect

connections but is even more difficult to forecast.
This mix must be handled by the capacity planning for the indirect connections
between EoL terminals and transshipment facilities. Three cases serve to illustrate
the difficulty.
Normal demand. There is no challenge as long as the demand occurs just as ex-
pected in the network. Low volume and expected overflow can be transported
with the existing capacity; excess demand does not exist.
Peaks of high demand. If the direct connections face an exceptionally high de-
mand, the planned overflow is likely to increase as well. Furthermore, this is
the situation in which the unplanned excess capacity will have to be trans-
ported through the transshipment terminals. Assuming the likely case that in
this situation the low volume markets will as well have a higher demand, this
situation will lead to the necessity for more transport capacity on the links to
and from the transshipment terminals.
Drops of low demand. The situation with exceptionally low demand actually ap-
pears more difficult for the planning of the indirect connections than the sit-
uation with high demand. Unplanned excess capacity will not be channeled
through the transshipment terminals as the capacity on the direct connection
is sufficient for the entire demand. Even worse, the planned overflow is likely
to be transported on the direct connection to utilize the empty capacity. The
indirect connection, therefore, faces a far more dramatic drop in demand than
the direct connection.
147
Hence, it is challenging to plan the correct capacity for the indirect connections
in the network. This aspect has been introduced in section 3.2.6 as system capac-
ity flexibility for the two prototypes of network topologies: purely central and purely
decentral network. The networks designed by the algorithm have some medium de-
gree of centrality. The direct connections by themselves form a decentral network,
whereas the transshipment network is a central network when considered separately.
The interplay between the two networks is interesting to look at in detail as it promises
improvements in capacity planning.
The situation can be modeled analytically from different perspectives for a suitable
capacity planning. Various assumptions on the distribution of the expected demand
may then be tested and evaluated.
Inventory model The situation can be modeled as an inventory problem.159 Inter-

preting the available capacity as the stock of goods and the demand for transport as
the demand in the model, the capacity planning is the standard situation of defin-
ing the level of safety stock for a given service level. This approach will have to be
extended as to cover the aspect that demand may be stored for a limited amount of
time; for example a shipment could be loaded to the over-next departure of a vehicle
and still receive the expected service.
Queueing model The other important foundation to model the situation is queue-
ing theory.160 Queueing models allow to describe the empirical situation that is of in-
terest here. The departing vehicles on the indirect network are the servers that serve
batches of size c a p in a deterministic service time equaling the interval between two
consecutive departures. The arrival process of demand can be modeled by one or
three independent arrival processes to the queue whose parameters may vary. The
service level is directly deducted from the model.
Queueing theory offers many closed form solutions for various models, but this
model creates two difficulties. Deterministic service times contradict basic Markov
chain assumptions regarding birth-death-processes (Winston, 2004, p. 1064). Fur-
thermore, the practical situation of having shipments wait in queue while a vehicle
is available but it is not yet time for departure, must be modeled with care and is not
part of standard solutions.
The queueing model formulation appears to describe the situation better than the
inventory model. Searching for an analytical model is certainly interesting for fu-
ture research, but does not, in its detailed form, support the general and strategic
approach that is chosen for the algorithm.
159 Winston (2004, ch. 16) introduces probabilistic inventory models.

160 Winston (2004, ch. 17 and ch. 20) introduces Markov chains and queuing theory. The interested reader
is referred to this helpful starting point for further details.
148
5.4. The bigger picture: Contributions to strategic network design
Shipment Priorities
It is possible to adapt the algorithm so that it is able to simulate the transport of ship-
ments with different priorities in the network. This aspect has been touched on early
in the algorithm development. Early results revealed that the impact of different ship-
ment priorities appears to be independent of the chosen network design, especially
of the number of transshipment terminals that are available in the network.
It remains an open question to verify how the capacity planning would benefit
from different shipment priorities in the network. The main effect that was shown
previously was that the benefit in transportation time which high priority shipments
have is strongly related to the decrease in transportation time by the low priority ship-
ments. This result did not justify further research in the topic. Yet, the interplay be-
tween direct connections and back-up transshipment network has so far not been
tested empirically in depth.
Some considerations will illustrate what the impact of high and low priority ship-
ments on the network may be. Assuming low priority shipments to be less sensitive
to variations in overall transportation time, these shipments will require a lower min-
imal frequency of transport and a longer accumulation period for direct connections.
This characteristic will allow the network on the one hand to speed up high priority
shipments by transporting them earlier than the low priority shipments. On the other
hand, the capacity planning will be simplified as the low priority shipments can be
shifted in time and transported in times of low demand, as long as a certain threshold
of accepting the delay is not breached. Future work in this field will certainly benefit
from and might contribute to peak-load pricing and yield management research.
5.4. The bigger picture: Contributions to strategic network

design
The entire algorithm is depicted in figure 5.8. All elements were presented in the pre-
ceding sections. Network data and parameters were treated as input to the algorithm.
Section 5.3 presented the algorithm network design procedure and the network test-
ing that generates the network KPIs.
The development of the algorithm was guided by the goal to depict realistic situ-
ations and provide a reasonable picture of network appearance and operations that
allow to strategically evaluate at a high level various impact factors of network design.
With this goal in mind, a heuristic approach was chosen.
Notwithstanding, a heuristic approach must have various limitations. The α-rule
provides fast and rather good decisions, but drawbacks on the pragmatic handling
of specific situations were already mentioned above. Further, the capacity planning
in the algorithm is at a very high level. It was shown that future research can further
149
Network data
Potential Shipment
Terminals Shipments Distances
hubs forecast
Parameters Split direct and hub

connections
Minimal
frequency
Networkk design
Identify right number of hubs
Maximal per terminal
accumulation
time
Ensure full connectivity
Costs
Calculate necessary number
of trucks
Truck
capacity
work testing
Load shipments
Max load
Netw
Performance indicators
Figure 5.8.: Overview of the developed algorithm.
improve this, but as for heuristic insights, the chosen modeling already proved to be
helpful.
Certain aspects were not included in the scope of the algorithm but would be nec-
essary for the tactical planning of a network. The schedule deducted with the algo-
rithm is only day-specific. A detailed analysis will have to model terminal operations
and create real schedules regarding arrival and departure times. Similarly, vehicle
drivers were not modeled at all. Work regulations, driving time restrictions and oth-
ers are included in the parameter of traveled km per day but not considered in order
to create work schedules. Moreover, any P&D operations are not taken into account.
Closely related to this, all transports bypassing the EoL terminals were re-routed for
the simplicity of the algorithm. Further, the algorithm does not consider location de-
cisions. It is able to take different locations as input and depict the impact of using
these to construct the network. This approach was chosen to model the real life situ-
ation at an LSP that will not start the network design from scratch. So, there probably
is an efficiency gap that is not depicted by the algorithm at all. Lastly, the algorithm
does not model perturbative effects in the LTL network; the impact of operational
challenges in LTL operations cannot be assessed directly, but may be approached in-
directly by the centrality indices included in the KPIs.
The algorithm does not take into account any changes in demand due to offering
the better service of stable transportation time. It can depict network performance
150
5.4. The bigger picture: Contributions to strategic network design
when changes in demand occur, but it cannot answer questions concerning how the
market will react to the availability of stable transportation time.161
The algorithm has a clear structure and the decisions made throughout the process
reflect the constraints in LTL operations. Hence, a decision-maker can easily follow
each step. Discussing the results is facilitated by these preconditions. Extensions or
alternations of scenarios can be easily implemented or specified for further research
which additionally supports the possibility to apply the algorithm. In sum, it proved
to be valuable guidance for further thinking about network configurations.
The case study in the following chapter will report on an application and some
results from a project. General insights for network design will be provided following
the illustrative case study.
The algorithm was employed in a project with an LSP and served its purpose to
show high level relations for strategic network design. The generated networks are
tested with sets of shipments. The network performance is shown in different KPIs.
Network concentration indices are directly drawn from the algorithm and give further
information on the networks. Classical performance indicators support the deeper
comparison of network alternatives; cost, transportation time, and number of ve-
hicles being only few examples. This allows for comparing the scenarios and guide
further decision-making on a long term perspective.
161 The aspect of how demand reacts to changes in the network appearance has been discussed for passen-
ger airlines (e.g. Wei and Hansen (2003)), but has not been a prominent research focus in the context of
freight transportation networks. The algorithm does not attempt to fill this research gap.
151
6. Generated networks and their
performance
The first part of this chapter presents a project that was conducted with a logistics
service provider (LSP) targeting at a less-than-truckload (LTL) network redesign. This
case study allows not only to demonstrate how the developed algorithm (cf. chapter 5)
was applied, but - more importantly - what type of results were obtained with its help.
The LSP is introduced very briefly, while information for the industry was already
given in chapter 4. The short introduction supports the understanding of how the
algorithm was trimmed to fit the requirements of the company. The obtained data is
presented subsequently. The focus lies on comparing the obtained results with the
base case scenario representing the LSP’s operations prior to the project.
General insights for LTL network design are deducted based on further network
scenarios that were simulated. This second part of the chapter highlights strategic
relations of parameters that go beyond the application in the case study and are of
interest for transportation networks in general. The presentation includes the com-
parison of the scenarios amongst each other, independent of the situation at the LSP.
The chapter closes by explicitly addressing the importance of network centrality
for transportation network design. The empirical demonstration of the theoretically
formulated relations in chapter 3 remained an open point and that loose end shall
now be tied.
Some points regarding the terminology in this chapter are worth mentioning.
A scenario refers to a setting of potential hubs given to the algorithm, e.g. the 3-hub
scenario {Frankfurt, London, Moscow}.
The adjectives direct and indirect are relevant for this chapter. In the following, a
direct link, connection, or vehicle refers to one that connects two end-of-line (EoL)
terminals where neither of them is a potential hub (or transshipment terminal) for
the network scenario. An indirect link, connection, or vehicle has a potential hub
(or transshipment terminal) either as origin, destination, or both. When referring to
the number of indirect links, connections, and vehicles, the sum of all these in the
networks is understood. Concerning direct and indirect shipments (or shipment vol-
ume), it is possible to account for actual transshipment operations. Direct shipments
(or shipment volume) are all shipments that do not require handling at a transship-
ment terminal on their way. Indirect shipments are handled once or twice at a trans-

153
6. Generated networks and their performance
shipment terminal. One aspect of this terminology is important: direct shipments

can be transported on indirect vehicles.162
6.1. Case study

The data provided for the case study stems from a project conducted jointly between
the University of Cologne, Department of Business Policy and Logistics and a large
LSP based in Germany but operating worldwide. The LTL segment, that was the target
of the project, is but one amongst different segments served by the LSP. The LSP is a
successful player on the European LTL market.163
6.1.1. Input and parameters to the algorithm

The LSP provided comprehensive data from the current LTL operations. This dataset
was used for two purposes: parts of it served to estimate expected demand for the
LTL network and the entire dataset was used to test the simulated network scenarios
in the last step of the algorithm.
The dataset covers all European shipments that are transported in the LTL segment.
It contains data from a four week operating period. After initial data cleansing, the set
is of 204,308 distinct shipments in the network. Detailed information is available for
each shipment; an excerpt of the data structure is shown in table 6.1.
The European LTL network consist of 154 EoL terminals. About half of these func-
tion as transshipment terminals for some operations.164
Transportation volume could be assessed by cubic volume or by weight. Yet, all
measures of transportation volume are related to pay weight for the remainder of this
chapter. Pay weight is the weight of the shipment adjusted to suit extraordinary di-
mensions or special requirements. Pay weight equals the real weight if the shipment
does not have any special requirements associated with it. It is the basis used by LSPs
for billing purposes and, therefore, available for every shipment. So, it is an accept-
able measure for transportation volume for the project.
162 An example illustrates this. A vehicle runs from the potential hub Frankfurt to the EoL terminal Cologne.
Hence, it is an indirect vehicle. However, there may be some shipments on that vehicle with their origin
in Frankfurt and their destination in Cologne. These shipments are accounted for as direct shipments
since they do not require handling apart from (un-) loading at the terminals.
163 The industry would most likely not label the LTL operations as being conducted on a network. The term
network is typically associated with various attributes such as one central coordinating institution for the
entire operations, standard procedures for the entire operations no matter where they are conducted,
and one single profit and loss responsibility. None of this is true for the LTL segment at the LSP.
Nevertheless, the operations can be classified as being conducted on a network from an academic view-
point. There are transshipment terminals (nodes), and truck connections (links). Further, the graph
theoretic interpretation of the therewith spanned network is possible. Moreover, the scientific literature
on network design is pertinent to the discussion in the project. Thus, network will be used to describe
the LTL operations for the case study as this terminology was used throughout the entire thesis so far and
is not contradictory to the pertinent academic literature.
164 The general operating procedures as described in chapter 4 apply for the LSP.
154
6.1. Case study
Information Value
Origin EoL terminal One of 154 EoL terminals
Destination EoL terminal One of 154 EoL terminals
Departure date A distinct day
Arrival date A distinct day
Used transshipment terminals One or two terminal (-s)
Shipment pay weight With a mean of 905 kg
Table 6.1.: Relevant shipment information.

umber of shipments
Number
30 2500
Shipment pay weight [kg]
Figure 6.1.: Shipment pay weight distribution.
The shipments have a pay weight in the range of up to 24,000 kg, which represents
the exceptionally wide variety of shipments in the LTL market. Figure 6.1 sketches
the distribution of the shipment pay weights in the dataset. It indicates that most
shipments are in the window from 30 to 2,500 kg, which would be what is typically
defined to be LTL shipments.
Routing information can be read from the dataset as well. The routing in the initial
situation in the dataset is always one of the three basic types (direct, 1-stop or 2-stop
routing).165 Table 6.2 depicts the routing shares in terms of number of shipments and
of pay weight in the initial situation.
Distance matrix
The distance matrix was provided by the LSP. It was generated by a routing software
and provides real road distances between most of the EoL terminals. The few missing
connections were added by applying the estimation suggested by Ballou et al. (2002),
as well as distances obtained from Google Maps.
165 Figure 4.3 shows four routing schemes in LTL operations. Case (ii) and (iii) both represent 1-stop routings
which are shown as one type of routing in the following.
155
Routing type Share by shipment Share by pay weight

Direct 69% 75%
1-stop 28% 23%
2-stop 4% 2%
Table 6.2.: Shares of initial routing schemes.
Parameter Value
Minimal number of trucks per week 3
Maximal accumulating days 4
Transport cost per km per full truck 0.781 Euro
Handling cost at transshipment terminals 0.019 Euro / kg
Handling cost at EoL terminals 0.013 Euro / kg
Percentage of sold empty capacity 70 %
Capacity per truck 16,000 kg
Kilometers per day 600
Set of 8 EoL terminals
Potential hubs
and all permutations.
Table 6.3.: Case study algorithm parameters.
Parameters
The relevant parameters for the algorithm are listed in table 6.3. It suffices here, to
give the values chosen for the case study.166
The transport costs per km are based on the values used by Krause (2007, p. 200)
and adjusted to price increases since 2007. Handling costs are estimates based on the
LSP’s operations. The capacity per truck is purposely set to be low. Yet, as the entire
case study is based on pay weight information, it is reasonable to assume that a truck
loaded with 16 metric tonnes (t) pay weight has reached its capacity.
The cost function per km depending on the utilized capacity on a transport vehicle
that is applied reads as:
16,000−c a p u s e d
(0.781 − 0.7 ∗ 0, 781 ∗ 16,000
)
c k m (c a p u s e d ) = (6.1)
c a pu sed
Table 6.4 shows the values of the cost function at four positions. These values were
validated with expert knowledge and picture the situation in practice.
166 Refer to section 5.2.1 for the meaning of the parameters.
156
6.1. Case study
Load Cost per t per km

4t 0.093 Euro
8t 0.063 Euro
12 t 0.054 Euro
16 t 0.049 Euro
Table 6.4.: Transport cost at different load factors.
Data cleansing
A base case is established that represents the operations at the LSP prior to the
project. The base case can be derived out of the extensive dataset provided by the
LSP. It was used to identify an as-is situation for cost, transportation time, stability of
transportation time, and network concentration. Unfortunately, some values had to
be approximated as they were missing in the set. Incoming or outgoing scans provide
valuable date and routing information, but they were sometimes lacking for the ship-
ments and had to be filled in. This typically involved retracing the shipment routing
and the exact arrival and departure dates at different transshipment terminals along
the way. With in-depth knowledge of the process, large shares of these could be
reconstructed and anchored to existing scan events. Yet, for some shipments it
was impossible to anchor the routing and related time data points to any event. It
also appeared that existing data values were contradictory, which might result from
inconsistent synchronization between different data sources. If the reconstruction
was infeasible, the shipments were excluded for the base case scenario. In particular,
retracing the transportation time was difficult as information about the delivery data
was not always consistent.167
It should be mentioned that the dataset does not reflect any waiting time periods
prior to the departure of the shipment from the origin EoL terminal, as the first rele-
vant scan event occurs at the departure from the EoL terminal. Therefore, the trans-
portation time is actually underestimated for the cases where shipments had to wait
before the departure.168 All instability in overall transportation time that is related to
this waiting time, cannot be included in the base case.
167 Table A.3 in the appendix summarizes the order of magnitude of data cleansing to establish the base case
in terms of transportation time, as well as transportation and handling cost.
168 The transportation time in the base case only partially covers what was introduced in section 5.1.1 as
overall transportation time. The waiting before initial transport is lacking. However, all intermediary
waiting time is accounted for.
157
Figure 6.2.: Coefficient of variation of transportation time per OD-market in base

case.
6.1.2. Generated results on stable transportation time

The overall goal of the project was to suggest network designs that lead to a stable
overall transportation time (cf. chapter 5). Therefore, the results concerning the sta-
bility of transportation time are the first to be presented.
The first challenge was to obtain the fluctuations in transportation time for the base
case scenario. Extensive data cleansing was carried out as described above. Neverthe-
less, the results are only an approximation of the situation and should be interpreted
as such.
Figure 6.2 visualizes the coefficient of variation of the overall transportation time
per origin-destination (OD)-market in the base case.169 The mean of the coefficients
of variations is 0.232. Out of the 9,398 markets, 7,169 have more than one shipment.170
The figure highlights the vast amount of OD-markets in the dataset and their diver-
gent behavior. While the transportation times on some OD-markets are very stable,
others are not.
The same type of figure can be generated for all scenarios simulated by the algo-
rithm. One example of a 1-hub scenario is shown in figure 6.3 for illustrative purposes.
Again, the coefficients of variation for all OD-markets are depicted. Out of the 9,398
markets, 7,426 markets have more than one shipment.171 The mean of the coefficients
169 The coefficient of variation is used as a measure of stability; the lower the coefficient of variation, the
more stable the data, cf. section 5.1.3.
170 Markets with only one shipment will have a standard deviation and a coefficient of variation of 0.
171 The algorithm will split large shipments virtually into several pieces and thereby increase the number of
shipments but not the overall transportation volume.
158
6.1. Case study
Figure 6.3.: Coefficients of variation of transportation time per OD-market in an ex-

emplary 1-hub scenario.
of variation is only 0.121, already hinting at the larger stability of transportation time
in the algorithm.
A visual comparison of figures 6.2 and 6.3 reveals that the stability of transporta-
tion times in the base case is lower than in the exemplary algorithm scenario. The
figure for the base case appears more scattered. The data points in the algorithm sce-
nario are strikingly compact at the low end of the scale - a good visual indicator of the
stability of transportation times.
In order to depict the stability of transportation times in all 255 scenarios calculated
for the project, the mean transportation time and its standard deviation for direct
and indirect shipments across all OD-markets per scenario were calculated.172 The
coefficient of variation of transportation time is depicted in figure A.2 in the appendix.
The mean of the direct coefficients of variation is very stably at 0.513 (with a standard
deviation of 0.002), the mean of the indirect coefficients of variation is more dispersed
(standard deviation of 0.017) at a level of 0.301.
The generated results confirm that the algorithm creates networks and routings
that will, indeed, increase the stability of transportation time for the entire network.
This is a direct result derived from the stronger concentration of traffic flows to central
transshipment facilities by the algorithm.
172 This averages out all transportation distances impacting the transportation time. It will increase the stan-
dard deviation of the transportation time. This is the reason why the measured coefficient of variation is
higher than the values in figures 6.2 and 6.3.
159
200%
n % of comparative base case

180%
160%
140% Transport cost full vehicles
120%
100% Transport cost 0.7 level
80%
Transport cost no empty capacity
60%
40%
H dli att ttransshipment
Handling hi t terminals
t i l
Cost in
20%
0% Handling at EoL terminals
Worst base case Comparative base case Best base case
Figure 6.4.: Costs in the base case scenario.
6.1.3. Generated results on cost

It is important to be aware of how costly an increase in transportation time stability
would be. Therefore, the costs of the base case scenario were assessed for compari-
son. To do so, the vehicle movements in the dataset were retraced; transportation dis-
tance, transshipment operations, and shipment pay weight were accounted for. This
data allowed to calculate the overall costs with the same cost rates as assumed for the
algorithm. Figure 6.4 summarizes the results.173 Three cases were calculated within
the base case scenario, as different price levels may be included in the algorithm. The
three depicted cases are a worst base case where all transport vehicles have to be paid
at full price, the comparative base case assumes that 70% of the empty capacity may
be sold on the spot market, while the best base case presumes the entire empty capac-
ity to be sold on the spot market. The comparative base case is the setting applied in
the heurisic results as presented here.174 The comparison between the three cases is
interesting as it shows the impact of different cost assumptions on the results. LSP ex-
perts agreed that worst and best case are far from realistic, yet, these two cases depict
how strongly the costs are impacted by unused vehicle capacity.
The comparison of different cost categories is important as well. The handling cost
at EoL terminals accounts for 23% of the comparative base case. They will not vary
between the scenarios, as they depend directly on the total pay weight of shipments
in the network. All shipments are presumed to be handled once at the origin EoL ter-
minal and once at the destination EoL terminal. The handling costs at transshipment
terminals represent the degree to which transshipment takes place in the network. In
all cases in figure 6.4 they arise to 4% of the costs and contribute only at a minor de-
gree to the overall costs. The volume-distance-related transportation cost is the most
173 When retracing the real operations, some surprisingly low vehicle utilizations were identified. It is likely
that these shipments were subcontracted to another carrier or co-loaded to a part load (PTL) vehicle.
This is not feasible for the algorithm. Hence, the transports are assumed to have taken place at the
applicable rate.
174 The cost axis is set so that the comparative base case is the 100% level.
160
6.1. Case study
100%
st in % of comparative base case

90%
80%
70%
60%
50%
Transport Cost
40%
30%
20% Handling at transshipment terminals
10%
Cost
0% Handling at EoL terminals
Scenarios
Figure 6.5.: Comparison of least cost scenarios to base case.
important part of the overall transportation cost. 73% of the comparative base case is
transportation cost.
Figure 6.5 compares the ten scenarios with the least sum of costs to the compar-
ative base case. The results indicate that the networks and schedules generated by
the algorithm are actually less costly than the base case, yet, all top 10 cost solutions
are within a gap of 0.56% of the least cost scenario. The key difference between the
generated networks and the base case is consolidation. The handling costs at trans-
shipment terminals increase to 5% of the comparative base case. Table 6.5 supports
this finding by identifying that the percentage of direct shipments is lower in the re-
sults generated by the algorithm than in the base case. Furthermore, the share of
1-stop shipments is always higher than in the comparative base case. Thus, the in-
crase in handling cost at transshipment terminals results from the increase in ship-
ments handled only once, while the pay weight of shipments handled twice on their
way actually decreases. Consolidation is exploited by the algorithm, and this leads
to the decrease of direct shipments as well as an increase in vehicle utilization. The
higher vehicle utilization creates lower volume-distance-related transportation cost.
This lower volume-distance-related transportation cost offsets by far the increase in
handling cost at the transshipment terminals.
6.1.4. The role of network centrality for the case study

The algorithm was pre-structured with the help of knowledge of network appearance
and network performance. The desired network performance was related to a cer-
tain network appearance, that of a hybrid hub-&-spoke (HH&S) network, that sets
the frame in which the algorithm creates networks. The results shown support that
161
Scenario (num-
ber of transship- Share direct Share 1-stop Share 2-stop
ment terminals)
A (4) 0.713 0.273 0.014
B (4) 0.710 0.273 0.017
C (5) 0.715 0.263 0.022
D (3) 0.711 0.279 0.010
E (3) 0.708 0.283 0.009
F (4) 0.713 0.269 0.018
G (4) 0.708 0.273 0.019
H (4) 0.711 0.272 0.017
I (3) 0.706 0.282 0.012
J (2) 0.703 0.293 0.004
Comparative
0.751 0.227 0.022
base case (79)
Table 6.5.: Shipment shares by pay weight.
this is a suitable approach. The presented specifications concerning stability of trans-

portation time and cost are met.
Measuring the concentration of the generated networks after the simulation al-
lows to compare in detail the different scenarios. Network centrality was measured
by the McShan-Windle (MW) index, the network concentration (NC) index, the hub-
bing concentration (HC) index, and the ratio of HC index over connecting share that
were modified to the LTL context earlier.175
The results obtained for the ten least cost scenarios generated by the algorithm as
well as the comparative base case are summarized in table 6.6. The networks are very
similar in their cost and some potential hubs are used in all solutions. Differences in
centrality between these networks are limited as well, but the difference to the com-
parative base case is striking.
• The importance of the top 3% hubs in the network is depicted by the MW in-
dex. The relevance of these high volume EoL terminals is impacted their by
origin and destination traffic. Nevertheless, their traffic shares are increased
in the results generated by the algorithm as compared to the comparative base
case. Aditional flow is consolidated through already highly used EoL terminals.
The risk of congestion must be kept in mind in this situation. Scenario J has a
175 Chapter 4 presented the European LTL market as well as the modifications necessary for applying the
indices to LTL networks.
162
6.1. Case study
Scenario (num-
MW HC index
ber of transship- NC index HC index connecting share
index
ment terminals)
A (4) 0.355 0.723 0.249 0.869
B (4) 0.342 0.720 0.252 0.869
C (5) 0.314 0.717 0.248 0.870
D (3) 0.375 0.724 0.250 0.866
E (3) 0.380 0.726 0.253 0.867
F (4) 0.336 0.718 0.249 0.866
G (4) 0.315 0.717 0.249 0.855
H (4) 0.344 0.719 0.250 0.867
I (3) 0.372 0.722 0.254 0.865
J (2) 0.408 0.728 0.257 0.863
Comparative
0.225 0.678 0.169 0.679
base case (79)
Table 6.6.: Centrality indices per scenario.
comparably high MW index value. Its risk of congestion is higher than for the
other scenarios.
• The NC index is higher for the generated networks than for the comparative
base case. Therefore, the differences in traffic shares at the EoL terminals in the
generated networks are higher than in the base case scenario. This indicates
that traffic flows in the generated networks were successfully channeled to few
hubs that, in turn, increased their traffic share in comparison to the other EoL
terminals.
• The HC index exhibits that hubbing is of higher importance in the generated

networks than in the comparative base case. This is driven by two influence
factors: firstly, the share of direct shipments decreases (cf. table 6.5). Secondly,
the possible routings on each market are decreased which can directly be read
from the ratio of HC index over connecting share.
To various degrees, all networks were shown to be rather decentral by the NC index.
This is an indication that direct shipments dominate in the networks. As these do not
require additional handling, schedule reliability will be comparably high. Moreover,
events such as congestion and late arrivals will not spread across the network. On
contrast, their system capacity flexibility will be of a low degree, since the bundling
effects of central networks cannot be exploited.
163
The HC index also indicates that hubbing takes place only to a minor extend in
the network, confirming the general conclusion of the NC index. However, the ratio
of HC index over the connecting share provides a perspective that is not identified
by the other metrics. It reveals that for indirect shipments, hubbing is of high impor-
tance and few alternatives for routing exist for OD-markets. This conclusion concerns
only the back-up network that serves the indirect shipments. For many OD-markets,
only one routing path is allowed by the algorithm. This is done with the intention of
exploiting consolidation effects. The backup network is concentrated. For this spe-
cific sub-network, the planner must pay special attention to create schedules with
enough slack to ensure their reliability. The back-up network is vulnerable: if the few
transshipment terminals in the back-up network will break down, its viability will be
substantially diminished. As opposed to these threats, the indirect shipments will
benefit from a high system capacity flexibility in the back-up network. By consolidat-
ing shipments for different OD-markets on vehicles, the back-up network can react
more easily to changes in demand than the network for direct shipments.
The generated networks use less different routing paths per OD-market. This in-
creases the potential for consolidation. It may also be seen as a risk: if one path is
disturbed by link or node failure, only very few detour options are available. The ten
depicted scenarios differ only slightly in this aspect, but the difference to the com-
parative base case cannot be neglected. Emergency plans should be devised for this
situation.
Summing up, all above mentioned indices point at the general conclusion that the
ten above mentioned generated networks are more concentrated than the compara-
tive base case scenario. This information is consistent for all 255 generated network
scenarios.
6.2. General Insights

This section compares the results obtained with the help of the algorithm to each
other and, therewith, allows to draw more general insights independent from the set-
ting at the LSP. All results are obtained on the same dataset as used above; compara-
bility to the afore stated data is ensured.
The following sections exhibit various parameter modifications and the related
changes in network key performance indicators (KPIs). Explanations on the reasons
behind the changes in performance are provided. The presentation for each parame-
ter modification focuses on different changes in KPIs as not every parameter is related
to all KPIs. The most illustrative data is disclosed in figures and tables; additional in-
formation can be found in the appendix and is referenced in the text.
The parameter modifications are structured as follows. Firstly, the impact of setting
the minimal frequency is discussed as this parameter is fundamental for the algo-
rithm. Secondly, the analog for direct connections, the maximal accumulating days,
164
is varied. Then, the low impact of setting distance-related transportation cost is dis-
closed before the more influential aspect of the percentage of sold empty capacity
is evaluated in detail. Fifthly, the daily distance by truck proves to be of lesser im-
pact. Lastly, the impact of selecting certain potential hubs in a scenario is reviewed.
The section closes with a brief summary of the most important points. Discussing
the impact of transportation network centrality will be the subject of the subsequent
section.
6.2.1. Minimal frequency

Imposing a minimal frequency in order to obtain a more stable transportation time
is the key idea behind the implemented algorithm. Therefore, the impact of defining
the minimal frequency is again the first to be presented here.
The effects in the network related to the minimal frequency constraint become
clear when taking a detailed look at how the minimal frequency impacts the network.
Two levers are relevant: the number of runs to and from transshipment terminals and
the transshipment terminals connected to each EoL terminal.
The minimal frequency constraint is restrictive for all indirect connections in the
network. The number of runs per week is set to match the minimal frequency unless
the accumulated transportation volume requires more transports.176 The minimal
frequency constraint increases the number of indirect transports in the network.
The second lever is the number of transshipment terminals connected to each EoL
terminal. The algorithm may limit the number of transshipment terminals (in and
out independently) if the accumulated shipment volume is too low to support the
minimal frequency to several transshipment terminals.177 The higher the minimal
frequency, the more constraining it is (cf. figure A.4 in the appendix). For instance,
if an EoL terminal can accumulate 3 weekly departures, it will be able to connect to
2 transshipment terminals if the minimal frequency is of 1 run per week. At a min-
imal frequency of 2 runs per week, only one single transshipment terminal will be
connected to the EoL terminal.
Reducing the number of connected transshipment terminals increases the detour
for the indirect shipments that would have been routed through the transshipment
terminal that is no longer available. Thus, the average travel distances increase when
imposing more weekly runs (cf. figure A.6 in the appendix). By raising the travel dis-
tance, the travel time is also raised slightly. Yet, the stability of transportation times
remains untouched (cf. figure A.5 in the appendix).178
176 See section 5.3 for details on the network construction by the algorithm.
177 The reasons and the process for this were extensively explained in section 5.3.2.
178 The minor decrease in share of direct shipments (cf. figure A.3 in the appendix) is related to those ship-
ments that had their origin or destination at a transshipment terminal initially connected to the EoL
terminal was then eliminated for that EoL terminal. In the cases where the shipment volume on this OD-
market is too low, these shipments are channeled through one of the remaining available transshipment
terminals. This effect results in a slight decrease of direct shipments.
165
Minimal runs per week

1 2 3 4 5
Indirect links 952 709 593 509 448
Direct links 509 512 515 523 532
Weekly indirect runs (total) 3090 2881 3009 3147 3298
Mean weekly runs per indirect link 3.25 4.06 5.07 6.18 7.36
Direct weekly runs (total) 1864 1872 1879 1913 1947
Mean weekly runs per direct link 3.66 3.66 3.65 3.66 3.66
Table 6.7.: Connections and runs in the 5-hub example.
Table 6.7 lists the number of direct and indirect links for an exemplary 5-hub sce-
nario.179 It furthermore contains the information about the number of vehicle runs
on these links during one week. The first observation in the table is the outstanding
decrease of indirect links as the minimum of runs per week increase. This results from
decreasing the number of transshipment terminals connected to the EoL terminals.
Secondly, the number of direct connections increases slightly. This is also caused by
constraining the number of transshipment terminals per EoL terminal, but mainly is
an issue of bookkeeping by the algorithm.180
The next information to highlight from table 6.7 is the stability of the average
weekly runs on the direct links. The demand on the direct links does not change,
neither does the number of runs on the links.181
The minimal frequency impacts the runs on the indirect links. Comparing the av-
erage weekly runs per indirect link amongst each other reveals the increase in runs
that could be expected (cf. table 6.7). At first sight, the weekly indirect runs at a mini-
mum of 2 appears to be too low for practical application. All the same, this is evidence
for consolidation. The algorithm is not able to strive for terminal consolidation by it-
self, but forces it by limiting the number of potential hubs per EoL terminal. At the
transition from 1 to 2 minimal runs, some transshipment terminals are dropped as
potential hubs and the flows of these terminals are consolidated and channeled to an
179 This scenario will serve as the exemplary scenario throughout the section. It is a typical scenario that
does not have any outstanding features but blends in with the other scenarios. The graphical representa-
tions become clearer if only one scenario is depicted. The general relations are the same for all scenarios.
180 It concerns only OD-demand that had a potential hub either as origin or destination and was connected
to the potential hub at a low constraint for minimal runs. This link was counted as an indirect link as it
has a transshipment terminal. However, limiting the number of transshipment terminals may drop that
very transshipment terminal from the list of transshipment terminals the EoL terminal is connected to.
If demand on that OD-market is sufficiently high to justify the direct transportation, nothing will actually
change for the EoL terminal. The only difference is that the algorithm perceives this connection now as
a direct one as both terminals, origin and destination, act as mere EoL terminals.
181 The variations in the total numbers result from the algorithm that accounts for them differently, see
footnote 180.
166
68%
67%
zation of indirect links

66%
65%
64%
63%
Utilization
62%
61%
60%
59%
1 2 3 4 5
Minimal frequency in runs / week
Figure 6.6.: Vehicle utilization against minimal frequency.
other transshipment terminal. Figure 6.6 highlights that the utilization of the indi-
rect vehicles was very low at 1 weekly run. By dropping transshipment terminals and
forcing terminal consolidation, empty capacity is used. Thereby, the number of lowly
utilized vehicles that were saved by reducing the number of transshipment terminals
is larger than the required additional vehicles to cope with demand and minimal fre-
quency restrictions to the remaining transshipment terminals. Furthermore, figure
6.6 shows that the utilization of the indirect links decreases after exceeding the num-
ber of 2 runs per week, a sign that an efficient situation has been reached for the given
dataset.
The centrality metrics (cf. figure A.7 in the appendix) show a slight increase in net-
work centrality as more weekly runs are planned. Then again, the indices are based on
the traffic shares in transportation volume but not on the shares of vehicles. There-
fore, the metrics will only react indirectly to the routing of shipments changing in
function of the number of connected transshipment terminals per EoL terminal. No
direct reaction to the increase of runs by themselves can be found.
Summing up, the main contribution of the minimal frequency to network design
lies not in increasing the number of trips but in restraining the number of allocated
transshipment terminals. Nonetheless, if no minimal frequency were to be guaran-
teed, the entire HH&S network would not be able to operate the way it is devised by
the algorithm.
6.2.2. Maximal accumulating days

The maximal accumulation days describe how long shipments are allowed to wait for
the direct transport. The α-rule takes this into consideration when establishing di-
rect connections between EoL terminals. Direct connections between EoL terminals
should be opened if enough shipment volume can be accumulated over time to jus-
167
0.85
0.80
are of direct shipments

0.75
0.70
0.65
0.60
Share
0.55
0.50
0 2 4 6 8 10
Maximal accumulation time [days]
Figure 6.7.: Share of direct shipments against maximal accumulation time in 255
scenarios.
tify reasonable load factors. The data shown in figure 6.7 confirms that the algorithm
behaves as it can be expected. It depicts four different cases an accumulation period
of 2, 4, 6, or 8 days and their respective share of direct shipments per scenario. It can
be seen that the share of direct shipments raises from about 0.64 (2 days of accumula-
tion) to 0.81 (8 days of accumulation) the longer the accumulation period is. Thus, if
the LSP allows for longer accumulation periods, more shipments may be transported
directly without transshipment.
Stability of the total transportation time of the indirect shipments is, furthermore,
impacted by the maximal accumulation time. Figure 6.8 depicts the coefficients of
variation of the indirect transportation times in the networks against the maximal
accumulation time for direct transports. The coefficients of variation rise as the accu-
mulation time for direct transports increases. When comparing transportation time
and its standard deviation, it becomes apparent that the increase of the coefficients
of variation is mainly driven by the increase in the standard deviation of the trans-
portation time, not by changes in the mean.182 The longer the accumulation period,
the more shipments are transported directly. When shipments are re-routed from in-
direct to direct transportation, this is done for all demand on an OD-market. These
shipments all have basically the same transportation time, but are no longer part of
the sample for calculating the standard deviation of the transportation times. There-
with, the standard deviation must rise as more heterogeneous shipments remain in
the sample. It must be noticed that the stability of the indirect transportation time
decreases as the maximal accumulation time for direct shipments increases.
One last interesting aspect to be considered in relation to the maximal accumula-
tion time is the centrality of the generated networks. Table 6.8 summarizes the data
obtained on the generated networks.183 MW index, NC index, and HC index all agree
182 The coefficient of variation is the standard deviation of the indirect transportation time over the mean of
the indirect transportation time (cf. figure A.9 in the appendix).
183 Figure A.10 in the appendix displays the obtained results graphically.
168
0.40
ansportation time indirect

0.35
0.30
CV transportation
0.25
0.20
0 20
0.15
0 2 4 6 8 10
Figure 6.8.: Coefficient of variation of indirect transportation time against maximal

accumulation time in 255 scenarios.
that the concentration of the network decreases as the maximal accumulation time
increases. This behavior is in accordance with the observation that more shipments
are transported directly as the maximal accumulation time increases.
Furthermore, the indices highlight that the networks become more similar to each
other as the maximal accumulation time increases: the standard deviation in concen-
tration values decreases. If there is only a very short time to accumulate shipments, it
depends strongly on the setting in the distinct scenario how the routing takes place.
The availability and location of potential hubs will have an impact on the decision
since the accumulation time cannot contribute to increasing load factors. As the ac-
cumulation time increases, this effect is decreased and the networks are more alike in
their concentration.
The ratio of HC index over connecting share behaves slightly different from the
other three metrics (cf. figure 6.9 for details). It depicts the degree of possibilities of
different routing options for indirect shipments. For the case of a maximum of 2 ac-
cumulation days, the ratio displays two types of cases. The upper cloud contains only
scenarios where two distinct EoL terminals are amongst the potential hubs that both
have a high OD-demand by themselves. This aspect will be revisited again when dis-
cussing the impact of potential hubs in the network. In the case of a maximum of 4
accumulation days, three groups of scenarios are depicted. The top cloud contains
scenarios where both of the EoL terminals are in the set of potential hubs. The mid-
dle cloud has either one of them. Finally, the scenarios in the lower cloud are the ones
that have neither of the two in their sets of potential hubs. Again, this effect dimin-
ishes as the maximal accumulation time increases and the scenarios become more
similar.
169
Maximal accumulation time

2 4 6 8
x̄ 0.388 0.359 0.327 0.324
MW index σ 0.028 0.025 0.019 0.019
CV 0.072 0.071 0.059 0.059
x̄ 0.734 0.720 0.703 0.701
NC index σ 0.005 0.004 0.002 0.002
CV 0.007 0.006 0.003 0.003
x̄ 0.307 0.249 0.165 0.166
HC index σ 0.007 0.004 0.001 0.001
CV 0.023 0.016 0.009 0.006
x̄ 0.855 0.867 0.843 0.871
HC index
connecting share
σ 0.018 0.014 0.009 0.009
CV 0.021 0.016 0.011 0.011
Table 6.8.: Centrality indices of 255 generated networks in relation to maximal accu-
mulation time.
170
0.90 0.90
0.89 0.89
0.88 0.88
0.87 0.87
0.86 0.86
0.85 0.85
0.84 0.84
0.83 0.83
0.82 0.82
0.81 0.81
HC index / connecting share (2 days)

0.80 0.80
Scenarios by increasing cost Scenarios by increasing cost
(a) 2 days max accumulation (b) 4 days max accumulation
0.90 0.90
0.89 0.89
0.88 0.88
0.87 0.87
0.86 0.86
0.85 0.85
0.84 0.84
0.83 0.83
0.82 0.82
0.81 0.81

0.80 0.80
Scenarios by increasing cost Scenarios by increasing cost
(c) 6 days max accumulation (d) 8 days max accumulation
Figure 6.9.: Ratio of HC index / connecting share depending on the maximal accumulation time in 255 scenarios.
171
6.2.3. Distance-related transportation cost

The distance-related transportation cost was set to different levels in comparison to
the value applied for the case study.184 Yet, the only impact that was deducted from
the simulation runs is the obvious: transportation cost increases the higher the cost
per distance.
Even the factor of 200 as compared to the transportation cost in the above demon-
strated case study only slightly impacts the decisions taken by the algorithm as table
A.4 in the appendix highlights with some KPIs for the 5-hub scenario. The impact is
negligible.
Even though it appears astonishing to see almost no impact of varying the trans-
portation cost on the network at first, this can be explained. The α-rule decides
whether an OD-market is served directly or indirectly. Equations 5.1 (direct cost) and
5.2 (indirect cost) describe the flow-distance-related transportation cost created by a
transport. Furthermore, equation 6.1 explains how the cost is calculated for the case
study. The cost per km c k m is an important input to these equations. When the trade-
off direct vs. indirect is decided by the α-rule, it will balance the costs against each
other. Thereby, most of the flow-distance cost cancels out. The impact of a higher
flow-distance cost must actually be of limited impact, even though this seems con-
tradictory at first.
This result is of particular interest for an LSP. It highlights that the quality of a net-
work appearance is not sensitive to increases in flow-distance-related transportation
cost, such as an increase in fuel cost.
6.2.4. Percentage of sold empty capacity

The impact of the percentage of the empty capacity that is sold on the market is very
high; actually higher than the impact of the variation of the transportation cost per
distance. Several relations can be seen. The exemplary 5-hub scenario will be used to
present the relations.185
The share of empty capacity sold on the market can be seen as a proxy for the value
given to avoid empty capacity. If the LSP is able to sell significant shares of its empty
capacity to some other player on the spot market for transportation, it reduces its own
costs. This decreases the cost per transported unit of shipment pay weight even if the
volume accumulated by the LSP does not by itself fill up an entire vehicle.
The following will show interesting insights on the network when the share of
empty capacity sold to the market varies. One aspect is obvious (cf. figure A.8 in
the appendix): if empty capacity is sold to the market, the flow-distance-related
transportation cost is reduced. But this is certainly of minor importance here.
184 The applied factors were of 0.6, 0.8, 1.0, 1.2, 1.4, 1.6, 1.8, 2.0, and 200.
185 The data points in this section are connected by an interpolated line to give a clearer impression of the
connection between the data points. The abscissa in this section is the share of empty capacity; connect-
ing the individual data points is feasible.
172
101%
ling cost at transshipment

100%
99%
terminals
98%
97%
Handling
96%
95%
0.0 0.2 0.4 0.6 0.8 1.0
Share of sold empty capacity
Figure 6.10.: Handling cost at transshipment terminals against share of sold empty
capacity.
The handling cost at transshipment terminals is a linear relation to the shipment

volume transshipped at the terminals. Figure 6.10 reveals that the more empty capac-
ity is sold to the market, the less transshipment operations take place.186 The higher
share of sold empty capacity leads to a decrease in flow-distance-related cost, but
does not impact the flow-related handling cost at transshipment terminals. The re-
lation between these two cost categories is shifted. Thus, the α-rule tends to allow
more directly conducted transports between EoL terminals.
Figure 6.11 gives an indication of the network concentration in relation to the share
of sold empty capacity. MW index, NC index, and HC index all agree that the network
becomes less concentrated as the share of sold empty capacity increases. This is not
surprising, since it was already outlined that more direct transportation takes place
in this case. As both, HC index and connecting share (cf. figure 6.10) decrease but at
different rates, the ratio of the two depicts, at first, even an increase before it stabilizes
at a level of 0.865.187
186 The 100% cost level is set to the cost induced by the 5-hub senarios serving as an example at a level of 0.0
share of sold empty capacity.
187 Figure 6.10 displays the decrease in transshipment cost that is directly related to a decrease in trans-
shipped pay weight of shipments.
173
174
0.350 0.720
0.349 0.719
0.348 0.718
NC index
MW index
0.347 0.717
0.346 0.716
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Share of sold empty capacity Share of sold empty capacity
(a) MW index (b) NC index

0.258 0.868
0.256
0.867
0.254
0.252 0.866
0.250 0.865
HC index
0.248
0.864
0 864
0.246
HC index / connecting share

0.244 0.863
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
Share of sold empty capacity Share of sold empty capacity
HC index
(c) HC index (d) connecting share
Figure 6.11.: Centrality indices against share of sold empty capacity.

2.000
Number off transshipment terminals

1.980
1.960
1.940
1.920 Average in
1.900 Average out
1.880
1.860
0.0 0.2 0.4 0.6 0.8 1.0
Figure 6.12.: Number of transshipment terminals connected to each EoL terminal

against share of sold empty capacity.
One parameter that gives a notion of network topology is worth noting in this con-
text; the number of transshipment terminals an EoL terminal is connected to.188 The
average number of transshipment terminals (outgoing and incoming) connected to
each EoL terminal in the network is presented in figure 6.12. The number of con-
nected transshipment terminals decreases only slightly, still, a clear relation to the
share of sold empty capacity exists. The increase in direct transportation, hence the
increase in shipment volume that is not transshipped, leads to the decrease of the
number of transshipment terminals connected to an EoL terminal.189
All outlined relations to sold empty capacity depict no further changes once the
share of empty capacity exceeds 0.4 or 0.5. This is an interesting point to note, as all
scenarios show one level of critical share of empty capacity. The critical share varies
among the scenarios, but one such level can always be identified.
6.2.5. Daily distance by vehicle

The daily distance a transport vehicle travels is another parameter to the algorithm,
but its impact is limited. It does not impact any parameters significantly apart from
the transportation time. Transportation time is related to the distance per day by
the transportation distance of each transport. For the case of direct shipments, the
transportation time simply is the distance divided by the daily distance. For the in-
direct transportation, an additional day per transshipment is added. In both cases,
additional waiting before the departure (-s) might occur. The coefficients of varia-
tion of the direct and the indirect transportation time against the distance per day are
presented in figure 6.13. An increase in daily distance stabilizes the transportation
188 Section 5.3 explained that the algorithm will limit the number of transshipment terminals if an EoL ter-
minal does not have sufficient incoming or outgoing volume.
189 The same relation is apparent in figure A.11 in the appendiy. It presents the average number of direct
vehicles from a EoL terminal per week simulated period. 71 to 73 of the 154 EoL terminals in the scenario
do not have any direct connections at all.
175
0.65
cient of variation of direct

0.60
transportation time
0.55 max
Coefficient mean
0.50 min
0.45
0 200 400 600 800 1000 1200
Kilometer per day
(a) Direct transports
0.40
ariation off indirectt
0.35
me
tation time
nt of variation
ansportation
transportation
0.30 max
mean
Coefficient
0 25
0.25 min
0.20
0 200 400 600 800 1000 1200
Kilometer per day
(b) Indirect transports
Figure 6.13.: Coefficient of variation of transportation times against daily distance.
time for indirect transports slightly. Direct transports are more impacted. The more
distance is traveled daily, the shorter the transportation time becomes and the more
similar they will be.
Setting this parameter to a realistic value allows the algorithm to create data com-
parable to real operations, yet the impact of modifying this parameter is limited.
6.2.6. The impact of the potential hubs in the network

The potential hubs specified for the algorithm impact the solution. Their number and
the different characteristics are a strong lever on cost, stability of transportation time,
and network centrality.
Number of potential hubs

The number of potential hubs in the network firstly impacts the total cost. Figure 6.14
shows the minimum, the maximum, and the mean of the total cost incurred by the
176
125%
Total cost (% of 8-hub scenario)

120%
115%
110%
max
105%
mean
100%
min
95%
90%
0 1 2 3 4 5 6 7 8 9
Potential hubs
Figure 6.14.: Total cost against number of potential hubs.
x-hub scenarios.190 Two conclusions can be taken from it. Firstly, the cost minimal
scenario is a 4-hub scenario, but this naturally depends on the dataset and cannot be
seen as a general conclusion. Secondly, the difference between minimum and maxi-
mum cost shrinks as the number of potential hubs in the scenario increases. There-
fore, the negative effect of using a suboptimal transshipment terminal is much higher
if only few transshipment terminals are available in the network.
Direct transportation time and distance are only marginally impacted by the num-
ber of potential hubs in the network (cf. figure A.13 in the appendix). But indirect
transportation is influenced.
Indirect transportation distance decreases the more transshipment terminals are
in the network (figure 6.15). The more potential hubs are available in the network,
the higher the chances for a shipment to have only a small detour to a transshipment
terminal. Offering more transshipment terminals will never lead to an increase in
average indirect transportation distance.
The stability of transportation time as measured by the coefficient of variation ac-
tually decreases with the increase of potential hubs in the networks. This results
from a rather stable standard deviation of indirect transportation times at the same
time that the average transportation time tends to decrease (cf. figure A.14 in the ap-
pendix).
Furthermore, the number of potential hubs in the network influences the centrality
of the networks that are generated. Table 6.9 summarizes the mean, the standard de-
viation and the coefficient of variation for all scenarios.191 MW index and the ratio of
HC index over connecting share react to the number of potential hubs in the network.
NC index and HC index show only a very slight decrease as the number of potential
hubs in the network increases.
190 There is only one 8-hub scenario. Hence, minimum, maximum, and mean are the same for it.
191 Figure A.12 in the appendix displays all data points available for all of the generated networks.
177
210%
ect transportation distance
190%
% of 8-hub scenario)
170%
150% max
130% mean
min
Indircect
(%
110%
90%
0 1 2 3 4 5 6 7 8 9
Potential hubs
Figure 6.15.: Indirect transportation distance against number of potential hubs.
178
Number of potential hubs
1 2 3 4 5 6 7 8
# scenarios 8 28 56 70 56 28 8 1
x̄ 0.409 0.390 0.372 0.357 0.344 0.332 0.323 0.317
MW index σ 0.007 0.014 0.016 0.017 0.017 0.016 0.013 0.000
CV 0.018 0.036 0.043 0.046 0.050 0.047 0.040 0.000
x̄ 0.730 0.726 0.723 0.720 0.718 0.716 0.714 0.712
NC index σ 0.001 0.002 0.002 0.002 0.002 0.002 0.001 0.000
CV 0.002 0.003 0.003 0.003 0.003 0.003 0.002 0.000
x̄ 0.259 0.255 0.251 0.249 0.247 0.245 0.244 0.243
HC index σ 0.002 0.002 0.002 0.002 0.002 0.001 0.001 0.000
CV 0.008 0.009 0.009 0.008 0.007 0.006 0.004 0.000
x̄ 0.850 0.854 0.860 0.867 0.874 0.881 0.888 0.895
HC index
connecting share
σ 0.007 0.010 0.011 0.012 0.012 0.010 0.008 0.000
CV 0.008 0.011 0.013 0.014 0.013 0.012 0.009 0.000
Table 6.9.: Centrality indices of generated networks against number of potential hubs.
179
This highlights that an increase of the number of potential hubs leads to a decrease
in network concentration. Whilst this is intuitive, it points to the aspect that the algo-
rithm actually makes use of all potential hubs. Even outlying potential hubs are used
for operations when offered to the algorithm, as there still is a benefit to be gained
from consolidation.
Characteristics of potential hubs

Which transshipment terminal is used in a scenario plays a role in the generated net-
works. Taking a closer look at the cost minimal networks that served as examples
earlier (e.g. table 6.5) reveals that five potential hubs figure within all of the scenarios,
while none of the three other potential hubs does.192 All good solutions have 2 to 5
potential hubs. The combined potential hubs mostly form circles around the center
of Europe. But the difference in terms of cost between the solutions is very small, they
all figure within 0.56% of the cost optimal solution. This difference in cost cannot, by
itself, help the planner to prefer one solution over the other. Moreover, the mere loca-
tion of the potential hub is of minor importance for the generated scenarios as long
as it is not located in the outskirts of the geographical area to be covered.
It is interesting to note that two EoL terminals seem to influence the distance trav-
eled indirectly strongly. An Austrian and a Dutch EoL terminal are present as potential
hubs in the top 67 scenarios with the lowest indirect travel distance. Their availability
is one of the factors leading to low indirect transportation time. Table 6.10 summa-
rizes the span of mean indirect transportation time, standard deviation of indirect
transportation time, and the coefficient of variations for all scenarios either contain-
ing both of the EoL terminals as potential hubs or lacking at least one of them. The
mean transportation time per scenario differs between those scenarios containing
both potential hubs and those who do not contain both.193 The coefficients of vari-
ation of the indirect transportation time tend to be higher for these as well. Again,
it is not only the number of potential hubs per scenario but more importantly the
question of which potential hubs are included.
A similar performance exists concerning network centrality. While MW index, NC
index, and HC index hardly react to the potential hubs in the scenario, the ratio of HC
index over connecting share does. If a potential hub in the United Kingdom of Great
Britain and Northern Ireland (UK) and in the Netherlands are available in the sce-
nario, the ratio reacts. Figure 6.16 compares minimum, mean and maximum value of
the ratio for the scenarios that either contain none, one, or both of these two potential
hubs. Quite interestingly, the intervals do not even overlap. It is graphically obvious
192 The five hubs are located in the Netherlands, the north of France, the north of Italy, the east of Austria,
and Denmark.
193 A Wilcoxon-Mann-Whitney test allows to reject the hypothesis that the mean of the mean indirect trans-
portation time for the two samples are equal at the 1% niveau.
180
Contains both EoL terminals Lacks at least one EoL terminal

x̄ 6.345 - 6.615 6.513 - 9.437
σ 1.973 - 2.114 1.873 - 2.233
CV 0.303 - 0.331 0.212 - 0.318
Table 6.10.: Indirect transportation time indicators as influenced by two explicit EoL
terminals.
0.90
0.89
ex / connecting share
0.88
0.87 max
0.86 mean
min
HC index
0.85
0.84
0.83
None, one, or both potential hubs in the network
Figure 6.16.: The impact of two distinct EoL terminals on the ratio of HC index / con-
necting share.
that these two potential EoL terminals influence the value of the ratio strongly.194 This
effect is much stronger than the number of potential hubs in the scenario as listed in
table 6.9. It must be pointed out that these two EoL terminals have the highest origin
and destination demand of all tested potential hubs.195
194 The linear regression that estimates the influence of having a certain potential hub in the scenario to the
ratio of HC index over connecting share is summarized in table A.5 in the appendix. It clearly reveals, as
well, that the two potential hubs F and G contribute strongly to the value of the ratio. These are the two
above mentioned potential hubs.
195 Grove and O’Kelly (1986) find that the demand at a hub in an airline network is very important for de-
termining the delays in the network; more demand at the hub leads to less delays. “Locally originating
demand at the hub appears to be a crucial determinant of delay” (Grove and O’Kelly, 1986, p. 116). Delay
was not measured by the algorithm. However, it was conceptually shown in chapter 3 that network con-
centration will lead to less reliable schedules, a different notion of delays. Thus, the identified trends by
the authors and in the algorithm’s results are comparable.
The location of transshipment terminals for freight transport is usually found by facility location ap-
proaches. In passenger transport the role of OD-traffic is more important (Woxenius, 2007, p. 737). The
identified importance of OD-demand in freight transport may question this approach.
181
6.2.7. Summarizing the most influential findings

The presented results show the impact of several of the algorithm’s parameters on rel-
evant network KPIs. The identified performance was explained and related to the de-
cisions taken by the algorithm as outlined in the previous chapter 5. Some of the iden-
tified and discussed interdependencies of parameters and network KPIs are of out-
standing importance for general network design. Four of these aspects are brought to
attention again below to highlight their outstanding role.
Minimal frequency The minimal frequency on the indirect connections influences

the generated networks in two different ways. Firstly, its mere existence allows the
routing to have a back-up solution in case the direct line does not exist or does not
provide enough capacity. Providing the back-up routing that interconnects the entire
LTL network enables a stable transportation time.
Secondly, the minimal frequency on indirect links actually forces consolidation to
take place on a heuristic basis. The number of transshipment terminals connected
to each EoL terminal will be lower the more weekly runs of vehicles between EoL ter-
minal and transshipment terminals are required by the minimal frequency. This is a
strong lever for exploiting terminal consolidation. Moreover, it also shows that this
effect will decline from a certain level on. A minimal frequency of 2 runs per week
is, on average, most beneficial in terms of vehicle utilization for the specific case of
the dataset used to generate and test the networks. This number will vary with the
demand for transportation.
It can be stated as a general insight for network design that considering the im-
plementation of a minimal frequency constraint is worthwhile. Furthermore, the fre-
quency needs not even be of many runs during a week. Merely guaranteeing its exis-
tence is beneficial for several aspects of network performance.
Critical level of sold empty capacity The cost of transporting goods is of course
an impact factor in network design. However, it was shown that the price of empty
vehicle capacity has a strong of influence on the network while the transportation
cost per distance hardly influences the network design.
The price of unused capacity was modeled as the share of empty capacity that is
(re-) sold to the market. Different effects on how the network changes as this param-
eter varies were shown earlier. One aspect is a key message for network design: all
scenarios showed one level of sold empty capacity from which on, no further changes
in the generated networks are perceived. Up to a certain point, the willingness to
accept empty capacity will influence the appearance of the network. Identifying the
critical level of sold empty capacity provides important knowledge of the sensitivity
of the network appearance to this assumption.
182
6.3. Indicators of network centrality to compare networks
Impact of potential hubs The number and the location of potential hubs in the
network is a relevant parameter to be set for network design. The impact of the num-
ber of potential hubs is consistent with intuitive reasoning. The more potential hubs
are in the network, the easier it is to send shipments to well located transshipment
terminals. Hence, the transportation distance decreases, so does transportation time.
What is more important to note, though, is the question of which potential hubs
to choose. Several aspects were shown based on the simulated network scenarios.
Firstly, several potential hubs form different networks that are very similar in terms
of cost. From this perspective, it seems unreasonable to optimize transportation net-
work design solely based on cost considerations. Secondly, not only the location of
the potential hubs seems to impact the network KPIs but also other characteristics.
One example shown earlier is that of the high influence of the two EoL terminals with
the highest origin and destination demand. In scenarios were either one or both of
these EoL terminals were in set of potential hubs, the ratio of HC index over connect-
ing share was significantly higher than in the other scenarios.
Therefore, the simulated networks and their performance confirm, once over, the
conclusion of chapter 2. Network design should not follow solely the dictate of cost
orientation but strive to include other, more expressive, decisive elements.
Expressiveness of the ratio of HC index over connecting share The ratio of

HC index over connecting share is one network KPI that proved to depict a different
aspect of network centrality than the other three applied network centrality indices.
It measures the degree of having different routing options at hand in a network for
the indirect shipments. Thereby, it points out a very specific viewpoint of network
centrality but does so with a high sensitivity.

The general relations between transportation network centrality and various network
KPIs were conceptually outlined in chapter 3. The algorithm is devised based on
knowledge of qualitative network centrality. It was decided that an HH&S network
with central and decentral elements will be beneficial for the market environment for
European LTL transport and the perceived market by the LSP. Therefore, the algo-
rithm creates hybrid networks made up of a decentral direct and a central back-up
network. The algorithm was used to generate networks varying by the set of potential
hubs. It is possible to draw comparisons between the measured KPIs and network
centrality as assessed by the algorithm in order to illustrate the relations between the
KPIs in the generated data. Notwithstanding, what is compared are the generated
networks, which all stem from the general frame provided by qualitative network cen-
trality. Still, general relations may be depicted amongst the networks.
183
0.430
0.410
0.390
0.370
MW index
0.350
0.330
0.310
0.290
0.270
0.250
0.60 0.65 0.70 0.75 0.80 0.85 0.90
Indirect vehicle utilization
(a) MW index
0.735
0.730
NC index
0.725
0.720
0.715
0 715
0.710
0.60 0.65 0.70 0.75 0.80 0.85 0.90
(b) NC index
0.265
0.260
HC index
0.255
0.250
0.245
0 245
0.240
0.60 0.65 0.70 0.75 0.80 0.85 0.90
(c) HC index
Figure 6.17.: Centrality metrics against indirect vehicle utilization.
184
Consolidation Consolidation in the algorithm is best measured by vehicle utiliza-

tion. When shipments are consolidated, the most apparent result is that vehicle uti-
lization increases. For the case of terminal consolidation, this concerns the indirect
links. When inventory consolidation is performed, there must be an effect on the di-
rect links. The setup of the algorithm does not allow for vehicle consolidation.
Figure 6.17 indicates that less concentrated networks tend to have lower indirect
vehicle utilization. The indirect vehicle utilization would be expected to increase if
more terminal consolidation takes place. Concentrated networks should have more
terminal consolidation than decentral networks. This is confirmed by the shown data.
Table 6.8 earlier summarized the results obtained for the relation between maximal
accumulation time and the centrality of the generated networks. It can be stated that
the longer the maximal accumulation time the less the networks are concentrated.
This is in accordance with the conceptually outlined relation. Decentral networks
tend to employ inventory consolidation for achieving efficiencies. The algorithm, too,
generates more decentral networks when it is allowed to store shipments over a longer
period of time.
Transportation distance and time Transportation time and transportation dis-

tance are directly connected by the logic applied in the algorithm. Therefore, it suf-
fices to discuss the relation of network centrality and one of the two. Figure 6.18 de-
picts the three concentration metrics against the indirect transportation time for all
255 scenarios. The conceptual reasoning that central networks tend to have longer
transportation distances is generally acceptable, even though the effect is marginal.
Link frequency Link frequency is a key element in the entire algorithm. Thus, it
is worthwhile to illustrate the relation between network centrality and frequency on
the indirect connections in the network. Figure 6.19 shows clearly that the more con-
centrated the networks are the more vehicles are operated on indirect lines.196 This
confirms the considerations outlined in chapter 3.
Cost The total cost in the generated networks is strongly driven by the mean indi-
rect transportation distance. Figure 6.20 confirms this relation clearly. It is not sur-
prising that a certain relation between the centrality of the network and the total cost
of each network is found. Even though only indicatively, more concentrated networks
tend to be costlier than their less concentrated counterparts. This seems, at first, to
contradict the result from chapter 3.
196 It was discussed in depth above how the minimal frequency impacts the number of connected hubs in
the network. The x-axis in figure 6.19 is the average number of vehicles per indirect connection. More-
over, the networks shown in the figure are simulated with a minimal frequency of only 1 vehicle per
indirect connection which gives the algorithm the freedom to plan the necessary number of vehicles it
needs to operate the network. The effects are not constraint by any minimum requirements.
185
0.430
0.410
0.390
0.370
MW index
0.350
0.330
0.310
0.290
0.270
0.250
1,500 2,000 2,500 3,000 3,500 4,000
Indirect shipment distance [km]
(a) MW index
0.735
0.730
NC index
0.725
0.720
0.715
0 715
0.710
1,500 2,000 2,500 3,000 3,500 4,000
(b) NC index
0.265
0.260
HC index
0.255
0.250
0.245
0 245
0.240
1,500 2,000 2,500 3,000 3,500 4,000
(c) HC index
Figure 6.18.: Centrality metrics against indirect transportation distance.
186
0.430
0.410
0.390
0.370
MW index
0.350
0.330
0.310
0.290
0.270
0.250
5 7 9 11 13 15 17
Mean number of vehicles (arrival + departure) per indirect link
(a) MW index
0.735
0.730
0.725
0.720
NC index
0.715
0.710
0.705
0.700
0.695
5 7 9 11 13 15 17
(b) NC index
0.270
0.260
0.250
HC index
0.240
0.230
0.220
0.210
0.200
5 7 9 11 13 15 17
(c) HC index
Figure 6.19.: Centrality metrics against number of arriving and departing vehicles at a
terminal per indirect connection.
187
3,500
3,300
3,100
rect distance [km]

2,900
2,700
2,500
2,300
Indirect
2,100
1,900
1,700
1,500
Total cost
Figure 6.20.: Indirect transportation distance against cost.
It was pointed out that the question on whether central or decentral networks are
more costly depends on the setting in the network. For the given setting, the flow-
distance-related transportation cost dominates the total costs of the network. It is
only normal that central networks with longer transportation distances generally are
more costly. This finding does, thereby, not contradict the general conceptual con-
sideration that the relation between network centrality and network cost depends on
the given setting.
6.4. The benefit of including network centrality in the

network design process
This chapter presented the results obtained with the help of the algorithm outlined
in chapter 5. Firstly, it conveyed type of results for a project conducted with a large
LSP. The algorithm allowed to compare different network scenarios to a base case
by their network KPIs, including cost, transportation time, and network centrality.
Various advantages and disadvantages of the scenarios were pointed out. The final
decision concerning which network scenario should best be implemented for the LSP
is supported by the analysis but must eventually be taken by the LSP.
The chapter secondly exhibited interdependencies of parameters applied in the al-
gorithm and aspects of network performance. These allowed to draw general insights
for network design and support strategic decision-making. The impact of different
parameters on the final solution was assessed and those parameters with outstand-
ing importance to the final network design were pointed out.
Research limitations
Perturbative effects were not simulated; the previously developed conceptual rela-
tions between network centrality and network performance when the network faces
188
6.4. The benefit of including network centrality in the network design process
0.430
0.410
0.390
0.370
MW index
0.350
0.330
0 310
0.310
0.290
0.270
0.250
Total cost
(a) MW index
0.735
0.730
0.725
NC index
0.720
0.715
0.710
Total cost
(b) NC index
0.265
0.260
0.255
HC index
0.250
0.245
0.240
Total cost
(c) HC index
Figure 6.21.: Centrality metrics against total cost.
189
perturbations were not revisited. Capturing perturbative effects in network perfor-

mance requires a different approach in modeling.
Furthermore, the shipment data that was used in the application of the algorithm
stems from one single LSP and reflects its operations. The entries in the dataset are
consistent with general information available for the LTL market. Therefore, it seems
reasonable to assume that the shipment dataset is a good example for European LTL
networks. Yet, the shipment data impacts the network performance, and further
datasets from other LSPs on the market would certainly be interesting as modified
inputs to the algorithm.
Finally, the algorithm is specific to the European LTL market. In a way, this is nat-
ural for the chosen approach. Qualitative network centrality set the frame in which
the algorithm generates networks. The frame is based on the environment in the Eu-
ropean LTL market. Consequently, the algorithm must be specific to this context.
Nevertheless, it will be interesting to see how other networks behave. For example,
intermodal or air cargo networks have many similarities to LTL networks, but differ
in some aspects; e.g. the handling operations at transshipment terminals are very dif-
ferent from those in LTL transportation. The algorithm might be a good starting point
for research in this field.
None of the above was intended to identify any interpretation of an optimal degree
of network centrality. Some general tendencies may be indicated regarding a more or
less central network, given an environmental setting and the expectations of a net-
work provider towards a future transportation network. Network centrality was iden-
tified as the link between network appearance and network performance for network
design. Thus, it allows to indicate probable performance of networks when they take
a certain appearance. It allows to match the environmental setting and the provider’s
expectations with a network appearance. Detailed analysis on how to design the net-
work must follow this first step. Hence, there is no optimal degree of network central-
ity, and it does not appear to be beneficial to optimize networks to their centrality.
Network centrality and its relation to network performance to pre-structure

the algorithm
Knowledge of qualitative network centrality as developed in section 3.2 served to pre-
structure the algorithm. The present chapter depicted the type of results obtained
by applying the algorithm to a dataset. In particular section 6.1 exhibited that the
algorithm achieves to generate networks whose overall performance matches the de-
sired performance: stable transportation time while respecting cost constraints. The
generated networks prove a higher stability in transportation time than the base case.
Moreover, their cost is competitive in comparison to the base case as well.
The 255 generated scenarios differ in their KPIs. Some of them are unattractive
from a cost perspective. Others have a stable transportation time but these are too
long to provide an interesting service offer on the European LTL market. Still, many
190
6.4. The benefit of including network centrality in the network design process
of the generated networks in the case study are applicable alternatives for the LSP
to continue to evaluate as options for network redesign. None of the generated net-
works is far beyond thought for the application to the European LTL context. All net-
works match the desired network performance to a certain degree. The number of
255 scenarios remains handleable for a network design process. This is largely due to
the purposeful pre-structure of a decentral network combined with a central back-up
network. Knowledge of qualitative network centrality led to this structure.
Summing up, the qualitative understanding of network centrality and its relation
to network performance proved to be a powerful tool to describe the appearance that
a potential European LTL network should have in order to achieve certain elements of
network performance. It allowed to develop and implement an algorithm to generate
and test various network scenarios. The interpretation of the generated data served as
an illustrative example that the networks match the desired performance. Qualitative
network centrality pre-structures design decisions for different network types: the
desired performance of a national rail network will certainly differ from European LTL
networks. All the same, allowing the strategic planing phases to be guided by network
centrality will focus the network design, no matter what the application.
Network centrality and its relation to network performance to compare

network scenarios
Measuring the centrality of the simulated networks was easy. The network perfor-
mance for perturbations was not simulated. Still, general conclusions to these as-
pects were be drawn based on the reasoning indicated in chapter 3. This is the addi-
tional benefit of metrics of network centrality. Conclusions can be drawn for unmea-
sured elements of network performance based on the conceptual relations between
network centrality and network performance. Hence, when comparing different sce-
narios regarding their network centrality, strengths and weaknesses of the scenarios
amongst each other may be identified that go beyond the KPIs accounted for in the
algorithm. The power of this approach is based on the wide perspective of network
centrality. Firstly, it is measured without additional assumptions. Secondly, centrality
is related to various elements of network performance and can be interpreted from
different viewpoints.
Measures of a network’s centrality are obtained based on its layout and its traffic
flows. No additional assumptions are necessary, e.g. to node or link capacity or to
levels of demand uncertainty.
Network centrality is related to various elements of performance, so that it is not
necessary to specify a complete list of elements of network performance before de-
signing the algorithm. Network centrality provides a comparison of networks on a
more general level.
Insights on general network performance for stable situations and when perturba-
tions in the network occur can be found based on the networks’ centrality as mea-
191
sured by different metrics. This quantitative understanding provides a benefit for the
comparison of network scenarios in network design. Comparing network scenarios
based on different indicators of network performance offers an additional perspec-
tive to the existing and often applied KPIs. The easy accounting for network centrality
provides information on diverse aspects of network performance that are more dif-
ficult to keep track of, especially as their number can be large and certainly varies
with specific interests of a network planner. The transportation networks perform ac-
cording to their network appearance. Centrality is at the core of many of aspects of
network performance as was shown conceptually and contrasted with data generated
by the algorithm. It is the concept connecting network appearance and network per-
formance. Thereby, centrality - in its qualitative and its quantitative understanding -
is an influence factor for network performance.
192
7. Closing remarks
It was the goal of this thesis to emphasize the outstanding role of network centrality
for transportation network design. This was undertaken by, firstly, conceptualizing
the relations of network centrality to other transportation network key performance
indicators (KPIs) capturing network performance: consolidation, link frequency,
transportation distance and time, schedule reliability, network vulnerability, and
system capacity flexibility. This provided a qualitative understanding of trans-
portation network centrality. An algorithm to generate and test less-than-truckload
(LTL) transportation networks was presented, making use of this qualitative under-
standing. Secondly, quantitative measures to account for LTL network centrality
were furthermore suggested and applied in interpreting the data generated by the
algorithm.
The thesis was structured in seven chapters. The first chapter introduced the back-
ground and the goal of the thesis. Chapter 2, Transportation networks and their opti-
mal design, presented transportation, transportation networks, and Operational Re-
search (OR) approaches to network design. These approaches are often cost-oriented
optimizations. Additional research streams including more service-oriented aspects
into objective functions of OR models were touched on. Some examples of research
striving to identify general trade-offs in transportation network design were pointed
out. Summing up, this chapter concluded that a cost-oriented approach in trans-
portation network design may fail to create good networks that satisfy user or provider
expectations in terms of service, flexibility, risk, and others on a strategic planning
horizon. Thus, more general approaches are necessary.
The third chapter, Transportation network centrality, suggested the inclusion of
transportation network centrality in network design decisions. The chapter firstly
explained that topological network centrality is a descriptive way to measure a net-
work’s layout. Network concentration refers to how flows travel through the network.
These two perspectives form the overall concept of transportation network central-
ity. Transportation networks differ in their appearance; hub-&-spoke (H&S) networks
and point-to-point (P2P) networks are often seen as two ends of the scale of possible
network appearances. It influences the way a network will perform when shipments
are transported. For example, a central H&S network will have longer transportation
distances on average as compared to a decentral P2P network. The chapter then con-
tinued to conceptually develop qualitative relations between network centrality and
various aspects of network performance. It secondly presented quantitative network
centrality indices. Having introduced the relations between network performance

193
7. Closing remarks
and network centrality, it stood to reason to identify ways to expressively measure the
degree of network centrality. Being able to measure network centrality is a precon-
dition for comparing different networks based on it. Various centrality metrics from
social sciences were presented. Furthermore, measuring passenger airline network
centrality proved to be a beneficial example to highlight how these measures have
already been applied to the specific context of transportation networks.
Chapter 4, Less-than-truckload network centrality, led over from the part of the the-
sis that is general to all transportation networks to the concrete context of LTL trans-
portation networks. To achieve this, the chapter firstly introduced logistics service
providers (LSPs) since these companies operate LTL transportation networks. The
chapter continued by giving background information on the European LTL market in
terms of market size, players, operations, service, and demand. Finally, the McShan-
Windle (MW) index, the network concentration (NC) index, the hubbing concentra-
tion (HC) index, and the ratio of HC index over connecting share were presented as
expressive measures to account for different aspects of LTL network concentration.
The fifth chapter, An algorithm for less-than-truckload network design, illustrated
an algorithm to generate LTL transportation networks and simulate the transport of
shipments on these. The algorithm structure is fundamentally influenced by insights
on qualitative network centrality. It is an example of applying this conceptual knowl-
edge for strategic transportation network design. Considering the LTL market envi-
ronment and the desired future network performance in terms of stable transporta-
tion time, the algorithm strives to find a topologically decentral network for direct
transports between end-of-line (EoL) terminals and a concentrated back-up network
making use of transshipment terminals. This approach is based on the argumenta-
tion concerning network performance and its relation to transportation network cen-
trality. The reasoning behind the algorithm, input, output, and parameters, as well as
the algorithmic logic were presented in detail.
Finally, chapter 6, Generated networks and their performance, presented the results
obtained with the help of the algorithm. It firstly depicted the comparison to a base
case scenario established based on the realistic situation from the LTL market and,
inter alia, showed that transportation time becomes more stable when applying the
logic implemented in the algorithm. Thus, making use of qualitative network central-
ity to pre-structure the algorithm in order to achieve the desired performance served
its purpose. Secondly, the chapter presented extensive simulations that allow to draw
general conclusions for network design in the given context. Amongst others, it was
highlighted that the minimal frequency has an impact on the generated networks.
Surprising at first, this impact does not directly relate to the number of vehicles on
indirect links, but is a much stronger lever by limiting the number of transshipment
terminals per EoL terminal. Furthermore, the assumptions about empty capacity in-
fluence the network appearance highly; even more than the level of flow-distance-
related transportation cost. Lastly, the generated results indicate that the location of
potential hubs in the network is important, but other characteristics such as origin-
194
destination (OD)-demand at the potential hubs will also modify the network KPIs ob-
tained with the help of the algorithm. From a methodological point of view, the ratio
of HC index over connecting share proved to be helpful for summarizing the degree
of alternative routing options in the indirect back-up network. Consequently, it is a
beneficial tool to analyze LTL networks. Research limitations and directions for fu-
ture work were pointed out. In summary, the chapter illustrated how applying the
thinking in dimensions of network centrality is beneficial for transportation network
design. Quantitatively measuring network centrality allowed conclusions as to net-
work performance in elements that were not simulated, e.g. the risk of congestion.
Hence, the chapter is the applied counterpart to the findings of chapter 3.
Summing up, strategic network design nowadays has to find good networks that
reach multiple objectives in desired performance. Traditional, mono-objective
optimization is challenged. Transportation network centrality can be applied as a
higher level criterion to pre-structure strategic network design decisions according to
the network’s desired performance. By its encompassing nature, network centrality
allows to support strategic decision-making. The conceptually devised relations of
transportation network centrality and network performance are fundamental for
this insight. This idea was applied in an algorithm that generates and tests network
scenarios for European LTL networks. Qualitative network centrality set the frame-
work of network appearance that most likely achieves performance expectations.
Within the framework set by network centrality, well-established decision criteria
such as cost may be used for creating the networks. Reviewing the generated data
allowed to conclude that the generated networks fulfill the desired performance
as expressed prior to the development. Consequently, the inclusion of qualitative
network centrality to pre-structure the algorithm served its purpose.
The quantitative aspect of measuring transportation network centrality allows to
compare the generated network scenarios in detail. This is of importance especially
regarding conclusions on aspects of network performance that were not covered in
the scope of the simulations. For instance, comparing the concentration indices of
the generated networks allows to conclude on the risk of congestion in the network.
If one of these aspects of performance is identified as a major opportunity or threat
to the design, it can then be analyzed in detail.
This thesis set out to present network centrality as a valuable perspective in trans-
portation network design. Network centrality is a core concept connecting the ap-
pearance of a transportation network with its performance. This relation as well as
the outstanding role of network centrality has thus far been mostly neglected by re-
searchers.
Having in mind how strongly societies rely on transportation networks, designing
good networks in multifarious perspectives is crucial. Eyjafjallajökull’s eruption, that
served as the example to open this thesis, only provided us with a glimpse of what
may happen when today’s transportation networks fail.
195
7. Closing remarks
Granting a more prominent position to transportation network centrality leads to

enriching traditional, cost-oriented transportation network design by supporting the
inclusion of additional performance indicators in the design process due to its en-
compassing nature. Designing networks to meet various elements of desired perfor-
mance will rise in importance in the future; and this thesis will hopefully serve to
inspire that research.
196
Maybe, then, more often, there will be something in the air.
197
A. Appendix

199
Edition KWV, https://doi.org/10.1007/978-3-658-24241-1
A. Appendix
100
B
ve % of y
A
Cumulativ
C
0 C
Cumulative
l i % off x 100
Figure A.1.: The Lorenz curve of a distribution (Rodrigue et al., 2009, p. 185).
200
LSP Website Geographical coverage Delivery time nat. /
neighbor
DB Schenker Logistics www.schenker.de Europe 24/48
Dachser www.dachser.com Europe, East-Europe, 24/48
Maghreb
Deutsche Post DHL www.dhl.com Europe, CIS, Middle East, 24/48
North Africa
Geodis www.geodis.com 21 European countries proof of delivery
within 48 hours
Kühne & Nagel www.kn-portal.com 38 European countries fixed transit times
DSV www.dsv.com Europe 24/48
Hellmann Worldwide www.hellmann.net cf. System Alliance
Mory Group SA (F) www.morygroup.com France 24/48
IDS Logistik GmbH www.ids-logistik.de European coverage 24/48
System Alliance www.systemalliance.de Europe 24/48
CargoLine GmbH www.cargoline.de Europe 24/48
24plus Systemverkehre www.24plus.de Europe 24 (nat.
Heppner SA (F) www.groupe-heppner.com Europe 24/72
Ziegler Group www.zieglergroup.com Europe 24/48
Table A.1.: European LTL: Coverage and delivery times (information compiled from the websites as specified; last access on
September 13, 2010).
201
202
A. Appendix
LSP Guaranteed delivery Track & trace Specials

DB Schenker Logistics yes yes yes
Dachser yes yes yes
Deutsche Post DHL yes yes yes
Geodis * yes *
Kühne & Nagel * yes Industry solutions
DSV Special transport services
Hellmann Worldwide cf. System Alliance
Mory Group SA (F) * yes *
IDS Logistik GmbH yes yes Procurement logistics
System Alliance yes yes Procurement logistics
CargoLine GmbH yes yes Warehousing etc.
24plus Systemverkehre yes yes Procurement, distribution
Heppner SA (F) yes yes Packaging
Ziegler Group * * Special transport services
Table A.2.: European LTL: Services (information compiled from the websites as specified; last access on September 13, 2010),
* = no information on website.
Number of shipments
Initial 204,308
Lacking departure EoL terminal data 22,205
Lacking arrival EoL terminal data 83,403
Lacking and infeasible delivery data 22,292
Dummy values in departure EoL terminal date 7,666
Dummy values in arrival EoL terminal date 1,838
Dummy variables in delivery date 135
Excluded from base case (transportation time) 22,485
Excluded from base case (cost) 13
Table A.3.: Value approximation for base case scenario.
0.60
0.50
icient of variation of
nsportation time
0.40
transportation
0.30
Coefficient
0 20
0.20
0.10
0.00
Scenarios
(a) Direct coefficients of variation
0.60
0.50
icient of variation of
nsportation time
0.40
transportation
0.30
Coefficient
0 20
0.20
0.10
0.00
Scenarios
(b) Indirect coefficients of variation
Figure A.2.: Coefficient of variation of transportation times in 255 scenarios.
203
A. Appendix
0.76
0.75
are of direct shipments
0.74
0.73
0.72 max
75% quartile
0.71
25% quartile
Share
0.70
min
0.69
0.68
0 1 2 3 4 5 6
Figure A.3.: Share of direct shipments against minimal frequency.
4.0
Number off transshipment terminals
3.5
3.0
2.5
2.0
Average out
1.5
Average in
1.0
0.5
0.0
0 1 2 3 4 5 6
Minimum frequency in runs / week
Figure A.4.: Number of connected transportation terminals per terminal against min-
imal frequency constraint.
204
7
ansportation time in days

6
5
Transportation 4
3 Average direct shipments

2 Average indirect shipments
0
0 1 2 3 4 5 6
(a) Mean transportation time
0.6
0.5
CV transportation time
0.4
0.3
CV_direct
0.2
CV_indirect
0.1
01
0.0
0 1 2 3 4 5 6
(b) Coefficient of variation
Figure A.5.: Coefficients of variation of transportation time against minimal fre-

quency constraint.
2,000
1,800
1,600
1,400
Distance [km]
1,200
1,000
800 Mean indirect shipment distance
600 Mean direct shipment distance
400
200
0
0 1 2 3 4 5 6
Minimum frequency in runs / week
Figure A.6.: Mean shipment distance against minimal frequency constraint.
205
A. Appendix
1.0
0.9
0.8
0.7
0.6
0.5 MW index
0.4 NC index
0.3 HC index
0.2 HC index / connecting share
0.1
0.0
0 1 2 3 4 5 6
Figure A.7.: Centrality metrics against minimal frequency constraint.
206
Factor of transportation cost
0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 200
Share direct volume 0.717 0.717 0.717 0.717 0.717 0.717 0.717 0.717 0.711
Mean weekly trucks (dep.) 12.4 12.4 12.4 12.4 12.4 12.4 12.4 12.4 12.1
Mean # of hubs (in) 1.920 1.920 1.920 1.920 1.920 1.920 1.920 1.920 1.947
Mean # of hubs (out) 1.867 1.867 1.867 1.867 1.867 1.867 1.867 1.867 1.873
MW index 0.347 0.347 0.347 0.347 0.347 0.347 0.347 0.347 0.348
NC index 0.717 0.717 0.717 0.717 0.717 0.717 0.717 0.717 0.718
HC index 0.245 0.245 0.245 0.245 0.245 0.245 0.245 0.245 0.250
Table A.4.: KPIs against distance-related transportation cost.
207
A. Appendix
120%
100%
80%
Transport cost
60%
40%
20%
0%
0.0 0.2 0.4 0.6 0.8 1.0
Figure A.8.: Transportation cost in the network against share of empty capacity sold.
208
3.0
2.8
ndard deviation indirect
2.6
transportation time
2.4
2.2
2.0
1.8
Standard
1.6
14
1.4
1.2
1.0
0 2 4 6 8 10
(a) Standard deviation
11
direct transportation time
10
8
Mean indirect
5
0 2 4 6 8 10
(b) Mean
Figure A.9.: Indirect transportation time per scenario against maximal accumulation
time.
209
A. Appendix
0.50
0.45
MW index
0.40
0.35
0.30
0 30
0.25
0 2 4 6 8 10
(a) MW index
0.80
0.75
NC index
0.70
0.65
0.60
0 2 4 6 8 10
(b) NC index
0.35
0.30
HC index
0.25
0.20
0.15
0 15
0.10
0 2 4 6 8 10
(c) HC index
Figure A.10.: Centrality indices against maximal accumulation time.
210
12.5
12.4
Number of direct vehiclees
12.3
12.2
12.1
12 0
12.0
11.9
11.8
11.7
0.0 0.2 0.4 0.6 0.8 1.0
Figure A.11.: Number of direct vehicles in the network against share of sold empty
capacity.
Variable Coefficient Standard error

Constant 0.8410*** 0.0003
Potential hub A 0.0026*** 0.0002
Potential hub B 0.0046*** 0.0002
Potential hub C 0.0019*** 0.0002
Potential hub D 0.0038*** 0.0002
Potential hub E 0.0016*** 0.0002
Potential hub F 0.0184*** 0.0002
Potential hub G 0.0206*** 0.0002
Potential hub H -0.0011*** 0.0002
Adjusted R 2 0.9913
F-statistic 3601.3222***
Table A.5.: Regression results.
211
A. Appendix
1.0 1.0
0.9 0.9
0.8 0.8
MW index MW index
0.7 0.7
0.6 NC index 0.6 NC index
0.5 0.5
0.4 HC index 0.4 HC index
0.3 0.3
HC index / connecting HC index / connecting
0.2 share 0.2 share
0.1 0.1
0.0 0.0
Scenarios by increasing total cost Scenarios by increasing total cost
(a) 1-hub scenario (b) 2-hub scenario
1.0 1.0
0.9 0.9
0.8 0.8
MW index MW index
0.7 0.7
0.5 0.5
0.3 0.3
0.2 share 0.2 share
0.1 0.1
0.0 0.0
(c) 3-hub scenario (d) 4-hub scenario
1.0 1.0
0.9 0.9
0.8 0.8
MW index MW index
0.7 0.7
0.5 0.5
0.3 0.3
0.2 share 0.2 share
0.1 0.1
0.0 0.0
(e) 5-hub scenario (f) 6-hub scenario
1.0 1.0
0.9 0.9
0.8 0.8
MW index MW index
0.7 0.7
0.5 0.5
0.3 0.3
0.2 share 0.2 share
0.1 0.1
0.0 0.0
(g) 7-hub scenario (h) 8-hub scenario
Figure A.12.: Centrality indices against number of hubs.
212
102%
ct transportation distance
101%
100% max
mean
99% min
(%
Dircect
98%
0 1 2 3 4 5 6 7 8 9
Potential hubs
(a) Distance
102%
ect transportation time
101%
100% max
mean
Direct
99% min
(%
98%
0 1 2 3 4 5 6 7 8 9
Potential hubs
(b) Time
Figure A.13.: Direct transportation against number of potential hubs.
213
A. Appendix
2.30
dard deviation of indirect
2.25
2.20
transportation time
2.15
2.10
2.05 max
2.00
mean
1.95
Standard
min
1 90
1.90
1.85
1.80
0 1 2 3 4 5 6 7 8 9
Potential hubs
(a) Standard deviation
0.34
direct transportation time
0.32
0.30
0.28
max
0.26
mean
CV indirect
0.24
min
0.22
0.20
0 1 2 3 4 5 6 7 8 9
Potential hubs
(b) Coefficient of variation
Figure A.14.: Indirect transportation times against number of potential hubs.
214
References
Mostafa Abd-El-Barr. Topological network design: A survey. Journal of Network and

Computer Applications, 32(3):501–509, 2009.
Ravindra K. Ahuja, Thomas L. Magnanti, and James B. Orlin. Network flows: theory,
algorithms, and applications. Prentice Hall, Englewood Cliffs, N.J., 1993.
Sascha Albers. Kooperationsorganisation von Stückgutnetzwerken. In Sascha Albers

and Markus Reihlen, editors, Management integrierter Wertschöpfungsnetzwerke,
pages 113–144. Kölner Wissenschaftsverlag, Köln, 2009.
Marco Alderighi, Alessandro Cento, Peter Nijkamp, and Piet Rietveld. Assessment of
New Hub-and-Spoke and Point-to-Point Airline Network Configurations. Transport
Reviews, 27(5):529–549, 2007.
Sibel Alumur and Bahar Y. Kara. Network hub location problems: The state of the art.
European Journal of Operational Research, 190(1):1–21, 2008.
Daniela Ambrosino and Maria Grazia Scutellà. Distribution network design: New
problems and related models. European Journal of Operational Research, 165(3):
610–624, 2005.
Jardar Andersen, Teodor Gabriel Crainic, and Marielle Christiansen. Service network
design with management and coordination of multiple fleets. European Journal of
Operational Research, 193(2):377–389, 2009.
Henrik Andersson, Arild Hoff, Marielle Christiansen, Geir Hasle, and Arne Løkketan-
gen. Industrial aspects and literature survey: Combined inventory management
and routing. Computers & Operations Research, 37(9):1515–1536, 2010.
Anthony B. Atkinson. On the measurement of inequality. Journal of Economic Theory,

2(3):244–263, 1970.
Anthony B. Atkinson and Andrea Brandolini. On Analyzing the World Distribution of

Income. World Bank Econ Rev, 24(1):1–37, 2010.
Turgut Aykin. The hub location and routing problem. European Journal of Opera-
tional Research, 83(1):200–219, 1995.

215
Edition KWV, https://doi.org/10.1007/978-3-658-24241-1
References
Michael Ball, Cynthia Barnhart, George Nemhauser, and Amedeo R. Odoni. Air Trans-
portation: Irregular Operations and Control. In Cynthia Barnhart and Gilbert La-
porte, editors, Transportation, Handbooks in Operations Reserach and Manage-
ment Science, pages 1–67. North-Holland, Amsterdam et al., 2007.
Ronald H. Ballou. Logistics Network Design: Modeling and Informational Considera-

tions. International Journal of Logistics Management, 6(2):39–54, 1995.
Ronald H. Ballou. Business logistics/supply chain management: planning, organizing,

and controlling the supply chain. Pearson/Prentice Hall, Upper Saddle River, N.J.,
5th edition, 2004.
Ronald H. Ballou, Handoko Rahardja, and Noriaki Sakai. Selected country circuity
factors for road travel distance estimation. Transportation Research Part A: Policy
and Practice, 36(9):843–848, 2002.
James Banks, Wade Driscoll, and Robert Stanford. Design Methodology for Airport
Limousine Service. Transportation Science, 16(2):127, 1982.
Philippe Barla and Christos Constantatos. Airline network structure under demand
uncertainty. Transportation Research Part E: Logistics and Transportation Review,
36(3):173–180, 2000.
Philippe Barla and Christos Constantatos. Strategic interactions and airline network
morphology under demand uncertainty. European Economic Review, 49(3):703–
716, 2005.
Cynthia Barnhart, Niranjan Krishnan, Daeki Kim, and Keith Ware. Network Design
for Express Shipment Delivery. Computational Optimization and Applications, 21
(3):239–262, 2002.
Cynthia Barnhart, Peter Belobaba, and Amedeo R. Odoni. Applications of Opera-

tions Research in the Air Transport Industry. Transportation Science, 37(4):368–391,
2003a.
Cynthia Barnhart, Amy M. Cohn, Ellis L. Johnson, Diego Klabjan, George L.

Nemhauser, and Pamela H. Vance. Airline Crew Scheduling. In Randolph W.
Hall, editor, Handbook of Transportation Science, International series in operations
research and management science, pages 517–560. Kluwer Academic Publishers,
Boston, 2nd edition, 2003b.
Leonardo J. Basso and Sergio R. Jara-Díaz. Are Returns to Scale with Variable Network
Size Adequate for Transport Industry Structure Analysis? Transportation Science,
40(3):259–268, 2006a.
216
References
Leonardo J. Basso and Sergio R. Jara-Díaz. Distinguishing Multiproduct Economies

of Scale from Economies of Density on a Fixed-Size Transport Network. Networks
and Spatial Economics, 6(2):149–162, 2006b.
Richard Batley, Joyce Dargay, and Mark Wardman. The impact of lateness and re-
liability on passenger rail demand. Transportation Research Part E: Logistics and
Transportation Review, 47(1):61–72, 2011.
Helmut Baumgarten and Jack Thoms. Trends und Strategien in der Logistik: Supply
Chains im Wandel; Ergebnisse 2002. TU Berlin, Bereich Logistik, Berlin, 2002.
Alex Bavelas. A Mathematical Model for Group Structures. Human Organization, 7

(3):16–30, 1948.
Alex Bavelas. Communication Patterns in Task-Oriented Groups. The Journal of the

Acoustical Society of America, 22(6):725–730, 1950.
Mokhtar S. Bazaraa, John J. Jarvis, and Hanif D. Sherali. Linear programming and
network flows. Wiley, New York, 2nd edition, 1990.
BBC News. Iceland volcano: Nissan and BMW suspend some production, 2010. URL
news.bbc.co.uk/2/hi/8631676.stm. (last accesssed November 1, 2010).
Wolfgang Berens and Werner Delfmann. Modellbildung und quantitative Methoden

zur Lösung logistischer Probleme. In Herbert Jacob, editor, Logistik, volume 32 of
Schriften zur Unternehmensführung, pages 31–55. Gabler, Wiesbaden, 1984.
Wolfgang Berens and Franz-Josef Körling. Estimating road distances by mathematical

functions. European Journal of Operational Research, 21(1):54–56, 1985.
Magnus Berglund, Peter van Laarhoven, Graham Sharman, and Sten Wandel. Third-
Party Logistics: Is There a Future? International Journal of Logistics Management,
10(1):59–70, 1999.
Oded Berman and Qian Wang. Inbound Logistic Planning: Minimizing Transporta-
tion and Inventory Cost. Transportation Science, 40(3):287–299, 2006.
Dennis E. Blumenfeld, Lawrence D. Burns, J. David Diltz, and Carlos F. Daganzo. Ana-
lyzing trade-offs between transportation, inventory and production costs on freight
networks. Transportation Research Part B: Methodological, 19(5):361–380, 1985.
Gunter Bolch. Queueing networks and Markov chains: modeling and performance
evaluation with computer science applications. Wiley, New York, 1998.
217
References
Sebastian Bollig. Preiskampf: Konzerne setzen Mittel-

stand unter Druck, 2009. URL www.verkehrsrundschau.de/
preiskampf-konzerne-setzen-mittelstand-unter-druck-844287.html. (last ac-
cessed November 1, 2010).
Yemisi A. Bolumole. The Supply Chain Role of Third-Party Logistics Providers. Inter-
national Journal of Logistics Management, 12(2):87–102, 2001.
Phillip Bonacich. Factoring and weighting approaches to clique identification. The

journal of mathematical sociology, 2(1):113–120, 1972.
Phillip Bonacich. Power and Centrality: A Family of Measures. The American Journal
of Sociology, 92(5):1170–1182, 1987.
Phillip Bonacich. Some unique properties of eigenvector centrality. Social Networks,

29(4):555–564, 2007.
Severin Borenstein. The Dominant-Firm Advantage in Multiproduct Industries: Ev-

idence from the U. S. Airlines. The Quarterly Journal of Economics, 106(4):1237–
1266, 1991.
Severin Borenstein. The Evolution of U.S. Airline Competition. The Journal of Eco-
nomic Perspectives, 6(2):45–73, 1992.
Stephen P. Borgatti. Centrality and network flow. Social Networks, 27(1):55–71, 2005.
Stephen P. Borgatti and Martin G. Everett. A Graph-theoretic perspective on centrality.

Social Networks, 28(4):466–484, 2006.
François Bouffard and Daniel S. Kirschen. Centralised and distributed electricity sys-
tems. Energy Policy, 36(12):4504–4508, 2008.
John Bowen. Network Change, Deregulation, and Access in the Global Airline Indus-
try. Economic Geography, 78(4):425–439, 2002.
Donald J. Bowersox, David J. Closs, and M. Bixby Cooper. Supply chain logistics
management. The McGraw-Hill/Irwin series operations and decision sciences.
McGraw-Hill/Irwin, Boston, 3rd edition, 2010.
Larry A. Bowman and Mark A. Turnquist. Service frequency, schedule reliability and
passenger wait times at transit stops. Transportation Research Part A: General, 15
(6):465–471, 1981.
Jan K. Brueckner. Network Structure and Airline Scheduling. The Journal of Industrial
Economics, 52(2):291–312, 2004.
218
References
Arno Bruns, Andreas Klose, and Paul Stähly. Restructuring of Swiss parcel delivery
services. OR Spectrum, 22(2):285–302, 2000.
Deborah L. Bryan and Morton E. O’Kelly. Hub-and-Spoke Networks in Air Transporta-

tion: An Analytical Review. Journal of Regional Science, 39(2):275–295, 1999.
Guillaume Burghouwt. Airline network development in Europe and its implications for
airport planning. Ashgate, Aldershot, Hampshire, England; Burlington, VT, 2007.
Guillaume Burghouwt and Jaap de Wit. Temporal configurations of European airline

networks. Journal of Air Transport Management, 11(3):185–198, 2005.
Guillaume Burghouwt, Jacco Hakfoort, and Jan Ritsema van Eck. The spatial config-
uration of airline networks in Europe. Journal of Air Transport Management, 9(5):
309–323, 2003.
Lawrence D. Burns, Randolph W. Hall, Dennis E. Blumenfeld, and Carlos F. Daganzo.

Distribution Strategies that Minimize Transportation and Inventory Costs. Opera-
tions Research, 33(3):469–490, 1985.
Kenneth Button. Debunking some common myths about airport hubs. Journal of Air
Transport Management, 8(3):177–188, 2002.
Christian Butz, Wendelin Groß, Cristina Hayden, and Felix Zesch. Der Einfluss des
Ölpreises auf Distributionsnetzwerke von Industrie und Handel - Ein Mythos wird
entzaubert. 4flow AG, Berlin, 2010.
Alberto Caprara, Matteo Fischetti, Paolo Toth, Daniele Vigo, and Pier Guida. Algo-
rithms for railway crew management. Mathematical Programming, 79(1):125–141,
1997.
Valentina Carbone and Marilyn A. Stone. Growth and relational strategies used by
the European logistics service providers: Rationale and outcomes. Transportation
Research Part E: Logistics and Transportation Review, 41(6):495–510, 2005.
Raymond K. Cheung and B. Muralidharan. Dynamic Routing for Priority Shipments

in LTL Service Networks. Transportation Science, 34(1):86–98, 2000.
Sunil Chopra. Designing the distribution network in a supply chain. Transportation

Research Part E: Logistics and Transportation Review, 39(2):123–140, 2003.
Sunil Chopra and Peter Meindl. Supply chain management : strategy, planning, and
operation. Prentice Hall, Upper Saddle River, N.J., 2nd edition, 2004.
Gérard C. Clarens and Vanolin F. Hurdle. An Operating Strategy for a Commuter Bus
System. Transportation Science, 9(1):1–20, 1975.
219
References
G. Clarke and J. W. Wright. Scheduling of Vehicles from a Central Depot to a Number

of Delivery Points. Operations Research, 12(4):568–581, 1964.
Pedro Conçeicão and Pedro Ferreira. The Young Person’s Guide to the Theil Index:
Suggesting Intuitive Interpetations and Exploring Analytical Applications. Techni-
cal report, UTIP Working Paper, 2000.
Tiago F. G. Costa, Gui Lohmann, and Alessandro V. M. Oliveira. A model to identify

airport hubs and their importance to tourism in Brazil. Research in Transportation
Economics, 26(1):3–11, 2010.
Council of Supply Chain Management Professionals. Glossary of terms, 2010. URL

cscmp.org/digital/glossary/document.pdf. (last accessed November 1, 2010).
Teodor Gabriel Crainic. Service network design in freight transportation. European

Journal of Operational Research, 122(2):272–288, 2000.
Teodor Gabriel Crainic. Long-Haul Freight Transportation. In Randolph W. Hall, edi-

tor, Handbook of Transportation Science, volume 56 of International Series in Oper-
ations Research & Management Science, pages 451–516. Springer, Boston, 2nd edi-
tion, 2003.
Teodor Gabriel Crainic and Gilbert Laporte. Planning models for freight transporta-
tion. European Journal of Operational Research, 97(3):409–438, 1997.
Teodor Gabriel Crainic, Michel Gendreau, and Jean-Yves Potvin. Intelligent freight-
transportation systems: Assessment and the contribution of operations research.
Transportation Research Part C: Emerging Technologies, 17(6):541–557, 2009.
Ludwig-Michael Cremer. "Runair" hat Stückgut im Blick. Deutsche Verkehrszeitung,

64(126):11, 2010.
Michael A. Crew, Paul R. Kleindorfer, and Marc A. Smith. Peak-Load Pricing in Postal
Services. Economic Journal, 100(402):793–807, 1990.
Frans Cruijssen, Martine Cools, and Wout Dullaert. Horizontal cooperation in logis-
tics: Opportunities and impediments. Transportation Research Part E: Logistics and
Carlos F. Daganzo. The Break-Bulk Role of Terminals in Many-To-Many Logistic Net-

works. Operations Research, 35(4):543–555, 1987.
Carlos F. Daganzo. Logistics systems analysis. Springer, Berlin, New York, 4th edition,
2005.
220
References
Carlos F. Daganzo and Gordon F. Newell. Physical distribution from a warehouse: Ve-
hicle coverage and inventory levels. Transportation Research Part B: Methodologi-
cal, 19(5):397–407, 1985.
Carlos F. Daganzo and Gordon F. Newell. Configuration of Physical Distribution Net-

works. Networks, 16(2):113–132, 1986.
Donna F. Davis, Susan L. Golicic, and Adam J. Marquardt. Branding a B2B service:
Does a brand differentiate a logistics service provider? Industrial Marketing Man-
agement, 37(2):218–227, 2008.
Keith G. Debbage and Dawn Delk. The geography of air passenger volume and lo-
cal employment patterns by US metropolitan core area: 1973-1996. Journal of Air
Transport Management, 7(3):159–167, 2001.
Rommert Dekker, Moritz Fleischmann, Karl Inderfurth, and Luk N. Van Wassenhove.
Reverse logistics: quantitative models for closed-loop supply chains. Springer, Berlin,
New York, 2004.
Werner Delfmann. Das Steiner-Weber-Modell. WISU, 16(6):291–293, 1987.
Werner Delfmann. Logistikkonzeption, Kernelemente der. In Peter Klaus and Win-

fried Krieger, editors, Gabler Lexikon Logistik, pages 322–326. Gabler, Wiesbaden,
2nd edition, 2000.
Werner Delfmann and Sascha Albers. Supply Chain Management in the Global Con-
text. Technical report, Working papers of the Department of Business Policy and
Logistics at the University of Cologne, 2000.
Werner Delfmann and Natalia Nikolova. Strategische Entwicklung der Logistik - Di-
enstleistungsunternehmen auf dem Weg zum X-PL? In Bundesvereinigung Logistik
e.V., editor, Wissenschaftssymposium Logistik der BVL, pages 421–435. Huss, Mu-
nich, 2002.
Werner Delfmann, Sascha Albers, and Martin Gehring. The impact of electronic com-
merce on logistics service providers. International Journal of Physical Distribution
& Logistics Management, 32(3):203, 2002.
Werner Delfmann, Wilhelm Dangelmaier, Willibald Günthner, Peter Klaus, Ludger

Overmeyer, Werner Rothengatter, Jürgen Weber, and Joachim Zentes. Towards a
science of logistics: cornerstones of a framework of understanding of logistics as an
academic discipline. Logistics Research, 2(2):57–63, 2010.
Ulrich Derigs, Stefan Friederichs, and Simon Schäfer. A New Approach for Air Cargo
Network Planning. Transportation Science, 43(3):370–380, 2009.
221
References
Ben Derudder and Frank Witlox. The impact of progressive liberalization on the spa-
tiality of airline networks: a measurement framework based on the assessment of
hierarchical differentiation. Journal of Transport Geography, 17(4):276–284, 2009.
Hannah Devlin. Iceland’s Eyjafjallajökull volcano rears back into life with new erup-
tion. Times Online, (April 14, 2010), 2010.
Frédéric Dobruszkes. An analysis of European low-cost airlines and their networks.

Journal of Transport Geography, 14(4):249–264, 2006.
Rigas Doganis. The airport business. Routledge, London, New York, 1992.
Rigas Doganis. Flying off course: airline economics and marketing. Routledge, Lon-
don, New York, 4th edition, 2010.
Wolfgang Domschke and Andreas Drexl. Logistik: Rundreisen und Touren. Olden-
bourgs Lehr- und Handbücher der Wirtschafts- und Sozialwissenschaften. Olden-
bourg, München, Wien, 3rd edition, 1990.
Wolfgang Domschke and Andreas Drexl. Logistik: Standorte. Oldenbourgs Lehr- und
Handbücher der Wirtschafts- und Sozialwissenschaften. Oldenbourg, München,
Wien, 4th edition, 1996.
Wolfgang Domschke and Andreas Drexl. Einführung in Operations Research. Springer,

Berlin, Heidelberg, 6th edition, 2005.
Wolfgang Domschke and Armin Scholl. Grundlagen der Betriebswirtschaftslehre.

Springer, Berlin, Heidelberg, 2000.
Harald Dyckhoff, Richard Lackes, and Joachim Reese. Supply chain management and
reverse logistics. Springer, Berlin, New York, 2004.
Economist. Volcanic fallout. Economist print edition, 395(8679):63, 2010.
Samuel Eilon. Management strategies: a critique of theories and practices. Kluwer

Academic, Boston, 1999.
Alan Erera, Burak KaracIk, and Martin Savelsbergh. A dynamic driver management
scheme for less-than-truckload carriers. Computers & Operations Research, 35(11):
3397–3411, 2008.
Steven Erlanger, Jack Ewing, and Nicola Clark. Volcano lays waste to plans and busi-
ness. The International Herald Tribune, (April 19, 2010):6, 2010.
Eva Elisabeth Ernst. DHL holt sich BMW-Auftrag. Verkehrsrundschau, (25):26–27,

2005.
222
References
European Union. EU-US open skies: a new era in transatlantic aviation starts on Sun-
day, 2008. URL www.eurunion.org/News/press/2008/2008026.htm. (last accessed
November 1, 2010).
Kerry Ezard. Back in the ring. Airline Business, 24(5):66–68, 2008.
Kerry Ezard. I want to break free. Airline Business, 25(1):32–33, 2009.
FAA. Federal Aviation Administration, 2009. URL www.faa.gov. (last accessed Novem-
ber 1, 2010).
Eran Feitelson and Ilan Salomon. The implications of differential network flexibility
for spatial structures. Transportation Research Part A: Policy and Practice, 34(6):
459–479, 2000.
Marshall L. Fisher. What Is the Right Supply Chain for Your Product? Harvard Business
Review, 75(2):105–116, 1997.
Ricardo Flores-Fillol. Airline competition and network structure. Transportation Re-

search Part B: Methodological, 43(10):966–983, 2009.
Ricardo Flores-Fillol. Congested hubs. Transportation Research Part B: Methodologi-

cal, 44(3):358–370, 2010.
Mogens Fosgerau and Anders Karlström. The value of reliability. Transportation Re-
search Part B: Methodological, 44(1):38–49, 2010.
Anthony S. Fowkes, Paul E. Firmin, G. Tweddle, and Anthony E. Whiteing. How highly
does the freight transport industry value journey time reliability–and for what rea-
sons? International Journal of Logistics: Research & Applications, 7(1):33–43, 2004.
Markus Franke. Competition between network carriers and low-cost carriers–retreat

battle or breakthrough to a new level of efficiency? Journal of Air Transport Man-
agement, 10(1):15–21, 2004.
Linton C. Freeman. Centrality in Social Networks: Conceptual Clarification. Social

Networks, 1:215–239, 1979.
Noah E. Friedkin. Theoretical Foundations for Centrality Measures. The American

Journal of Sociology, 96(6):1478–1504, 1991.
Shahin Gelareh, Stefan Nickel, and David Pisinger. Liner shipping hub network de-
sign in a competitive environment. Transportation Research Part E: Logistics and
223
References
Atef Ghobrial and Adib Kanafani. Future of Airline Hubbed Networks: Some Policy
Implications. Journal of Transportation Engineering, 121(2):124–134, 1995.
Billy E. Gillett and Leland R. Miller. A Heuristic Algorithm for the Vehicle-Dispatch
Problem. Operations Research, 22(2):340–349, 1974.
James N. Giordano. Economies of scale after deregulation in LTL trucking: a test case
for the survivor technique. Managerial & Decision Economics, 29(4):357–370, 2008.
Balaji Gopalakrishnan and Ellis Johnson. Airline Crew Scheduling: State-of-the-Art.

Annals of Operations Research, 140(1):305–337, 2005.
Siva Govindasamy and Lori Ranson. Open Skies fuel Japan-US pacts. Airline Business,
26(1):10–10, 2010.
Bas Groothedde, Cees Ruijgrok, and Lóri Tavasszy. Towards collaborative, intermodal
hub networks: A case study in the fast moving consumer goods market. Transporta-
tion Research Part E: Logistics and Transportation Review, 41(6):567–583, 2005.
Peter G. Grove and Morton E. O’Kelly. Hub networks and simulated schedule delay.
Papers in Regional Science, 59(1):103–119, 1986.
Tony H. Grubesic, Timothy C. Matisziw, Alan T. Murray, and Diane Snediker. Com-
parative Approaches for Assessing Network Vulnerability. International Regional
Science Review, 31(1):88–112, 2008.
Tore Grünert and Stefan Irnich. Optimierung im Transport, volume 1: Grundlagen.

Shaker, Aachen, 2005.
Valérie Guihaire and Jin-Kao Hao. Transit network design and scheduling: A global
review. Transportation Research Part A: Policy and Practice, 42(10):1251–1273, 2008.
Angappa Gunasekaran, C. Patel, and Ronald E. McGaughey. A framework for supply

chain performance measurement. International Journal of Production Economics,
87(3):333–347, 2004.
Randolph W. Hall. Consolidation Strategy: Inventory, Vehicles and Terminals. Journal

of Business Logistics, 8(2):57–73, 1987a.
Randolph W. Hall. Direct versus terminal freight routing on a network with concave
costs. Transportation Research Part B: Methodological, 21(4):287–298, 1987b.
Randolph W. Hall. Comparison of strategies for routing shipments through trans-

portation terminals. Transportation Research Part A: General, 21(6):421–429, 1987c.
224
References
Randolph W. Hall and Hongsheng Zhong. Decentralized inventory control policies

for equipment management in a many-to-many network. Transportation Research
Part A: Policy and Practice, 36(10):849–865, 2002.
Árni Halldórsson and Tage Skjøtt-Larsen. Developing logistics competencies through

third party logistics relationships. International Journal of Operations & Production
Management, 24(2):192–206, 2004.
Árni Halldórsson and Tage Skjøtt-Larsen. Dynamics of relationship governance in

TPL arrangements - a dyadic perspective. International Journal of Physical Distri-
bution & Logistics Management, 36(7):490–506, 2006.
Árni Halldórsson, Paul D. Larson, and Richard F. Poist. Supply chain management: a
comparison of Scandinavian and American perspectives. International Journal of
Physical Distribution & Logistics Management, 38(2):125–142, 2008.
Hawaiian Airlines. Hawaiian Airlines, 2010. URL www.hawaiianair.com. (last ac-

cessed November 1, 2010).
David A. Hensher. Determining passenger potential for a regional airline hub at Can-
berra International Airport. Journal of Air Transport Management, 8(5):301–311,
2002.
Susanne Hertz and Monica Alfredsson. Strategic development of third party logistics
providers. Industrial Marketing Management, 32(2):139–149, 2003.
Jeff Hoi Yan Yeung, Willem Selen, Sum Chee-Chuong, and Huo Baofeng. Linking
financial performance to strategic orientation and operational priorities. Inter-
national Journal of Physical Distribution & Logistics Management, 36(3):210–230,
2006.
Kaj Holmberg, Martin Joborn, and Jan T. Lundgren. Improved Empty Freight Car Dis-
tribution. Transportation Science, 32(2):163, 1998.
Åke J. Holmgren. A Framework for Vulnerability Assessment of Electric Power Sys-

tems. In Alan T. Murray and Tony H. Grubesic, editors, Critical infrastructure: reli-
ability and vulnerability, Advances in spatial science, pages 31–55. Springer, Berlin,
New York, 2007.
Chaug-Ing Hsu and Yuh-Horng Wen. Reliability evaluation for airline network design
in response to fluctuation in passenger demand. Omega, 30(3):197–213, 2002.
Hans Huber. Comparing spatial concentration and assessing relative market struc-
ture in air traffic. Journal of Air Transport Management, 15(4):184–194, 2009a.
225
References
Hans Huber. Spatial structure and network behaviour of strategic airline groups: A
comparison between Europe and the United States. Transport Policy, 16(4):151–
162, 2009b.
Anthony S. Humphrey, G. Don Taylor, John S. Usher, and Gary L. Whicker. Evaluating
the efficiency of trucking operations with weekend freight leveling. International
Journal of Physical Distribution & Logistics Management, 37(5):360–374, 2007.
INFORMS. The Science of Better, 2010. URL www.scienceofbetter.org. (last accessed

November 1, 2010).
International Transport Forum. International Transport Forum, 2010. URL www.

internationaltransportforum.org. (last accessed November 3, 2010).
Benjamin Jansen, Pieter C. J. Swinkels, Geert J. A. Teeuwen, Babette van Antwerpen de

Fluiter, and Hein A. Fleuren. Operational planning of a large-scale multi-modal
transportation system. European Journal of Operational Research, 156(1):41–53,
2004.
Sergio R. Jara-Díaz and Leonardo J. Basso. Transport cost functions, network expan-
sion and economies of scope. Transportation Research Part E: Logistics and Trans-
portation Review, 39(4):271–288, 2003.
Seung-Ju Jeong, Chi-Guhn Lee, and James H. Bookbinder. The European freight rail-
way system as a hub-and-spoke network. Transportation Research Part A: Policy
and Practice, 41(6):523–536, 2007.
Jari Juga, Saara Pekkarinen, and Heli Kilpala. Strategic positioning of logistics service
providers. International Journal of Logistics: Research & Applications, 11(6):443–
455, 2008.
Ilona Jüngst. Cargo Concept hat abgeschlossen. trans aktuell, (18):9, 2008.
Louis Kaplow. Why measure inequality? Journal of Economic Inequality, 3(1):65–79,

2005.
Bahar Y. Kara and Barbaros Ç. Tansel. The Latest Arrival Hub Location Problem. Man-
agement Science, 47(10):1408–1420, 2001.
Aneel Karnani. The Trade-Off Between Production and Transportation Costs in De-
termining Optimal Plant Size. Strategic Management Journal, 4(1):45–54, 1983.
Mark H. Keaton. Are there economies of traffic density in the less-than-truckload

motor carrier industry? An operations planning analysis. Transportation Research
Part A: Policy and Practice, 27(5):343–358, 1993.
226
References
John Kidd and Marielle Stumm. From pre-Medieval to post-Modern times: Logistics
routes and their modalities have not changed much. Management Decision, 43(9):
1249–1261, 2005.
Hyun Kim and Morton E. O’Kelly. Reliable p-Hub Location Problems in Telecommu-
nication Networks. Geographical Analysis, 41(3):283–306, 2009.
Christine Kiss and Martin Bichler. Identification of influencers – Measuring influence

in customer networks. Decision Support Systems, 46(1):233–253, 2008.
Thorsten Klaas-Wissing and Sascha Albers. Cooperative versus Corporate Gover-

nance of LTL Networks. International Journal of Logistics: Research & Applications,
13(6):493–506, 2010.
Peter Klaus. TOP 100 in European transport and logistics services . Deutscher
Verkehrs-Verlag, Hamburg, 2004.
Peter Klaus and Christian Kille. TOP 100 in European Transport and Logistics Services.
Deutscher Verkehrs-Verlag, Hamburg, 2007.
Peter Klaus, Evi Hartmann, and Christian Kille. TOP 100 in European Transport and
Logistics Services. Deutscher Verkehrs-Verlag, Hamburg, 2009.
Walid Klibi, Alain Martel, and Adel Guitouni. The design of robust value-creating
supply chain networks: A critical review. European Journal of Operational Research,
203(2):283–293, 2010.
Andreas Klose and Andreas Drexl. Facility location models for distribution system
design. European Journal of Operational Research, 162(1):4–29, 2005.
Kerstin Kloss. Logistikmarkt bleibt hart umkämpft. Deutsche Verkehrszeitung, 64(44):

7, 2010.
A. Michael Knemeyer and Paul R. Murphy. Is the glass half full or half empty? Inter-
national Journal of Physical Distribution & Logistics Management, 35(10):708–727,
2005.
David Knibb. Breaking the bilateral web. Airline Business, 26(5):32–37, 2010.
Jens Kohagen. Audi stellt auf Cross Docking um. Deutsche Verkehrszeitung, 64(37):5,
2010.
Niklas Kohl and Stefan E. Karisch. Airline Crew Rostering: Problem Types, Modeling,
and Optimization. Annals of Operations Research, 127(1):223–257, 2004.
227
References
Dirk Koschützki, Katharina Anna Lehmann, Leon Peters, Stefan Richter, Dagmar
Tenfelde-Podehl, and Oliver Zlotowski. Centrality Indices. In Ulrik Brandes and
Thomas Erlebach, editors, Network Analysis, Lecture Notes in Computer Science,
pages 16–61. Springer-Verlag, Berlin Heidelberg, 2005a.
Dirk Koschützki, Katharina Anna Lehmann, Dagmar Tenfelde-Podehl, and Oliver Zlo-
towski. Advanced Centrality Concepts. In Ulrik Brandes and Thomas Erlebach, edi-
tors, Network Analysis, Lecture Notes in Computer Science, pages 83–111. Springer-
Verlag, Berlin Heidelberg, 2005b.
Kai Krause. Organisation und Steuerung von Transportnetzwerken. Logistik und Un-
ternehmensführung. Kölner Wissenschaftsverlag, Köln, 2007.
Leo Kroon and Gaby Vrijens. Returnable containers: an example of reverse logis-
tics. International Journal of Physical Distribution & Logistics Management, 25(2):
56, 1995.
Francine Lafontaine and Laura Malaguzzi Valeri. The deregulation of international

trucking in the European Union: form and effect. Journal of Regulatory Economics,
35(1):19–44, 2009.
Sophie D. Lapierre, Angel B. Ruiz, and Patrick Soriano. Designing Distribution Net-
works: Formulations and Solution Heuristic. Transportation Science, 38(2):174–187,
2004.
Gilbert Laporte. Fifty Years of Vehicle Routing. Transportation Science, 43(4):408–416,

2009.
Paul D. Larson and Britta Gammelgaard. Logistics in Denmark: A Survey of the In-
dustry. International Journal of Logistics: Research & Applications, 4(2):191 – 206,
2001.
Lutz Lauenroth. Mittelständler arbeiten lieber mit mittelständischen Spediteuren.

Deutsche Verkehrszeitung, 60(28), 2006.
Lutz Lauenroth. Palletways zieht Stückgutnetz auf. Deutsche Verkehrszeitung, 64(123):

9, 2010a.
Lutz Lauenroth. Stückgut hat seine Reize. Deutsche Verkehrszeitung, 64(126):2, 2010b.
Harold J. Leavitt. Some effects of certain communication patterns on group perfor-

mance. The Journal of Abnormal and Social Psychology, 46(1):38–50, 1951.
Phillip J. Lederer and Ramakrishnan S. Nambimadom. Airline Network Design. Op-

erations Research, 46(6):785–804, 1998.
228
References
Darin Lee. Concentration and price trends in the US domestic airline industry: 1990-
2000. Journal of Air Transport Management, 9(2):91–101, 2003.
Hau L. Lee, Venkata Padmanabhan, and Seungjin Whang. Information Distortion in

a Supply Chain: The Bullwhip Effect. Management Science, 43(4):546–558, 1997.
Olga W. Lemoine and Lars Danæs. Globalisation strategies and business organisation
of a network of logistics service providers. International Journal of Physical Distri-
bution & Logistics Management, 33(3):209, 2003.
Mark G. Lijesen. Adjusting the Herfindahl index for close substitutes: an application
to pricing in civil aviation. Transportation Research Part E: Logistics and Transporta-
tion Review, 40(2):123–134, 2004.
Lufthansa. Annual Report 2009, 2010. URL http://investor-relations.lufthansa.com/

en/finanzberichte.html. (last accessed November 1, 2010).
Kenth Lumsden and Fabrizio Dallari. Improving the efficiency of the Hub and Spoke
system for the SKF European distribution network. International Journal of Physi-
cal Distribution & Logistics Management, 29(1/2):50–66, 1999.
Jenny Mageean and John D. Nelson. The evaluation of demand responsive transport
services in Europe. Journal of Transport Geography, 11(4):255–270, 2003.
Konstantin Makukha and Richard Gray. Logistics Partnerships Between Shippers and
Logistics Service Providers: The Relevance of Strategy. International Journal of Lo-
gistics: Research & Applications, 7(4):361–377, 2004.
Nicholas Gregory Mankiw. Principles of economics. Harcourt College Publishers, Fort

Worth, TX, 2nd edition, 2001.
Riccardo Manzini and Filippo Bindi. Strategic design and operational management
optimization of a multi stage physical distribution system. Transportation Research
Part E: Logistics and Transportation Review, 45(6):915–936, 2009.
Alessandra Marasco. Third-party logistics: A literature review. International Journal

of Production Economics, 113(1):127–147, 2008.
Vassilis Markides and Matthias Holweg. On the diversification of international freight

forwarders. International Journal of Physical Distribution & Logistics Management,
36(5):336–359, 2006.
Juan Carlos Martín and Augusto Voltes-Dorta. Theoretical Evidence of Existing Pit-
falls in Measuring Hubbing Practices in Airline Networks. Networks and Spatial
Economics, 8(2/3):161–181, 2008.
229
References
Juan Carlos Martín and Augusto Voltes-Dorta. A note on how to measure hubbing
practices in airline networks. Transportation Research Part E: Logistics and Trans-
portation Review, 45(1):250–254, 2009.
Gabriela Mayer. Strategische Logistikplanung von Hub & Spoke Systemen. Gabler Edi-
tion Wissenschaft, Produktion und Logistik, Wiesbaden, 2001.
Scott McShan and Robert Windle. The Implications of Hub-and-Spoke Routing for
Airline Costs and Competitiveness. Logistics and Transportation Review, 25(3):209–
230, 1989.
Yongyut Meepetchdee and Nilay Shah. Logistical network design with robustness and
complexity considerations. International Journal of Physical Distribution & Logis-
tics Management, 37(3):201–222, 2007.
Mary J. Meixell and Vidyaranya B. Gargeya. Global supply chain design: A literature
review and critique. Transportation Research Part E: Logistics and Transportation
Review, 41(6):531–550, 2005.
M. Teresa Melo, Stefan Nickel, and Francisco Saldanha da Gama. Facility location and
supply chain management - A review. European Journal of Operational Research,
196(2):401–412, 2009.
John T. Mentzer, William DeWitt, James S. Keebler, Min Soonhoong, Nancy W. Nix,
Carlo D. Smith, and Zach G. Zacharia. Defining Supply Chain Management. Journal
of Business Logistics, 22(2):1–25, 2001.
Jack Meredith. Theory Building through Conceptual Methods. International Journal

of Operations & Production Management, 13(5):3–11, 1993.
Karlheinz Mevissen. Mehrwegsysteme für Verpackungen - Probleme und Gestal-

tungsansätze in der Konsumgüterwirtschaft. Integrierte Logistik und Un-
ternehmensführung. Deutscher Universitätsverlag Gabler Vieweg Westdeutscher
Verlag, Wiesbaden, 1996.
William Miehle. Link-Length Minimization in Networks. Operations Research, 6(2):

232–243, 1958.
Michel Minoux. Networks synthesis and optimum network design problems: Models,
solution methods and applications. Networks, 19(3):313–360, 1989.
Diane Mollenkopf, Hannah Stolze, Wendy L. Tate, and Monique Ueltschy. Green, lean,
and global supply chains. International Journal of Physical Distribution & Logistics
Management, 40(1/2):14–41, 2010.
230
References
Edward K. Morlok and David J. Chang. Measuring capacity flexibility of a transporta-

tion system. Transportation Research Part A: Policy and Practice, 38(6):405–420,
2004.
Steven A. Morrison. Deregulation of US Air Transportation. In Kenneth J. Button and

David A. Hensher, editors, Handbook of Transport Strategy, Policy and Institutions,
pages 405–420. Elsevier, Amsterdam, Boston et. al, 2005.
Steven A. Morrison and Clifford Winston. Intercity Transportation Route Structures

Under Deregulation: Some Assessments Motivated by the Airline Experience. The
American Economic Review, 75(2):57–61, 1985.
Ole Mortensen and Olga W. Lemoine. Integration between manufacturers and third
party logistics providers? International Journal of Operations & Production Man-
agement, 28(4):331–359, 2008.
Karl Mosler and Friedrich Schmid. Beschreibende Statistik und Wirtschaftsstatistik.

Springer, Berlin, 3rd edition, 2006.
Paul R. Murphy and Richard F. Poist. Third-Party Logsitics: Some User versus Provider
Perspectives. Journal of Business Logistics, 21(1):121–133, 2000.
Alan T. Murray and Tony H. Grubesic. Overview of Reliabilty and Vulnerability in

Critical Infrastructure. In Alan T. Murray and Tony H. Grubesic, editors, Critical
infrastructure: reliability and vulnerability, Advances in spatial science, pages 1–8.
Springer, Berlin, New York, 2007.
Gábor Nagy and Saïd Salhi. Location-routing: Issues, models and methods. European
Journal of Operational Research, 177(2):649–672, 2007.
Alfons Niebuer. Qualitätsmanagement für Logistikunternehmen. Deutscher Univer-

sitätsverlag Gabler, Wiesbaden, 1996.
Kevin O’Connor. Global air travel: toward concentration or dispersal? Journal of

Transport Geography, 11(2):83–92, 2003.
Włodzimierz Ogryczak. Inequality measures and equitable locations. Annals of Oper-

ations Research, 167(1):61–86, 2009.
Morton E. O’Kelly. Routing Traffic at Hub Facilities. Networks and Spatial Economics,
10(2):173–191, 2010.
Morton E. O’Kelly and Harvey J. Miller. The hub network design problem: A review
and synthesis. Journal of Transport Geography, 2(1):31–40, 1994.
231
References
Morton E. O’Kelly, Deborah L. Bryan, Darko Skorin-Kapov, and Jadranka Skorin-

Kapov. Hub network design with single and multiple allocation: A computational
study. Location Science, 4(3):125–138, 1996.
Tae Hoon Oum and William G II Waters. A Survey of Recent Developments in Trans-
portation Cost Function Research. Logistics and Transportation Review, 32(4):423–
463, 1996.
Tae Hoon Oum, Anming Zhang, and Yimin Zhang. Airline Network Rivalry. The Cana-
dian Journal of Economics / Revue canadienne d’Economique, 28(4a):836–857, 1995.
Oxford English Dictionary. "Vehicle" 7a, 2010. URL dictionary.oed.com. (last accessed
November 1, 2010).
Okan Örsan Özener, Özlem Ergun, and Martin Savelsbergh. Lane-Exchange Mecha-
nisms for Truckload Carrier Collaboration. Transportation Science, forthcoming.
Photis M. Panayides. Logistics service providers: an empirical study of marketing

strategies and company performance. International Journal of Logistics: Research
& Applications, 7(1):1–15, 2004.
Photis M. Panayides and Meko So. Logistics service provider-client relationships.

Transportation Research Part E: Logistics and Transportation Review, 41(3):179–200,
2005.
Andreas Papatheodorou and Pavlos Arvanitis. Spatial evolution of airport traffic and
air transport liberalisation: the case of Greece. Journal of Transport Geography, 17
(5):402–412, 2009.
Eric Pels. Airline network competition: Full-service airlines, low-cost airlines and
long-haul markets. Research in Transportation Economics, 24(1):68–74, 2008.
Eric Pels and Piet Rietveld. Cost functions in transport. In David A. Hensher and
Kenneth J. Button, editors, Handbook of Transport Modelling, pages 321–333. Perg-
amon, Amsterdam et al., 2000.
Hans-Christian Pfohl. Logistiksysteme: Betriebswirtschaftliche Grundlagen. Springer,

Berlin et al., 7th edition, 2004a.
Hans-Christian Pfohl. Logistikmanagement: Konzeption und Funktionen. Springer,

Berlin et al., 2nd edition, 2004b.
Hans-Christian Pfohl, Holger Köhler, and David Thomas. State of the art in sup-
ply chain risk management research: empirical and conceptual findings and a
roadmap for the implementation in practice. Logistics Research, 2(1):33–44, 2010.
232
References
Thomas Pliakas, George Kormentzas, and Charalabos Skianis. End-to-end QoS is-
sues of MPEG-4 FGS video streaming traffic delivery in an IP/DVB/UMTS network.
Warren B. Powell. A Local Improvement Heuristic for the Design of Less-than-

Truckload Motor Carrier Networks. Transportation Science, 20(4):246–257, 1986.
Warren B. Powell and Yosef Sheffi. The load planning problem of motor carriers: Prob-
lem description and a proposed solution approach. Transportation Research Part
A: General, 17(6):471–480, 1983.
Psychometrika. Psychometrika, 2010. URL www.psychometrika.org. (last accessed

November 1, 2010).
James P. Rakowski. Marketing Economies and the Results of Trucking Deregulation in

the Less-Than-Truckload Sector. Transportation Journal, 27(3):11–22, 1988.
Kerry Reals. Phase two of EU-US Open Skies to be signed tomor-

row , 2010. URL www.flightglobal.com/articles/2010/06/23/343593/
phase-two-of-eu-us-open-skies-to-be-signed-tomorrow.html. (last accessed
November 1, 2010).
Aura Reggiani, Sara Signoretti, Peter Nijkamp, and Alessandro Cento. Network Mea-
sures in Civil Air Transport: A Case Study of Lufthansa. In Ahmad K. Naimzada, Sil-
vana Stefani, and Anna Torriero, editors, Networks, Topology and Dynamics: Theory
and Applications to Economics and Social Systems, volume 613 of Lecture Notes in
Economics and Mathematical Systems, pages 257–282. Springer, Berlin, Heidelberg,
2009.
Charles S. ReVelle and Horst A. Eiselt. Location analysis: A synthesis and survey. Eu-
ropean Journal of Operational Research, 165(1):1–19, 2005.
Aisling J. Reynolds-Feighan. The Impact of U.S. Airline Deregulation on Airport Traffic

Patterns. Geographical Analysis, 30(3):234–253, 1998.
Aisling J. Reynolds-Feighan. Traffic distribution in low-cost and full-service carrier

networks in the US air transportation market. Journal of Air Transport Manage-
ment, 7(5):265–275, 2001.
Jean-Paul Rodrigue, Claude Comtois, and Brian Slack. The geography of transport
systems. Routledge, London, New York, 2nd edition, 2009.
Gert Sabidussi. The centrality index of a graph. Psychometrika, 31(4):581–603, 1966.
Lisa F. Saunders and William G. Shepherd. Airlines: Setting constraints on hub domi-
nance. Logistics and Transportation Review, 29(3):201–220, 1993.
233
References
Aaron Scholz. Assesment of spatial network configuration for cargo airlines. In ATRS,
Abu Dhabi, 2009.
Darren M. Scott, David C. Novak, Lisa Aultman-Hall, and Feng Guo. Network Robust-
ness Index: A new method for identifying critical links and evaluating the perfor-
mance of transportation networks. Journal of Transport Geography, 14(3):215–227,
2006.
Robert Sedgewick. Algorithms in C++: Fundamentals, data structures, sorting, search-

ing. Addison-Wesley, Amsterdam, 3rd edition, 1998.
Konstantinos Selviaridis and Martin Spring. Third party logistics: a literature review
and research agenda. International Journal of Logistics Management, 18(1):125–
150, 2007.
Alfonso Shimbel. Structural parameters of communication networks. Bulletin of

Mathematical Biology, 15(4):501–507, 1953.
David Simchi-Levi, Derek Nelson, Narendra Mulani, and Jonathan Wright. Crude Cal-
culations. The Wall Street Journal, (September 22, 2008):R8, 2008.
Paul Simons. Weather eye: Dormant volcano erupts in Iceland. Times Online, (March
25, 2010), 2010.
Charalambos Skianis. Performance evaluation of QoS-aware heterogeneous systems.

SmartWings. SmartWings, 2010. URL www.smartwings.com. (last accessed Novem-

ber 1, 2010).
Social Networks. Social Networks, 2010. URL www.elsevier.com/wps/find/

journaldescription.authors/505596/description. (last accessed November 1, 2010).
Haiqing Song, Vernon N. Hsu, and Raymond K. Cheung. Distribution Coordination

Between Suppliers and Customers with a Consolidation Center. Operations Re-
search, 56(5):1264–1277, 2008.
Spiegel Online. Volcanic Eruption in Iceland Disrupts Air Travel, 2010a. URL www.
spiegel.de/international/europe/0,1518,684991,00.html. (last accessed November
1, 2010).
Spiegel Online. Vulkan-Eruptionen locken Touristen an, 2010b. URL www.spiegel.

de/reise/aktuell/0,1518,686981,00.html. (last accessed November 1, 2010).
Ralph H. Sprague and Hugh J. Watson. Decision support systems : putting theory into
practice. Prentice Hall International, London, 2nd edition, 1989.
234
References
Gunnar Stefansson. Collaborative logistics management and the role of third-party

service providers. International Journal of Physical Distribution & Logistics Man-
agement, 36(2):76–92, 2006.
James R. Stock and Douglas M. Lambert. Strategic logistics management. The

McGraw-Hill/Irwin series in marketing. McGraw-Hill/Irwin, Boston, 4th edition,
2001.
Paul M. Swamidass. Empirical Science: New Frontier in Operations Management

Research. The Academy of Management Review, 16(4):793–814, 1991.
Edward J. Taaffe and Howard L. Gauthier. Geography of transportation. Prentice-Hall

foundations of economic geography series. Prentice-Hall, Englewood Cliffs, N.J.,
1973.
Edward J. Taaffe, Howard L. Gauthier, and Morton E. O’Kelly. Geography of transporta-

tion. Prentice-Hall, Englewood Cliffs, N.J., 2nd edition, 1996.
Wayne K. Talley. Regulatory Issues: The Role of International Maritime Institutions. In

Kenneth J. Button and David A. Hensher, editors, Handbook of Transport Strategy,
Policy and Institutions, pages 421–434. Elsevier, Amstedam, Boston et. al, 2005.
G. Don Taylor and Gary L. Whicker. Extended regional dispatch for truckload carriers.
International Journal of Physical Distribution & Logistics Management, 40(6):495–
515, 2010.
Michael A. P. Taylor and Glen M. D’Este. Transport Network Vulnerability: a Method

for Dignosis of Cirtical Locations in Infrastructure Systems. In Alan T. Murray and
Tony H. Grubesic, editors, Critical infrastructure: reliability and vulnerability, Ad-
vances in spatial science, pages 9–30. Springer, Berlin, New York, 2007.
Atsushi Tero, Seiji Takagi, Tetsu Saigusa, Kentaro Ito, Dan P. Bebber, Mark D. Fricker,
Kenji Yumiki, Ryo Kobayashi, and Toshiyuki Nakagaki. Rules for Biologically In-
spired Adaptive Network Design. Science, 327(5964):439–442, 2010.
Henri Theil. Economics and information theory, volume 7 of Studies in mathematical

and managerial economics. North-Holland Pub. Co.; Rand McNally, Amsterdam,
Chicago, 1967.
Jean Tirole. The theory of industrial organization. MIT Press, Cambridge, Mass., 1988.
Rex S. Toh and Richard G. Higgins. The Impact of Hub and Spoke Network Centraliza-
tion and Route Monopoly on Domestic Airline Profitability. Transportation Journal,
24(4):16–27, 1985.
235
References
Michael Toth and Axel Wagenitz. Ganzheitliche Versorgungsplanung mithilfe der Dig-
italen Logistik Integration verschiedener Planungsansätze in eine übergreifende
Logistikplanung. Industrie Management, (2):37–40, 2010.
Mark A. Turnquist. Research opportunities in transportation system characteristics

and operations. Transportation Research Part A: General, 19(5-6):357–366, 1985.
Satish V. Ukkusuri and Gopal Patil. Multi-period transportation network design under
demand uncertainty. Transportation Research Part B: Methodological, 43(6):625–
642, 2009.
U.S. Department of Transportation. U.S. Gross Domestic Product (GDP) Attributed

to For-Hire Transportation Services, 2010. URL http://www.bts.gov/publications/
national_transportation_statistics/html/table_03_01a.html. (last accessed Novem-
ber 1, 2010).
Hans van den Berg, Michel Mandjes, and Rudesindo Núñ Queija. Pricing and dis-
tributed QoS control for elastic network traffic. Operations Research Letters, 35(3):
297–307, 2007.
Remko I. van Hoek. The Role of Third-Party Logistics Providers in Mass Customiza-
tion. International Journal of Logistics Management, 11(1):37, 2000.
Peran Van Reeven. Transport Policy in the European Union. In Kenneth J. Button and
pages 705–724. Elsevier, Amstedam, Boston et. al, 2005.
Manish Verma and Vedat Verter. A lead-time based approach for planning rail-truck
intermodal transportation of dangerous goods. European Journal of Operational
Research, 202(3):696–706, 2010.
Bernd Wagner. A Note on "The Latest Arrival Hub Location Problem". Management
Science, 50(12):1751–1755, 2004.
Wenbin Wei and Mark Hansen. Cost Economics of Aircraft Size. Journal of Transport
Economics and Policy, 37(2):279–296, 2003.
Douglas Brent West. Introduction to graph theory. Prentice Hall, Upper Saddle River,
N.J., 2nd edition, 2001.
Hans Westlund. An interaction-cost perspective on networks and

territoryInteraction-cost perspective on networks. The Annals of Regional
Science, 33(1):93–121, 1999.
Nicole Wieberneit. Service network design for freight transportation: a review. OR

Spectrum, 30(1):77–112, 2008.
236
References
Robert J. Windle. Economic Deregulation in the USA. In Kenneth J. Button and

pages 49–64. Elsevier, Amstedam, Boston et. al, 2005.
Wayne L. Winston. Operations Research - Applications and Algorithms. Thomson,

Toronto, 4th edition, 2004.
S. Chan Wirasinghe and Upali Vandebona. Route layout analysis for express buses.
Transportation Research Part C: Emerging Technologies, In Press, Corrected Proof,
2010.
Wizz Air. Wizz Air, 2010. URL wizzair.com. (last accessed November 1, 2010).
Oliver W. Wojahn. Airline networks. Volkswirtschaftliche Analysen. Peter Lang, Frank-

furt am Main et al., 2001.
Isaac Woungang. Foreword to special issue on wireless ad hoc, sensor and mesh net-
works. Telecommunication Systems, 44(1/2):1–2, 2010.
Johan Woxenius. Generic Framework for Transport Network Designs: Applications

and Treatment in Intermodal Freight Transport Literature. Transport Reviews, 27
(6):733–749, 2007.
Feng Xie and David Levinson. Measuring the Structure of Road Networks. Geograph-
ical Analysis, 39(3):336–356, 2007.
Hai Yang and Michael G. H. Bell. Models and algorithms for road network design: a
review and some new developments. Transport Reviews, 18(3):257–278, 1998.
Jianxin Yao, Lei Xiao, Chun Nie, David Tung Chong Wong, and Yong Huat Chew. Re-
source allocation for end-to-end QoS provisioning in a hybrid wireless WCDMA
and wireline IP-based DiffServ network. European Journal of Operational Research,
191(3):1139–1160, 2008.
Günther Zäpfel and Michael Wasner. Planning and optimization of hub-and-spoke

transportation networks of cooperative third-party logistics providers. Interna-
tional Journal of Production Economics, 78(2):207–220, 2002.
Guanghui Zhou, Yer Van Hui, and Liang Liang. Strategic alliance in freight consoli-
dation. Transportation Research Part E: Logistics and Transportation Review, 47(1):
18–29, 2011.
Hans-Jürgen Zimmermann. Operations Research. Vieweg und Teubner, Wiesbaden,

2nd edition, 2008.
237

Centrality in Strategic Transportation Network Design: Anne Lange

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Centrality in Strategic Transportation Network Design: Anne Lange

Uploaded by

Copyright:

Available Formats

Edition KWV

Weitere Bände in der Reihe http://www.springer.com/series/16033

Bis 2018 erschien der Titel im Kölner Wissenschaftsverlag, Köln

Library of Congress Control Number: 2019934949

Indes sie forschten, röntgten, ﬁlmten, funkten,

(c) Atrium Verlag, Zürich 1950 and Thomas Kästner

1. Introducing network centrality for strategic transportation network de-

2. Transportation networks and their optimal design 9

3. Transportation network centrality 53

3.2. Impact of network centrality on network performance . . . . . . . . . . . . . 59

4. Less-than-truckload network centrality 95

5. An algorithm for less-than-truckload network design 119

6. Generated networks and their performance 153

7. Closing remarks 193

1.1. Line of argument. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2.1. A typology of transportation networks. . . . . . . . . . . . . . . . . . . . . . . . . 16

3.1. Topological centrality scale of networks. . . . . . . . . . . . . . . . . . . . . . . . 55

4.1. Sizes of European LSP markets in bn Euro. . . . . . . . . . . . . . . . . . . . . . . 97

5.1. Impact of waiting time on overall transportation time. . . . . . . . . . . . . . 121

5.6. Algorithm: Ensuring full connectivity in the network. . . . . . . . . . . . . . . 142

6.1. Shipment pay weight distribution. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

A.1. The Lorenz curve of a distribution. . . . . . . . . . . . . . . . . . . . . . . . . . . . 200

A.6. Mean shipment distance against minimal frequency constraint. . . . . . . 205

2.1. US GDP attributed to for-hire transportation services. . . . . . . . . . . . . . 11

3.1. Relevance of cost categories associated with network types. . . . . . . . . . 72

4.1. LSP literature contributions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

5.1. Stability of transportation time: One departure on Thursday. . . . . . . . . 125

6.1. Relevant shipment information. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

A.1. European LTL: Coverage and delivery times. . . . . . . . . . . . . . . . . . . . . . 201

3PL third party logistics.

DSS decision support system.

FAA Federal Aviation Administration.

GDP gross domestic product.

ICT information communication technology.

LCC low cost carrier.

UK United Kingdom of Great Britain and Northern

VRP vehicle routing problem.

WLP warehouse location problem.

B Set of links to and from break-bulk terminals.

cap Capacity of a vehicle.

di The weight of a node v i , typically a demand at this

E Set of edges or links in a graph.

L Set of direct links.

m A number of nodes in a graph or of terminals in a

n Number of all nodes in a graph or network, may be

ω Incidence relation mapping edges in a graph to

Q Overall trafﬁc in a network.

2 2-dimensional real space.

S Any arbitrary subset of V .

T Set of transshipment terminals.

u Average vehicle utilization.

V Set of vertices or nodes in a graph.

Nothing is in the air.

The Eyjafjallajökull outbreak is a perfect example of how strongly modern societies

© Springer Fachmedien Wiesbaden GmbH, part of Springer Nature 2011

and prominent example of research on transportation networks (West, 2001, p. 26).

1.1. Emphasizing the role of network centrality in

Enriching OR by generating strategic insights for ill-deﬁned decision situations.

Implementing an algorithm to simulate network scenarios. The theoretically de-

Generating conclusions on non-measured network performance. It is challeng-

1.2. Structure of the thesis

Demonstrate the relevance of efficient