Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
1
Aspects of spatial proximity and
sustainable travel behaviour in Flanders:
A quantitative approach
2
Cover: The picture on the cover (© Rocky Zutterman) is a remake of the
painting “Il ciclista attraversa la città”, by the futurist Fortunato Depero
(1945). The Futurism movement revered speed, technology and industry,
and was thus not particularly advocating sustainable mobility, despite
the iconic use of a bicycle in this work of art.
Copyright © Kobe Boussauw, Department of Geography, Faculty of
Sciences, Ghent University, 2011. All rights reserved. No part of this
publication may be reproduced, stored in a retrieval system, or transmit-
ted, in any form or by any means, electronic, mechanical, photocopying,
recording, or otherwise, without permission in writing from the copyright
holder(s).
ISBN: 978-94-906-9553-8
Legal deposit: D/2011/12.134/6
NUR: 755/901/904/976
The research reported in this dissertation was conducted at the Social
and Economic Geography research unit, Department of Geography,
Faculty of Sciences, Ghent University, and funded by the Policy Research
Centre on Regional Planning and Housing - Flanders (Steunpunt Ruimte
en Wonen 2007-2011).
3
Aspects of spatial proximity and
sustainable travel behaviour in Flanders:
A quantitative approach
Aspecten van ruimtelijke nabijheid en
duurzaam verplaatsingsgedrag in Vlaanderen:
Een kwantitatieve benadering
Proefschrift
Proefschrift aangeboden tot het behalen van de graad van
doctor in de wetenschappen: geografie
vrijdag 4 maart 2011
door
ir. Kobe Boussauw
4
Promotor:
prof. dr. Frank Witlox, Universiteit Gent
Samenstelling examencommissie:
prof. dr. Georges Allaert, Universiteit Gent
prof. dr. David Banister, University of Oxford
prof. dr. Peter Cabus, Katholieke Universiteit Leuven
prof. dr. Ben Derudder (voorzitter), Universiteit Gent
prof. dr. Martin Dijst, Universiteit Utrecht
prof. ir. Dirk Lauwers, Universiteit Gent
prof. dr. ir. Jacques Teller, Université de Liège
dr. Veronique Van Acker, Universiteit Gent
5
Contents Preface 9
Chapter 1: Introduction 15
1.1 Summary 15
1.2 Sustainable mobility, climate change and peak oil 16
1.3 The time-distance-space relationship 25
1.4 The rebirth of distance 33
1.5 Flanders and Brussels: policy context 45
1.6 Research questions, conceptual framework and implementation 47
1.7 Overview of the research 51
References 55
Chapter 2: Introducing a commute-energy performance index 63
Abstract 63
2.1 Introduction 64
2.2 Energy use and urban spatial structure 64
2.3 Limitations of studying the home-to-work commute 66
2.4 Commute-energy performance (CEP) index 67
2.5 Geographical setting and data analysis 68
2.6 Results 72
2.7 Relation to spatial-morphological characteristics 81
2.8 Conclusions 84
References 85
Chapter 3: Minimum commuting distance as a spatial
characteristic in a non-monocentric urban system 89
Abstract 89
3.1 Introduction 90
3.2 Spatial variations in excess travel 92
3.3 Possible policy implications 94
3.4 Methodology 95
3.5 Case study area: Flanders and Brussels (Belgium) 102
3.6 Application and results of the case study 106
3.7 Possible biases 113
3.8 Conclusions 113
References 115
Contents
6
Chapter 4: Measuring spatial separation processes through
the minimum commute 119
Abstract 119
4.1 Introduction 120
4.2 Defining spatial separation processes 122
4.3 Measuring spatial separation by excess commuting characteristics 124
4.4 Spatial development and commuting in Flanders and Brussels 128
4.5 Data 131
4.6 Method 132
4.7 Results 134
4.8 Conclusions 143
References 145
Chapter 5: Excess travel in non-professional trips:
Why looking for it miles away? 149
Abstract 149
5.1 Introduction 150
5.2 Excess commuting and excess travel 152
5.3 Methodology 156
5.4 Determination of spatial classes 157
5.5 Developing a proximity map 159
5.6 Reported trip lengths 166
5.7 Excess travel 171
5.8 Possible biases in the results 174
5.9 Conclusions 176
References 178
Data sources 181
Chapter 6: Relationship between spatial proximity and
travel-to-work distance: The effect of the compact city 183
Abstract 183
6.1 Introduction 184
6.2 The relevant literature 187
6.3 Methods 190
6.4 Results 200
6.5 Discussion 212
6.6 Conclusions and pathways for further research 214
References 215
Contents
7
Chapter 7: Linking expected mobility production
to sustainable residential location planning 221
Abstract 221
7.1 Introduction 222
7.2 Study area 225
7.3 Methodology and data 225
7.4 Analysis 230
7.5 Forecasting model for Flanders 233
7.6 Discussion 235
7.7 Conclusion and directions for further research 237
References 238
Chapter 8: Conclusions and policy recommendations 243
8.1 General conclusions 243
8.2 Options for spatial planning policy 251
8.3 Some directions for further research 257
References 258
Addendum: The spatial component of air travel behaviour:
An exploration 261
A.1 The aeroplane: the forgotten transport mode 261
A.2 Urban versus rural lifestyle 264
A.3 Rebound effect and policy implications 265
References 266
Samenvatting 269
S.1 Overzicht 269
S.2 Onderzoeksopzet 270
S.3 Bevindingen 272
S.4 Aanbevelingen voor het ruimtelijk beleid 277
S.5 Verder onderzoek 279
Referenties 279
Curriculum vitae 281
Contents
8
9
Preface (English version below)
Na ruim drie jaar werken aan dit proefschrift is het tijd om er de laatste
hand aan te leggen. Hoewel het niet de bedoeling lijkt dat een “woord
vooraf” áchteraf geschreven wordt, is deze omgekeerde volgorde nood-
zakelijk om een terugblik te kunnen werpen. Want wat bracht mij er
eigenlijk toe om me te verdiepen in mobiliteit en ruimtelijke ordening?
Volgens de overlevering was mijn eerste woordje “mama” en mijn
tweede “lamborghini”, verwijzend naar een paars matchbox-autootje dat
deel uitmaakte van mijn eerste verjaardagsuitzet. Een volgende stap in
deze evolutie was dat ik rond mijn dertiende op zaterdag op de fiets
sprong om aan mijn nieuwe hobby te werken, die bestond uit het
verzamelen van prospectussen voor auto’s. Hoewel dat doorgaans vlot
ging bij garages die gespecialiseerd waren in Renault of Volkswagen,
werden er al eens wenkbrauwen gefronst toen ik op zekere dag de moed
had om de toonzaal van Rolls-Royce te betreden.
Deze vroege interesse voor auto’s heeft echter weinig opgeleverd: tot
op de dag van vandaag heb ik mij nooit een wagen aangeschaft. Met het
ouder worden groeide mijn belangstelling voor het leefmilieu en begon
mijn interesse zich te verschuiven naar fietstechniek. Handig, want die
nieuwe sportfiets was niet alleen een perfect vervoermiddel voor de stad,
maar ik kon er ook de weekends en de vakanties mee vullen.
Tijdens mijn opleiding in architectuur en planning maakte ik een
ontwerp voor een fietsvriendelijker heraanleg van het kruispunt “De
Sterre” in Gent. Dat werd uiteraard nooit uitgevoerd, en ironisch genoeg
is dit de plek waar ik de laatste drie jaar weer elke dag tweemaal op
lichtjes suïcidale wijze langs fiets.
Een volgende stap was mijn eerste job, als “consultant
verkeerskunde”, waarbij ik ingeschakeld werd in de opmaak van mobili-
teitsplannen, en waarin ik de aanleiding vond om mijn studie in de
ruimtelijke planning aan te vullen met nog een opleiding verkeerskunde.
Jaren later, in het UN-Habitat-team in Kosovo, zou deze achtergrond
weer zeer goed van pas komen bij het ontwikkelen van ruimtelijke en
mobiliteitsplannen in een heel andere context.
En uiteindelijk bleken mobiliteit en planning ook een domein te
vormen waarin nog heel wat theoretische onderzoeksmogelijkheden braak
Preface
10
lagen. Mijn beslissing om daarin te stappen heb ik me niet beklaagd,
getuige daarvan dit proefschrift.
Dan rest mij nog de eer en het genoegen om een reeks mensen te
bedanken, zonder wie dit werk er vandaag niet zou liggen. In de eerste
plaats komt natuurlijk professor Frank Witlox, mijn promotor, die mij
eind 2007 een goede reden gaf om Kosovo weer voor België in te ruilen.
Hij zorgde voor het onderzoekskader en gaf mij de tijd en ruimte om
cursussen te volgen en aan congressen deel te nemen. Van hem kreeg ik
het vandaag in academische kringen zo gewaardeerde peer review- en
publicatievirus te pakken. En met resultaat: de zes basishoofdstukken van
het voorliggende proefschrift hebben een “peer review”-proces doorstaan
en zijn (of worden) gepubliceerd in een reeks hoogstaande weten-
schappelijke tijdschriften. Een snel optelsommetje leert dat er in de loop
van het schrijven van deze verhandeling maar liefst achtentwintig
internationale reviewers hebben bijgedragen tot de kwaliteit van dit werk.
Hoewel deze experts doorgaans anoniem optraden, heb ik hun inspanning
zeer gewaardeerd.
Maar dichterbij huis, op de werkvloer van de sociaal-economische
geografie (SEG), zijn er wel meer mensen die een plaats verdienen in een
dankwoord. Ik denk vooral aan Tijs Neutens, Veronique Van Acker,
Thomas Vanoutrive, Enid Zwerts en Nathalie Van Nuffel die op tijd en
stond methodologische ondersteuning boden, in het bijzonder in het eerste
jaar, toen de wondere wereld van statistiek, dataverwerking en GIS zich
nog aan mij aan het openbaren was. Daarnaast heb ik ook de inbreng van
andere SEG-onderzoekers van het eerste uur enorm gewaardeerd: David
Bassens (die nog steeds op zoek is naar een manier om de kloof tussen het
wereldstedenonderzoek en het mobiliteitsonderzoek te dichten), Heidi
Hanssens (die altijd klaar stond om de sociale omkadering te verzorgen),
Sven Vlassenroot (die de koffiepauzes steevast met nuttige weetjes over
de academische wereld kwam opvrolijken), Lomme Devriendt (die de
SEG vorm gaf) en professor Ben Derudder (die zijn nuttige tips van de
zijlijn gaf). Daarmee wil ik de rest van de Vakgroep Geografie uiteraard
niet vergeten; één speciale vermelding nog voor Helga Vermeulen, zonder
wiens organisatorische en administratieve nauwgezetheid ik wellicht niet
eens aan de universiteit was kunnen beginnen. Maar ook buiten de
Vakgroep Geografie heb ik heel wat ondersteuning gekregen. De tientallen
mensen (onderzoekers, promotoren en coördinatoren binnen de
verschillende universiteiten, en begeleiders van de Afdeling Ruimtelijke
Planning) achter het Steunpunt Ruimte en Wonen (dat mijn onderzoek
Preface
11
financierde) verdienen hier dan ook een extra vermelding, net zoals Jos
Zuallaert en Dirk Lauwers die mij de mogelijkheid boden om van tijd tot
tijd nog wat in de planningspraktijk te gaan werken in Kosovo.
Tot slot nog een bedankje voor zij die mijn activiteiten van buiten de
werkvloer volgden, maar er niet voor terugschrokken om over mijn
onderzoek in discussie te gaan: eerst en vooral mijn ouders, Johan en
Kristien, maar ook mijn drie zussen (Anna, Marieke en Mathilde), oma
Simonne, het wekelijkse badmintongezelschap (Filip, Wouter, Bram en
Piet) waar de evolutie van de olieprijs een steeds weer opduikend thema
was, Rocky (die de lichtjes controversiële kaft van dit boek leverde),
Hermes (die bijna wekelijks naar de stand van zaken informeerde) en dan
nog een hele reeks van vrienden, familie, reisgenoten, huisgenoten en
facebook-friends die in de loop van de laatste jaren hun interesse lieten
blijken. Merci, allemaal!
___________________________________________
After more than three years working on this thesis, the time has come to
add the finishing touch. Although the word “preface” does not sound as if
it is intended to be written afterwards, this reverse order is necessary to
allow looking back. Because why did I bury myself in the study of
mobility and spatial planning?
According to tradition, my first word was “mama” and my second
“lamborghini”, referring to a purple matchbox-car that was one of my
first-birthday presents. As a next step in this evolution, when I was
thirteen, on Saturdays I jumped on my bike to work on my new hobby,
which consisted of collecting commercial prospectuses for cars. Although
this went generally smoothly in garages that specialized in Renault or
Volkswagen, some eyebrows raised when on a certain day I had the
courage to enter the showroom of Rolls-Royce.
However, this early interest in vehicles did not yield that much: up to
now, I never bought a car. With age, my interest in the environment
increased and I started concentrating on bicycle technology. That was
quite convenient, because my new touring bike was not only a perfect
transport means in the city, but I could also use it to fill my weekends
and holidays.
During my training in architecture and planning, I devised a bicycle-
friendly redesign of the intersection “De Sterre” in Ghent. Of course, my
design was never implemented, and ironically, during the last three years,
Preface
12
this is the place where I cycled through every day twice in a slightly
suicidal manner.
A further step was my first job, as a “traffic and mobility
consultant”, where I was called in to develop municipal mobility plans. In
this job, I also found a good reason to supplement my studies in spatial
planning with an advanced course on traffic and mobility. Some years
later, in the UN-Habitat team in Kosovo, this background proved again
very useful in developing urban plans and mobility plans in a quite
different context.
Ultimately, mobility and planning turned out to be a domain where
many theoretical research opportunities were still present. And I can say
that I have no regret on my decision to get into this, as is demonstrated
today by this dissertation.
Of course, I would like to use this opportunity to thank a number of
people, without whom this work would not be accomplished today. In the
first place there is of course Prof. Frank Witlox, my supervisor, who gave
me at the end of 2007 a good reason to exchange Kosovo for Belgium
again. Frank provided the research framework and gave me the time and
space to follow courses and participate in conferences. He transferred the
“peer review and publication virus” to me, which is highly valued in
academic circles. This was not without success: all of the six basic
chapters of the present dissertation have gone through a peer review
process and are (or will be) published in a series of high ranking academic
journals. A quick summation shows that twenty-eight international
reviewers have contributed to the quality of this work. Although these
experts usually performed anonymously, I greatly appreciate their efforts.
Closer to home, at the Social and Economic Geography (SEG)
research cluster, there are more people who deserve a place in this
expression of gratitude. Tijs Neutens, Veronique Van Acker, Thomas
Vanoutrive, Enid Zwerts and Nathalie Van Nuffel have all offered
methodological support to me, especially in the first year, when the
wonderful world of statistics, data processing and GIS was still in the
process of revealing itself to me. In addition, I also have greatly
appreciated the input from other SEG-researchers who were there in the
very early stage of my study: David Bassens (still looking for a way to
bridge the gap between world city research and mobility research), Heidi
Hanssens (who was always ready to care about the social environment),
Sven Vlassenroot (who was invariably cheering up coffee breaks with
interesting academic rumours), Lomme Devriendt (who was responsible
Preface
13
for the SEG house style) and Prof. Ben Derudder (who contributed with
useful tips from the sideline). This is of course not to forget the rest of
the Geography Department, while I would like to add a special note to
Helga Vermeulen, without whose organizational and administrative
accuracy I may not even have started working at Ghent University. But
even outside the Geography Department, I got a lot of support. The
dozens of people (researchers, supervisors and coordinators within the
different universities, and the coaches of the Ministry’s Spatial Planning
Department) working for the Policy Research Centre on Regional
Planning and Housing (which funded my research) deserve a special
mention here too, just as Jos Zuallaert and Dirk Lauwers who offered me
the possibility to do still some practical planning work in Kosovo.
Finally, a word of thanks to all those who observed my professional
activities from outside, but were not afraid of getting involved in some
debate on my research topic. First and foremost I should mention my
parents, Johan and Kristien, but also my three sisters (Anna, Marieke
and Mathilde), my grandmother Simonne, the weekly badminton
company (Filip, Wouter, Bram and Piet) (where the oil price trend was
an ever resurfacing theme), Rocky (who made the slightly controversial
cover of this book), Hermes (who informed almost weekly on the state of
affairs), and of course the whole series of friends, family, travel compan-
ions, housemates and facebook-friends who demonstrated their interest
during the last couple of years. Merci, everyone!
Preface
14
15
Chapter 1:
Introduction
1.1 Summary
An often-heard statement says that the interaction between mobility
policy and spatial planning practice needs more coordination. Any
concerned politician or citizen feels that the perceived increase in car
traffic, and the growth of problems that are associated with car use, have
“something” to do with unorganized urban expansion, sprawling new
housing and industrial allotments, and ribbon development. Nevertheless,
it is less clear how this relationship exactly looks like, leaving alone the
question how planning should be used as a tool to improve accessibility
and steer mobility to a more sustainable course.
This dissertation wants to gain insight in the reciprocal relationship
between mobility and spatial development, taking into account the
societal context of climate targets and imminent peak oil. This will be
done through the development of a number of quantitative research
methods, which are embedded in a literature review and will be applied
to the case study of Flanders (Belgium).1
This broadly defined research objective is narrowed to the mobility of
people, and will focus on exploring the sustainability of spatial structure
with respect to travel behaviour, with particular attention to the daily
distances travelled. Sustainability is defined in terms of resilience, not
only for growing mobility but also for a possible declining future mobility.
A generally declining mobility is a scenario that may develop due to
rising energy costs (e.g. peak oil scenario) or stringent climate policies,
while a selective shrinkage of mobility (only affecting parts of the popula-
tion) may occur through increased saturation of the traffic system.
Moreover, spatial structure plays a role in the potential steering of travel
behaviour in a more sustainable direction.
1 The reason for this research is found in the ‘mission statement’ of the funding
Policy Research Centre on Regional Planning and Housing - Flanders (2007-
2011), a policy-oriented research consortium that is administered by the De-
partment of Planning, Housing and Heritage of the Flemish government.
Chapter 1
16
We refer to the research line on the connection between spatial struc-
ture and travel behaviour, which has developed mainly in the English-
speaking, German, Dutch and Scandinavian world, and whose prelimi-
nary conclusions can be formulated as follows: “A sustainable travel
pattern can only be realized within an appropriate spatial framework, but
other measures (financial and regulating) are needed to effectively change
travel behaviour. In other words, planning is necessary but not sufficient”
(after Zhang, 2002, p. 3). The spatial quality that facilitates a travel
pattern based on short distances is called “spatial proximity”, even
though this concept has not been clearly defined in the exploratory phase
of this research.
This introductory chapter is structured as follows. Section 2 provides
an overview of climate change and peak oil, two global phenomena that
are directly related to the external effects of mobility. Section 3 deals
with some captivating aspects of time and space perception and increased
prosperity that underpin the growth of mobility. Section 4 gives an
overview of the possible role of spatial structure in a future-oriented
approach to mobility. Section 5 presents a snapshot of the spatial policy
context in Flanders and Brussels. Section 6 focuses on the research
questions, establishes a conceptual framework and states how the research
will be conducted. Section 7 gives an outline of Chapters 2 to 7, each of
which studies a separate aspect of the problem and can therefore be read
as an individual article.
1.2 Sustainable mobility, climate change
and peak oil
The objective of coming to a less car-dependent and, by extension, a less
oil-dependent transport system can be argued from different perspectives.
Local environmental and safety problems caused by transport have
already been in the spotlight for several decades. The principal issues in
this debate are air pollution, noise, deterioration of the livability of
residential areas, accidents and ecological and landscape fragmentation.
This environmental approach is part of what is called “sustainable
mobility” in the recent transport literature. Banister (2008) identifies four
components that may contribute to the transition to a more sustainable
transport system: (1) reducing the need to travel through substitution,
(2) achieving a modal shift through transport policy measures, (3)
Introduction
17
distance reduction through land-use policy measures, and (4) efficiency
increase through technological innovation.
Especially in the field of air pollution, accidents and livability, we can
say that in the western world a lot of progress has been made by a
combination of the mentioned policies and technological developments.
We do not elaborate on this: for an overview we refer to Gilbert and Perl
(2008, pp. 189-264).
However, in terms of climate change (an environmental problem) and
peak oil vulnerability (an economic problem), there seems to be much less
progress. Below, we examine these two global phenomena, considering
that these could constitute a major incentive to reduce the oil dependence
of the transport system.
1.2.1 Climate change
It seems that around 1995, in the scientific community, a consensus was
reached on the acknowledgement of climate change as an important
human-caused problem. With the Kyoto Protocol (1997), which was
ratified by almost all concerned countries except the US, it became clear
that these countries recognized anthropogenic climate change as a
problem and, at least in a rhetorical sense, wanted to commit themselves
to cut greenhouse gas emissions. In the Kyoto Protocol, Belgium, for
instance, is committed to limit its emissions levels by 2012 to 92% of the
1990 level. Since 2001, we see that climate change both in the peer-
reviewed literature and outside it has become an established phenomenon
(Weart, 2010).
Climate change is caused by greenhouse gases, of which carbon diox-
ide (CO2) is the most important. In 2006, transport was worldwide
responsible for 23% of the energy-related greenhouse gas emissions. One
fifth of the projected increase in emissions comes on account of transport,
mainly in the form of increasing ownership and use of passenger cars in
non-OECD countries and a general increase in international air travel
and overseas cargo shipping (IEA, 2008).
In 2005, in the European Union (EU) 80% of all greenhouse emissions
was energy-related, including 24% (equalling 19% of total emissions) that
was transport related. The international bunkers (fuel used by interna-
tional aviation and shipping, both intra-European and intercontinental)
are not yet included in these figures. In the period 1990-2005 greenhouse
Chapter 1
18
gas emissions in the EU decreased in all sectors, except in transport
(+26%) and international bunkers (+64%) (EEA, 2008).
In Flanders, greenhouse gas intensity in transport (this is the amount
of emissions per person-kilometre, or per ton-kilometre for freight)
decreased slightly in the period 1999-2005, but the absolute growth of
traffic offset this efficiency gain well and ensured an absolute increase in
emissions (by 1.50 megatons of CO2 equivalent per year, an increase of
12% over the period 1990-2007) that was the greatest in the transport
sector (again, international bunkers are not yet included). This sector was
therefore largely responsible for not achieving the intermediate emission
reduction targets of the Flanders Region for 2005, even though in 2007
the Kyoto target was still met. This last evolution was due to the manu-
facturing industry which recorded over the last period a reduction of 5.33
megatons of CO2 equivalent per year (VMM, 2008, pp. 95-96).
Regarding the transport sector, the Flanders Climate Policy Plan
(LNE, 2006) focuses on achieving a modal shift, increasing overall
efficiency, and improving the economy of the fleet. However, no measures
that would curb the autonomous growth of traffic are proposed and
international aviation and shipping are even completely out of focus.
Although in Flanders the growth rate of personal car travel has almost
reached zero in recent years (SVR, 2010), the complete traffic volume
seems to be mainly associated with economic dynamics, rather than with
climate policy. Moreover, both freight (SVR, 2010) and international air
traffic with origins and destinations in and around Flanders was growing
steadily up to 2008 (Brussels Airport, 2009).
Transport is clearly bottom of the class when it comes to greenhouse
gas emissions. Although efficiency improvement in vehicles is indeed
enforceable in terms of regulation, it seems that traffic growth itself is
directly linked with increasing prosperity. When mobility growth stag-
nates, this is usually due to a (temporary) economic recession. Also
reaching a high level of congestion can lead to stagnation within a
particular segment of the transport sector, e.g. in road traffic. Structural
congestion may suppress autonomous growth, but particularly increases
the pressure to guide the growth in another direction, for example by
building more road infrastructure, through measures that improve the
spread of traffic across the day, by increasing the capacity and the
attractiveness of public transport, or by shifting the growth from com-
muter traffic towards the segment of recreational and tourist travel
(including airplane use).
Introduction
19
Overall, proposed policy measures aimed to reduce greenhouse emis-
sions from traffic can be divided into four packages:
1. Steering mobility to a less car-dependent course by encouraging a
modal shift towards alternative transport modes (other than passen-
ger car or truck), a more efficient utilization of vehicles, but also
substitution of transport by telecommunication. Known examples in
passenger transport are: car sharing, carpooling, the provision of pub-
lic transport, encouraging cycling and walking, and teleworking and
videoconferencing (Robèrt and Jonsson, 2006).
2. Increasing, and possibly varying, financial charges on those types of
mobility which emit most greenhouse gases, while alternatives may be
supported in parallel. Examples in passenger travel are raising fuel
taxes, introducing registration charges corresponding to the emission
level of the vehicle, but also road tolls, “smart” charging, free public
transport and bicycle allowances (Chapman, 2007).
3. Increasing vehicle efficiency by accelerating fleet replacement by more
fuel efficient vehicles or vehicles that run on alternative fuels (includ-
ing switching to renewable organic fuels, known as biofuels) (Anable
and Bristow, 2007).
4. Intervening in the spatial structure through land use planning with
the aim of bringing potential destinations closer together and making
long distance transport needless. With regard to person mobility it is
generally assumed that increasing density and developing a high de-
gree of land use mix leads to less use of private cars and shorter daily
distances travelled (Newman and Kenworthy, 2006). The latter form
of proposed policy is the background of the research which is reported
in this dissertation.
The objectives of these four packages are clear, and throughout the
western world many examples can be found where this kind of measures
are implemented and measurable successes were reported. However, the
finding that overall traffic and transport emissions continue to grow
(Cervero and Murakami, 2010), leaves room for some healthy scepticism.
Below, we give a number of concerns, again grouped by package.
1. In terms of modal shift many local success stories are known. An
example is the historical centre of Bruges (Belgium), where in the
period 1999-2004 car traffic in the city core declined under the influ-
ence of a rigorous mobility policy, while the number of cyclists and
users of public transport increased considerably in the same period.
The number of registered cars in the entire municipality (which con-
Chapter 1
20
tains six times more inhabitants than the old town), however, con-
tinued to increase. Thus, the success story remains very local, and led
to a geographical shift in the growth of car traffic (City of Bruges,
2007). It is difficult, if not impossible, to find a clear example where a
purposive mobility policy on a regional scale has led to a significant
substitution of individual car use. Nevertheless, the finding that the
modal split largely varies throughout the western world, depending
on the city, the country and the general context, is promising for po-
tential sustainability gains in mobility.
2. The main success story in terms of taxation is perhaps the London
Congestion Charge, which has led two years after the introduction to
a reduction in CO2 emissions in the charged area by 19.5% (Beevers
and Carslaw, 2005). Again, it may be assumed that a part of the
suppressed traffic finds its way outside the demarcated charged area.
Moreover, public support for additional charges on car use (also out-
side the city centres) is hardly present anywhere in the western
world.
3. Anable and Bristow (2007) show that UK greenhouse gas emissions
from cars remained almost constant in the period 1990-2005, while
both the kilometrage and the average weight per vehicle increased
significantly. So we are talking about a major rebound effect, where
efficiency gains are associated with an increase in activity, and finally
with a status quo of energy consumption. Indications exist that the
rebound effect occurs at the macroeconomic level too, meaning that
efficiency gains may lead to accelerated economic growth with an
overall increase (instead of a decrease) in energy consumption
(whether or not outside the transport sector) as a consequence
(Saunders, 1992). Also, the effectiveness of the use of so-called biofu-
els has in recent years led to a major controversy (Righelato and
Spracklen, 2007).
4. Although many studies have found that spatial structures with an
urban character are associated with a lower per capita energy con-
sumption for transport (Newman and Kenworthy, 1999), the
correlation between density or spatial diversity and sustainability of
travel patterns is in fact quite weak. One possible reason is a geo-
graphical form of the rebound effect. An increase in the choice range,
manifested as a better internal accessibility which is typical of urban
structures, will partly offset the relatively small mutual distances,
making people less inclined to choose the nearest possible destination
Introduction
21
(Handy et al., 2005). A second reason is that a high density city is
often connected with a serviced area that is spread out very widely,
feeding people who travel long distances to the city (Mees, 2010, pp.
24-26). A third possible reason is the occurrence of the classic re-
bound effect: cost savings through more efficient daily travel patterns
can lead to more or longer recreational trips (Holden and Norland,
2005).
These outlined reservations contain a series of possible explanations for
the failure to meet emission reduction targets for transport. Decoupling
transport-related emissions and prosperity is not a technical but a social
problem, for which an univocal response has not been found yet. More-
over, in the preceding argument we confined ourselves to the western
world. The largest relative growth of transport-related emissions, how-
ever, is realized in developing countries.
1.2.2 Peak oil
Although it was in the 1950s introduced by M. K. Hubbert, peak oil has
only recently become a common term, describing the evolution of the
production of an oil well over time, from discovery to exhaustion.
Hubbert (1956) pictured this evolution as a bell-shaped curve, a theory
according to which the peak production of oil in the US was predicted to
occur between 1966 and 1972. This forecast became true in 1970 (Gilbert
and Perl, 2008, p. 126), after which the share of imported oil into the US
increased dramatically.
Today, the term peak oil is used to refer to phenomena caused by
global scarcity of fossil oil. Although the report of the Club of Rome in
1972 predicted for 1992 a depletion of all oil reserves that were known at
that time (Meadows et al., 1972), in practice peak oil is not to be seen as
depletion, but as a mismatch between supply and demand (Campbell and
Laherrère, 1998). When demand for oil continues to rise by a steadily
growing global economy, but production does not follow this trend, then,
in principle, the result is a drastic increase of the relatively inelastic oil
price.
Although the phenomenon of peak oil is much less known in common
than climate change (Bardi, 2009), it may be much more restrictive and
may therefore have more far-reaching consequences, both for the economy
in general and for the transport sector in particular. The reason is that
Chapter 1
22
peak oil is about physical limitation, which can virtually not be controlled
by any policy.
A survey by ASPO (2010) shows that most scenarios locate the peak
in world oil production somewhere between 2006 and 2012, although
information is highly dependent on the source. IEA (2008) estimates the
world oil production in 2030 to 101.5 million barrels per day (some 20%
more than the production level of 2008) while an estimate by Aleklett et
al. (2010) gives 75.8 million barrels per day (which is 10% less compared
to the production level of 2008). The debate seems characterized by an
optimistic view of the IEA versus the pessimistic view that is dominant in
the peer-reviewed literature.
But it is not just the point in time when peak oil will occur that is
uncertain: the potential consequences are not clear too. Before 2008 it
was generally assumed that reaching the production peak would be
followed by strong price increases. Indeed, during 2007 oil prices rocketed
up, to reach a top of about 147 USD per barrel in June 2008 (wich is
about 125 USD expressed in June 2010 US dollars). Then the price
dropped again very quickly to stabilize at a price of around 75 USD in
the course of 2010 (Fig. 1.1). Although the rate of the price evolution
may be due to speculation, the intrinsic explanation for the increase may
be found in a mismatch between supply and demand. But this does not
yet explain the last price drop: contrarily, peak oil theory predicts a
continuous price increase.
Hamilton (2008) presents a part of the solution, by showing that nine
out of ten recessions since World War II in the US immediately followed
a sudden surge in the oil price. This finding suggests that a surge in oil
prices is followed by recession, shrinking the demand for oil until a price
level is reached that can be worn by the current economic system. This
means that the oil price remains high, when expressed in terms of
prosperity but not necessarily in monetary terms. Kopits (2009) argues
that the oil price is linked to the share of oil consumption in the GDP of
the US. According to this thesis the oil consumption in the US is limited
to 4% of the GDP. If the consumption level exceeds this threshold, a
recession should be expected. In 2009 this would have meant that the US
economy was not ready to absorb oil prices of more than 80 USD per
barrel. Consequently, to avoid future recessions, oil dependence should
diminish. Kopits (2009) therefore proposes to increase the price artificially
by introducing a carbon tax in those periods where the market oil price is
Introduction
23
below 80 USD per barrel, in order to both ensure price stability and
encourage energy saving developments.
Within the euro area, based on a simulation by the macro-
econometric model of the Eurosystem, the economy of Belgium appears to
be the most sensitive to fluctuations in oil prices: a price increase of 10%
for Belgium would after three years result in a cumulative decrease of
0.4% in the GDP (ECB, 2010).
0
20
40
60
80
100
120
1946 1956 1966 1976 1986 1996 2006
US
D/b
arr
el
nominal
adjusted for inflation
Fig. 1.1. Evolution of the oil price, based on annual averages.
Source: Inflationdata (2010)
But what does this story mean for the transport system? According to
IEA (2008), in 2006, 95% of all energy used worldwide for transport
comes from petroleum. IEA (2008) believes this share to drop to 93% in
2015 and to 92% in 2030, provided that biofuel substitution is widely
implemented. This means that the transport sector remains extremely
dependent on the availability, and thus the price, of oil.
Rodrigue et al. (2009) present the following possible effects of high oil
prices on the transport system:
• reduction of both the speed and the total amount of mobility
• shifts to alternative modes that rely less on oil
• changes in the organization of transport and distribution networks in
favour of more fuel-efficient vehicles and shorter total distances
• in the long-term: changes in location choices as a function of facilitat-
ing the mentioned adjustments
Chapter 1
24
Although there is little evidence for long-term effects of a high fuel price
on the transport system, a number of indications can be deduced from
the responses to the surge in oil price in the second half of 2007 and the
first half of 2008.
In Europe, the fuel for cars and trucks is relatively heavily taxed. In
Belgium 50% to 60% of the customer price are taxes. In most US states
taxes make up only around 17% of the final fuel price. These differences
in taxation have developed historically, and are linked with oil depend-
ence. Oil price increases on the international market are quickly perceived
by consumers in countries where the tax level is low. Therefore, especially
in the US, the oil price surge in the period 2007-2008 had appreciable
impact on the traffic. INRIX (2008) found in the first half of 2008, when
the retail fuel price in the US increased by 28% on average, a significant
reduction of congestion. The correlation between the increase in price and
the decrease in automobile traffic was strongest in cities with a lot of
recreational traffic (tourist destinations such as Las Vegas and Miami)
and in cities where the car is dominant but public transport offers a
decent alternative (e.g., Atlanta, Los Angeles). In cities where public
transport was already heavily used (e.g. New York, Washington DC), the
effect was less pronounced. Apparently the strongest impact was reported
in the segment of tourism and recreational traffic, and in places where
room for behavioural change is present.
Hamilton (2009) points out that those houses that were most isolated
were hit hardest by the real estate credit crunch in 2007-2008. Easily
accessible homes that are located close to a wide range of potential
destinations remained relatively attractive.
The French study of Gonzalez-Feliu et al. (2010) shows that a distri-
bution system based on large-scale retail chains (hypermarkets) is much
more dependent on car use and fossil fuels, in comparison with a network
of local shops or pickup points. However, the latter category requires
more labour force. This means that small-scale distribution would gain
importance at the expense of hypermarkets, in case the relative cost of oil
would rise faster than the relative cost of labour.
Influenced by the recent volatility of the oil price and peak oil theory,
apparently some literature on the theme of urban resilience has emerged.
The research question is which factors determine the vulnerability of an
urban economy in the context of unstable oil prices and how the damage
could be minimized. Transport obviously plays a major role in this theme
Introduction
25
(Dodson and Sipe, 2008; Gleeson, 2008), to which this dissertation wants
to contribute.
1.3 The time-distance-space relationship
Based on the above-mentioned findings, it seems worth to take a closer
look at the mechanisms behind the continuous increase in mobility and
the associated oil dependence. The annual distance covered by an average
individual is increasing every year, and this is done with ever-faster
means of transport. Nevertheless, most of the possible destinations where
everyone is hurrying to remain relatively static: most town centres have
been in the same place for decades, if not for centuries, and although
cities are sprawling, they seem much less expansive than the mobility
itself.
1.3.1 Time distance and travel time budget
In general, mobility is considered as an achievement of modern man. In
the western world the ability to quickly move in any desired direction has
become a prerequisite for leading a full life. This principle has become
popular on every conceivable scale level. In Belgium, for example, one
who travels abroad for holiday less than once a year, is regarded as
underprivileged. The physical location of a job has become a minor
decision factor when choosing a home. Other activities such as education,
recreation and shopping have become virtually footloose in the course of
the twentieth century. Nevertheless, note that the importance of regional
and local identity is not disappearing, and is in some cases perhaps even
more appreciated than before. Increased mobility has made it possible to
operate as a cosmopolitan from your own village, with the rural family
history still fresh in mind. Specifically in Belgium decades of focused
mobility policies have led to an economic shift towards manufacturing
and service industry without causing a spatial division between vivid
metropolitan and empty rural areas. Thus, daily travel over rather large
distances has, to a certain extent deliberately, become an essential part of
the Belgian society.
Parallel to the increase of the footlooseness of activities, the meaning
of the term “distance” blurred. It seems that the increase of mobility has
resulted in a replacement of the notion of distance by the measure of
time. Already in 1791 the meter (m) has been defined very accurately as
Chapter 1
26
the ten-millionth of the distance between the North Pole and the equator.
But the need for setting a standard time was only felt with the arrival of
the train. In the Belgian cities, people only switched late 19th century to
the railway time, which was the same across the country (Reynebeau,
2003, p. 55). In municipalities that were not connected to a railway line,
and were thus deprived of rapid transport, it took some more decades
before distances were expressed precisely in units of time.
This anecdote illustrates the strange evolution in which the percep-
tion of a spatial measure - distance - was transformed into a temporal
concept: time. From a physics point of view, this obviously makes no
sense. If we express a distance in the SI unit2 s (second, or a secondary
unit such as minute or hour) (as in: “I live half an hour from my work.”),
and we stick to the definition of speed as the quotient of distance and
time, the inherent meaning of speed changes. Indeed, the unit of meas-
urement for speed is no longer m/s, but: s/s, which means that the unit is
just omitted. This is less enigmatic than it seems. The advent of rapid
transport and the declining importance of physical distance has resulted
in a perception of speed no longer behaving as a variable but as a con-
stant. This explains immediately why unexpected traffic jams and train
delays are so annoying: the supposedly constant average speed is sud-
denly a variable again, so that the stability of the whole system crashes.
The unexpected traffic jam makes a physical impossibility happen,
namely the sudden stretching of the distance between two fixed points.
To handle the perception of time as a function of distance, in trans-
port studies the concept of “time distance” was introduced (Mérenne-
Schoumaker et al., 1999, p. 90). This is actually the combination of the
variable distance and the unit s. In this context, the assertion above in
which the location of a job plays only a limited role in the search for a
home, should be reviewed. It is not the physical distance, but especially
the mutual time distance that determines the geographic location of the
various activities in which an individual or a household participates.
Common sense says that one wants to minimize the time distance
between origins and destinations, assuming that travel is a derived
demand (Mokhtarian and Salomon, 2001). This reasoning has been
largely prompted by the fact that people, within certain financial limits,
always choose the fastest available transport mode. Workers with a
decent income only take public transport to commute if this is faster than
2 Système International d’Unités (International System of Units)
Introduction
27
the car. The market share of international rail traffic has dropped steeply
since the advent of low cost airlines. Rail projects that have gained
market share are invariably those whose speed is competitive with the
airplane (Rodrigue et al., 2009). And biking is especially popular in dense
urban centres, where it takes hours to manoeuvre a car through or to find
some parking space (Verhetsel et al., 2007, p. 5).
However, the above, intuitive, argument is only a fallacy by which
many urban planners and policy makers have been caught in the past.
The Greek urban planner Doxiadis wrote in 1976 that the daily travel
time taken by an individual had evolved from several hours in prehistoric
times to a constant of half an hour in the first urban civilizations. In the
nineteenth century, however, this optimum constant would have been
abandoned because of the rapid expansion of cities and the inadequate
transport systems. But Doxiadis’ argument was not supported by any
data (Hupkes, 1977, p. 257). Time budget surveys, primarily those of
Szalai et al. (1972), shed new light on the matter. At the aggregate level
(e.g. all inhabitants of a region) the average amount of time spent on
travel appeared to be regarded as a constant, irrespective of the geo-
graphical location of the studied region. Since both the US, Western
Europe and Eastern Europe were involved in the investigation, we may
conclude that this time budget is independent of the economic develop-
ment stage or income. Szalai et al. (1972) found an average personal
travel time budget of 1h13min per day, or 444h per year.
In his thesis Hupkes (1977) formulates the so-called BREVER-law,
based on the time budget survey of Szalai et al. (1972): the law of
conservation of travel time and trips (in Dutch: Behoud van REistijd en
VERplaatsingen). The conclusions of Szalai et al. (1972) and Hupkes
(1977) were confirmed by those of Zahavi et al. (1980) and especially by
Schafer (2000), who also involved Asian and African rural regions in his
research and demonstrated in this way the general validity of the
BREVER law. However, more disaggregated research, like that of Joly
(2004), shows that the travel time budget strongly varies according to the
social group to which one belongs. Van Wee et al. (2002, p. 5) believe
that the travel time budget is less constant than claimed by Hupkes: in
the Netherlands over the years 1980-1990 they observed a slight annual
increase. In Flanders, Glorieux et al. (2005, p. 7) found 1h00min spent on
travel on an average day in 1999, and 1h04min in 2004. If only the
respondents who actually moved on the survey day are considered, travel
times rise to 1h24min and 1h27 min respectively. These results are in line
Chapter 1
28
with the previously mentioned studies, and are also not in contradiction
with the statement of Van Wee et al. (2002, p. 5).
1.3.2 Travel speed
The tendency to choose the fastest manner to make a trip has not led to
a reduction of the time spent on travel, but to an increase in kilometrage.
Within the constant travel time budget (and the financial travel budget),
speed is usually maximized. So, in the long run, our earlier argument that
the speed factor should be considered as a constant, does not hold. The
average travel speed is increasing over the years, perhaps since the
invention of the wheel (Ma and Kang, 2011). In the economic centres of
the western world congestion and traffic regulations have somewhat
constrained the increase of the average speed today, but air travel and
fast public transport systems are still growing unabated. Moreover, it
should not be forgotten that motorists who cause congestion just take the
car because it is still faster than the alternatives. On average, the indi-
vidual speed is thus higher in a congested situation than in a similar
travel pattern system without congestion.
From different perspectives explanations can be found for the human
inclination to travel ever faster. First, there is a simple economic explana-
tion. Maintaining a higher speed automatically means a wider radius of
action, or more precisely, a larger space-time prism (Hägerstrand, 1970).
The wider the radius of action, the more likely that within the associated
space-time prism a suitable home, one or two nice and well-paid jobs, a
decent school, a number of relatives and friends and the desired recrea-
tional and shopping facilities are located. High speed allows for spatial
optimization of the utility of one’s travel pattern within the constant
travel time budget. But this economic approach may not explain every-
thing. The desire for speed is as much a psychological phenomenon that is
related with sensation and status. The doctrine of “dromology” according
to Virilio (1977) is one of the possible approaches. Marinetti (1909, p. 1)
uses the beauty of speed as one of the statements of his futurist mani-
festo: “Nous déclarons que la splendeur du monde s’est enrichie d’une
beauté nouvelle: la beauté de la vitesse. Une automobile de course avec
son coffre orné de gros tuyaux tels des serpents à l’haleine explosive... une
automobile rugissante, qui a l’air de courir sur de la mitraille, est plus
belle que la Victoire de Samothrace.”
Introduction
29
The foregoing considerations cast a different light on the utility of
travel, compared to what is common in the public debate. From time to
time, in the popular press the cost of congestion is calculated. The
number of hours spent by motorists in traffic jams is considered as idle
time and is multiplied by the value of the activities that one could have
performed otherwise (Blauwens et al., 2002, p. 360). When we take into
account the fact that the travel budget is constant within certain limits,
then we must conclude that the time one could have saved by escaping
from the traffic jam would have been spent on travel anyway. In particu-
lar for structural congestion, the BREVER law indicates that the lost
time is already reckoned in by the traveller beforehand, and is thus part
of the maximized utility of the trip. The capacity of the road network
should be seen as a limiting factor, which is in nature not very different
from other limiting factors such as the fuel price, the purchase cost of the
vehicle or the socially desirable level of road safety (which determines the
speed limits). What distinguishes congestion from these other constraints
is primarily the idea that the government has a hold over the capacity of
the roads, and in addition, the relatively large and very annoying unpre-
dictability of the phenomenon.
The discussion on the value of time and the relationship between time
and space has often led to philosophical reflections. Hupkes (1977) writes:
“The desire to raise the marginal utility of time spent on things other
than paid work, has in the western world first led to the virtual disap-
pearance of doing nothing, and second, in terms of personal labour, to
disposable consumer items, ready-to-eat food, and motorized and prefera-
bly automatical housekeeping equipment that requires little maintenance.
The motorization of the housekeeping, however, has ... not lead to
significant time savings.” Rising affluence causes a pursuit for maximizing
the utility of one’s available (leisure) time by using it increasingly more
intensive. This story is as true for travel and explains the urge to keep
accelerating.
Illich (1974, p. 42) understands the increasing energy-intensity of
transport as a mechanism that feeds social inequality: “Beyond a critical
speed, no one can save time without forcing another to lose it.”
Coolsaet (1990) argues that influencing speed is the solution for many
negative aspects of the car. Since there is already a social basis for the
imposition of speed limits by the government, he suggests to adjust the
standards downwards. This would not only yield serious benefits for road
Chapter 1
30
safety, emission levels and energy consumption, but would also reduce
congestion.
Marte (2003) elaborates on this, and states that based on the
BREVER law a reduction of the (maximum allowed or effective) speed
would in the long run lead to a reduction in the amount of traffic. Indeed,
the pressure to search for a job closer to home, or to move house near the
workplace, or to choose schools, shops and leisure activities less far from
home, would increase. Moreover, the travel time ratio between private
and public transport would be adjusted in favour of the latter mode so
that a modal shift towards public transport is to be expected. Not only
the consumption of fuel would go down, probably also the congestion
problem would be influenced significantly in a positive sense.
1.3.3 Travel cost
The average travel time budget is not the only factor that is relatively
constant. A less well studied, but similar phenomenon, exists with respect
to the financial cost of travel. For Belgian households, the proportion of
household income that was spent on fuel remained constant at around 3%
in the period 1995-2004. For transport spending in general a slightly
upward trend is noticeable, but this is almost entirely due to the pur-
chase of vehicles (fixed costs) (Fig. 1.2) (Statistics Belgium, 2008).
Detailed historical data on household spending on cars is not available,
but based on figures for car ownership, it can be concluded that the retail
price of an average car must have declined over the past decades in
relation to purchasing power. As soon as a faster - and thus more expen-
sive in relative terms - transport mode became cheaper in absolute terms,
there was a growing market for it. The modal shift from the bicycle and
public transport towards the car in the 1970s can be explained in this
way. In the 1990s, a similar modal shift occurred in the specific field of
tourism travel: from the car and the train to the plane.
Introduction
31
Fig. 1.2. Evolution of the transport share of household expenditures in
Belgium. Source: Statistics Belgium (2008)
1.3.4 The combined travel time and cost budget
The substitution of the metre by the second as a measure for expressing
perceived distance is not absolute. In fact, we may talk about a combined
travel time-cost budget, in which both the kilometre-dependent cost and
the required time are reckoned. The ratio between the temporal and the
financial component in the time-cost budget is determinant for the
perception of physical distance by an individual. In Belgium, the cost per
kilometre has systematically declined over the years, with the introduc-
tion of the railways and tramways in the nineteenth century as a main
milestone. This trend was then accelerated by the democratization of the
car, which became cheaper in comparison with the general prosperity
level and provided - at least until the early 1980s - definitely the fastest
way to make any domestic trip.
Regarding the cost component, the base price of fuel is an important
element. The price elasticity for fuel is highly dependent on the kind of
trip. It is known that business travel and commuting are hardly affected
by price increases. However, in freight transport and recreational and
tourist travel, the connection is much clearer. The relationship between
the amount of touristic air traffic and the price of flights is even almost
linear (Litman, 2007). When fuel prices rise through the peak oil phe-
Chapter 1
32
nomenon, the importance of the cost component in the combined time-
cost budget will increase. Also any climate policy that is rationing the
consumption of fossil fuels (e.g. through emissions trading) would signifi-
cantly rise the importance of the cost component (Keppens, 2006).
Since variable transport costs are almost directly linked to distance
travelled,3 the importance of physical distance in the perception of the
traveller will increase when the share of the financial component in the
time-cost budget is rising. The oil crises of 1973 and 1979 caused a ripple
in the readily availability of fuels, so that over the period 1973-1985 the
share of cost in the time-cost budget was lifted to a higher level (Infla-
tiondata, 2010). Between 1985 and 2003 the price level of oil, adjusted for
inflation, was again quite low. In May 2008 oil prices had approximately
quadrupled since 2003. Oil prices reached a record of 147 USD per barrel
on July 11, 2008, after which the price quickly fell. In the period 2009-
2010, after some fluctuations it remained relatively constant at a level of
about 70-80 USD per barrel.
The ratio between the cost component and the time component in the
time-cost budget depends also on external factors such as the develop-
ment stage of the considered region, and the general state of the world
economy. In countries where the cost of motorized transport swallows, or
would swallow, a large share of household income, physical distance
continues to be determinant. The effort it takes to make a trip on foot or
by bicycle, or even the marginal cost of fuel in the case of a motorized
trip, remains largely linked to the kilometrage. In contrast, in the western
world the cost of having a car, and the fuel cost, has become a basic
expense that is almost always outweighing the perceived benefit of
intensifying the use of the available travel time budget.
The regional stage of development determines the generalization of
car ownership and the performance of public transport, but also the
congestion level and the social sensitivity to external impacts. During the
1970s, for example, in Belgium the average time spent on commuting
decreased, while the length of the commute simultaneously increased. The
massive shift from slow transport modes to the car allowed for a signifi-
cant intensification of the use of the travel time budget. So it became
3 Although the correlation varies with the speed: in steady-flow traffic at
medium speed, an average car operates in the most efficient mode. At high
speeds, but also in stop-and-go traffic, fuel consumption and emissions per
kilometre go up quickly (Deakin, 2007).
Introduction
33
possible to go living further away from one’s workplace, or to look for a
job far from home. Since 1981, in Belgium the time spent on a particular
commute is again in line with the distance. The generalization of car
ownership and the saturation of the road infrastructure does not any
longer allow to increase the average commuting speed (Fig. 1.3)
(Mérenne-Schoumaker et al., 1999; Verhetsel et al., 2007).
0
5
10
15
20
25
30
35
1970 1981 1991 2001
km
/day a
nd
min
/day
average home-to-work
distance (km)
average home-to-work time
(min)
Fig. 1.3. Evolution of the average commuting time and distance in
Belgium. Sources: Mérenne-Schoumaker et al. (1999); Verhetsel et al.
(2007)
The external effects of transport, as an umbrella for road safety and
environmental effects, play their role too. As there is more traffic on the
roads, public support for government action is growing. Speed limits are
in this very significant because they directly affect the average speed.
Financial policies, including fuel taxes and fixed or variable tolls, empha-
size again the cost component.
1.4 The rebirth of distance
1.4.1 Absorption of rising energy prices
In the future, potentially rising energy prices could lead to a reappraisal
of physical distance. In other words, the unit s will slowly but surely be
pushed aside by the unit m. The spatial structure and the coherence and
distance between different functions and services will be more pronounced
Chapter 1
34
in the individual perception, since one will be inclined to minimize not
only travel time but also travel distance. “Proximity”, a concept that is
associated with physical distance (as opposed to the concept of “accessi-
bility” that is rather associated with time distance), will gain in value,
just as slower and thus more energy-efficient transport modes.
To date, no recent evolutionary data sets that support this claim are
known. Leaving aside the oil crises of the 1970s, significant increases in
fuel prices have only very recently manifested, and were also short-lived.
The data available for Belgium appears to show that covered distances
continue to increase every year. However, for the number of car kilome-
tres travelled, there is a certain slowdown in growth since 2000
(Goossens, 2008). This is probably due to saturation of the road network,
rather than to an increased level of costs. In contrast, the number of train
passengers, and users of public transport in general, continues to grow
steadily. The distance between home and workplace certainly continued
to rise until 2001 (Verhetsel et al., 2007). Also, the number of registered
vehicles in Belgium continues to increase: in 2009, 5.5% more vehicles
were driving around compared to 2005 (Statistics Belgium, 2010).
Moreover, there is a technological buffer that may absorb for some
time a considerable part of the potential price increase, since it is possible
to make the fleet a lot more efficient. From 2002 to 2005 the average fuel
consumption of new cars sold in Belgium decreased already from 6.3 to
5.9 l/100 km. Second, the average weight of a newly sold car increased
from 1993 to 2004 by 30%, while the consumption level of a standard
passenger car with hybrid engine technology is only 4.3 l/100 km (De
Vlieger et al., 2006), and the most economical of all small cars even get
by on 3.4 l/100 km. So there is still a margin that allows reducing fossil
fuel consumption and vehicle emissions significantly by using more
efficient engines in smaller and lighter cars. Nevertheless, history teaches
us that any improvement in energy efficiency of motor vehicles is more
than offset by an increase in overall kilometrage. This pattern is consis-
tent with the Jevons paradox, and the macro-economic extension of it -
known as the Khazzoom-Brookes postulate - which states that, at
constant energy prices, any technological improvement in energy effi-
ciency at the micro level, may result in an overall increase (and not a
decrease) in energy consumption (Saunders, 1992). Since around 2000 in
Belgium we observe a tendency towards stabilization, at least with
respect to energy consumption by passenger transport (De Vlieger et al.,
Introduction
35
2006). Also, the use of alternative fuels should be taken into account,
even though these will have little impact on price making.
Research on price elasticity indicates a clear link between cost and
consumption of fuel, although there are noticeable differences depending
on the type of trip (Bogaert et al., 2006). Short-term price increases are
mainly captured by travelling less kilometres by car. In the longer term,
effects are more important in terms of vehicle efficiency and car owner-
ship. Long-term price elasticities are usually larger than short-term
values. Based on a meta-analysis approach, Brons et al. (2008) find -0.12
as the short-term price elasticity for mileage per car, while the value for
fuel efficiency is 0.14, and -0.08 for car ownership. In the long term, these
values amount to -0.29, 0.31 and -0.08, while the aggregate long-term
price elasticity is -0.84. The absolute value of these figures is less than 1,
meaning that demand for fuel is relatively inelastic.
An important explanation for the inelasticity is the virtual absence of
possible substitutes. There are only few, if any, fully-fledged alternative
fuels available at an equal or lower price. Furthermore, regional varia-
tions exist: North American, Canadian and Australian data show lower
elasticities, which is almost entirely due to the smaller perceived effect on
car ownership. On the other hand, based on comparative research
between similar cities in the US and Australia, Wegener and Fürst (1999)
found that Australians consumed per capita only a little more than half
the volume of fuel that was consumed in the US. Differences in taxation
make the cost of petrol and diesel in Australia about twice as much as in
the US, explaining most of the difference in consumption.
Regarding the distant future, we can say that existing analyses of
price elasticities do not take into account very long term effects, such as
shifts in terms of regional economic structure, real estate prices and
urban development. Brons et al. (2008) recognize that current analyses do
not constitute grounds for very long term predictions. In addition, they
show that mainly travel distance is sensitive to price rises as the studied
time span increases. Another important element is the ceteris paribus
assumption, which is common in price elasticity studies. How mobility
would evolve if a sharp rise in oil prices leads to an economic recession
cannot be inferred from these studies.
Chapter 1
36
1.4.2 Mobility, accessibility and spatial structure
It is clear that reducing the oil dependence of the (person) transport
system has an important technological component: cars need to become
more fuel efficient and alternative fuels should get adopted wherever
possible. The second component, however, has to do with adapting travel
behaviour: a reduction of oil dependence means less car kilometrage (both
by travelling less kilometres, and by a modal shift), and to some extent
less mobility.
In the economic approach, however, mobility is considered as a sec-
ondary activity, and not as a purpose in itself. The assumed objective is
to optimize the accessibility, where accessibility should be seen as a
measure of the interaction potential of an individual, a family or an
organization. According to Geurs and Ritsema van Eck (2001, p. 36)
accessibility reflects the extent to which the “land-use transport system
enables (groups of) individuals or goods to reach activities or destinations
by means of a (combination of) transport mode(s)”.
It is possible to realize a wide range of interaction possibilities, and
thus potential destinations, within short distance. Some of the densest
urban centres in the world, such as Manhattan or Hong Kong, illustrate
this. This mechanism underlies the formation of most cities, increasing
accessibility based on a minimum of transport. Engwicht (1992, p. 12)
puts it as follows: “Cities are an invention to maximise exchange and
minimise travel.”
The work of Newman and Kenworthy (1989, 1999) was the first that
managed to identify a link between the spatial structure of a variety of
world cities and energy consumption for transport. Their graph, of which
Fig. 1.4 is a reproduction, has over the years become popular among
urban and traffic planners. Newman and Kenworthy (1989, 1999) note
that population density shows an inverse relationship with energy
consumption for transport. Residents of densely populated cities cover
shorter distances, since their daily destinations are closer to home. In
addition, they use more often public transport, compared to cities with a
low population density. This phenomenon is strengthened by the higher
level of congestion and parking limitations that are associated with a high
density.
Introduction
37
Fig. 1.4. Fuel consumption per capita versus urban density (1980).
Source: Newman and Kenworthy (1989)
The discourse of Newman and Kenworthy (1989, 1999) has given rise to a
line of research that investigates the relationship between spatial struc-
ture and sustainability of travel patterns in more detail. Some studies
seem to confirm broadly the argument of Newman and Kenworthy (1989,
1999), while other researchers sharply criticise the methodology originally
used, particularly on the demarcation of the urban areas that were
examined in the original study (Mindali et al., 2004; Mees, 2010). Al-
though there is little debate about the existence of a relationship between
spatial structure and travel patterns (Ewing and Cervero, 2010), there is
less unanimity about the sense or nonsense of conducting spatial policies
Chapter 1
38
aimed at sustainable mobility. Gordon and Richardson (1997) for exam-
ple believe that the connection is too weak to justify interventions.
According to them, as cities are sprawling spontaneously, this would be
the most efficient form of urban growth within the current market
economy.
A more nuanced approach is the assumption that planning is neces-
sary, but will not suffice alone to make mobility more sustainable (Zhang,
2002, p. 3). However, the spatial structure is a hard constraint that
determines whether it is possible to shorten trip lengths or to take public
transport or the bike at all.
1.4.3 Urban sprawl
The acknowledgement that uncontrolled and unplanned urban growth
entails problems is far from new. In the peer-reviewed literature, several
types of quasi-unorganized development have been lumped together as
“(urban) sprawl”. Inspired by Harvey and Clark (1965), Ewing (1997)
distinguished three types of sprawl: (i) leapfrog or scattered development,
(ii) commercial strip development, and (iii) large expanses of low-density
or single-use development. In addition, Ewing (1997) states that indica-
tors based on accessibility are perhaps more suitable to detect sprawl
than morphological measures, which is an important contribution com-
pared to Harvey and Clark (1965), even without putting this statement
into practice. Nevertheless, in most of the literature, the definition of
sprawl has a strong emphasis on the morphological character. Torrens
and Alberti (2000, p. 4) call sprawl “a relatively wasteful method of
urbanization, characterized by uniform low densities”. They put forward
seven different variables as indicators for sprawl, including six based on
morphological characteristics (ranging from density gradient to fractal
dimension) and only one based on accessibility. Galster et al. (2001, p.
685) developed a more comprehensive definition: “Sprawl is a pattern of
land use in an urban area that exhibits low levels of some combination of
eight distinct dimensions: density, continuity, concentration, clustering,
centrality, nuclearity, mixed uses, and proximity.” Moreover, they focus
on the evolutionary nature, viewing sprawl as a process and not as a
state.
In the peer-reviewed literature, and in many policy documents, a
strong link between sprawl and increased traffic is assumed (Ewing et al.,
Introduction
39
2002). This is logical bearing in mind that separation processes between
origins and destinations inevitably entail larger distances to be covered.
If we restrict ourselves to commuter traffic, we see that trip lengths
have increased systematically over the past decades. This trend was
observed in Belgium (see before), the US and the UK, and we may
assume that this evolution is manifest throughout the Western world
(Aguilera, 2005). Even though there seems to be a link between the
presence of sprawl and the average trip length, this reasoning does not
necessarily imply a one-way causality. Sprawl is caused by increasing
individual mobility, i.e. a wealth-related phenomenon that is the basis of
an autonomous growth of traffic. So, we might explain the wave of
suburbanization as a materialization of this increased mobility, which has
itself a mutual reinforcing effect on the growth of traffic. Gilbert and Perl
(2008, p. 235) formulate this phenomenon as follows: “Sprawl is believed
to be facilitated by car ownership and use and also to contribute to it, in
a positive feedback loop that reinforces both low-density development and
motorization.”
Primarily in the US the debate on the advantages and disadvantages
of sprawl has been developed into a highly polarized, dichotomous
discussion. From a sociological perspective we can consider James How-
ard Kunstler as advocate of the anti-sprawl school. Kunstler (1993) uses
cultural, aesthetic and ecological reasons for rejecting suburbia as a
human habitat. On the other side of the spectrum, we find Robert
Bruegmann, who describes the American suburb as a natural materializa-
tion of the American Dream, yielding both many benefits for residents
and a number of social and ecological problems (Bruegmann, 2006).
Surprisingly, this ideological duality also appears to exist when we
view sprawl and suburbanization from a mobility perspective. Newman
and Kenworthy (1989) advocate compact cities with a high density, since
these would require less energy per capita than suburbanized regions. The
main reason for this is that distances are reduced at high densities, and
that low residential densities may never provide a sufficient basis for the
organization of efficient public transport. At the other side, Gordon and
Richardson (1989) argue that suburban areas eventually evolve into full-
fledged urban areas, while the intra-suburban traffic is much more
efficient and less sensitive to traffic congestion than inner-city traffic.
They find the more difficult organization of public transport in the
suburbs of minor importance, since car traffic is relatively smooth thanks
to the abundance of space, reducing the need for alternatives. They find
Chapter 1
40
the argument of Newman and Kenworthy (1989, 1999), which is built on
fuel dependence, too weak, since there is no scarcity (yet) and energy
consumption levels are far more influenced by economic factors than by
spatial structure.
Although both Newman and Kenworthy (1989, 1999) and Gordon
and Richardson (1989, 1997) rely on data and scientific methods to
underpin their argument, both of their analytical frameworks are strongly
ideologically biased. In their studies, sprawl is associated with a specific
morphology, particularly monotonous suburban districts with a strict
separation of functions, characterized by store strips, commercial archi-
tecture and large internal distances. However, the extent to which spatial
separation leads to an effective increase of distances that need to be
covered is less clear, since this cannot be derived from local morphological
characteristics. A monotonous residential lot embedded in a major
employment centre could possibly lead to a more sustainable travel
pattern than a compact town that is immersed in a rural area. Particu-
larly in a context where average trip lengths, and in particular average
commuting distances, have become very large in practice, it is hard to tell
which kind of spatial developments are problematic in relation to mobil-
ity, and which are rather beneficial. Banister (1999) observes the issue in
a non-morphological way, and suggests that it is utmost important that
new developments are sufficiently large and are located in or immediately
subsequent to existing urban areas. Local morphology, density and spatial
diversity come only in second place.
Over the last decade, research on the link between travel and land
use has evolved and improved in a number of areas. Datasets have
expanded and have become more reliable over the years, statistical
methods have improved, and additional hypotheses (e.g. on causality, but
also on behavioural aspects such as attitudes and (residential) self-
selection) were incorporated. The dichotomy that was introduced by
Newman and Kenworthy (1989) and Gordon and Richardson (1989),
evolved to a more nuanced understanding which provided the basis for a
series of recent textbooks such as Boarnet and Crane (2001), Glaeser and
Kahn (2003) and Levinson and Krizek (2008).
1.4.4 The paradigm of the compact city
In Europe, since World War II, in a number of agglomerations tackling
urban sprawl is considered an inherent part of the basic principles of
Introduction
41
what constitutes proper urban planning. The main examples that fit into
this tradition are the New Towns Act (UK, 1946), the Finger Plan of
Copenhagen (1947), the various Notes on Spatial Planning (The Nether-
lands, from 1960) the Villes Nouvelles around Paris (since 1965), et
cetera. In other, relatively densely populated European regions that did
not develop around a single city, such as the Ruhrgebiet (Germany) and
the Flemish Diamond (Belgium), the focus on steering spatial develop-
ment came much later. In these suburbanized regions, which have
developed between and around an historically polycentric structure,
traditionally less important economic problems were involved with
additional constructing. The more dispersed nature of the historical
spatial structure and the presence of a dense railway network resulted in
a relatively late manifestation of congestion problems. In the US, the
focus on thorough urban planning is also a relatively recent phenomenon,
which has resulted in, e.g., Florida’s anti-sprawl rule (Ewing, 1997) or
Portland’s urban growth boundary (Song and Knaap, 2004).
In 1990, the European Union took a position against the further de-
velopment of sprawl by endorsing the paradigm of the compact city in
the Green Paper on the Urban Environment (CEC, 1990). The compact
city is presented as the ideal to be pursued, ensuring a sustainable
development. Mobility is one of the key guiding principles that form the
basis of the compact city policy, since slow traffic and public transport
would get more opportunities through shorter distances and higher
densities. The compact city remained an important issue in the European
Spatial Development Perspective (ESDP, 1999, p. 22).
The origin of the European compact city policy can be found in the
existing post-war spatial strategies of a number of EU member states,
whose objective was to guide population growth while safeguarding a
sufficiently high quality of life. In the period when the foundation for
these plans was laid, still mainly benefits were expected from the auto-
mobile, so we can say that the sustainable mobility story was only
subsequently added to the concept of the compact city. A second track
that formed the basis of the compact city policy, is the nostalgic image of
the medieval city-state, which is located in the surrounding countryside
and where all possible destinations are within walking distance of each
other.
In one form or another, the compact-city policy turns up in most pro-
gressive spatial policy plans for cities and regions in Europe. The Spatial
Structure Plan for Flanders (RSV, 1997/2004) is no exception.
Chapter 1
42
Although by most urban planners the compact city policy is consid-
ered valid in a European context, also a number of critiques have been
developed. Perhaps the main criticism is the reasoning that the creation
of spatial proximity by increasing density and mixing functions does not
necessarily lead to a reduction in the number and the length of car trips,
but merely provides the appropriate spatial framework. Moreover, the
compact city policy does not take into account the size of the city. In a
city like London, for example, which is by its size much more compact in
a geometrical sense than a medium sized city, per capita energy consump-
tion for transport is higher than in medium sized British cities (Breheny,
1992). According to Banister (1992), in remote small towns with a limited
degree of self-sufficiency, transport energy consumption per capita is
easily one third higher than average, while in rural areas this may rise up
to three times the level of a regional city.
In Belgium, there has been only very limited research into the possi-
ble effects of a compact city policy on travel behaviour. Verhetsel (2001)
compared a policy scenario according to the Spatial Structure Plan for
Flanders (RSV, 1997/2004) with a trend scenario for the spatial devel-
opment of the Antwerp urban region, and concluded that the effects of
planning are very limited. This study considered newly expected devel-
opments over the period 1991-2010. Obviously, these are engrafted onto
the existing spatial structure of the Antwerp urban region, so that any
structural change would be necessarily small in relative terms. It is
therefore not surprising that the direct impact of an infill scenario is
limited within the considered planning horizon.
Breheny (1996) discusses the centralist and decentralist movements,
which in the 1970s strongly determined the debate in the field of spatial
planning. He concludes that eventually the compromise is the best
solution. Compactness and diversity is important but should not lead to a
loss of environmental quality. Moreover, it is not feasible to exclude
greenfield developments completely, and there should also be a role for
urban networks.
With regard to movement, it has become clear that the spatial struc-
ture should provide the framework that allows for a sustainable travel
pattern, with a substantial role for short-distance trips, public transport
and non-motorized modes. However, it cannot be expected that travel
behaviour will be steered only by the morphology of space.
Introduction
43
1.4.5 Public transport and spatial structure
Although this dissertation focuses on measuring spatial proximity in
relation to the amount of mobility, expressed in person kilometres, still
some comments should be made on the possible role of spatial structure
in relation to the potential for public transport.
The main spatial condition for efficient public urban transport con-
sists of the presence of a critical mass of potential travellers and
destinations, situated within walking distance of the stops in the network.
To ensure a competitive position compared to the car, it is also important
that car traffic is relatively slow and unattractive, for example due to
congestion or a limited number of available parking spaces. These
conditions are associated with thresholds, both in terms of density and in
relation to the size of the city. According to Newman and Kenworthy
(1989) a density of about 30-40 inhabitants per hectare (as an average for
the whole metropolitan area) is the threshold below which it is impossible
to develop an efficient public transport system. Levinson and Kumar
(1997) mention a minimum of 10,000 people per square mile (equivalent
to 39 persons per hectare), while Mees (2010, pp. 51-54) cites several
sources that discuss a variety of figures ranging from 48 to 275 people per
hectare. According to the 4th RTD Framework Programme of the
European Commission, the share of car use in a city could only decrease
in favour of public transport starting from a size of 750,000 inhabitants
(Wegener and Fürst, 1999).
Also at a higher scale level efficient public transport is possible, albeit
within a spatial network structure. The nodes of the network consist of
towns or villages with a high density and a limited size, concentrated
around the stops (stations). With a sufficiently large mass of potential
travellers, located within the radius of action of available pre- and post-
transport, public transport may also play an important role within this
network. Within the towns or villages which are the nodes, however,
public transport will hardly play any role, given the limited size of the
node. Ideally, the size of the settlement, therefore, will be limited to the
range of a pedestrian. To be competitive with automobile traffic, high
frequency and speed is required, in combination with relatively slow car
traffic (e.g., due to congestion or other speed restrictions).
In all other cases (local traffic in small and regional cities, or traffic in
and from vast and sparsely populated suburban areas and the outlying
area) public transport will never perform a significant share of all trips,
Chapter 1
44
and won’t be able to play a role in the sustainability and efficiency of the
transport system. In all these cases, public transport will continue to take
only a niche of the transport market and will have primarily a social
function.
However, a spatial structure which is completely responsive towards
public transport opportunities, is not necessarily the most suitable for
other forms of sustainable transport such as walking, cycling or even
short car trips. Because of the crowded traffic large cities are scoring
relatively poorly in terms of bicycle use. Due to their large size these
cities also often exhibit a more thorough separation of functions than is
the case in small towns. Therefore, the distances covered in big cities are
not necessarily shorter than in smaller cities. Also, big cities generate
significant amounts of commute from the surrounding area. The theoreti-
cal potential for walking is therefore suppressed, and an important
volume of car traffic remains present. The finding by Breheny (1992) that
the fuel consumption per capita in London is slightly higher than in the
British regional cities, despite the efficient London public transport
system, is significant in this regard. The assumption that an urban
structure that is suitable to be served by a high quality public transport
network is, by definition, a sustainable system, is one of the weaknesses
in the compact city theory.
In Flanders the railway-grafted historically polycentric structure ful-
fils the spatial conditions for the operation of a regional public transport
system. However, most spatial developments that occurred after World
War II no longer meet these, just as the many settlements where the
existing rail infrastructure was removed or is no longer operated. Regard-
ing the role of local public transport, Brussels is the only agglomeration
in our study area that meets both the density requirement and the
criterion of minimum size. The metropolitan areas of Antwerp and to a
lesser extent, Ghent, are in compliance with the density requirement of
30-40 persons per hectare, but lack the critical mass needed to justify the
implementation of a high end public transport system (such as a metro
network).
Of course, even this situation may change if energy prices would start
soaring extremely one day, or in case climate change would become top
political priority. Public transport is generally much more energy efficient
than private transport, so especially an extensive rail network could
ensure also in these conditions a high degree of accessibility for large
parts of the population, apart from the car. Mees (2010, p. 53) states that
Introduction
45
public transport should play a role anyway in a future-oriented vision of
mobility, even if most neighbourhoods do not meet the specified density
requirements: “Arguments that densities many times current levels are
needed before transport trends can change are really arguments for
continuing with automobile dependence.”
1.5 Flanders and Brussels: policy context
The research which is reported in this dissertation, is funded by the
Policy Research Centre on Regional Planning and Housing - Flanders,
and is limited to the territory of the Flanders Region. Geographically
speaking, this region, with over 6 million inhabitants and an area of
about 13,500 km2, covers the northern half of Belgium, but excludes the
Brussels Capital Region which is an enclave of only 161 km2 and is home
to over 1 million inhabitants. Given the geographical context, Brussels
was included in most analyses in this dissertation, unless data limitations
did not allow incorporation. It is however important to note that policies,
both in terms of planning and in terms of mobility, are in Flanders and
Brussels based on different policy plans that were drafted and adopted by
two individual governments.
In 1997, the Spatial Structure Plan for Flanders (RSV, 1997/2004),
which may be considered as the first full-fledged spatial policy plan for
the Flanders Region, chose resolutely for strengthening the dichotomy
between urban areas and the countryside. This plan introduces “decon-
centrated clustering” as the first main principle for steering spatial
development (Albrechts et al., 2003; Allaert, 2005, p. 10). “Clustering”
means selectively concentrating the growth of living, working and other
social functions in cities and centres, while “deconcentrated” means
accounting for the existing (deconcentrated) development pattern and the
spread distribution of dynamic functions throughout Flanders. The
protection of open space and the revitalization of the urban fabric are
clearly paramount. By pursuing a spatial concentration of development in
precisely those areas that already possess a significant density, fragmenta-
tion of the (open) space is supposed to be combated, while existing
facilities and infrastructure will be used in a more efficient and more
sustainable way.
A concrete instrument proposed to implement these objectives is the
demarcation of urban areas, meaning that a line is drawn around those
Chapter 1
46
areas that should be reserved for the development of new highly dynamic
activities. Additional supply of industrial and residential land is provided
in these urban areas, and in new residential developments a minimum
density of on average 25 dwellings per hectare is aimed for. Also in the
nuclei of the countryside a minimum density is required, although this
requirement is only 15 dwellings per hectare. From a spatial perspective,
the alleged advantages of demarcating the urban areas are as follows:
scarce space is dealt with in a more economical way, facilities end up
closer to homes and are therefore more accessible and cheaper, contiguous
agricultural areas do not continue to fragment and ecological relation-
ships are preserved.
The RSV suggests that the demarcation provides also a number of
advantages in terms of mobility: “The principle of deconcentrated
clustering leads to a spatial clustering of needs for movement. This leads
to broadening of individual travel options (e.g. different destinations
remain accessible on foot, by bicycle). Consequently, the accessibility of
various facilities is basically higher. Spatial clustering is a prerequisite for
common transport. Through spatial concentration of origins and destina-
tions, an overall reduction of the amount of traffic is also possible.”
(RSV, 1997/2004, p. 472) Regarding the spatial component of mobility
policy in Flanders, generally the RSV is referred to (MOW, 2001).
The clear option to demarcate the urban areas demonstrates that the
RSV is undeniably following the compact city model, although this is
supplemented by explicit attention to the functioning of the compact
urban areas as nodes in urban networks (e.g. the Flemish Diamond,
connecting the metropolitan areas of Antwerp, Ghent and Brussels
Capital Region with their surroundings). Such an urban network can be
considered as a spatial concept for an urban structure at a different scale
level and as an extension to the compact city model.
In Brussels, the RSV is not in force. The second version of the over-
arching spatial policy plan for Brussels (Gewestelijk Ontwikkelingsplan /
Plan Régional de Développement (GewOP/PRD, 2002)) is in fact not a
regional plan but rather an urban development plan. Compared with
RSV, the GewOP/PRD is much clearer in relation to the regulation of
density and land use mix. Moreover, the development of public transport
and a bicycle network are essential parts of the plan, and one of the
strategic objectives is to reduce car traffic by 20%.
Introduction
47
1.6 Research questions, conceptual
framework and implementation
1.6.1 Research questions
Based on the foregoing considerations, we formulate the general research
question that is underlying this thesis, as follows:
• To what extent is mutual proximity of potential destinations deter-
minant for daily travelled distances in Flanders, and what does this
mean in a context of peak oil and climate change? (A)
In order to make the basic question operational, we subdivide it into
three questions:
• How can the influence of spatial structure on daily travel distances be
quantified? (B1)
• How can spatial proximity be defined, measured and applied in
sustainable spatial planning practice? (B2)
• What locations should be selected for additional housing or jobs, if
we want to keep additional traffic generation as small as possible?
(B3)
1.6.2 Conceptual framework
The research questions summarize the research framework which stems
from the broader context outlined. In this section we propose a concep-
tual framework that clarifies the links between different aspects of the
research. Fig. 1.5 is a schematic representation of this conceptual frame-
work.
The conceptual framework is policy oriented and uses the method
that Mayer and Greenwood (1980) developed specifically for policy
research. The conceptual framework examines the opportunities that are
offered by the policy area of spatial planning to meet the stated objec-
tives, within Flanders (where possible extended to Brussels). Possible
influences of other policy areas (mobility, environment, climate change
and taxation) are regarded as exogenous. Two policy objectives to be
pursued are formulated: (1) improving the sustainability of daily travel
patterns, and (2) mitigating the spatial and economic vulnerability to
potentially rising transport costs, and thus strengthening resilience.
Chapter 1
48
Fig. 1.5. Conceptual framework
As an instrumental objective, which is assumed to be strongly impres-
sionable by spatial planning policy, we propose an increase in spatial
proximity. Spatial proximity is represented by a non-exhaustive set of
different measurable spatial variables such as the jobs-housing balance
(the ratio between the number of jobs and the number of working
population), density (e.g. population density, or job density), diversity
(the degree of spatial mix of different functions), and so forth. Part of the
research will be to select variables based on their representativeness of
the characteristic called spatial proximity, and to find ways to measure
these accurately.
The relationship between spatial proximity and the two proposed ba-
sic policy objectives is based on two assumptions. First, we know from
the literature that a higher population density and a better mix of
functions usually accompanies shorter trip distances and a lower share of
car drivers, factors indicating a more sustainable travel pattern. Although
the found relationships are often weaker than would be expected intui-
tively, they are usually statistically significant. This means that
strengthening the spatial proximity is expected to result in a more
sustainable travel pattern. How strong this effect is, depends inter alia on
personal socioeconomic characteristics, such as income, family composi-
Introduction
49
tion or personal preferences. In short, in this context spatial planning is
used to pursue environmental and climate objectives.
Second, we argue that increasing characteristics of spatial proximity
reduces the vulnerability of the spatial-economic system to increasing
transport costs (Dodson and Sipe, 2008). Although there is no extensive
literature available on this proposition, it is logical that increasing
transport cost is critical for interaction possibilities in a spatial system
where everything is far away from everything else. In urban structures,
characterized by a high degree of spatial proximity, however, interaction
possibilities remain ample even when transport is limited. The reason why
this view has hardly been addressed in the scholarly literature is the
almost continuous decline in transport costs throughout history. It is only
after acknowledging the sense of reality of the peak oil theory that an
increase in transport cost in the long term can be seen as a likely sce-
nario.
This brings us to the right part of the scheme depicted in Fig. 1.5,
where the influences that are considered as exogenous in this conceptual
model are located. The oil price is represented as the most important
factor, since it is highly autonomous and can only to a very limited
extent be controlled. Technological developments are important too in
the sense that a combination of more efficient engines and alternative
fuels could eventually substitute a part of the growing demand for oil.
Technology will have both an indirect (through transport costs), and a
direct effect on the two key policy objectives. Finally, also some other
policy areas play their role in determining the cost of transport. Never-
theless, in the overall picture, the impact of these policy factors may
perhaps be less important, since these are to some extent the result of a
social consensus, unlike the oil price which is mainly determined by
physical supply limits. The combination of all these exogenous influences
determines the relative cost of transport. This relative cost must be seen
in an objectified form: for example, the average cost to move one person
over one kilometre, expressed as a percentage of GDP. Ultimately, the
relative cost partly determines the sustainability of daily travel (distance
travelled and transport mode) and the extent to which the economy may
potentially suffer from spatial vulnerability.
A final element that occurs in the conceptual framework consists of
the so-called unintended effects of strengthening spatial proximity, which
are summarized under the heading of “land use restrictions”. The deliber-
ate strengthening of certain characteristics of spatial proximity, such as
Chapter 1
50
population density and functional diversity, inevitably entails implica-
tions that are not universally perceived as positive, such as higher land
prices, smaller houses or more potential nuisance. Unintended effects are
included in the conceptual framework for completeness, but are no
subject of further research in this dissertation.
1.6.3 Implementation
The implementation of the conceptual framework follows two tracks. The
first track is the placement of the research in the scholarly literature.
Although the thesis does not contain a separate chapter on existing
literature, each chapter is embedded in a review of the relevant literature.
This was done by consulting mainly international peer-reviewed journals,
in most cases included in the ISI / Thomson Reuters Web of Science
index, and complemented with international standard works and Belgian
or Flemish research reports and policy documents.
As may be presumed from the title of the dissertation, the second
track consists of the actual quantitative analysis. The sources for the
various quantitative analyses are several data sets that were made
available from the Policy Research Centre on Regional Planning and
Housing - Flanders, or by the Flemish regional government itself, sup-
plemented with data that is freely available on the internet. The origin
and content of the data sets used is described in the various chapters.
The applied methods are inspired from the literature, but were ad-
justed according to the proposed research questions and the specific
characteristics of the study area. Applied quantitative methods range
from simple calculations and cartographic representation to data manipu-
lation with own algorithms and simple (t-test, ANOVA and correlation
analysis) or more advanced statistical techniques (multivariate regression
analysis, spatial econometrics and the construction of a forecasting
model). Repeatedly attention is paid to the cartographic representation of
variables used, while the maps themselves are considered as a major form
of outcome. Analyses were performed to the whole study area (the
Flanders Region), if possible supplemented with the Brussels Capital
Region (depending on the availability of data).
Introduction
51
1.7 Overview of the research
Each of the following six chapters contains a clearly defined sub-study
that is providing an answer to a part of the problem. Since each chapter
can be read separately, below an overview is presented in order to
establish the linkages between the chapters (Fig. 1.6). All off these six
chapters have been published, or will be published in international peer-
reviewed journals. A basic version of each of these chapters was presented
at least at one scholarly international conference.
In the second chapter, which is published as Boussauw and Witlox
(2009), energy consumption levels for commuting are calculated and
mapped on the basis of residential location. Regional differentiations in
commuting distances, modal shares of non-car travel modes and aspects
of infrastructure and population densities are used to explain some
relationships between energy consumption, commuting behaviour and
spatial-economic structure in Flanders and Brussels. It is found that
mode choice appears to be of little impact for the energy performance of
home-to-work travel on a regional scale. At the other hand, proximity
between home and work locations is paramount. Residential density plays
a part in this, although much depends on the specific situation. This is
also the case for the accessibility of the main road and rail network. In
some regions these infrastructures induce long-distance commuting,
whereas in the economic core areas this effect is much less pronounced.
All these are factors that are very much determined by infrastructural
and spatial policies of the past.
Chapter 3, which will be published as Boussauw et al. (2011a), as-
sesses the possibility to use a spatially disaggregated form of the so-called
minimum commuting distance as a spatial proximity characteristic with
regard to the commute. This paper focuses on regional variations in
commuting trip lengths by calculating minimum (required) commuting
distances, along with excess commuting rates. So, this paper contributes
to the excess commuting research framework from a regional perspective,
both by stressing the specific characteristics of urban networks with
overlapping commute areas, and by putting forward an alternative
method for calculating spatially disaggregated values. In the study area,
large variations in minimum commuting distances occur. This in turn
identifies to a large extent opportunities for shrinking commuting dis-
tances by influences such as rising fuel prices, compact urban planning,
extreme congestion or dissuasive traffic policies.
Chapter 1
52
Fig. 1.6. Dissertation outline
The fourth chapter (Boussauw et al., 2011b) extends the excess commut-
ing framework to a longitudinal assessment, aiming to measuring spatial
separation processes between jobs and housing in Flanders. It is known
that the average distance covered by individual commuting trips increases
year after year, regardless of the travel mode. Although increasing
prosperity is often invoked as the main reason, the discipline of spatial
planning also points to the relevance of land-use policies that enable
processes of suburbanization and sprawl. By calculating time series of
spatially disaggregated theoretical minimum commuting distances, this
paper offers a method to identify and quantify the process of spatial
separation between the housing market and the job market. We consider
the detected spatial separation as an indicator for the contribution of
spatial processes to the growth of traffic.
In the study area, it is found that over time the minimum commuting
distance increased in many municipalities, especially where population is
growing faster than job supply, or where traditionally high concentrations
Introduction
53
of employment still increase. Decreases are noticed in suburban areas that
are getting a more urban character by acquiring a considerable functional
mix. For the study area in its entirety, we do indeed register an increas-
ing spatial separation between home and work locations. However, this
separation evolves less rapidly than the increase in commuting distances
itself.
Regarding the methodology, we find that the use of municipalities as
a spatial entity is suitable for grasping regional transformations of the
economy and intermunicipal forms of suburbanization and peri-
urbanization. However, a similar methodology, applied at a more detailed
geographical scale, could be used to detect processes of sprawl in the
morphological sense.
As a third contribution from the excess travel perspective, Chapter 5
(Boussauw et al., 2011c) is applying the excess commute framework on
non-commuting daily travel. Based on the spatial distribution of some
quasi-daily destination classes and survey-reported trip distances regional
variation in excess travel in non-professional trips in Flanders is assessed.
To this end, proximity to various quasi-daily destinations is compared
with the reported distance that is actually travelled to reach similar, but
alternative, facilities.
We note that in rural areas (compared with urban areas) larger dis-
tances are travelled, although the closest facility is chosen more often. In
the most urbanized areas, however, we note that spatial proximity is also
an important aspect in destination choice.
Quantification of these phenomena can support the practice of
sustainable spatial planning by distinguishing areas that are too mono-
functional or too remote, and therefore need more functional diversity,
and by identifying areas where densification is useful because the location
is closely to most quasi-daily destinations, reducing the need to travel
over large distances.
In Chapter 6 (Boussauw et al., 2011d), an assessment is made of the
relationship between selected aspects of spatial proximity (density,
diversity, minimum commuting distance, jobs-housing balance and job
accessibility) and reported commuting distances. Part of this paper is
built on results from Chapter 2 and Chapter 3. Results show that correla-
tions may depend on the considered trip end. For example, a high
residential density, a high degree of spatial diversity and a high level of
job accessibility are all associated with a short commute by residents,
while a high job density is associated with a long commute by employees.
Chapter 1
54
A jobs-housing balance close to one is associated with a short commute,
both by residents and by employees. In general, it appears that the
alleged sustainability benefits of the compact city model are still valid in
a context of continuously expanding commuting trip lengths.
The seventh chapter (Boussauw and Witlox, 2011) extends the
methodology of Chapter 6 to all daily travel, both commuting and non-
commuting, and includes results from Chapter 5. Based on a set of spatial
proximity characteristics this paper develops a model that estimates for
every neighbourhood in Flanders the amount of traffic that would be
generated by an additional residential unit when socioeconomic variables
are held constant. The results show that residential density, land use
diversity and proximity of facilities influence daily travelled distances
when these variables are measured in the immediate vicinity of the
residential location of the respondent (within a radius of 1 km). When
aggregating these variables at a larger geographical scale, in most cases
the impact proves no longer significant. Variables based on the spatial
distribution of jobs, or on the global accessibility of the entire population
in the study area, do not show significant effects on the travel distance.
Despite the statistical significance only a fraction of the observed
variance in reported distances is explained by characteristics of spatial
proximity. However, we can assume that the importance of spatial
structure in the genesis of mobility patterns will increase in case the cost
of transport would rise (cf. peak oil). For this reason, the application of
the mapped results of the proposed model could contribute to the prac-
tice of sustainable spatial planning.
As a conclusion, the eighth chapter summarizes the findings, makes
the necessary critical nuance, and formulates policy recommendations.
Finally, an addendum explores the possible relationship between air
travel behaviour and residential location, an issue that was not included
in the main theme of the dissertation but may be considered as an
important issue when putting travel demand into a global perspective.
Introduction
55
References
Aguilera, A. (2005) “Growth in commuting distances in French polycen-
tric metropolitan areas: Paris, Lyon and Marseille.” Urban Studies.
42(9), pp. 1537–1547.
Albrechts, L., P. Healey and K. R. Kunzmann (2003) “Strategic spatial
planning and regional governance in Europe.” Journal of the Ameri-
can Planning Association. 69(2), pp. 113-129.
Aleklett, K., M. Höök, K. Jakobsson, M. Lardelli, S. Snowden and B.
Söderbergh (2010) “The peak of the oil age - Analyzing the world oil
production reference scenario in World Energy Outlook 2008.” Energy
Policy. 38(3), pp. 1398-1414.
Allaert, G. (2005) Wegwijs in Ruimtelijke Economie: Doorkijk naar
Planning en Management van Ruimte. Gent: Academia Press.
Anable, J. and A. L. Bristow (2007) Transport and Climate Change:
Supporting document to the CfIT report. Aberdeen - Loughborough:
The Robert Gordon University - Loughborough University.
ASPO (2010) Association for the Study of Peak Oil & Gas from
http://www.peakoil.net.
Banister, D. (1992) “Energy use, transport and settlement patterns”. In:
M. Breheny (Eds.) Sustainable Development and Urban Form. Lon-
don: Pion, pp. 160-181.
Banister, D. (1999) “Planning more to travel less.” Town Planning
Review. 70(3), pp. 313-338.
Banister, D. (2008) “The sustainable mobility paradigm.” Transport
Policy. 15(2), pp. 73-80.
Bardi, U. (2009) “Peak oil: The four stages of a new idea.” Energy. 34(3),
pp. 323-326.
Beevers, S. D. and D. C. Carslaw (2005) “The impact of congestion
charging on vehicle emissions in London.” Atmospheric Environment.
39(1), pp. 1-5.
Blauwens, G., P. De Baere and E. Van de Voorde (2002) Transport
Economics. Antwerpen: De Boeck.
Boarnet, M. G. and R. Crane (2001) Travel by Design: The Influence of
Urban Form on Travel. Oxford: Oxford University Press.
Bogaert, S., L. De Smet, F. Verdonck, K. Van Biervliet, S. Logghe, L.
Franckx, J. Eyckmans and S. De Jaeger (2006) Verkennende Studie
Chapter 1
56
naar Prijs- en Inkomens-Elasticiteiten van Milieugerelateerde Goe-
deren en Diensten in Vlaanderen. Aalst: Flemish Environment
Agency.
Boussauw, K. and F. Witlox (2009) “Introducing a commute-energy
performance index for Flanders.” Transportation Research Part A.
43(5), pp. 580-591.
Boussauw, K. and F. Witlox (2011) “Linking expected mobility produc-
tion to sustainable residential location planning: Some evidence from
Flanders.” Journal of Transport Geography. in press.
Boussauw, K., T. Neutens and F. Witlox (2011a) “Minimum commuting
distance as a spatial characteristic in a non-monocentric urban sys-
tem: The case of Flanders.” Papers in Regional Science. 90(1), in
press.
Boussauw, K., B. Derudder and F. Witlox (2011b) “Measuring spatial
separation processes through the minimum commute: The case of
Flanders.” European Journal of Transport and Infrastructure Re-
search. 11(1), pp. 42-60.
Boussauw, K., V. Van Acker and F. Witlox (2011c) “Excess travel in
non-professional trips: Why looking for it miles away?” Tijdschrift
voor Economische en Sociale Geografie. in press.
Boussauw, K., T. Neutens and F. Witlox (2011d) “Relationship between
spatial proximity and travel-to-work distance: The effect of the com-
pact city.” Regional Studies. in press.
Breheny, M. (1992) “The contradictions of the compact city”. In: M.
Breheny (Eds.) Sustainable Development and Urban Form. London:
Pion, pp. 138-159.
Breheny, M. (1996) “Centrists, decentrists and compromisers: Views on
the future of urban form”. In: M. Jenks, E. Burton and K. Williams
(Eds.) The Compact City: A Sustainable Urban Form? London:
E&FN SPON, pp. 13-35.
Brons, M., P. Nijkamp, E. Pels and P. Rietveld (2008) “A meta-analysis
of the price elasticity of gasoline demand. A SUR approach.” Energy
Economics. 30(5), pp. 2105-2122.
Bruegmann, R. (2006) Sprawl: A Compact History. Chicago: University
of Chicago Press.
Brussels Airport (2009) Brutrends 2008. Brussel: Brussels Airport.
Campbell, C. J. and J. H. Laherrère (1998) “The end of cheap oil.”
Scientific American. March 1998, pp. 78-83.
Introduction
57
CEC (1990) Green Paper on the Urban Environment - COM(90)218.
Brussels: Commission of the European Communities.
Cervero, R. and J. Murakami (2010) “Effects of built environments on
vehicle miles traveled: Evidence from 370 US urbanized areas.” Envi-
ronment and Planning A. 42(2), pp. 400-418.
Chapman, L. (2007) “Transport and climate change: A review.” Journal
of Transport Geography. 15(5), pp. 354-367.
City of Bruges (2007) Evaluatie Mobiliteitsplan. Brugge: Stad Brugge -
Mobiliteitscel.
Coolsaet, W. (1990) Naar een Oplossing voor het Verkeersprobleem.
Gent: Kritiek.
De Vlieger, I., E. Cornelis, L. Int Panis, L. Schrooten, L. Govaerts, L.
Pelkmans, S. Logghe, K. Vanherle, F. Vanhove, G. De Ceuster, C.
Macharis, T. Festraets, E. Pekin, L. Turcksin, K. Van Bladel, M. de
Jong, J. Van Mierlo, J. Matheys, J.-M. Timmermans, C. De Geest
and E. Van Walsum (2007) MIRA (2007) Milieurapport Vlaanderen,
Achtergronddocument 2007, Transport. Aalst: Flemish Environment
Agency.
Deakin, E. (2007) “The environmental consequences of sprawl”. In:
Transport, Urban Form and Economic Growth - Round Table 137.
Paris: European Conference of Ministers of Transport - OECD Pub-
lishing.
Dodson, J. and N. Sipe (2008) “Shocking the suburbs: urban location,
homeownership and oil vulnerability in the Australian city.” Housing
Studies. 23(3), pp. 377-401.
ECB (2010) “Oil prices - Their determinants and impact on euro area
inflation and the macroeconomy.” ECB Monthly Bulletin. 12(8), pp.
75-92.
EEA (2008) Data and Maps: EN01 Energy Related Greenhouse Gas
Emissions. European Environment Agency from http://www.eea.eu
ropa.eu/data-and-maps/indicators/en01-energy-related-greenhouse-
gas-emissions.
Engwicht, D. (1992) Towards an Eco-City: Calming the Traffic. Sydney:
Envirobook.
ESDP (1999) European Spatial Development Perspective. Luxembourg:
European Commission.
Ewing, R. (1997) “Is Los Angeles-style sprawl desirable?” Journal of the
American Planning Association. 63(1), pp. 107-126.
Chapter 1
58
Ewing, R. and R. Cervero (2010) “Travel and the built environment: A
meta-analysis.” Journal of the American Planning Association. 76(3),
pp. 265-294.
Ewing, R., R. Pendall and D. Chen (2002) Measuring Sprawl and Its
Impact. Washington, DC: Smart Growth America.
Galster, G., R. Hanson, M. R. Ratcliffe, H. Wolman, S. Coleman and J.
Freihage (2001) “Wrestling sprawl to the ground: Defining and meas-
uring an elusive concept.” Housing Policy Debate. 12(4), pp. 681-717.
Geurs, K. T. and J. R. Ritsema van Eck (2001) Accessibility Measures:
Review and Applications. Bilthoven: National Institute for Public
Health and the Environment.
GewOP/PRD (2002) Gewestelijk Ontwikkelingsplan / Plan Régional de
Développement. Brussels: Brussels Capital Region.
Gilbert, R. and A. Perl (2008) Transport Revolutions: Moving People and
Freight Without Oil. London: Earthscan / James & James.
Glaeser, E. L. and M. E. Kahn (2003) “Sprawl and urban growth”. In: J.
V. Henderson and J.-F. Thisse (Eds.) Handbook of Urban and Re-
gional Economics. Amsterdam: Elsevier, pp. 2481-2527.
Gleeson, B. (2008) “Critical commentary. Waking from the dream: An
Australian perspective on urban resilience.” Urban Studies. 45(13),
pp. 2653-2668.
Glorieux, I., J. Minnen and J. Vandeweyer (2005) Tijdsbesteding in
Vlaanderen: Veranderingen in Tijdsbesteding tussen 1999 en 2004.
Brussels: Vrije Universiteit Brussel.
Gonzalez-Feliu, J., F. Toilier and J.-L. Routhier (2010) “End consumer
goods movement generation in French medium urban areas.” Proce-
dia - Social and Behavioral Sciences. 2(3), pp. 6189-6204.
Goossens, S. (2008) Verkeerstellingen 1e Jaarhelft 2008. Brussels: Minis-
try of the Flemish Community.
Gordon, P. and H. W. Richardson (1989) “Gasoline consumption and
cities: A reply.” Journal of the American Planning Association.
55(3), pp. 342-346.
Gordon, P. and H. W. Richardson (1997) “Are compact cities a desirable
planning goal?” Journal of the American Planning Association.
63(1), pp. 95-106.
Hägerstrand, T. (1970) “What about people in regional science?” Papers
in Regional Science. 24(1), pp. 6-21.
Introduction
59
Hamilton, J. D. (2008) “Oil and the macroeconomy”. In: S. N. Durlauf
and L. E. Blume (Eds.) The New Palgrave Dictionary of Economics.
Basingstoke: Palgrave Macmillan.
Hamilton, J. D. (2009) “Causes and consequences of the oil shock of 2007-
08”. In: D. H. Romer and J. Wolfers (Eds.) Brookings Papers on
Economic Activity: Spring 2009. Washington, DC: Brookings Institu-
tion Press, pp. 215-284.
Handy, S., L. Weston and P. L. Mokhtarian (2005) “Driving by choice or
necessity?” Transportation Research Part A. 39(2-3), pp. 183-203.
Harvey, R. O. and W. A. V. Clark (1965) “The nature and economics of
urban sprawl.” Land Economics. 41(1), pp. 1-9.
Holden, E. and I. Norland (2005) “Three challenges for the compact city
as a sustainable urban form: Household consumption of energy and
transport in eight residential areas in the greater Oslo region.” Urban
Studies. 42(12), pp. 2145–2166.
Hubbert, M. K. (1956) Nuclear Energy and the Fossil Fuels. Houston,
TX - Washington, DC: Shell Development Company - American Pe-
troleum Institute.
Hupkes, G. (1977) Toekomstscenario’s voor ons Vervoerssysteem. PhD
Thesis. Amsterdam: Universiteit van Amsterdam.
IEA (2008) World Energy Outlook. Paris: International Energy Agency.
Illich, I. (1974) Energy and Equity. New York, Evanston, San Francisco,
London: Harper & Row.
Inflationdata (2010) Historical Crude Oil Prices from http://inflation
data.com/Inflation/Inflation_Rate/Historical_Oil_Prices_Table.asp.
INRIX (2008) National Traffic Scorecard - Special Report. Kirkland,
WA: INRIX.
Jenks, M., E. Burton and K. Williams (Eds.) (1996) The Compact City:
A Sustainable Urban Form? London and New York: E&FN Spon.
Joly, I. (2004) Travel time budget - Decomposition of the worldwide
mean. International Association of Time-Use Research Annual Con-
ference 2004. Rome: Italian National Statistical Institute.
Keppens, M. (2006) Modelling Private Road Transport of Flemish
Households. The Mobility and Welfare Effects of Fuel Taxes versus
Tradable Fuel Rights. PhD Thesis. Hasselt: Universiteit Hasselt.
Kopits, S. (2009) “Oil: What price can America afford?” Douglas-
Westwood Research Note. 22nd June.
Kunstler, J. H. (1993) The Geography of Nowhere: The Rise and Decline
of America’s Man-Made Landscape. New York: Simon and Schuster.
Chapter 1
60
Levinson, D. M. and K. J. Krizek (2008) Planning for Place and Plexus:
Metropolitan Land Use and Transport. New York: Routledge.
Levinson, D. M. and A. Kumar (1997) “Density and the journey to
work.” Growth and Change. 28(2), pp. 147-172.
Litman, T. (2007) Transportation Elasticities: How Prices and Other
Factors Affect Travel Behavior. Victoria: Victoria Transport Policy
Institute.
LNE (2006) Vlaams Klimaatbeleidsplan 2006-2012. Brussels: Ministry of
the Flemish Community.
Ma, K.-R. and E.-T. Kang (2011) “Time-space convergence and urban
decentralisation.” Journal of Transport Geography. doi:10.1016/
j.jtrangeo.2010.06.016.
Marinetti, F. T. (1909) “Manifeste du futurisme.” Le Figaro. 20 Febru-
ary.
Marte, G. (2003) “Slow vehicle traffic is a more attractive alternative to
fast vehicle traffic than public transport.” Journal of World Trans-
port Policy & Practice. 9(2), pp. 18-23.
Mayer, R. and E. Greenwood (1980) The Design of Social Policy Re-
search. Englewood Cliffs, NJ: Prentice-Hall.
Meadows, D. H., D. L. Meadows, J. Randers and W. W. Behrens III
(1972) The Limits to Growth. New York: Universe Books.
Mees, P. (2010) Transport for Suburbia: Beyond the Automobile Age.
London: Earthscan.
Mérenne-Schoumaker, B., H. Van der Haegen and E. Van Hecke (1999)
Algemene Volks- en Woningtelling op 1 maart 1991: Werk- en
Schoolpendel. Brussels: Statistics Belgium.
Mindali, O., A. Raveh and I. Salomon (2004) “Urban density and energy
consumption: A new look at old statistics.” Transportation Research
Part A. 38(2), pp. 143–162.
Mokhtarian, P. L. and I. Salomon (2001) “How derived is the demand for
travel? Some conceptual and measurement considerations.” Transpor-
tation Research Part A. 35(8), pp. 695-719.
MOW (2001) Mobiliteitsplan Vlaanderen - Ontwerp. Brussels: Ministry of
the Flemish Community.
Newman, P. and J. Kenworthy (1989) Cities and Automobile Depend-
ence. A Sourcebook. Aldershot: Gower.
Newman, P. and J. Kenworthy (1999) Sustainability and Cities: Over-
coming Automobile Dependence. Washington, DC: Island Press.
Introduction
61
Newman, P. and J. Kenworthy (2006) “Urban design to reduce automo-
bile dependence.” Opolis. 2(1), pp. 35-52.
Reynebeau, M. (2003) Een geschiedenis van België. Tielt: Lannoo.
Righelato, R. and D. V. Spracklen (2007) “Carbon mitigation by biofuels
or by saving and restoring forests?” Science. 317(5840), pp. 902.
Robèrt, M. and R. D. Jonsson (2006) “Assessment of transport policies
toward future emission targets: A backcasting approach for Stock-
holm 2030.” Journal of Environmental Assessment Policy and
Management. 8(4), pp. 451-478.
Rodrigue, J.-P., C. Comtois and B. Slack (2009) The Geography of
Transport Systems. New York: Routledge.
RSV (1997/2004) Ruimtelijk Structuurplan Vlaanderen - Gecoördineerde
Versie. Brussels: Ministry of the Flemish Community.
Saunders, H. D. (1992) “The Khazzoom-Brookes postulate and neoclassi-
cal growth.” Energy Journal. 13(4), pp. 131-148.
Schafer, A. (2000) “Regularities in travel demand: An international
perspective.” Journal of Transportation and Statistics. 3(3), pp. 1-31.
Song, Y. and G.-J. Knaap (2004) “Measuring urban form - Is Portland
winning the war on sprawl?” Journal of the American Planning As-
sociation. 70(2), pp. 210-225.
Statistics Belgium (2008) Levensstandaard - verbruik van goederen en
diensten from http://www.statbel.fgov.be/figures/d323_nl.asp.
Statistics Belgium (2010) Grootte van het voertuigenpark 2010 from
http://statbel.fgov.be/nl/modules/publications/statistiques/verkeer_
vervoer/evolution_du_parc_de_vehicules_2010.jsp.
SVR (2010) Cijfers. Studiedienst van de Vlaamse Regering. Ministry of
the Flemish Community from http://www4.vlaanderen.be/dar/svr/
Cijfers/Pages/Excel.aspx. Retrieved 26/08/2010.
Szalai, A. and P. Converse (1972) The Use of Time: Daily Activities of
Urban and Suburban Populations in Twelve Countries. The Hague:
Mouton.
Torrens, P. M. and M. Alberti (2000) Measuring Sprawl. London: Centre
for Advanced Spatial Analysis - University College London.
Van Wee, B., P. Rietveld and H. Meurs (2002) A Constant Travel Time
Budget? In Search for Explanations for an Increase in Average
Travel. Amsterdam: Vrije Universiteit Amsterdam.
Verhetsel, A. (2001) “The impact of planning and infrastructure measures
on rush hour congestion in Antwerp, Belgium.” Journal of Transport
Geography. 9(2), pp. 111-123.
Chapter 1
62
Verhetsel, A., I. Thomas, E. Van Hecke and M. Beelen (2007) Pendel in
België. Deel I: De Woon-Werkverplaatsingen. Brussels: Statistics Bel-
gium.
Virilio, P. (1977) Vitesse et Politique: Essai de Dromologie. Paris:
Editions Galilée.
VMM (2008) “MIRA-T 2008 Flanders Environment Report: Indicator
Report”. In: MIRA-T 2008 Indicatorrapport: Flemish Environment
Agency.
Weart, S. R. (2010) “The idea of anthropogenic global climate change in
the 20th century.” Wiley Interdisciplinary Reviews: Climate Change.
1(1), pp. 67-81.
Wegener, M. and F. Fürst (1999) “Land-use transport interaction: State
of the art.” Berichte aus dem Institut für Raumplanung(46), pp. 1-99.
Zahavi, Y. and A. Talvitie (1980) “Regularities in travel time and money
expenditures.” Transportation Research Record. (750), pp. 13-19.
Zhang, M. (2002) Conditions and Effectiveness of Land Use as a Mobility
Tool. PhD Thesis. Cambridge, MA: Massachusetts Institute of Tech-
nology.
63
Chapter 2:
Introducing a
commute-energy
performance index
This paper has been published as Boussauw, K. and F. Witlox (2009)
“Introducing a commute-energy performance index for Flanders.”
Transportation Research Part A. 43(5), pp. 580-591. Copyright ©
Elsevier. All rights reserved.
Abstract
Based on 2001-census data for Belgium, energy consumption levels for
commuting were calculated and mapped on the basis of residential
locations in the administrative regions of Flanders and Brussels. Com-
parison with regional differentiations in commuting distances, modal
shares of non-car travel modes and aspects of infrastructure and popula-
tion densities clarifies some relationships between energy consumption,
commuting behaviour and spatial-economic structure in the suburbanised
historic-polycentric spatial structure which characterises the northern
part of Belgium. It is found that mode choice appears to be of little
impact for the energy performance of home-to-work travel on the scale of
the Flanders region. At the other hand, proximity between home and
work locations is paramount.
Residential density plays a part in this, although much depends on
the specific situation. This is also the case for the accessibility of the main
road and rail network. In some regions these infrastructures induce long-
distance commuting, whereas in the economic core areas this effect is
much less pronounced. All these are factors that are very much deter-
mined by infrastructural and spatial policies of the past.
Chapter 2
64
Keywords: sustainable spatial development; commuting; energy per-
formance; Flanders
2.1 Introduction
Following Newman and Kenworthy (1989), many researchers have put
forward the energy efficiency of urban transport as a sustainability
indicator. Although Newman and Kenworthy (1989) were repeatedly
criticized because of methodological reasons, the rationale for the use of
energy performance as an indicator for measuring the sustainability of
transport in relation to spatial structure kept upright.
This paper investigates the link between spatial structure and energy
consumption for home-to-work travel. To this end the concept of a
commute-energy performance (CEP) index will be developed and tested
for Flanders (Belgium). This indicator is not only considered as a proxy
for the sustainability of the transport system in itself, but by extension
for those of the spatial-economic structure as a whole. The results can
constitute a basis for further research, which aims to determine the
robustness of spatial structures in a climate of incipient fuel scarcity. A
better understanding of this matter will uncover social and spatial
evolutions, and leads to a policy that facilitates a more sustainable
development.
The paper is structured as follows. In Section 2, we briefly discuss the
relationship between energy use and spatial structure in order to concep-
tualize our CEP index. Section 3 puts the home-to-work commute in the
context of all personal travel. The CEP index is then formally developed
in Section 4. In Section 5, we introduce our data and geographical setting
and discuss some spatial differentiations. For this purpose, we map the
average energy consumption for home-to-work travel in Flanders and
Brussels. Based on obvious regional differences, a number of hypothetical
relationships with the underlying spatial and economic structures can be
put forward. Finally, in Section 6 and 7, we summarize our main findings.
2.2 Energy use and urban spatial structure
The main thesis of Newman and Kenworthy (1989, 1999) is the existence
of an inverse relationship between urban density and energy consumption
for transport. Their research was based on data from 32 world cities. In a
Introducing a commute-energy performance index
65
critical reaction to Newman and Kenworthy’s (1999) conclusions, based
on a new analysis of the same data, Mindali et al. (2004) argue that the
assumed general correlation between density and energy consumption for
transport is in fact only valid for certain aspects of the urban structure,
i.e. in the central business district. Banister (1992) and Banister and
Banister (1995) applied a similar methodology as Newman and Kenwor-
thy (1989) on British cities, using data from the National Travel Survey
(1985-1986) and the 1981 census. For London, the city with the highest
overall density, the analysis does not support Newman and Kenworthy’s
thesis: energy consumption per capita is slightly higher in the capital
than the average in the other surveyed cities (> 25.000 inhabitants).
Dodson and Sipe (2008) on the other hand introduced the concept of an
“oil vulnerability index” as a quantification of the robustness of a spatial
entity to rising oil prices, and also take social factors (such as income)
into account. They found that those parts of the outer urban fringe where
no public rail transport is available, are the most vulnerable.
It is also our aim to develop a spatial sustainability indicator that
enables the mapping of regional and urban differences in energy consump-
tion for transport. There are two important arguments that can be put
forward to develop such an index. First, there is the growing importance
of the energy factor in the public debate. The sharp fluctuations in oil
prices, the debate about peak oil and the efforts made to reduce emissions
of greenhouse gases play a role in this discussion (Witze, 2007). Second,
the relationship between spatial structure and travel is a vexed issue.
Travel patterns are highly heterogeneous, and vary with the morphology
and the use of the space. Although concepts such as high density and
diversity are classically considered characteristics of a spatial structure
with a high potential for sustainable trips (Cervero and Kockelman,
1997), it is in fact very difficult to isolate spatial parameters and to
demonstrate causal relationships (Van de Coevering and Schwanen, 2006;
Van Acker et al., 2007; Hammadou et al., 2008). Existing literature
envisages almost always clearly demarcated urban areas. Newman and
Kenworthy (1989) for example, did not consider external flows, while in
some of the cities they studied a significant proportion of the jobs are
filled by commuters. Cervero and Kockelman (1997) and Schwanen and
Mokhtarian (2005) studied the San Francisco Bay Area, which is morpho-
logically much more homogeneous and extensive than, for example, an
average European urban area. Little is known however about the rela-
tionship between the suburbanised historic-polycentric spatial structure
Chapter 2
66
which characterises densely populated European regions such as Flanders,
and the travel patterns of its users. Moreover, travel behaviour is to a
large extent determinant for the energy consumption, distances and used
transport mode being the main parameter. In addition, the rate of car
ownership and - for car owners - the chosen type of car play their part
(Keppens, 2006). Hence, it is worthwhile to investigate the spatial
differentiation in energy consumption for transport in spatially highly
heterogeneous areas. The indicator to be developed should be able to
make explicit the relationship with spatial qualities, such as density,
characteristics of proximity or remoteness, or the presence or absence of
major transport infrastructure at different geographical scales.
2.3 Limitations of studying the
home-to-work commute
Accurate data on the home-to-work commute is more often available than
data on other trips. This is probably the main reason why many studies,
such as those of Dodson and Sipe (2008), focus on home-to-work travel to
quantify the sustainability of travel patterns. However, in studies focusing
on a small enclosed area, it is easier to incorporate different kinds of
trips, as was done by Saunders et al. (2008).
This paper too is based on home-to-work commuting data. This
commute is not representative of all trips, but does affect significantly
non-work related trips. From the theory of the constant travel time
budget (Schafer, 2000), we can say that commuters who spend a lot of
time travelling to work will spend less time on other trips. This means
that they will make more efficient chained trips and that they will look
for destinations closer to home, but also that they will choose more
frequently for fast means of transport (i.e. the car). Moreover, the home-
to-work commute is more inert than other trips are, which can be
illustrated on the basis of price elasticities (Mayeres, 2000). Given this
rigidity, changes in preconditions, such as fluctuations in fuel prices, will
be more problematic for commuting patterns than for non-work trips.
Furthermore, commuting trips cover more often large distances (Zwerts
and Nuyts, 2004), and thus contribute significantly to the negative effects
of traffic. The last two arguments indicate that the study of the home-to-
work commute remains very important.
Introducing a commute-energy performance index
67
It is important to understand that the rigidity of the commute, both
with respect to distance covered and with respect to modal choice, is not
only a spatial issue. The attitude of the commuter towards the destina-
tion and itinerary, and in particular its habits, determine this rigidity to
a large extent. Consequently, we should consider the travel pattern as a
result of the interaction between space, motivation and habit (Gardner,
2009).
2.4 Commute-energy performance
(CEP) index
In order to exemplify the relationship between the spatial configuration of
an urban region and energy use we develop a commute-energy perform-
ance (CEP) index. This index is obtained by dividing the total amount of
energy consumption for home-to-work travel per census block (i.e. the
smallest geographical research unit) by the working population (active
workforce) that lives in the census block. More formally
s
isiis
sN
cED
CEP∑ ⋅⋅
=
,
(2.1)
In which CEPs is the energy performance per member of the active
workforce for home-to-work travel from the considered (statistical) census
block s; Ds is the total distance travelled (one way) for home-to-work
travel from the considered census block s; iE is the mean energy con-
sumption per passenger for the considered mode i; ci,s is the correction
factor for the considered mode i, within the census block s; Ns is the
number of members of the active workforce in the considered census
block s.
In order to take into account the differences in energy efficiency be-
tween the different transport modes used, the home-to-work travel trips
are split up into motorized (fuel consuming) trips (car, public transport)
and non-motorized trips (on foot, bicycle). For public transport there are
significant differences in energy efficiency between bus, tram, metro, and
train. Hence, we formulate the mean energy consumption per passenger in
relation to the type of public transport used
∑ ⋅= iibtm EKE (2.2)
In which =iE mean energy consumption per passenger for the considered
mode i (bus, tram, metro); Ki is the share of the considered mode i (bus,
Chapter 2
68
tram, metro) in the total number of kilometres made by these three
modes.
For the train mode we can further distinguish between electric and
diesel trains (where α denotes the fraction of electric/diesel trains),
resulting in
trdtretr EEE ⋅−+⋅= )1( αα (2.3)
For car use a comparable subdivision in view of the type of combustion
fuel used (petrol, diesel, LPG) should be made as well. However, our
commuting data (see further) does not allow to distinguish between
different types of car trips, so although further subdividing iE is useful,
it will not add to the analysis. For all non-motorized trips iE is of course
equal to zero.
To keep the relationship between the mode and the distance trav-
elled, for each mode a correction factor is derived from the average trip
length
∑ ⋅
⋅
=
iisi
msm
smDS
DSc
,
,
, (2.4)
In which cm,s is the correction factor for the considered mode m, within
the census block s; Sm,s is the share of the considered mode m as main
transport mode in the total number of home-to-work trips from the
considered census block s; mD is the mean distance of a home-to-work
trip with the considered mode m; i is each of the considered modes.
Finally the resulting number of person kilometres per mode is multi-
plied with a standardized value for the energy consumption per mode.
2.5 Geographical setting and data analysis
Our developed CEP index is tested for Flanders and the Brussels Capital
Region. Fig. 2.1 is a schematic representation of the spatial structure of
Flanders, as defined in the Spatial Structure Plan for Flanders (Ruim-
telijk Structuurplan Vlaanderen (RSV), 1997/2004). The RSV is the
overarching spatial policy plan for the Flanders region. The RSV selects
three “metropolitan areas”, with more than 300,000 inhabitants, being
Ghent, Antwerp and the Flemish urban area around Brussels. Because of
the consistency of the research, we also integrate the Brussels Capital
Region.
Introducing a commute-energy performance index
69
Furthermore the RSV selects ten “regional urban areas” with a mag-
nitude which is situated between 50,000 and 150,000 inhabitants, as well
as five “urban networks”, an “economic network” along the Albert canal
and five “gates” (ports and airports). The main urban network is that of
the Flemish Diamond, which is bounded by the three metropolitan areas
and the regional urban area of Leuven, and is the economic core of
Flanders. The other areas are considered as countryside, still including a
number of small urban areas and economic nodes with rather limited
development perspectives.
Fig. 2.1. Schematic representation of the spatial vision on Flanders.
Source: RSV (1997/2004)
Chapter 2
70
Table 2.1. Average trip length for home-work travel by transport mode,
for the purpose of determining the correction factor for the trip length
(based on OVG 2001)
bicycle =bcD 4.07 km
moped, motorbike =mbD 10.85 km
transportation organized by the employer or school =emD 18.86 km
car driver =cdD 20.33 km
car passenger =cpD 17.43 km
train =trD 48.48 km
bus, tram, metro (underground) =btmD 18.86 km
on foot only =ofD 2.15 km
Table 2.2. Itemization of passenger kilometre share for urban and
regional public transport by mode, Flanders and Brussels
tram and trolley bus =tK 11.3%
metro (underground) =meK 2.0%
bus =bK 86.7%
Table 2.3. Default values for energy consumption per person kilometre
(source: De Vlieger et al. (2006))
tram =tE 0.06 kWh/pkm
metro (underground) =meE 0.08 kWh/pkm
city bus =cbE 0.25 kWh/pkm
coach =coE 0.32 kWh/pkm
aggregated: urban and regional public
transport
=btmE 0.26 kWh/pkm
electric train =treE 0.13 kWh/pkm
diesel train =trdE 0.18 kWh/pkm
aggregated: train =trE 0.14 kWh/pkm
city car =ccE 0.43 kWh/pkm
family car =cfE 0.53 kWh/pkm
aggregated: car =cE 0.48 kWh/pkm
bicycle, on foot =ofbc EE , 0.00 kWh/pkm
Introducing a commute-energy performance index
71
Also within the infrastructure network, especially the road network, a
selection and classification of roads is made. The metropolitan and
regional urban areas are all opened up by the main road network, and
served by at least one train station.
The data used to calculate the CEP index for Flanders and Brussels
are drawn from various sources. The so-called General Socio-Economic
Survey 2001 (SEE 2001, see: Verhetsel et al., 2007) is a comprehensive
survey of the Belgian population (excluding children younger than six
years old), which has its origin in the 10-yearly census. The response to
the survey is 95%. The questionnaire of SEE 2001 contains a number of
mobility related questions. The distance between home and work or
school is assessed. In addition, questions are raised about the means of
transport used for home-to-work or home-to-school trips, the number of
daily trips whether or not combined, car ownership and the perception of
the supply of public transport around.
Data on the average trip length per mode is based on the Travel Be-
haviour Research project in Flanders (OVG 2001) (Zwerts and Nuyts,
2004) (Table 2.1).
The standardized values for the energy consumption per mode are
taken from De Vlieger et al. (2006), and are based on the French research
by Enerdata (2004) (Table 2.3). All energy values are converted to
kilowatt hour per person kilometre (kWh/pkm). In each case the final
energy consumption by the vehicle is considered, meaning that for electric
vehicles the losses that occur in the production and transmission of the
electricity, which depend greatly on how it is generated, are not taken
into account. For the category “car as passenger” the same value is
applied as for the category “car driver”, since the default value is set per
person and already takes into account the average occupancy rate of the
vehicle. More specific variations in energy consumption, such as the
distinction between diesel and gasoline cars, or regional differences in the
composition of the fleet of personal cars or the ridership of buses and
trains, are not taken into account.
Data on the use of local public transport was split up on the basis of
the passenger statistics of the urban and regional public transport
companies in Flanders (De Lijn) and Brussels (STIB) for the year 2006
(De Lijn, 2006; STIB, 2006) (Table 2.2). The mode train is itemized into
78% electric and 22% diesel, based on De Vlieger et al. (2006).
Chapter 2
72
2.6 Results
2.6.1 Spatial distribution of the CEP index:
a first glance
We calculate the CEP index for home-to-work travel, based on the
departure zones. Because of the limitations of the available data, the
resulting map should only be interpreted as an approximation, which
aims to uncover the gradients with regard to energy consumption for
home-to-work travel in Flanders and Brussels. In a next step the relation-
ship with a number of spatial characteristics will be unveiled (see
further).
In order to interpret the CEP index, the relevant spatial characteris-
tics are mapped as well, based on data from SEE 2001. The main road
network and the railways are added to all maps. Fig. 2.3 shows the
average home-to-work distance per member of the active work force. For
this, data on home-to-work distances are aggregated for each neighbour-
hood (census block) and divided by the working population. It is possible
to draw up a similar map for the home-to-school distances. Figs 2.4-2.6
present the frequency with which the modes, that are known to be
energy-efficient, are used as main transport mode for home-to-work trips.
For the purpose of ease of interpretation, these maps are simplified in the
sense that, for example, any pre- and post-transportation is not consid-
ered while evaluating the use of the various modes. Moreover, not all
modes are included: pedestrians, car passengers and transport which is
organised by the employer are not incorporated in the set. The purpose of
the maps is thus to give a global overview of the variation of these
parameters on the territory of Flanders and Brussels.
According to the mapped CEP index, energy consumption for home-
to-work travel seems to be particularly high in those regions which in
spatial planning terminology are defined as the countryside (A1-8) (Fig.
2.2)1. These regions have in common that they possess a relatively rural
character, compared to the labour markets where they are focused on.
The regions A1 and A3, for example, are influenced by the labour
markets in the metropolitan and urban areas of Brussels, Ghent and
Leuven, even if those are relatively distant (Van Nuffel, 2007). In addi-
1 The applied codes refer to the respective zones on Figs 2.2-2.5.
Introducing a commute-energy performance index
73
tion, commuters in these rural regions have on average higher incomes
which allows them to live outside the city centres in the relatively quiet
and green environments, being less sensitive to the financial impact of the
large daily commuting distances.
Apart from that, some corridors along the motorways are strongly
reflected in the map. The locations B1-4 catch the eye. It is clear that in
these cases the increased accessibility by the presence of a motorway has
contributed to enlarge commuting distances and the increased importance
of the car as a transport mode. The area, in which the energy consump-
tion is pre-eminently low, is the Brussels Capital Region (C1). The
Flemish urban area around Brussels has a more or less comparable
pattern, but still scores worse than the Brussels’ municipalities. This
result concurs with what might be expected, as the Brussels region
represents the largest job market of the country, and also has the highest
population densities. It is therefore consistent with the idea that the
match in the labour market supply and demand is achieved within short
distances. Moreover, the metropolitan spatial structure is responsible for
the relatively large influence of other parameters on the energy consump-
tion, such as modal split and vehicle ownership. This will be discussed
below.
Similar patterns occur in the two other metropolitan areas of Ant-
werp (C2) and Ghent (C3), in which the effect of the metropolitan
structure of Antwerp clearly outreaches the case of Ghent. In all regional
urban areas, we also find lower energy consumption than the average.
But also outside the metropolitan and regional urban areas, there are a
number of regions that come out on the right side by their significantly
lower energy consumption. The most contiguous region we find at D1-2.
This region is characterized by a strong sprawl of less specialized labour,
and a strong spatial interweaving of the labour market with the residen-
tial structure. The importance of location-bound industries, in particular
in the agricultural sector, probably plays a part in this. So, the distance
between home and workplace remains relatively confined.
Furthermore, also the corridor Brussels-Mechelen-Antwerp (D3), an
important transport artery, scores remarkably well on the map, as well as
a part of the economic network of the Albert canal (D4). These economi-
cally strong areas have high concentrations of employment in a - on the
scale of Flanders - relatively good mix with the residential structure. We
see the same phenomenon, albeit on a smaller scale, arising in D5-D7.
Chapter 2
74
The rural areas D8-D11 show rather low figures. Apparently, the rela-
tively poor accessibility of these regions has caused only a few long-
distance commuters to settle here. In addition, the low population and
building density in these regions makes that a rather large share of the
population is still working in the local agribusiness.
Fig. 2.2. Daily energy consumption per capita for home-work travel
(kWh). Source: SEE 2001
2.6.2 Spatial patterns and relation to
home-to-work distance
The CEP map (Fig. 2.2) shows a remarkable resemblance with the map
that visualizes the home-to-work distances (Fig. 2.3). The Pearson’s
correlation coefficient between the two sets of indicators is close to 0.95.
We can therefore conclude that the energy consumption for home-to-work
travel is first and foremost determined by the distance between home and
workplace. Contrary to what is generally assumed, it appears that the
used transport mode plays only a very limited role. This can partly be
explained by the fact that the average distance covered by train commut-
ers (on average 48 km in 2000) is much larger than the average journey
that is made by car (on average 20 km). Secondly, the bicycle is only an
alternative for short trips, which makes this mode only marginally
represented in the total number of kilometres.
Introducing a commute-energy performance index
75
Fig. 2.3. Mean distance for home-work travel (km). Source: SEE 2001
2.6.3 Relation to the use of the bicycle
In Belgium, particularly in Flanders, the bicycle has always been an
important means of transport in short distance commuting. Belgium is
ranked third in Europe (after The Netherlands and Denmark (1997)) with
regard to the use of the bicycle (Witlox and Tindemans, 2004). This is
chiefly explained by the flat topography, as well as the strong railway-
bounded pre-war spatial development (in which the bicycle appeared to
be the perfect pre- and post-transportation means). The relief shows an
unmistakable determining factor with respect to the use of the bicycle in
home-to-work travel. The cycling map (Fig. 2.4) has therefore bright
spots in the hilly regions A1, A2, A3, D10 and E1. Also, in the Brussels
Capital Region and the Flemish urban area around Brussels, cycling to
work hardly occurs. The very dense traffic during peak-hours is an
important explanatory variable.
In all regional urban areas, the bike is still very important in the
home-to-work travel. That is also the case, albeit somewhat less explicit,
in the metropolitan areas of Antwerp and Ghent. Furthermore those
regions where a significant mix of living and working occurs, and the
distances are accordingly short, catch the eye (D1-7). In addition, there
are some more isolated areas that also score well in terms of bicycle use,
Chapter 2
76
such as the urban network of the coast (E2) and the peripheral regions
E3 and E4.
When looking at the cycling map (Fig. 2.4) and the distance map
(Fig. 2.3) at one glance, we see that the largest share of cyclists occurs at
first in those regions where the home-to-work distances are the shortest,
with the Brussels Capital Region and the Flemish urban area around
Brussels as a major exception to this rule. According to the OVG 2001,
the average length of a cycle trip in home-to-work travel amounts to 4
km. This figure is based on a survey, and its reliability as a reported
distance is relatively small given the small distance range (Witlox, 2007).
Fig. 2.4. Share of bicycle in home-work trips. Source: SEE 2001
Clearly, the bicycle plays a major role in those regions where the com-
muting distances are of this small magnitude. The influence of the bicycle
on the total energy consumption for home-to-work travel is very limited.
Even in regions with a relatively high share, the bike use share in com-
muting is less than 20% of the total number of trips. As trips with other
modes cover much larger distances per trip (see Table 2.1), the gain in
terms of energy consumption made by cyclists is of little significance. The
low energy consumption in areas with a high proportion of cyclists is
largely on the account of the small absolute distances, regardless of the
transport mode used. But the positive impact of the short distances in
Introducing a commute-energy performance index
77
these areas is reinforced by the larger share of cyclists that substitutes car
use on short distances, in comparison with other regions.
2.6.4 Relation to the use of the train
Figs 2.5 and 2.6 visualize the level of use of public transport. Fig. 2.5
deals with the share of train passengers, while Fig. 2.6 demonstrates the
share of tram, bus and metro travellers. High concentrations of train
commuters can be found along the railway axes which provide since a
long time a fast connection with the capital, particularly to the west and
south-west of Brussels (F1-F7). Also some railway axes that are focused
on Antwerp, catch the eye (F7-8). Apart from this, some local concentra-
tions of train commuters (F9-12) can be distinguished. The inhabitants of
the Brussels Capital Region and the Flemish urban area around Brussels
hardly use the train for home-to-work travel. This applies in general to
the regions which are relatively remote from Brussels too. In the outlying
regions of Flanders, and particularly the northeast, commuters hardly use
the train.
Interestingly, a significant portion of the concentrations of train
commuters is located in those areas where commuting distances are the
largest and corresponding energy consumption is the highest. This is
particularly the case in the area southwest of Brussels (G1), and the
infrastructure axes towards Ghent (G2) and Leuven (G3). This can be
explained by the finding from the OVG 2001 that the average length of a
home-to-work trip by train amounts to 48 km, which is very high (Table
2.1). The train is therefore mainly an alternative to long car trips.
Consequently, it stands to reason that the train is popular in regions with
large commuting distances.
However, this does not apply to all areas where the train is well posi-
tioned. In particular, a number of regional urban areas (G4-7) and Ghent
show relatively low energy consumption combined with a rather high
proportion of rail commuters. This applies to a certain extent also to the
railway axes that focus on Antwerp (G8-G11).
It is however not possible to make unequivocal statements about the
impact of the use of the train on energy consumption. First, it is true
that the average rail passenger covers larger distances than the average
car driver (48 km compared with 20 km). When we multiply the average
energy consumption per kilometre (0.14 kWh/pkm respectively 0.48
kWh/pkm, see Table 2.3) by the average number of kilometres per mode,
Chapter 2
78
it appears that in home-to-work travel the energy consumption of a train
trip per person is only 30% less than the energy consumption of a trip by
car. This difference is less impressive than the expectations raised by the
sustainable image of the train. Fast rail transport supply also induces
long-distance commuting, which means no gain in terms of energy
consumption. Finally, the majority of train journeys entail travel to and
from the station, which often means an additional trip by car.
On the other hand, the train substitutes long car trips, at least where
rail transport supply is present. Furthermore, the train is doing this -
calculated per kilometre - in a more energy-efficient way. Nevertheless,
the share of train commuters, even in the concentration areas, is limited
to a maximum of approximately 20%. In general it may be said that
there occurs only a positive impact on energy consumption by use of the
train in those regions where the average home-to-work distances are
already large. But the effect is still too small to be visible on the energy
performance map (Fig. 2.2).
Fig. 2.5. Share of train in home-work trips. Source: SEE 2001
2.6.5 Relation to the use of urban and regional
public transport
Fig. 2.6 shows that tram, metro and bus as a main transport mode in
home-to-work trips only play a significant part in the three metropolitan
Introducing a commute-energy performance index
79
areas. In Ghent, the smallest of these metropolitan areas, this mode
scores only in a few neighbourhoods higher than 10%. In Antwerp the
influence of the urban and regional transport reaches beyond that, with
values that are often well above 10%. In Brussels, where urban transport
is built on the backbone of the metro, the situation is completely differ-
ent. Many neighbourhoods show a share of around 30%.
Given the preponderance of Brussels in the use of this mode, we can-
not draw any conclusions about the average trip distance. This is because
the values in Table 2.1 are derived from a sample survey from Flanders.
The impact on energy consumption is difficult to predict. There is a
significant discrepancy between the energy consumption of electric rail
transport and that of (diesel) buses. Only in Brussels an extensive metro
and tram network exists.
On the other hand, those parts of Brussels and Antwerp that show
lower energy consumption per capita have also a high to very high
proportion of public transport users. It is clear that the high density of
inhabitants and jobs in those cities combines the positive effects of the
proximity of functions with an energy-efficient and well exploited urban
transport system. The relative good energy efficiency and the high
patronage of the metro, supplemented with the tram, account for this.
Fig. 2.6. Share of bus, tram, metro in home-work trips.
Source: SEE 2001
Chapter 2
80
In the rest of Flanders the urban and regional transport, based on diesel
buses, hardly plays a part in the energy performance of home-to-work
commuting. The supply, which is limited in comparison with the metro-
politan areas, and the non competitive speed of this kind of public
transport contributes to the limited success of the urban and regional
transport in the market segment of home-to-work travel. Although we
will not discuss this more profoundly, the relatively low ridership outside
the urban areas also works to the detriment of the energy performance of
public transport.
2.6.6 Relation to car ownership
Car ownership is expressed in number of available cars per household. In
the historic centres of the study area, in particular those that are part of
the metropolitan areas, car ownership is low. Among other things, spatial
factors are at the basis of this. The high density, the significant mix of
functions and good supply of public transportation makes it relatively
easy to live without a car in the city centres. Also, a number of social
elements play a part, since it is precisely in these areas that the family
sizes are small and household revenues are rather low. However, it is
difficult to isolate environmental and social factors. The environment and
the rent and real estate prices in the city are often not adapted to the
lifestyle of families with children. In addition, those families need more
often combined trips, for which the car is usually the appropriate means
of transport.
In spatial terms the zones with a remarkably high car ownership have
following characteristics: they are suburban areas of major employment
centres, which also have a quick access to the main road network. In
social terms those are areas with high incomes. All main concentrations
are located in the suburban belts around the three metropolitan and some
regional urban areas.
Despite those regions that catch the eye, the regional differentiation is
in fact very limited. The resulting map is fairly homogeneous. The
differences are usually too small to be able to determine unique relation-
ships between car ownership and energy consumption for home-to-work
travel. However, this is possible in the above mentioned areas that attract
the attention. Where the car ownership is significantly lower (the urban
centres), also energy consumption is lower, and vice versa in zones where
the car ownership is significantly higher (the mentioned suburban areas).
Introducing a commute-energy performance index
81
In the rest of Flanders the relationship is much less pronounced. In the
urban network of cities situated along the coast, with its atypically aged
demographic composition, there even seems to be an inverse relationship
between car ownership and energy consumption.
2.7 Relation to spatial-morphological
characteristics
The CEP index is also useful in interpreting the relation to a range of
spatial-morphological characteristics. As an exploration, we outline the
relationship between energy consumption and two notable spatial charac-
teristics: residential density and proximity to the main road and railway
network.
2.7.1 Relation to the population density
Density is perhaps the most widely used measure, which is also easily
quantifiable. In their research about the relationship between urban
density and fuel efficiency, Newman and Kenworthy (1989) included
Brussels. The study area was limited to what is today called the Brussels
Capital Region. External commuting flows were not considered, while we
know that most jobs in Brussels are taken up by people who commute.
Let us now repeat the Newman and Kenworthy (1989) exercise for
Flanders and Brussels based on our developed CEP. Fig. 2.7 plots the
energy consumption for home-to-work travel on the basis of residence in
relation to the population density of the block. The image is of course
only a facet. Further research will also provide relations with the density
of jobs on employment locations, and the focus could be expanded to all
types of travel (school, shopping, recreation, visiting, et cetera). Despite
the limited coverage of the research field, we can see a number of inter-
esting similarities and differences with what Newman and Kenworthy
(1989) have found.
Chapter 2
82
Fig. 2.7. Plot of residential density and energy consumption for home-
work commuting, per census block
In general, it appears that the basic argument is still valid: the energy
performance improves with increasing density. However, the large
variation at the root of the graph catches the eye. Even though the trend
line for energy consumption is rising with decreasing density, for a part of
Flanders, in fact, an inverse relationship applies. The part of the dotted
cloud that is situated near the root of the graph (A) represents the rural
areas (i.e. countryside). These areas are typified by sparsely populated
regions with a non-intensive home-to-work travel pattern. In these areas,
most people tend to work at or near their home: farmers, and local
economy workers.
Introducing a commute-energy performance index
83
The part of the dotted cloud that is situated above the left part of
the trend line (B) represents the peripheral suburban sprawl. These areas
have developed as a result of counter-urbanised delocalization, and are
mostly inhabited by people who exhibit a strong urban lifestyle. Their
professional lives are also mainly concentrated in the urban areas; hence,
they represent an important travel demand.
The areas which are situated around the trend line roughly represent
the urban areas. The suburban blocks that are adjacent to the traditional
city centres are represented around the left part of the trend line (C),
while the traditional city centres themselves can be found on the right
side of the graph (D). In contrast to the analysis of Newman and Ken-
worthy (1989), which considered only urban areas, we can thus deduce a
spatial typology from regional variations in the efficiency of the transport
system.
2.7.2 Relation to the access to the main
road network
In the analysis of the spatial variation in energy performance of home-to-
work travel we already referred to a possible link with the location of the
motorways. In order to gain further insight into the potential effects of
the present infrastructure, the main road network and the rail network
was already presented in Fig. 2.2. Interestingly, certain main roads stand
out on the map, while others do not. Increased energy consumption is
clearly visible in the regions I1-10. All of these cases are rural regions that
became very well connected by the construction of a main road, but do
not have any large or diversified employment. The new access to the
main road network has encouraged moving to the countryside, has
generated traffic, has increased commuting distances and has put the
focus more exclusive on car travel. Eventually, all this has led to in-
creased energy consumption. For the other main roads, which are
traversing a structurally different socio-economic landscape, this finding
does not apply. Overall, it seems that two types of roads can be identified
using our framework. The first type of main road crosses an area with a
fairly diverse economy, where the residential structure is to a large extent
mixed with the job market. This is particularly true between Brussels and
Antwerp (I11, I12), but also in a number of other places (I13-15). More-
over, all main road segments crossing the metropolitan areas, harbour
and airport areas are of this type. The second type of main road runs
Chapter 2
84
through a relatively remote area, without providing a smooth connection
with a major employment centre in the urban network of the Flemish
Diamond. Examples of these corridors can be found in I16-20.
2.7.3 Relation to the access to the rail network
The link with the rail network is also very visible. Stations are located
next to the historic centres, which have the highest residential density
and often also the largest functional mix. Of the lower energy consump-
tion for home-to-work travel that we often observe in the vicinity of the
railway stations, only a tiny share can be attributed to the train com-
muters. This is explained in that the average train trip in home-to-work
travel still represents an energy quantity of 70% of that of an average car
commuting trip.
But the link is not unambiguous. A number of important rail lines are
flanked by an - in terms of traffic volume - much more important motor-
way, which eventually leaves its heavy mark on local commuting
behaviour. This is illustrated in the counter-urbanized area around J1,
where good train and motorway connections are present. The energy
consumption is thus high.
In the region southwest of Brussels (A1), a different phenomenon
emerges. Despite good rail links with the Flemish Diamond and the
absence of a main road, we find very high energy consumption levels. The
census blocks in which the main stations are located, score a lot better in
terms of energy than the surrounding areas. The blocks in the immediate
vicinity of Ghent’s main railway station (J2) have a very large share of
train commuters as well as relatively high energy consumption. Note that
this station has a very good train connection to Brussels. Areas which
manage to combine a large share of train commuters with low energy
consumption in home-to-work travel are rare: J3-8 fall into this category.
2.8 Conclusions
We have argued that the energy performance of the transport system is
an important approximate indicator for the sustainability of a spatial
structure. This is certainly true when advocating a so-called low carbon
economy is put increasingly higher on the political agenda. Obviously the
link with the spatial or urban (re)development of cities should be made as
well. Having a better understanding of the mechanisms that cause the
Introducing a commute-energy performance index
85
major observed regional variations in energy consumption will lead to
better land use planning in practice.
The issue of proximity in planning remains very important. In home-
to-work travel, the distance between home and workplace is to a very
large extent determinant for the energy performance of the commuting
system. Contrary to the conventional belief, the mode used is of much
less importance. In this respect we notice a discrepancy with the current
mobility policy of the Flemish government, which is very much focused
on the reduction of the share of car drivers, but much less on a reduction
of the number of kilometres, despite an increase by 10% of the average
commuting distance between 1991 and 2001 (Verhetsel et al., 2007).
Hence, the opportunity for someone to find a suitable job nearby his
or her living environment, or the ease with which someone can move in
the vicinity of his or her work will increasingly determine the robustness
of a spatial economic system in a climate of rising oil prices. It appears
that travel behaviour remains largely determined by the rigidity of the
housing stock, which makes short term policy intervention not easy. This
applies too - mutatis mutandis - for non-work related trips, although
these show higher elasticities. Therefore, we argue that it is important to
develop a more profound insight into the regional variations in the energy
performance of the whole transport system. So we will get a better
understanding of the processes that led to the current situation, and we
will be able to assess the policies that played a part or can still play a
part in this. In this respect the development and implementation of a
commute-energy performance index seems a useful indicator to assess
both transport and spatial planning policies with respect to inducing
sustainable development.
References
Banister, D. (1992) “Energy use, transport and settlement patterns”. In:
M. Breheny (Eds.) Sustainable Development and Urban Form. Lon-
don: Pion, pp. 160-181.
Banister, D. and C. Banister (1995) “Energy consumption in transport in
Great Britain: Macro level estimates.” Transportation Research Part
A. 29(1), pp. 21-32.
Chapter 2
86
Cervero, R. and K. Kockelman (1997) “Travel demand and the 3Ds:
Density, diversity, and design.” Transportation Research Part D.
2(3), pp. 199-219.
De Lijn (2006) Jaarverslag 2006. Mechelen: De Lijn - Flemish Public
Transport Agency.
De Vlieger, I., E. Cornelis, L. Int Panis, L. Schrooten, L. Govaerts, L.
Pelkmans, S. Logghe, K. Vanherle, F. Vanhove, G. De Ceuster, C.
Macharis, T. Festraets, E. Pekin, L. Turcksin, K. Van Bladel, M. de
Jong, J. Van Mierlo, J. Matheys, J.-M. Timmermans, C. De Geest
and E. van Walsum (2006) MIRA (2006) Milieurapport Vlaanderen,
Achtergronddocument 2006, Transport. Aalst: Flemish Environment
Agency.
Dodson, J. and N. Sipe (2008) “Shocking the suburbs: Urban location,
homeownership and oil vulnerability in the Australian city.” Housing
Studies. 23(3), pp. 377-401.
Enerdata S.A. (2004) Efficacité Energétique des Modes de Transport:
Rapport Final. Grenoble-Gières: Enerdata S.A.
Gardner, B. (2009) “Modelling motivation and habit in stable travel
mode contexts.” Transportation Research Part F. 12(1), pp. 68-76.
Hammadou, H., I. Thomas, A. Verhetsel and F. Witlox (2008) “How to
incorporate the spatial dimension in destination choice models: The
case of Antwerp.” Transportation Planning and Technology. 31(2),
pp. 153-181.
Keppens, M. (2006) Modelling Private Road Transport of Flemish
Households. The Mobility and Welfare Effects of Fuel Taxes versus
Tradable Fuel Rights. PhD Thesis. Hasselt: Universiteit Hasselt.
Mayeres, I. (2000) “The efficiency effects of transport policies in the
presence of externalities and distortionary taxes.” Journal of Trans-
port Economics and Policy. 34(2), pp. 233-260.
Mindali, O., A. Raveh and I. Salomon (2004) “Urban density and energy
consumption: a new look at old statistics.” Transportation Research
Part A. 38(2), pp. 143–162.
Newman, P. and J. Kenworthy (1989) Cities and Automobile Depend-
ence. A Sourcebook. Aldershot: Gower.
Newman, P. and J. Kenworthy (1999) Sustainability and Cities: Over-
coming Automobile Dependence. Washington, DC: Island Press.
RSV (1997/2004) Ruimtelijk Structuurplan Vlaanderen - Gecoördineerde
Versie. Brussels: Ministry of the Flemish Community.
Introducing a commute-energy performance index
87
Saunders, M. J., T. Kuhnimhof, B. Chlond and A. N. R. da Silva (2008)
“Incorporating transport energy into urban planning.” Transportation
Research Part A. 42(6), pp. 874-882.
Schafer, A. (2000) “Regularities in travel demand: An international
perspective.” Journal of Transportation and Statistics. 3(3), pp. 1-31.
Schwanen, T. and P. L. Mokhtarian (2005) “What affects commute mode
choice: Neighborhood physical structure or preferences toward
neighborhoods?” Journal of Transport Geography. 13(1), pp. 83-99.
STIB (2006) Activiteitenverslag 2006. Brussels: Société des Transports
Intercommunaux de Bruxelles / Maatschappij voor het Intercommu-
naal Vervoer te Brussel.
Van Acker, V., F. Witlox and B. van Wee (2007) “The effects of the land
use system on travel behavior: A structural equation modeling ap-
proach.” Transportation Planning and Technology. 30(4), pp. 331-
353.
Van de Coevering, P. and T. Schwanen (2006) “Re-evaluating the impact
of urban form on travel patterns in Europe and North-America.”
Transport Policy. 13(3), pp. 229-239.
Van Nuffel, N. (2007) “Determination of the number of significant flows
in origin-destination specific analysis: The case of commuting in
Flanders.” Regional Studies. 41(4), pp. 509-524.
Verhetsel, A., I. Thomas, E. Van Hecke and M. Beelen (2007) Pendel in
België. Deel I: De Woon-Werkverplaatsingen. Brussels: Statistics Bel-
gium.
Witlox, F. and H. Tindemans (2004) “Evaluating bicycle-car transport
mode competitiveness in an urban environment. An activity-based
approach.” Journal of World Transport Policy & Practice. 10(4), pp.
32-42.
Witlox, F. (2007) “Evaluating the reliability of reported distance data in
urban travel behaviour analysis.” Journal of Transport Geography.
15(3), pp. 172-183.
Witze, A. (2007) “That’s oil, folks….” Nature. 445(4), pp. 14-17.
Zwerts, E. and E. Nuyts (2004) Onderzoek Verplaatsingsgedrag Vlaan-
deren 2000-2001. Brussels-Diepenbeek: Ministry of the Flemish
Community.
Chapter 2
88
89
Chapter 3:
Minimum commuting
distance as a spatial
characteristic in a
non-monocentric
urban system
This paper will be published as Boussauw, K., T. Neutens and F. Witlox
(2011) “Minimum commuting distance as a spatial characteristic in a
non-monocentric urban system: The case of Flanders.” Papers in Re-
gional Science 90(1). Copyright © Regional Science Association
International - Blackwell Publishing. All rights reserved.
Abstract
This paper focuses on regional variations in commuting trip lengths by
calculating minimum (required) commuting distances, along with excess
commuting rates. The study contributes to the excess commuting re-
search framework from a regional perspective, both by stressing the
specific characteristics of urban networks with overlapping commute
areas, and by putting forward an alternative method for calculating
spatially disaggregated values. A case study in the north of Belgium
shows that large variations in minimum commuting distances occur. This
in turn identifies to a large extent opportunities for shrinking commuting
distances by influences such as rising fuel prices, compact urban planning,
extreme congestion or dissuasive traffic policies.
Keywords: excess commuting; spatial proximity; sustainable spatial
development; Flanders
Chapter 3
90
3.1 Introduction
The concept of excess commuting or wasteful commuting was initially
introduced by Hamilton (1982), and has become a well-established line of
inquiry in transportation research in the last decades (Ma and Banister,
2006). Hamilton (1982, p. 1040) defined wasteful commuting as the
difference between the actual commuting distances and the theoretical
minimum (required) commuting distances, which are suggested by the
spatial structure of a considered city. Hamilton’s interest for minimized
commuting distances probably stemmed from the consecutive oil crises of
1973 and 1979-1980, when the availability, and in particular, the afforda-
bility of fossil oil products was at stake. Daily trips over large distances
were suddenly considered problematic, because of their particularly high
energy consumption.
There is a broad consensus that spatial structure as a combination of
morphological elements and activities (e.g. size, shape and functional
mix) is a key determinant in explaining travel pattern generation (Giuli-
ano and Small, 1993; Van Acker et al., 2007). In many policy documents
on mobility and transport much hope is set on achieving an “adequate
spatial planning” as an effective means to improve the efficiency and
sustainability of mobility. However, while the spatial structure is gener-
ally recognized as a prerequisite for trip generation, observed travel
behaviour and trip distances in particular are additionally induced by
other factors.
We know that the observed average commuting distances in Euro-
pean and North American metropolitan areas increase year after year
(Banister et al., 1997; Aguilera, 2005). There is no doubt that the in-
creased prosperity implicitly or explicitly plays a role in the growth of the
travelled distances. However, it appears that the possibility to travel over
increasingly larger distances is systematically materialized in the form of
the physical separation of functions. Given that the public debate ac-
knowledges that an important part of the traffic problems is related to
land use policies, it is important to estimate what share of the actual
travel is caused by the spatial structure itself and what share is in fact an
extrapolation that originated from other elements, such as the general
prosperity, but also the level of congestion, the quality of roads, or the
price of fuel. Hamilton (1982) explored only home-work commuting, but
the concept may, mutatis mutandis, be extended to all daily travel
categories (Horner and O’Kelly, 2007).
Minimum commuting distance as a spatial characteristic
91
Researchers in this field usually consider excess commuting as a char-
acteristic of a specific city, regarded as a monocentric, polycentric or
dispersed urban system. Building on the exploratory work by Niedzielski
(2006) and Yang and Ferreira (2008) we want to put forward an exten-
sion to this line of reasoning, in which minimum commuting distance and
excess commuting are to be seen as properties of specific spatial entities
(spatially homogeneous areas) within a non-monocentric spatial system.
To this end, the local minimum commuting distance is considered as a
measure for spatial proximity of each relevant area, as embedded in the
studied region. This measure can be used to quantify relations to other
spatial characteristics such as density, spatial diversity or accessibility.
Quantification of traffic volumes is essential to assess the extent of
excess travel. As a proxy for traffic volumes we will use the number of
kilometres travelled per person to or from a considered zone within a pre-
defined time frame. As shown in Boussauw and Witlox (2009), at least at
the regional level (macro scale), distance travelled per person can be
deemed a good approximation of energy consumption, sustainability of
travel behaviour and total traffic volume. Depending on the spatial
characteristics of the study area and the research scale level, other
elements such as modal split, the composition of the fleet and the level of
congestion might be incorporated to avoid running the risk of oversimpli-
fication.
In relation to Niedzielski (2006) and Yang and Ferreira (2008), the
novelty of our approach is twofold. First, we develop a method that
extends the use of linear programming techniques and is more adequate
in studying spatial variations. In respect to the latter we are able to
simulate the case where all commuters start to look simultaneously for a
job closer to their homes. Second, we apply the spatially disaggregated
approach at a regional scale, on an urban network with overlapping
commuter areas. Ultimately, the methodology is applied to the study area
(Flanders and Brussels) and the results are interpreted.
Chapter 3
92
3.2 Spatial variations in excess travel
As a starting point, Hamilton (1982) considered a monocentric urban
model, as developed by Mills (1967) and Muth (1969). Within this model,
there is a balance between accessibility to the central business district
(CBD) and the bid rent, which is materialized in a density gradient of
both jobs and houses. The monocentric model allows predicting the
minimum home-work distances on the basis of the geometric characteris-
tics of a circle. Hamilton (1982) calculated these minimum distances for a
number of American and Japanese cities. He found that these distances
differed substantially from the actual commuting distances drawn from
survey data of the considered cities, questioning the validity of the
monocentric city model. Hamilton (1982) found an average actual
commuting distance of 8.7 miles (13.92 km), corresponding to a minimum
commuting distance of only 1.1 miles (1.76 km). Thus, in this case the
excess rate amounts to 7.9 or, put differently, 87% (= (8.7 - 1.1)/8.7) of
the actual commute is excessive. Hamilton’s (1982) methodology proved
very controversial, and was criticized by White (1988) and Small and
Song (1992), whose results showed large deviations compared to the
values calculated by Hamilton. The common motivation behind these
studies is the investigation of the predictive power of the monocentric
urban model (later extended to polycentric and dispersed urban models;
Song, 1995), with respect to commuting cost minimizing behaviour.
A major dichotomy appearing among studies on excess commuting is
the choice of either travel distance or travel time as a proxy for the travel
cost. While travel time is an intuitively appealing proxy for travel cost,
there are, from environmental policy perspectives, well-grounded reasons
to use distance as a parameter. First, the adverse external effects of travel
are more closely related to travel distance than to travel time. Second,
there is the constant nature of the personal travel time budget, which
means that over time urban travellers spend a constant amount of time
on their daily travel (Schafer, 2000).
Among the studies in which distance is used as a parameter, extreme
variations are recorded across various study areas (Ma and Banister,
2006). Frost et al. (1998) found an excess commuting rate of only 18.9%
for London (including all inward commuting), while the ratio found by
Song (1995) amounts to 81.6% for Los Angeles. Given the lack of uni-
Minimum commuting distance as a spatial characteristic
93
formity in the methodologies used, comparing results of different studies
can hardly be justified. It was found that calculated excess commuting
rates decrease as the number of constraints which are explicitly controlled
increases; examples include limitations originating from two-worker
households and the divide between tenants and home-owners (Kim,
1995), or accounting for the expectation of future job locations (Crane,
1996). Furthermore, the modifiable areal unit problem (MAUP), intrinsic
to the practice of spatial analysis, adds to the difficulty of comparing the
results of different studies in the sense that the size and configuration of
the zones used may bias the excess commuting rate (Horner and Murray,
2002).
Ma and Banister (2007) examined the theoretical relationship be-
tween variations in the spatial distribution of population and jobs, and
excess commuting. They showed that the excess commuting rate is a
good proxy for the potential reduction of commuting distances within a
specified study area, but that the minimum commuting distances (and by
extension also maximum commuting distances) are more suitable to use
in comparisons between different cities or points in time. Niedzielski
(2006) and Yang and Ferreira (2008), studied for the first time spatially
disaggregated minimum commuting distances within an urban area.
None of the above-mentioned contributions, however, analysed em-
pirically the spatial variations of the minimum commuting distance and
excess commuting rate on a regional scale. In the monocentric city model,
access to the CBD is the main determinant of the urban structure, while
in a polycentric or dispersed model accessibility to a diverse range of job
and service locations is determinant. For the detection of spatial varia-
tions of excess travel within the suburbanized historically polycentric
spatial structure that characterizes many urbanized regions in Europe,
the application of the monocentric city model to derive minimum com-
muting distances makes little to no sense.
We hypothesize that there exist important regional variations of
minimum commuting distances and excess travel, for which a link with
spatial characteristics (e.g., density, functional mix or proximity to major
transportation infrastructure), can be found. Earlier research shows that
there are important differences between cities (Charron, 2007), even if it
is still unclear how these values evolve through a region with multiple
centres and suburban areas.
Chapter 3
94
3.3 Possible policy implications
In a number of studies, such as Scott et al. (1997), Frost et al. (1998) and
O’Kelly and Niedzielski (2008), the focus shifts from the explanatory
nature of an urban economic model to the possible relevance for spatial
policies, aimed at assessing the potential reduction of undesired external
effects of the traffic, such as emissions or energy consumption. In particu-
lar, Frost et al. (1998) examined the evolution of the minimum
commuting distance and excess commuting rate for several British cities,
and linked this evolution with spatial developments.
For a given area, the excess commuting rate can be considered as an
indicator for the extent to which the particular spatial structure of this
area is able to absorb shrinkage of the total traffic volume, without
incurring severe economic damage. We assume that such a reduction can
be caused by external factors such as rising fuel prices, extreme conges-
tion or dissuasive policy measures (e.g., to limit emissions; Scott et al.,
1997, or congestion). Alternatively, we can also state that shrinkage of
excess travel volumes might become part of a scenario that, in the long
run, will ensue from an expected continued increase of oil prices (peak
oil). In addition, minimum commuting distances are an indicator of the
imbalance between the residential function (considered as the origin of
travel) and other functions, such as jobs, schools, shops and recreational
attraction poles (which may be regarded as destinations). The effect of
encouraging or just discouraging certain new developments can thus be
tested against the expected impact on minimum commuting distances. A
reduction of these theoretical minimum commuting distances will eventu-
ally make an effective reduction of travelled distances under the
aforementioned changing external influences much easier.
A final interesting feature of our approach is the ability to assess the
impact of various autonomous development scenarios on commuter
traffic. In this way, a significant growth or decline in both population and
employment under the influence of external factors can lead to a better
understanding of the need for investments in infrastructure or public
transport.
Minimum commuting distance as a spatial characteristic
95
3.4 Methodology
3.4.1 Premises
The premise of our method implies that any observed departure will be
matched to the nearest observed arrival, within a pre-defined time frame
(e.g. morning of a working day). For each trip purpose (e.g., work), the
number of departures per zone, as well as the number of arrivals, is
retained, but the observed connection between origins and destinations
will be cut through with the aim of minimizing the distance between
those two. To apply the methodology on a real case, we assume that an
origin-destination (OD) matrix is available for the selected zones, as well
as a distance matrix, providing the shortest physical network distance
between each pair of zones.
This theoretical exercise does not take into account the match
between origin and destination that exists in the real world. For home-
work travel, this would imply that everyone who is part of the active
population can be considered suitable to perform any job. Although this
theoretical assumption does not necessarily correspond to a real world
situation, we have made this assumption deliberately because we want to
gain an insight into the theoretically maximum reduction of travelled
distances.
Apart from job qualification, there are some more possible biases that
should be kept in mind. We do not consider residential self-selection
(Mokhtarian and Cao, 2006; Van Acker et al., 2010), the trade-off
between real estate prices, accessibility and environmental quality, or the
income and the composition of households (Van Ommeren, 2000). Also,
the presence or absence of rapid transport infrastructure, such as com-
muter rail, was not included in the calculation. In Section 7 we will
discuss the possible impact of these simplifications for our case study.
Furthermore, chained trips or detours are not simulated. As a conse-
quence, results should be considered an underestimation. In the discussion
of the case study (infra), a comparison will be made between the calcu-
lated distances travelled and those reported by survey respondents.
Chapter 3
96
3.4.2 Linear programming and the Monge-
Kantorovich mass transportation problem
White (1988) first suggested to define the problem of the minimum
required commuting distance as a conventional Monge-Kantorovich mass
transportation problem, to be solved by standard techniques of linear
programming (e.g., a simplex algorithm). The definition of this type of
problem is often illustrated by mines that have to supply factories. All
mines together produce the ore needed by all factories. Because of
differences in location, the transport cost to deliver one shipment from a
mine to a factory varies across the possible pairs of mines and factories.
The problem is solved when an OD matrix is calculated, which yields the
smallest total transport cost. Today several commercial and non-
commercial solvers are available, offering good approximate solutions for
this global minimum cost (in our case: minimum distance) and yielding a
corresponding OD matrix. A popular non-commercial solver, based on the
simplex algorithm, is for example lp_solve (Berkelaar et al., 2003).
In most of the excess commuting literature, the focus is on the calcu-
lated minimum commuting distance, which is usually compared with the
observed total commuting distance. In his spatially disaggregated ap-
proach Niedzielski (2006) also uses the calculated OD matrix to obtain
local values for the minimum commuting distance.
However, this procedure has some disadvantages. First, there is no
unique solution for the mass transportation problem (Feldmann and
McCann, 2002), making the resulting OD matrix dependent on the
software used and, to a certain extent, on the sort order of rows and
columns in the cost matrix. Second, and perhaps more problematic, is
that the algorithm fills as many cells as possible with zeros, mostly
assigning traffic flows from one zone to only one corresponding zone. This
situation is incompatible with our approach which aims at simulating the
implications of the case where all commuters start to look simultaneously
for a job closer to their homes. Apart from minimizing the total travelled
distance (which is equivalent to the cost in the transportation problem),
it is in this case also important that the optimization process is done in a
geographically balanced way, obtaining the global optimum through
multiple parallel local optimization processes. Because of this additional
premise, we develop our own procedure instead of using a standard solver
software package. Our method results in a geographically much more
Minimum commuting distance as a spatial characteristic
97
“smooth” solution, avoiding results where adjacent zones show extremely
differing values.
3.4.3 Algorithm
The algorithm developed here has been implemented to assess the extent
of excess commuting. The approach adopted is as follows. First, per zone
trips are made as much as possible intra-zonal. Thereafter in each run,
one trip from the remaining surplus or deficit is exchanged with the
nearest matching zone. The distance covered by the exchanged trip is
recorded. This cycle is repeated until all departures are matched with an
arrival. We give a more detailed description of the process in Fig. 3.1.
Initially, the physical distance dij over the network between any pair
of zones (i, j) is calculated. Each zone is represented by its centroid.
Using the Dijkstra algorithm, as implemented in the Network Analyst
extension of ESRI ArcGIS 9.2, all shortest paths between any pair of
centroids over the network were calculated. The resulting product is a
symmetrical distance matrix in which both rows and columns represent
every zone of the dataset. The distance for an intrazonal trip, which is
originally calculated as zero, is simulated by taking half the network
distance between the centroid of the considered zone and the centroid of
the nearest zone. In this way, intrazonal network distance is also taken
into account, which is not always done by alternative methods such as
developed in O’Kelly and Lee (2005).
The combination of the OD matrix and the distance matrix provides
the total distances travelled between each pair of zones, both viewed as
outbound (every zone considered as origin) as well as inbound (every zone
considered as destination).
The departures and arrivals within the OD matrix are then summed,
so that we obtain for each zone i the total number of departures Oi as
well as the total number of arrivals Di. The minimization process starts
by equalizing the number of internal trips Ii to the number of departures
Oi (in the case where there are fewer departures than arrivals) or to the
number of arrivals Di (in the opposite case). This number is then multi-
plied by the distance of the simulated internal trip HIi, and stored as a
basic travelled distance, both outbound (HOi) and inbound (HD
i). For the
next step in the process, the smaller of those two values is subtracted
from the number of departures from and arrivals at the considered zone.
Trips in one of the two directions are thus reduced to zero.
Chapter 3
98
For the remaining surplus (if originally there were more arrivals than
departures) or deficit (in the opposite case), we question whether a deficit
or otherwise a surplus exists in the nearest zone j. If a deficit (or surplus)
can be found in zone j, then one trip from the surplus (or deficit) of zone
i is used to fill up (or to receive) a part of the deficit (or surplus) of zone
j. The distance dij, which should be covered to eliminate this trip, will be
added to the outgoing (or incoming) distances (HOi or HD
i) of the consid-
ered zone as well as to the incoming (or outgoing) distances (HDj or HO
j)
of the nearby zone.
This process is pursued step by step until all zones i are given an ini-
tial chance to exchange trips ends. Then the cycle is repeated until all
surpluses and deficits are eliminated. We are particularly interested in the
total minimized outbound distance TOi and the total minimized incoming
distance TDi. In combination with the observed outbound distance
(TOi(obs)) and the observed inbound distance (TD
i(obs)) for the consid-
ered zone i, which we previously calculated, we can also map the
corresponding excess rates.
The main advantage of our algorithm is that the structure of the cal-
culated OD matrix is not arbitrary, but that the global optimum is
achieved through many cycles of local optimization. Tests show that the
impact of the sort order of the rows and columns on the resulting matrix
is negligible. However, there is also a disadvantage to taking this ap-
proach. In comparison with a solution generated by lp_solve, the total
minimum commuting distance yielded by our algorithm is a bit larger (11
to 15% in our tests), making it suboptimal in a mathematical sense,
although still preferable from a geographical point of view.
Minimum commuting distance as a spatial characteristic
99
for every zone i characterized by iO
departures and
iD arrivals:
is ii OD < or ii DO < ?
if ii OD < : if ii DO < :
internal movements ii DI = ,
0=iD , iii IOO −=
internal movements ii OI = ,
0=iO , iii IDD −=
external covered distance for departures 0=OiH ,
external covered distance for arrivals 0=DiH ,
select zone i nearest by j
look up network distance ijd in the distance matrix
calculate the internally covered distance 2
ijiIi
dI .H =
store IiH , iD and iO
select the first zone i
select zone i that is nearest by j for which:
not jj OD == 0
if 0=jD then 0≠iD
if 0=jO then 0≠iO
look up network distance ijd in the distance matrix
is 0=jD or 0=jO ?
if 0≠jD (and 0=jO ): if 0≠jO (and 0=jD ):
calculate the externally covered distances:
ijOi
Oi dHH += ,
ijDj
Dj dHH += ,
1−= ii OO ,
1−= jj DD
calculate the externally covered distances:
ijDi
Di dHH += ,
ijOj
Oj dHH += ,
1−= ii DD ,
1−= jj OO
store the calculated values
is 0== ii OD ?
if not 0== ii OD :
select the next zone i in the list if 0== ii OD :
is for all zones i : 0== ii OD ?
if not for all zones i : 0== ii OD :
select the next zone i in the list if for all zones i : 0== ii OD :
for all zones i :
total minimized outgoing covered distance Ii
Oi
Oi HHT +=
total minimized incoming covered distance Ii
Di
Di HHT +=
Fig. 3.1. Minimization algorithm
Chapter 3
100
3.4.4 Excess rate
The excess rate E is defined as the ratio between the observed travelled
distance T(obs) (in the model) and the calculated minimum commuting
distance T, and is calculated both for outbound (EOi) and inbound (ED
i)
trips of every zone:
Oi
OiO
iT
obsTE
)(= (3.1)
Di
DiD
iT
obsTE
)(= (3.2)
We calculate the excess rate E for the average departure from (EO) or the
average arrival at (ED) a typical zone i as follows:
∑∑ ⋅=
ii
i
iOi
OiO
O
O
T
obsTE
)( (3.3)
∑∑ ⋅=
ii
i
iDi
DiD
D
D
T
obsTE
)( (3.4)
While the average excess rate over the whole study area is given by:
∑
∑
∑
∑===
i
Di
i
Di
i
Oi
i
Oi
DO
T
obsT
T
obsT
EE
)()(
(3.5)
In case the data set is acquired by different sub models, as described in
the “data” section infra, a specific application of the MAUP might occur
leading to a small error so that in practice DO
EE ≠ . The reason is that
every sub model, of which the results will be combined into one entity,
provides only detailed zoning in its own focus area (in the case of our
data: a province). We overcome the aggregation error by means of the
following approximation:
∑ ∑
∑∑
+
+
=
i i
Di
Oi
i
Di
i
Oi
TT
obsTobsT
E
)()(
(3.6)
We define the observed average distance covered per trip )(obsh as
follows:
∑
∑
∑
∑===
ii
i
Di
ii
i
Oi
DO
D
obsT
O
obsT
obshobsh
)()(
)()( (3.7)
while we define the minimized average distance per trip:
∑
∑
∑
∑===
ii
i
Di
ii
i
Oi
DO
D
T
O
T
hh (3.8)
Minimum commuting distance as a spatial characteristic
101
Again, an error may occur by aggregating sub models, so that in practice
)()( obshobsh DO≠ and DO hh ≠ . In that case, we approximate )(obsh
and h as follows:
∑∑
∑ ∑
+
+
=
ii
ii
i i
Di
Oi
DO
obsTobsT
obsh
)()(
)( (3.9)
and
∑∑
∑ ∑
+
+
=
ii
ii
i i
Di
Oi
DO
TT
h (3.10)
3.4.5 Spatial distribution and density ratio of
arrivals and departures
The minimum commuting distance consists of a combination of both the
spatial separation of functions and a difference in density between typical
residential areas and typical employment centres. We will quantify these
properties by defining density ratios between departures and arrivals.
We calculate the density C of the departures O and arrivals D for the
typical zone i with area A from which an average trip departs or at which
an average trip arrives as:
∑∑
⋅=
i
ii
i
i
iO
O
O
A
OC (3.11) and ∑
∑⋅=
i
ii
i
i
iD
D
D
A
DC (3.12)
While the average density over the entire study area is given by:
∑
∑
∑
∑===
ii
ii
ii
ii
DO
A
D
A
O
CC (3.13)
Spatial separation of the considered functions can thus be measured by
calculating the ratio between CO and CD:
D
OOD
C
CR = (3.14)
Chapter 3
102
3.5 Case study area: Flanders
and Brussels (Belgium)
Within the context of this research we want to test our model for the
Flanders and Brussels region. It is important to take into account that in
this region commute areas of different cities overlap (Van Nuffel, 2007),
and that many jobs are located far outside the CBDs, such as in port
areas, small towns, historically developed businesses outside of urban
areas or peripheral industrial sites. Even though spatial dispersal is
usually larger in the residential function than in other functions, a
polycentric and partly dispersed spatial distribution of both jobs and
other destinations should be taken into account. The polycentric nature
of regional employment has a historical basis, while sprawl is mainly a
post-war phenomenon that is still developing (Vandenbulcke et al., 2009).
Riguelle et al. (2007) note that contemporary polycentric development, in
the form of sub-centres in the periphery of major cities (“edge cities”) has
hardly occurred in Belgium. Moreover, Brussels, and to a lesser extent
Antwerp and Ghent, are dominant employment centres, putting the
importance of other historical centres into perspective (Aujean et al.,
2005).
Brussels (with more than 1 million inhabitants) is the main centre of
service industries and government activities and is also the largest
employment centre in Belgium. The economy of Antwerp (about 500,000
inhabitants), the second largest city in Belgium and one of the largest
ports in Europe, is based on port activities and industries (e.g., petro-
chemicals). Ghent (about 240,000 inhabitants) is the next urban area in
the ranking, with significant activity in industry, port operations and
research and development. Major centres in the immediate sphere of
influence of Brussels and Antwerp are Mechelen, Leuven, Aalst and Sint-
Niklaas. In the east we find the double-centre of Hasselt-Genk, which
developed around the long gone mining industry, but managed to attract
new businesses. In the southwest we find the Roeselare-Kortrijk region
characterized by smaller-scale industrial activities, while Bruges (in the
northwest) is oriented towards tourism and limited port handling. The
coastal area is dominated by an elongated urban network which is mainly
based on tourism.
The framework for the description of spatial structures at the macro
scale in this region is the so-called Spatial Structure Plan for Flanders
Minimum commuting distance as a spatial characteristic
103
(RSV, 1997/2004). The RSV is the overarching spatial policy plan for the
Flanders region. Among other issues, the RSV selects and demarcates
urban areas, for which specific urban planning policies are defined.
3.5.1 Data
The methodology can be applied at different scale levels. Within the
scope of this paper we focus on the level of census wards, corresponding
with neighbourhoods. This unit division allows the development of a
detailed analysis of the theoretical minimum commuting distances and
excess traffic generation. This implies having detailed data. We use the
OD matrices of the multimodal model for Flanders (MMM) (Verhetsel,
1998). MMM is a macroscopic traffic simulation model that was commis-
sioned by the Flemish government and has been developed since 1998.
The model is essentially made up of five sub models, one for each prov-
ince of Flanders, including the Brussels Capital Region for consistency.
Every sub model consists of a GIS map that divides the province into
small traffic analysis zones (TAZs). In most places, TAZs correspond to
standardized census wards. To obtain homogeneous densities, some
repartitioning was done. In sparsely populated areas different census
wards were regrouped into one TAZ. In other places, such as the port
areas, a more refined zoning was applied. The surrounding areas (the
other sub modelled provinces, the Walloon Region, France, the Nether-
lands, Germany and Luxembourg) are also part of the GIS map, but are
divided into TAZs of which the size increases with the distance from the
study area. Properties of the used TAZs can be found in Table 3.1.
Table 3.1. Properties of the used TAZs
total mean st. deviation
number of TAZs 6,652.00 - -
area (km2) 13,751.36 2.07 2.80
number of departures
TAZ (4-11 a.m.)
2,356,461.00 354.00 406.00
number of arrivals
TAZ (4-11 a.m.)
2,356,461.00 354.00 692.00
The GIS map is linked to an OD matrix which indicates for a certain
period of time how many trips occur from every zone to any other zone in
the model. Each provincial model contains roughly between 1,400 and
Chapter 3
104
3,300 zones, and the associated matrices have as many rows as columns.
The matrices that were available for this study simulate traffic on an
average weekday between 4 a.m. and 11 a.m. (i.e., morning traffic). By
aggregating the data for a largely extended morning rush period, we
avoid inaccuracies caused by the calibration of the MMM, which is
designed to calculate traffic flows on an hourly basis.
The matrices were first built on data on home-work commuting and
home-school commuting from the General Socio-Economic Survey 2001
(SEE 2001; Verhetsel et al., 2007), that are available at the level of
census wards. SEE 2001 is an exhaustive survey of the Belgian population
(excluding children younger than six years), which has its origin in the
decennial census. The questionnaire of SEE 2001 gathers each individual’s
residence address and the address of the workplace or school. While
83.2% of the respondents provided the name of the municipality of the
workplace or school, only 56.4% filled out the full address. This is
significantly lower than the overall SEE 2001 response rate of 95%
(Verhetsel et al., 2007). The processed data were aggregated by
neighbourhood and supply a picture of the daily travelled distances to
and from each neighbourhood. This information can also be aggregated
for analysis at the municipal level. This data was geocoded; errors and
deficiencies such as the lack of addresses were corrected wherever possi-
ble. To this end, alternative socio-economic databases were used, which
were supplied by the Crossroads Bank for Enterprises1 (for home-work
travel) and the Flemish Department of Education (for home-school
travel). For other kinds of travel, grouped as recreation, shopping and
other traffic, synthetic matrices were built using a gravity model of which
the parameters were derived from the Travel Behaviour Survey in
Flanders and other relevant surveys conducted in Belgium. In this way, a
complete OD matrix for the base year 2007 was obtained.
Given the significant share of home-work and home-school travel dur-
ing the morning peak hours, one can state that the OD matrices
represent an adequate simulation of personal mobility in the morning.
Nevertheless, caution is required with respect to the interpretation of the
results for trip purposes other than work or school because for these trips
the data is of lower accuracy. To illustrate, on the basis of the datasets
used, we cannot make sufficiently accurate simulations for the traffic
1 The Crossroads Bank for Enterprises is the public data management service
of the Belgian social security system.
Minimum commuting distance as a spatial characteristic
105
outside the peak hours (which consists mainly of other than home-work
and home-school based trips), particularly during weekends or vacation
periods.
The zoning of the MMM, which is based on census wards, was done
in a more refined way for densely populated areas than for those areas
that are sparsely populated. While it is logical to define zones with a
homogeneous density, this has the disadvantage that the effect of the
specific travel behaviour of a small population in a large but sparsely
populated zone could easily result in a disproportionately prominent spot
on the map. This problem can be partly alleviated by using a predeter-
mined density threshold below which no data is represented.
Even though the MMM provides data for non-work travel too (Ham-
madou et al., 2008), within the framework of this paper, the calculation
and discussion is done for home-work travel only. This is plausible since
home-work trips are in an economic sense the most crucial of all personal
trips. This can be illustrated using price elasticities. Home-work travel in
particular is a lot less price elastic, and thus more inert, than other types
of trips, such as leisure or shopping travel (Mayeres, 2000). An additional
reason to focus on home-work travel is the fact that the availability of
data about this commuting class is generally better, and that those
datasets usually are more complete and reliable (Witlox, 2007).
3.5.2 Network
We use Streetnet as a network to connect the various zones. Streetnet is
a detailed topological representation of the Belgian road network built up
by links and nodes to which various attributes, for example, road classifi-
cation, are attached. To calculate the shortest network-based path
between each pair of centroids of the considered zones, we use the seven
highest functional road classifications from the Streetnet network data.
These categories include all regional and local roads that could be used as
a connection. The two lowest categories which cover alleys and rural
roads are not included. The public transport network is also not included,
since we assume that the search for the shortest road between two points
usually results in a shorter path than a search for the shortest link
through a line of public transport.
For travel outside Belgium, where a lower accuracy is acceptable, we
manually extend the Streetnet file to areas outside the country borders
Chapter 3
106
with a square grid with a mesh of 5 km in the immediate vicinity of the
borders, and a mesh of 10 km in the more remote areas.
3.6 Application and results of the case study
From the available MMM data we deduce a data set containing only
home-work travel. The calculation is done for each sub model separately,
for the time frame of a weekday morning. The resulting maps of the five
sub models are then combined into one map, covering Flanders and
Brussels.
The detailed zoning of the MMM allows the detection of variations
both in the observed commuting distances and the minimum commuting
distances on Flanders’ scale, but also allows us to make a more detailed
analysis and to discover relations to spatial characteristics of neighbour-
hoods.
Further, it is also possible to make a typological classification (e.g.,
industrial area, suburban allotment, business district, nineteenth-century
belt, historic town centre, ribbon development, peripheral built-up area,
et cetera), and then to seek explanations for variations of (minimum)
commuting distances within a selected typology (e.g., distance to eco-
nomic cores, distance to the main road network, supply of public
transport, et cetera). This form of analysis, however, falls outside the
scope of this paper and will be subject to further research.
3.6.1 Spatial variations of the minimum
home-work distance
Figs 3.2 and 3.3 show the calculated minimum commuting distances,
based on departures and arrivals in the morning traffic. The zones with a
density of departures or arrivals below the 10th percentile were omitted
to avoid disproportional dominance of large but sparsely populated areas
on the map.
For the entire study area the following values were found:
=)(obsh 16.2 km and =h 6.9 km
The value for )(obsh (16.2 km) differs from the reported value of 19.0
km which is given by SEE 2001 (Verhetsel et al., 2007). Hence, an
underestimation of the reported situation by our model is found, which is
caused by chained trips and detours that were not taken into account.
Minimum commuting distance as a spatial characteristic
107
3.6.1.1 Origins
Taking Fig. 3.2 as a reference, the following areas, considered as depar-
ture areas or origins, display low values (which we situate in the lowest
distance class on the map, e.g., 0 - 3.20 km).
This can be interpreted in a positive sense, since low values indicate a
high degree of spatial proximity between home and work locations, and
thus marks opportunities for short travel distances.
The metropolitan region of Brussels (1), Antwerp (2) and a wide belt
between these two cities stand out. In Antwerp, the port plays an
important role. Parts of the metropolitan region of Ghent (3) show rather
low values. However, some rather mono-functional residential areas score
relatively poorly, including some densely populated parts of the nine-
teenth-century belt. Like in Antwerp, in the northern part of Ghent the
nearby port is decisive. The highly dispersed southwestern region of
Roeselare-Kortrijk (4), the urban network of the coast (5) and the
regional urban areas and small urban areas are in general in the same
case. There is a wide variety of ranges of influence between the different
regional urban areas. A number of specific areas show low scores as well.
These are usually characterized by a rather low population density in
combination with a local concentration of employment.
Following areas, among others, display high values (which we situate
in the higher distance classes on the map, i.e., > 8.20 km). The broad
north-south oriented belt located between Ghent (3) and Brussels (1)
stands out, with the highest values being recorded in the south (6). Also
a corresponding belt located east of Brussels shows high values (7).
Finally, the rather remote parts of the east-west oriented axis of the E40
highway (8, 9) are in the same case. The high values, obtained for these
areas, can be interpreted in a negative sense, since these indicate a poor
degree of spatial proximity and actual travel distances that are necessar-
ily large.
Chapter 3
108
Fig. 3.2. Theoretical minimum commuting distance in home-work travel,
origin zones during morning traffic
Fig. 3.3. Theoretical minimum commuting distance in home-work travel,
destination zones during morning traffic
Minimum commuting distance as a spatial characteristic
109
3.6.1.2 Destinations
Also in the arrival zones (destinations), in this case to be considered as
employment centres, there are significant differences in the distance
within which employees can be found for the available jobs. From the
explained perspective, low values (situated in the lowest distance class on
the map) can be interpreted as positive. Typically residential areas, where
the supply of active labour force is high but only few jobs are available,
show low values. This is the case in most areas because jobs occur in
stronger spatial concentrations than houses do. Urban neighbourhoods
with a good balance between housing and jobs are in the same case.
Some specific areas display high values (situated in the higher dis-
tance classes on the map), which can be interpreted as negative. Port
areas and other industrial areas with a high concentration of jobs stand
out in this sense. Also parts of administrative city centres, and especially
the districts in the Brussels Capital Region, that comprise a large share of
offices, show high values.
3.6.2 Spatial variations of the excess rate
Figs 3.4 and 3.5 show the excess rate E for departure zones and arrival
zones. The same density thresholds were used as in Figs 3.2 and 3.3.
For the study area the following values were found:
=OE 16.9 , =
DE 16.0 and =E 2.33
On the one hand, the high values which we find for EO and ED indi-
cate that for a typical trip a lot of profit could be made. On the other
hand, optimization of commuting in areas with a relatively high density
would in the first instance have a negative impact on the areas with low
density, which explains the rather low figure we find for the global excess
rate E (= 2.33, or, put differently, 57% of the actual commute is exces-
sive). Despite the proportionally large profits which could be obtained in
high density areas, a minimization of the commuting distances would lead
to an overall reduction of only a factor of 0.43.
3.6.2.1 Origins
Areas with a high excess rate are typically located in urban areas or,
more specifically, near major concentrations of employment. That is
because in these regions the minimum commuting distances are very
small. At the same time the accessibility of jobs is usually higher than in
Chapter 3
110
the more remote areas, so that the physical distance criterion will be less
preponderant in job choice.
These findings are in line with what Hamilton (1982) has found,
namely, that city-dwellers go to work many times further from home than
is suggested by the spatial distribution of housing and jobs. The explana-
tion is found, on the one hand, in the theory of the constant travel time
budget and, on the other hand, in the financial travel budget being a
constant share of the household income (Schafer, 2000). Departing from
an urban area, there are a lot more jobs available within reach of the
available generalized personal travel budget than departing from a more
rural area. Viewed from the urban area, the job that yields the greatest
benefit will often not be the nearest job, compared to the viewpoint from
the countryside.
Nevertheless, net commuting distances departing from the urban ar-
eas are still shorter than average. In summary, the spatial structure of
urban areas ensures that the commuting distances are relatively low, but
that there still exists a wide margin that allows an additional reduction of
travelled distances.
Note that some more suburban regions can be found that have both a
high excess rate and long commuting distances. Those are generally areas
that are easily accessible and have high incomes so that the barrier to
travel over long distances is far lower there.
As opposed to these urban and suburban areas are the more remote
municipalities, mostly belonging to the rural areas. Most of these munici-
palities have an excess rate of around 1, or often even lower than 1.
Again, these low values could be explained by the high excess rate in the
core municipalities. Many workers who live in those core municipalities
still go to work far from home, and thus make the nearby jobs available
for residents of the surrounding municipalities. The very high excess rates
in the core municipalities are responsible for the counterintuitively low
excess rate in the surrounding municipalities.
Minimum commuting distance as a spatial characteristic
111
Fig. 3.4. Excess rate in home-work travel, origin zones
during morning traffic
Fig. 3.5. Excess rate in home-work travel, destination zones
during morning traffic
Chapter 3
112
The municipalities with a very low excess rate are the most vulner-
able to changes in external factors that steer travel behaviour. When the
generalized cost of trips would increase (e.g., by rising fuel prices, conges-
tion, road pricing or deterioration of the supply of public transport),
employees will be inclined to look for a job closer to home. At the time
that residents of core municipalities are going to work closer to home, this
would mean that the residents of the surrounding municipalities with a
low excess rate would have to go to work even further from home.
3.6.2.2 Destinations
Zones with a low excess rate are typically found in areas with high
concentrations of employment, such as the port areas and other large-
scale industrial cores and city centres. In the case of large-scale industry,
this is somewhat remarkable because in most cases the observed commut-
ing distances to such locations are already considerably higher than
average. The physical separation of functions plays here: for the indus-
tries that are established at these, often remote, locations it appears
difficult to attract employees from a small recruitment area. A similar
phenomenon as in the large industrial cores occurs in the office centres in
Brussels.
Also in the region Roeselare-Kortrijk we notice low values, which are
in this case, however, linked to short observed commuting distances. In
this region the actual travelled distances are more often than average
approaching the optimum. In the north-south oriented belt between
Ghent and Brussels, we notice again low values, but given the limited
employment there, this should not be evaluated positively.
High excess rates are typically found in both sparsely and densely
populated residential areas, where the relatively limited number of
available jobs is often occupied by employees who do not necessarily live
in the vicinity.
3.6.3 Spatial distribution and density ratio
of arrivals and departures
The density of a zone from which an average trip departs CO is 1,187
departures/km2, while for the average arrival CD is 4,756 arrivals/km2.
The ratio between the two density measures ROD = 0.25. The average
density of arrivals and departures DO
CC = = 171 trips/km2.
Minimum commuting distance as a spatial characteristic
113
The large average disparity in concentrations of jobs and houses sheds
light on the spatial background of the minimum commute distance. In the
specific case of home-work travel, this ratio is of course strongly related
to the more commonly used jobs-housing balance parameter (Peng, 1997).
3.7 Possible biases
In the context of Flanders and Brussels, the mentioned possible biases,
originating from our premises, will be more serious when we consider the
larger cities, particularly Brussels. In Belgium, highly specialized, well-
paid jobs are mostly centralized in the Brussels region, where the CBD
plays an important role. The long distance rail accessibility to this CBD
is excellent, while the geographically central location in the Brussels
agglomeration ensures the interaction with a large number of potential
employees who live in the surroundings. These factors result in specialist
workers in Brussels not living in the city. Instead, they prefer the green
suburban neighbourhoods in Brussels’ periphery, or in the less densely
built municipalities of the large commuting region. The spatial mismatch
between locations of work and residence of specialized employees is an
additional obstacle to a possible reduction of commuting distances,
occurring particularly in the large cities, and especially in Brussels. A
similar bias is found in the spatial variation of household sizes. One-
person households, which are encountered more often in cities, face more
freedom regarding the choice of job and residence location than families
with two breadwinners.
3.8 Conclusions
In this paper, we have studied the spatial variation of the minimum
commuting distance and the excess rate as indicators of spatial proximity
of functions, in particular housing and jobs. We have elaborated a
methodology to calculate these indicators, and indicated their relevance
for spatial planning policies. Methodological problems associated with the
use of those indicators still occur, especially when different regions or
cities are compared with each other. However, the spatial variation of the
minimum commuting distance seems suitable to measure the extent to
which a given area can operate on the basis of short distances. The main
Chapter 3
114
reason for these variations is the systematic differences in the spatial
distribution of the job market with respect to the housing market.
In Flanders and Brussels jobs, but also services, show a much
stronger pattern of concentration than dwellings do. Moreover, subur-
banization of both functional groups happens in a different way.
Employment is mainly situated in or in the immediate sphere of influence
of urban areas. Furthermore, extremely high local concentrations of
employment exist, such as in the Brussels office districts, the seaports or
the national airport. Employers who are located in these areas are usually
unable to recruit workers living on average close to the company.
The suburbanization of dwellings is for a major part located in mu-
nicipalities in the countryside, often far away from the economic core
areas. Although we find the highest concentrations of this residential
function in the urban areas, housing is much more homogeneously spread
over the entire study area than jobs. This means that for inhabitants of
the more remote regions with a low jobs-housing ratio it is very difficult
to find a job close to home.
The minimum commuting distance can be considered as a measure of
proximity to the labour market, viewed from the housing market, or vice
versa, as a measure of proximity to the housing market, viewed from the
labour market. The model that is discussed in this paper suggests that
the minimum average distance for a home-work trip within the current
job and housing market in Flanders and Brussels is fixed at 6.9 km, to be
compared to a calculated real world value of 16.2 km. However, in this
variable important regional gradients can be observed. Employees living
in the vicinity of the economic core areas can easily find a job close to
home, whereas the inhabitants of remote rural municipalities must
necessarily commute over long distances.
The regional variations of the excess rate show that people who live
in the vicinity of major employment centres could still significantly
reduce their daily commuting distance, relatively spoken. However, for
residents of outlying regions this would be difficult or even impossible:
they will instead be required to travel even longer distances in case a
general contraction of commuting distances would happen. Such a
reduction of commuting distances is a scenario that may occur when the
absolute cost of transportation would increase.
To date, in the Western world we have only observed a trend of in-
creasing commuting distances, an evolution originating in the
democratization of the car. Since the fastest way to make a typical
Minimum commuting distance as a spatial characteristic
115
journey to work is by car, this has led to higher speeds and ultimately to
trips covering increasingly longer distances.
The spatial segregation of functions in Flanders and Brussels mainly
developed in the era of upcoming cheap, fast transportation. We can
therefore say that the longer distances travelled partly materialized in the
form of the physical separation of functions. To a certain extent, this
spatial development shuts the door to a potential shrinkage of commuting
distances. Regions with a large minimum commuting distance are there-
fore very sensitive to rising transport costs, leading to a reduction of
mobility. In the long run, however, such an increase in costs is likely to
occur in the light of peak oil.
Although the paper only deals with home-work commuting, a similar
logic may be valid for services such as schools and shops. Spatial prox-
imity of functions is a paramount prerequisite for a sustainable travel
pattern on the basis of relatively short distances. Towards policy making,
this can be translated as the importance of providing an adequate spatial
and functional mix. Given the relatively large average daily distance that
is covered per person, the role of this spatial mix is probably more
important at a regional level than at the level of, for example, a historical
urban structure (compact city). Concretely, this could mean that the
stimulation of additional jobs and services in areas with a low jobs-
housing balance should be given priority, but also that suburbanization of
the residential function in remote rural municipalities should be discour-
aged.
Mapping the development of minimum commuting distances is an
important research field. For a better understanding of the evolution of
this indicator, it is necessary to compare longitudinal data. Further
research should also include non-work-related trips, and should take into
account other biases, such as the influence of accessibility (in terms of
travel time), the role of chained trips, household composition and income,
and modal choice.
References
Aguilera, A. (2005) “Growth in commuting distances in French polycen-
tric metropolitan areas: Paris, Lyon and Marseille.” Urban Studies.
42(9), pp. 1537–1547.
Chapter 3
116
Aujean, L., E. Castiau, M. Roelandts and C. Vandermotten (2005)
Central Belgium: Commuting & the Definition of Functional Urban
Regions. London: Insitute of Community Studies/The Young Foun-
dation & Polynet Partners.
Banister, D., S. Watson and C. Wood (1997) “Sustainable cities, trans-
port, energy and urban form.” Environment and Planning B. 24(1),
pp. 125-143.
Berkelaar, M., K. Eikland and P. Notebaert (2003) lp_solve, version 4.0
from http://groups.yahoo.com/group/lp_solve/.
Boussauw, K. and F. Witlox (2009) “Introducing a commute-energy
performance index for Flanders.” Transportation Research Part A.
43(5), pp. 580-591.
Charron, M. (2007) “From excess commuting to commuting possibilities:
More extension to the concept of excess commuting.” Environment
and Planning A. 39(5), pp. 1238-1254.
Crane, R. (1996) “The influence of uncertain job location on urban form
and the journey to work.” Journal of Urban Economics. 39(3), pp.
342-356.
Feldman, M. and R. J. McCann (2002) “Uniqueness and transport
density in Monge’s mass transportation problem.” Calculus of Varia-
tions and Partial Differential Equations. 15(1), pp. 81-113.
Frost, M., B. Linneker and N. Spence (1998) “Excess or wasteful com-
muting in a selection of British cities.” Transportation Research Part
A. 32(7), pp. 529-538.
Giuliano, G. and K. A. Small (1993) “Is the journey to work explained by
urban structure?” Urban Studies. 30(9), pp. 1485-1500.
Hamilton, B. W. (1982) “Wasteful commuting.” Journal of Political
Economy. 90(5), pp. 1035-1053.
Hammadou, H., I. Thomas, A. Verhetsel and F. Witlox (2008) “How to
incorporate the spatial dimension in destination choice models: The
case of Antwerp.” Transportation Planning and Technology. 31(2),
pp. 153-181.
Horner, M. W. and A. T. Murray (2002) “Excess commuting and the
modifiable areal unit problem.” Urban Studies. 39(1), pp. 131-139.
Horner, M. W. and M. E. O’Kelly (2007) “Is non-work travel excessive?”
Journal of Transport Geography. 15(6), pp. 411-416.
Kim, S. (1995) “Excess commuting for two-worker households in the Los
Angeles metropolitan area.” Journal of Urban Economics. 38(2), pp.
166-182.
Minimum commuting distance as a spatial characteristic
117
Ma, K. R. and D. Banister (2006) “Excess commuting: A critical review.”
Transport Reviews. 26(6), pp. 749-767.
Ma, K. R. and D. Banister (2007) “Urban spatial change and excess
commuting.” Environment and Planning A. 39(3), pp. 630-646.
Mayeres, I. (2000) “The efficiency effects of transport policies in the
presence of externalities and distortionary taxes.” Journal of Trans-
port Economics and Policy. 34(2), pp. 233-260.
Mills, E. S. (1967) “An aggregative model of resource allocation in a
metropolitan area.” American Economic Review. 57(2), pp. 197-210.
Mokhtarian, P. and X. Cao (2006) “Examining the impacts of residential
self-selection on travel behavior: A focus on methodologies.” Trans-
portation Research B. 42(3), pp. 204-228.
Muth, R. F. (1969) Cities and Housing: The Spatial Pattern of Urban
Residential Land Use. Chicago: University of Chicago Press.
Niedzielski, M. A. (2006) “A spatially disaggregated approach to com-
muting efficiency.” Urban Studies. 43(13), pp. 2485-2502.
O’Kelly, M. E. and M. A. Niedzielski (2008) “Efficient spatial interaction:
Attainable reductions in metropolitan average trip length.” Journal
of Transport Geography. 16(5), pp. 313-323.
O’Kelly, M. E. and W. Lee (2005) “Disaggregate journey-to-work data:
Implications for excess commuting and jobs-housing balance.” Envi-
ronment and Planning A. 37(12), pp. 2233-2252.
Peng, Z. R. (1997) “The jobs-housing balance and urban commuting.”
Urban Studies. 34(8), pp. 1215-1235.
Riguelle, F., I. Thomas and A. Verhetsel (2007) “Measuring urban
polycentrism: A European case study and its implications.” Journal
of Economic Geography. 7(2), pp. 193-215.
RSV (1997/2004) Ruimtelijk Structuurplan Vlaanderen - Gecoördineerde
Versie. Brussels: Ministry of the Flemish Community.
Schafer, A. (2000) “Regularities in travel demand: An international
perspective.” Journal of Transportation and Statistics. 3(3), pp. 1-31.
Scott, D. M., P. S. Kanaroglou and W. P. Anderson (1997) “Impacts of
commuting efficiency on congestion and emissions: Case of the Hamil-
ton CMA, Canada.” Transportation Research Part D. 2(4), pp. 245-
257.
Small, K. A. and S. Song (1992) “Wasteful commuting: A resolution.”
Journal of Political Economy. 100(4), pp. 888-898.
Chapter 3
118
Song, S. (1995) “Does generalizing density functions better explain urban
commuting? Some evidence from the Los Angeles region.” Applied
Economics Letters. 2(5), pp. 148-150.
Van Acker, V., B. van Wee and F. Witlox (2010) “When transport
geography meets social psychology: Toward a conceptual model of
travel behaviour.” Transport Reviews. 30(2), pp. 219-240.
Van Acker, V., F. Witlox and B. van Wee (2007) “The effects of the land
use system on travel behavior: A structural equation modeling ap-
proach.” Transportation Planning and Technology. 30(4), pp. 331-
353.
Van Nuffel, N. (2007) “Determination of the number of significant flows
in origin-destination specific analysis: The case of commuting in
Flanders.” Regional Studies. 41(4), pp. 509-524.
Van Ommeren, J. (2000) “Job and residential search behaviour of two-
earner households.” Papers in Regional Science. 79(4), pp. 375-391.
Vandenbulcke, G., T. Steenberghen and I. Thomas (2009) “Mapping
accessibility in Belgium: A tool for land-use and transport planning?”
Journal of Transport Geography. 17(1), pp. 39-53.
Verhetsel, A. (1998) “The impact of spatial versus economic measures in
an urban transportation plan.” Computers, Environment and Urban
Systems. 22(6), pp. 541-555.
Verhetsel, A., I. Thomas, E. Van Hecke and M. Beelen (2007) Pendel in
België. Deel I: De Woon-Werkverplaatsingen. Brussels: Statistics Bel-
gium.
White, M. J. (1988) “Urban commuting journeys are not ‘wasteful’.”
Journal of Political Economy. 96(5), pp. 1097-1110.
Witlox, F. (2007) “Evaluating the reliability of reported distance data in
urban travel behaviour analysis.” Journal of Transport Geography.
15(3), pp. 172-183.
Yang, J. and J. Ferreira (2008) “Choices versus choice sets: A commuting
spectrum method for representing job-housing possibilities.” Envi-
ronment and Planning B. 35(2), pp. 364-378.
119
Chapter 4:
Measuring spatial
separation processes
through the minimum
commute
This paper is published as Boussauw, K., B. Derudder and F. Witlox
(2011) “Measuring spatial separation processes through the minimum
commute: The case of Flanders.” European Journal of Transport and
Infrastructure Research 11(1), pp. 42-60. Copyright © The authors. All
rights reserved.
Abstract
The average distance covered by individual commuting trips increases
year after year, regardless of the travel mode. The causes of this phe-
nomenon are diverse. Although increasing prosperity is often invoked as
the main reason, the discipline of spatial planning also points to the
relevance of land-use policies that enable processes of suburbanization
and sprawl. By calculating time series of spatially disaggregated theoreti-
cal minimum commuting distances, this paper offers a method to identify
and quantify the process of spatial separation between the housing
market and the job market. We identify the detected spatial separation
as one of the possible indicators for the contribution of spatial processes
to the growth of traffic.
In the case study area of Flanders and Brussels (Belgium), it is found
that over time the minimum commuting distance increased in many
municipalities, especially where population is growing faster than job
supply, or where traditionally high concentrations of employment still
increase. Decreases are noticed in suburban areas that are getting a more
urban character by acquiring a considerable functional mix. For the study
Chapter 4
120
area in its entirety, we do indeed register an increasing spatial separation
between home and work locations. However, this separation evolves less
rapidly than the increase in commuting distances itself.
Regarding the methodology, we find that the use of municipalities as
a spatial entity is suitable for grasping regional transformations of the
economy and intermunicipal forms of suburbanization and peri-
urbanization. However, a similar methodology, applied at a more detailed
geographical scale, could be used to detect processes of sprawl in the
morphological sense.
Keywords: excess commuting; Flanders; sustainable spatial develop-
ment; urban sprawl
4.1 Introduction
Over the last century, there has been a mutually reinforcing relation
between rising prosperity and a general increase in individual mobility:
the growth of traffic volumes are both an effect and a cause of mounting
affluence. For instance, faster transport modes (especially but not exclu-
sively, the car) have consistently gained ground because the individual
budget spent on transport has in absolute terms continuously been
growing. The net result is that the average distance covered per person
has been systematically increasing (Bleijenberg, 2003).
With regards to the commute, this means that, in general, workers
have been looking for daily occupations increasingly further away from
their home, or - conversely - that they have been moving house further
away from their jobs. The logical result is a stronger separation between
house and job location: if travel costs decrease, then travel consumption
increases (Rietveld and Vickerman, 2003). This is especially true when
the spent amount of time can be kept constant by increasing average
travel speed, which seems generally to be the case in the Western world
(Schafer, 2000).
Interestingly, this evolution apparently has an important spatial
component. That is, it looks like changing travel behaviour is partly
materialized in suburban and peri-urban developments, implying that
possible origins and destinations lost relative proximity to each other. In
many regions in the developed world, this materialization may be respon-
sible for a certain degree of irreversibility of the expansion of travel
Measuring spatial separation processes through the minimum commute
121
patterns, as this limits the possibility of travel distances being reduced
again in the future, even if environmental or congestion policies would
aim for this. Since the adverse external effects of traffic are linked to the
distance travelled (especially when travel is by car), this evolution makes
it harder to address traffic problems at the source. In addition, spatial
separation of functions is probably a long term economic disadvantage,
since it makes the economic system vulnerable to possible future circum-
stances where transport costs would increase, for example by rising oil
prices (peak oil) (Boussauw and Witlox, 2009) or growing structural
congestion.
The perspective of this paper is the study of the mobility component
of spatial separation of the job and housing market. Our research ques-
tion is: how can spatial separation between residential locations and job
locations be measured by means of a spatial proximity indicator, and
what is the connection with the growth of observed commuting distances?
We hereby hypothesize that the well-established line of inquiry on excess
commuting (Ma and Banister, 2006) can provide a methodology to
quantify evolutions of spatial separation in a functional sense, and to
provide better insight in the contribution of spatial structure to the
growth of traffic volumes over the last decades.
In practice, this will be done through calculating spatially disaggre-
gated evolutions in minimum commuting distances, defined as the
theoretical minimum distance that each worker would have to cover in
order to find a job as close to home as possible under the assumption that
actual residential locations and job locations are maintained and the total
distance travelled (by all workers together) is minimized. After applying
the method to a case study area in northern Belgium, we evaluate the
observed evolution based on both occurring sprawl and regional-economic
shifts in the labour market, aiming to explain regional variations.
Although assessing time series of the minimum commuting distance in
order to describe spatial separation between the housing market and the
labour market has been done before (Frost et al., 1998; Horner, 2007;
Yang, 2008), the approach adopted in his paper is novel for two reasons.
First, we adopt a study area which could be qualified as a suburbanized
historically polycentric spatial structure - as there are many in Europe -
and shows therefore significant spatial variations (i.e. it includes many
historical urban areas, suburban and peri-urban developments, and major
industrial zones), although it can be demarcated in a fairly consistent
way based on economic and political criteria. Second, we refine the
Chapter 4
122
method in order to detect variations in the minimum commuting distance
in a spatially disaggregated way, which certainly contributes to a critical
understanding of the role of the present variations of spatial structure in
comparison with Frost et al. (1998), Horner (2007) or Yang (2008) who
define only one figure for the entire city or region.
The remainder of this paper is structured as follows. First, we clarify
the possible reasons behind evolutions in spatial proximity by putting
this in the perspective of the existing literature. Second, we present an
overview of the case study area (Flanders and Brussels), including the
main socio-economic and spatial processes and changes that may play a
role in our analysis. Then we develop our methodology, including calcula-
tion and evaluation methods, and discuss the available data. After
applying the method, we present the results, discuss possible biases and
draw some final conclusions.
4.2 Defining spatial separation processes
One can think of several possible reasons leading to changes in spatial
proximity between homes and jobs. For planning policies, the most
relevant possible causes are suburbanization and sprawl. Ewing (1994)
assumes a strong link between sprawl and increased traffic. But also
thorough zoning policies may lead to increasing mutual distances. These
phenomena are local in nature and manifest themselves at a small
geographical scale, meaning that in quantitative analysis, these will
particularly manifest when data are available at a detailed spatial
division (e.g. traffic analysis zones).
Newman and Kenworthy (1989) and Gordon and Richardson (1989)
associate sprawl with a specific morphology, particularly consisting of
monotonous suburban districts with a strict separation of functions,
characterized by store strips, commercial architecture and large internal
distances. However, the extent to which spatial separation leads to an
effective increase of distances that need to be covered is less clear, since
this cannot be derived from local morphological characteristics. For
instance, a monotonous residential lot embedded in a major employment
centre could possibly lead to a more sustainable travel pattern than a
compact town that is immersed in a rural area.
Therefore it is important to view spatial separation not only in a
morphological way but also in a functional sense. Particularly in a
Measuring spatial separation processes through the minimum commute
123
context where average trip lengths, and in particular average commuting
distances, have become very large in practice, it is hard to tell which kind
of spatial developments are problematic in relation to mobility, and
which are rather beneficial. Banister (1999) therefore observes the issue at
hand in a non-morphological way, and suggests that it is of the utmost
importance that new developments are sufficiently large and are located
in or immediately subsequent to existing urban areas. As a consequence,
local morphology, density and spatial diversity come only in second place.
Another phenomenon that could cause this kind of functional separa-
tion processes, although particularly at a larger geographical scale level
and consequently only partly related to suburbanization processes,
consists of regional economic shifts within the labour market. An eco-
nomic transformation, e.g. towards a more service-based industry, could
lead to a different spatial distribution of jobs, e.g. through centralization.
But also an absolute increase in the number of jobs in one zone can lead
to a spatial distribution where work locations are on average closer to
residences of potential employees. Since these kinds of regional transfor-
mations take place at a higher geographical scale than suburbanization
processes, these will rather become clear when relatively large zones (e.g.
districts or municipalities) are studied.
If we restrict ourselves to commuter traffic, we see that trip lengths
have increased systematically over the past decades. This trend was
observed in Belgium (see below), the US and the UK, and we may
assume that this evolution is manifest throughout the Western world
(Aguilera, 2005). The basic mechanism that underlies this growth is an
increase in travel speed. Basically, people do not spend more time
commuting than they did before, nor do they spend a greater proportion
of their income on transport (Schafer, 2000). Rather, it is the general
prosperity growth that has led to an absolute increase in resources
devoted to transport, resulting in more car ownership, more car use (at
the expense of slower transport modes), an extension of the motorway
network and a pushed up speed of public transport. In the surroundings
of large agglomerations, congestion has slightly slowed down the overall
increase of travel speed, although it did certainly not stop it (Van Wee et
al., 2002).
Even though spatial separation processes and increasing trip lengths
appear to be associated, this does not necessarily imply a one-way
causality. The fact that employees live on average further from their work
location than before, a phenomenon which occurs partly in the form of
Chapter 4
124
sprawl, is caused by increasing individual mobility, i.e. a wealth-related
phenomenon that is the basis of an autonomous growth of traffic. So, we
might explain these spatial separation processes as a materialization of
this increased mobility, which has itself a mutual reinforcing effect on the
growth of traffic.
4.3 Measuring spatial separation by
excess commuting characteristics
The perceived increase in traffic volume due to spatial separation proc-
esses, whether or not perceived as sprawl, remains a main concern in
most spatial policy plans (Sultana and Weber, 2007). Given the impor-
tance attached to mobility, it may be surprising that quantification of
this phenomenon is usually confined to measuring morphological charac-
teristics. However, in this paper we want to measure the process of
spatial dispersion that results in further separating origins and destina-
tions over the years, or, as might be the case in some areas, decreasing
separation between origins and destinations. By measuring this develop-
ment over a certain time interval, we have a good idea what proportion
of the traffic growth in that period is due to spatial expansion. An
important methodological issue concerns the definition of these origins
and destinations. This obstacle disappears when restricting ourselves to
the commute, where origins (residential locations of workers) and destina-
tions (job locations) can be clearly defined.
Cross-sectional analyses that compare travel behaviour of commuting
residents or workers in suburban areas with commuting behaviour in
areas with urban characteristics are common in the literature. Sultana
and Weber (2007), for instance, recorded large differences in actual travel
times and trip lengths, depending on the direction of the commuter flow.
Flows within urban areas appear to be the shortest, both in distance and
in time. Commuting trips within the suburban areas, however, appear
shorter than commuting trips from the suburbs towards the city.
In order to map the spatial separation between the housing market
and the labour market, regardless of actual commuting patterns, other
methods are needed. An elementary indicator of the spatial distribution
of housing and jobs with an alleged impact on mobility is the jobs-
housing balance (the ratio between the number of jobs in an area and the
number of workers living in the same zone). In its simplest form, this
Measuring spatial separation processes through the minimum commute
125
indicator is measured by zone, making it sensitive to variations in zoning
and insensitive to effects of embeddedness in the surrounding region or
the presence or absence of transport infrastructure. Peng (1997) tried to
overcome these problems partly by calculating the jobs-housing balance
based on the surrounding area of which every considered zone is the
centre.
A more advanced line of inquiry is the field of “excess commuting”,
which originally focuses on the study of spatial structure in relation to
commuting behaviour by comparing observed commuting distances with
theoretical calculated minimum commuting distances. An overview of
existing research on this topic is given by Ma and Banister (2006). One of
the spatial characteristics, provided by the excess commuting literature,
is the minimum commuting distance, which incorporates both the spatial
distribution of jobs and housing, and the infrastructure network. It is
related to the job-housing balance, but is a more sophisticated indicator
as it also takes into account the embeddedness of the study area in the
wider region, as well as the transport network (Horner and Murray,
2003). We consider the minimum commuting distance as a measure of the
proximity of the housing market to the labour market (Horner, 2004), or
as a variable that indicates to what extent the system could absorb
shrinkage of commuting distances. The latter is relevant when we want to
understand the potential impact of the commute getting more expensive,
e.g. under the influence of rising fuel prices (Boussauw and Witlox, 2009).
The calculation of the minimum commuting distance implies that the
existing residential and job locations are retained, but that workers are
assigned to jobs in a way that leads to a minimization of the total
distance travelled. More formally:
minimize ∑∑= =
=
n
i
n
jijijtdH
1 1
(4.1)
given: j
n
iij Dt =∑
=1
(4.2), i
n
jij Ot =∑
=1
(4.3), and 0≥ijt (4.4)
in which: H = total distance travelled within the system by
matching workers and jobs
n = number of zones
Oi = number of workers living in zone i
Dj = number of jobs in zone j
dij = network distance between centroids of zone i and zone j
tij = number of trips between zone i and zone j
Chapter 4
126
Ma and Banister (2007) published an in-depth study on the relationship
between excess commuting and urban form. Horner (2007) and Yang
(2008) used also the maximum commuting distance as a measure of
spatial dispersion, while Charron (2007) and Yang (2008) introduced the
terms “proportionally matched commute” and “random average com-
mute”, based on a more extensive theoretical exploration on possible
commuting ranges, and proposed as more realistic alternatives to the
maximum commuting distance.
The maximum commuting distance should be seen as the conceptual
inverse of the minimum commuting distance: it is the value obtained
when all employees within the study area simultaneously would exchange
jobs or houses aiming to live as far as possible from their jobs. In an
urban system consisting of a single city that is isolated from neighbouring
cities, this measure is a good complement to the minimum commuting
distance. In the context of our assessment, however, the notion of maxi-
mum commuting distance is rather abstract. Not only will employees
never be inclined to spontaneously maximize home to work trip length,
also outcomes seem to be largely dependent on the considered region
frontier and are highly correlated with the size of the study area
(Charron, 2007). Moreover, the amount of traffic that is associated with a
maximized commuting distance is not compatible with the existing
transport infrastructure or any form of pursuing economic efficiency. In
this paper, we will calculate the global value of the maximum commuting
distance as an additional support of the observed trends in minimum
commuting distance, but we will not elaborate on spatially disaggregated
values.
The contribution of a time series approach of the proportionally
matched commute is mainly in quantifying the extent to which a regional
system evolves from mono-centric to dispersed (Yang, 2008). Since the
possible interpretation of the meaning of spatially disaggregated values of
the proportionally matched commute in a polycentric region is subject to
further research that is beyond the scope of this paper, we will no further
elaborate on this.
In the literature, the average minimum commuting distance is usually
calculated as a single measure for an entire city or region (Ma and
Banister, 2006). However, it is also possible to calculate this in a spatially
disaggregated way in order to reveal regional variations. In the latter
case, for each zone (traffic analysis zone, municipality, district ...) the
distance outcome is calculated twice: once for the outgoing commute, and
Measuring spatial separation processes through the minimum commute
127
once for the incoming commute. Niedzielski (2006) elaborated on a
spatially disaggregated analysis of Warsaw on the basis of minimum
commuting distances. When taken the locations of residence as a view-
point, the largest minimum commuting distances were recorded in areas
with suburban characteristics. Viewed from the perspective of job loca-
tions, the highest values are recorded downtown (CBD). Yang and
Ferreira (2009) included spatially disaggregated values in their informa-
tion system for planners in the Boston area. A similar analysis, albeit on
a regional scale, was made for Flanders and Brussels (Belgium) by
Boussauw et al. (2011). A spatially disaggregated calculation of the
minimum commuting distance on cross-sectional data allows obtaining
these values per zone for a given point in time. In a next step, a similar
analysis based on time series data would then reveal trends, indicating for
a certain zone if the housing market moved on average further away from
the job market over the chosen time span, or if, maybe, the opposite has
happened.
Horner and Murray (2002) devoted considerable attention to the im-
pact of the underlying zoning system on the calculated minimum
commuting distance, both with regard to size and to location. This issue
is known as the modifiable areal unit problem (MAUP). The use of larger
zones means that more trips become intrazonal, leading to an overestima-
tion of the minimum commuting distance (Horner and Murray, 2002).
Both interzonal and intrazonal trips originating in large zones can only be
simulated with limited precision, leading to significant inaccuracies.
However, when excess commuting is studied, this problem can partly be
accommodated by recalculating observed commuting distances using the
same origin-destination matrix that is used to calculate the minimum
commuting distance.
For our research, however, the main problem associated with the se-
lection of zone sizes is the geographical scale at which the studied
developments occur. This means that the spatial separation between jobs
and housing can both be caused by sprawl (at a small geographical scale)
and by regional economic shifts within the labour market (at a large
geographical scale). The first phenomenon will particularly manifest when
detailed zoning (e.g. traffic analysis zones) is used, while the second
phenomenon will become clear when relatively large zones (e.g. districts)
are studied.
While a considerable volume of research on excess commuting has
been published yet, the literature is still fast evolving. Recent inquiries
Chapter 4
128
include attainability of trip length reductions (O’Kelly and Niedzielski,
2008), differentiation in excess rate depending on mode choice (Murphy,
2009), the relationship with the jobs-housing balance as a suitable
indicator for measuring spatial expansion (Layman and Horner, 2010),
the influence of uncertainties in travel time measurements when minimiz-
ing time distance (Horner, 2010) and optimization of spatially
disaggregated calculation methods (Boussauw et al., 2011).
Based on the properties of the excess commuting framework, we find
a time series approach of the minimum commute useful to study proc-
esses of suburbanization. Existing research in this field is rare, however.
Frost et al. (1998) described the evolution of the minimum commuting
distance for a set of demarcated British cities. They compared the 1981
situation with 1991. Yang (2008) reported on similar research in the
metropolitan areas of Boston and Atlanta, based on data from 1980, 1990
and 2000. Both studies find an increase of both minimum commuting
distances and actual commuting distances, in all involved cities. Horner
(2007) compared commuting data from Tallahassee, Florida, from 1990
and 2000. The average reported commuting distance and maximum
commuting distance increased during this period, while the minimum
commuting distance showed only a statistically non-significant increase.
However, neither Frost et al. (1998) nor Horner (2007) or Yang (2008)
applied a spatially disaggregated method, so it is not clear which parts of
the urban areas are the most affected by spatial separation. Furthermore,
no typical peri-urban areas, which are relatively far from the CBD, were
incorporated in the analyses. As stated in the introduction, our approach
consists of an extension to the empirical diversity of the three aforemen-
tioned authors by including suburban and rural areas and an extension of
their methodology by applying a spatially disaggregated method.
4.4 Spatial development and commuting
in Flanders and Brussels
The case study area consists of the administrative regions of Flanders and
Brussels, which together form the northern part of Belgium (Fig. 4.1).
Historically, the area is built around three major cities (Brussels, Ant-
werp and Ghent), a dozen regional cities and several dozen smaller towns
and central municipalities. The economic core is borne by the wide
surroundings of the axis Brussels-Antwerp, a region known as the Flemish
Measuring spatial separation processes through the minimum commute
129
Diamond. The rapid development of an extensive railway network in the
nineteenth century and the introduction of cheap season tickets formed
the backbone of a policy aimed at an industrialization of the country
based on limited urbanization (Verhetsel et al., 2010). The new working
class was able to continue living outside major cities, in a house with a
garden and under the watchful eye of the village priest, while commuting
every day to the factory or the office. This form of institutionalized
commute laid the foundation for the spatial separation between housing
and work locations that has only developed more distinctly since the
advent of the motorway and general car ownership. In the period before
World War II, this phenomenon has materialized in the form of spatial
developments that were strongly clustered around the railway stations. In
the post-war period, however, this structure fanned out into a network of
suburbanized and peri-urbanized areas around the ancient settlements.
Fig. 4.1. Situation of infrastructure and urban areas
In this context, suburbanization refers to sprawl that is occurring in or
adjacent to urban areas, while peri-urbanization consists of developments
in the countryside, free-standing or in line with existing villages. These
postwar developments gradually led to the emergence of a variety of
forms of urbanization, ranging from historical densely populated cities
and sparsely populated suburban residential belts to commuter areas and
Chapter 4
130
pure countryside (Vanneste et al., 2008). Depending on the definition,
57% (Vanneste et al., 2008) to 97% (Antrop, 2004) of the Belgian popula-
tion lives in urbanized areas.
The problem of spatial separation, supposedly mainly in the form of
sprawl, is not a new issue in Flanders. In the period between 1980 and
2000 planning became a major political topic, as open space became
scarce, and urban flight appeared to feed a wave of suburban and peri-
urban development. Newly emerging issues in the field of landscape
ecology, water pollution, increasing distribution costs, road safety,
congestion and social segregation were partly attributed to sprawl. The
political response, in the form of the Spatial Structure Plan for Flanders
(RSV, 1997/2004) had to wait until 1997, but did offer an answer in the
form of the demarcation of urban areas and the focus on encouraging
additional building in the cities and existing settlements.
The policy measures that are imposed by this plan are based on a
rather intuitive analysis of the problem, with often no thorough scientific
rationale behind it. Since then, several quantitative spatial analyses have
been carried out. The main research report on sprawl, which was issued
by the Flemish Environmental Agency (Gulinck et al., 2007), approaches
the phenomenon mainly from a landscape-ecological perspective. It
focuses on morphological changes that are related to spatial fragmenta-
tion. In the process, sprawl detection is based on maps and satellite
images from different periods. The main indicator is the overall total
built-up area, which for Flanders increased over the period 1990-2000
from 13.2% to 14.6%. Other indicators were the length of the road
network (increasing by 6.4% between 1991 and 2001), morphological
grain size and global proximity to buildings and infrastructure. A time
series analysis of these indicators points to a systematic increase in
sprawl.
The successive censuses show that commuting distances in Belgium
systematically increase. Based on an assessment of the respondents, the
perceived average distance between home and workplace evolved from
11.9 kilometres in 1970 (Mérenne-Schoumaker et al., 1999) to 19.0
kilometres in 2001 (Verhetsel et al., 2007), an increase of no less than
60%. Although non-commuting trips and freight transport grew even
faster over this period, it is clear that the increase in trip length is largely
responsible for the overall growth in traffic.
Measuring spatial separation processes through the minimum commute
131
4.5 Data
The decennial census, organized by the Belgian federal government,
assesses the municipality of residence and the work municipality of all
working citizens. Furthermore, the (perceived) distance travelled every
day to work is also registered. Commuting data for 1981, 1991 and 2001
is available in the form of origin-destination matrices, with the municipal-
ity as a spatial unit. The small municipalities of the Brussels Capital
Region were combined into one zone, in order to avoid biases in relation
to the two other metropolitan municipalities of Antwerp and Ghent. As a
result, the study area is divided into 309 zones with a mean area of 44
km2 (standard deviation: 29 km2).
For the particular purposes of our research, the matrices were cleaned
by removing all trips that have their origin or destination outside the
study area (Wallonia and the neighbouring countries). Apart from these
inter-regional commuters, also home workers (including teleworkers) were
removed from the matrices, as well as respondents with an itinerant
occupation. The exclusion of these records is justified because we only
want to measure the extent to which residential and labour structure
separate within Flanders-Brussels.
Although the method of data collection in the three survey moments
was performed in a similar manner, there are rather large differences
between the three data sets. Important variations in the number of
commuting trips are found, as well as in the number of unknown or
incomplete registrations. Moreover, the number of home workers declined
dramatically over the three consecutive survey moments, and the struc-
ture of the inter-regional commute (between Flanders-Brussels and the
neighbouring region and countries) changed. Possibly more important is
the influence of the regional economic transformations that occurred. In
the period 1981-2001 the economy shifted towards a more service-based
system, and lost a number of important industrial employment centres
such as the Kempen coal mines in the east of Flanders and the Hainaut
steel industry south of the study area. Also, the port of Antwerp and the
airport of Brussels went through an era of major growth, and a high
standard service industry (especially in information technology and
international public institutions) developed in the urban areas, in particu-
lar in the Brussels agglomeration. This restructuring of the labour market
cannot be separated from the suburbanization of the residential structure:
Chapter 4
132
both transformations add to the spatial separation that we want to
assess.
Regarding the development of sprawl in the studied period, also the
growth of the infrastructure network is undoubtedly important. Although
most motorways that exist today where constructed in the period prior to
1981 (roughly between 1961 and 1981, see Fig. 4.1), it can be expected
that many developments along the motorways have been built in the
period 1981-2001.
4.6 Method
To answer the research question, we investigate the evolution of the
spatial distribution of residential locations of workers and job locations in
Flanders and Brussels (Belgium), mainly relying on the minimum com-
muting distance concept, and comparing the trend found with the
evolution of observed commuting distances. To this end we first take the
study area as a whole, then we reiterate the calculation in a spatially
disaggregated way (for each municipality separately). Regarding indica-
tors for the study area as a whole, we will also provide the global
maximum commuting distance.
As we consider the minimum commuting distance as a measure of
spatial proximity between job market and housing market, the possible
mismatch between qualifications and job preferences of workers and
requirements of employers is not taken into account, which is an impor-
tant deviation from reality. Although in the literature attempts were
made to disaggregate excess commuting characteristics based on a
classification of workers and jobs (O’Kelly and Lee, 2005), this approach
falls outside the scope of our paper. Nevertheless it is important to keep
this constraint in mind when interpreting the results. O’Kelly and Lee
(2005) found e.g. that service workers are subject to shorter minimum
commuting distances than industrial workers, but that actual commuting
behaviour of the former group gives rise to larger excess rates than the
latter. However, it is unclear whether these findings also apply outside
the surveyed American cities, e.g. in a Belgian context. Also, the unem-
ployed population and home workers do not affect the results since they
are excluded from the dataset, which is important to keep in mind when
interpreting the results.
Measuring spatial separation processes through the minimum commute
133
An increase of the minimum commuting distance towards or from a
certain area is usually indicative of the spatial expansion of the residen-
tial market or a trend of concentration in the labour market, while a
decline usually will point to a greater functional mix. The underlying
causes may be both of spatial development nature (sprawl), and of
regional-economic nature (e.g. the economic shift towards service indus-
tries, increased tertiarisation).
According to our hypothesis, the rate at which the minimum com-
muting distance increases is an indicator for the impact of spatial
transformations on the actual average commuting distance. The difference
between the growth rate of the minimum commuting distance and the
growth rate of the actual commuting distance is a measure of the non-
spatial component in the overall growth of the commute. Consequently,
an increase in commuting distances that would develop faster than the
increase of the minimum commuting distances points to a mobility
growth that is relatively independent of changes in spatial structure. In
the opposite case, we would be dealing with a spatial evolution that is in
and by itself responsible for the entire growth in commuter traffic
volume.
To calculate the minimum commuting distance, most authors make
use of one of the minimization algorithms for the so-called “transporta-
tion problem”, as available in various software packages. However, we
applied an alternative method that has been used before for a cross-
sectional analysis of the same study area, and is explained in Boussauw et
al. (2011). This method is conceived as an iterative process that simulates
the behaviour of commuters who simultaneously look for a job closer to
home. The disadvantage of this method is that the achieved “optimum”
remains slightly higher than the mathematical minimum. The advantage
of the method is that the produced origin-destination matrix holds a
much greater realism, since it is not influenced by algebraic tricks that
are applied to find the mathematical minimum but do not make sense in
a real world approach. An example is the allocation of all resident
workers from one zone to jobs in only one corresponding zone. Moreover,
the structure of origin-destination matrices that are produced by the
iterative process on the basis of various data sets (representing different
points in time) is similar, meaning that comparing these matrices using
statistical methods (such as Student’s t-test) makes sense. This is not
necessarily true when applying a standard “transportation problem”
Chapter 4
134
algorithm, where the structure of the resulting origin-destination matrix
depends on the used software package.
Confronting the minimum commuting distance for an area with the
observed (calculated) commuting distance is known as the study of excess
commuting. In practice we implement this confrontation by calculating
the excess rate, defined as the quotient of the mean observed commuting
distance and the mean minimized commuting distance.
The mean observed distance used in the analysis is actually a calcu-
lated (thus not perceived by the respondent) distance. For each
municipality, the centre of gravity (centroid) is determined. Then a
shortest-path matrix is calculated by means of a skeleton file of the
Belgian road network, making the shortest network distance between
each possible pair of municipalities available. Intra-municipal distances
are simulated by taking for every municipality half of the network
distance to the centroid of the nearest municipality. Given that the road
network in 1981 was already very dense, we use the same skeleton file for
the three time periods.
4.7 Results
4.7.1 Evolution of the mean observed
commuting distance
We calculate for each municipality the mean trip length for both outgo-
ing (outbound mean calculated distance: omcd) and incoming (inbound
mean calculated distance: imcd) commuting trips. A weighted average
can be found in Table 4.1, along with the commuting distance as per-
ceived by the respondents (mean perceived distance: mpd).
Table 4.1. Observed commuting trips and commuting distances
1981 1991 2001
number of commuting trips
(matrix)
1,950,477 2,331,090 1,907,197
mcd (matrix) 12.2 kma 12.6 kmb 14.7 kmb
mpd (Belgium) 14.6 kma 17.2 kmb 19.0 kmb a source: Mérenne-Schoumaker et al. (1999), p. 80 b source: Verhetsel et al. (2007), p. 60
Measuring spatial separation processes through the minimum commute
135
The important differences between the calculated (mcd) and the observed
mean commuting distance (mpd) can be explained as follows:
• mcd is calculated based on the shortest network distance while the
shortest route is usually not the fastest, and thus not the path that is
chosen by the commuter;
• mcd is calculated from centroids of municipalities, which is an
important simplification, particularly with regard to intra-municipal
trips;
• mcd ignores interregional, usually relatively long, commuting trips;
• mpd includes commuting trips in the south of Belgium, which are
longer on average.
Other possible biases in perceived trip length are discussed by Witlox
(2007).
Taken together, the mcd variable should therefore be regarded as a
theoretical measure, which only makes sense when compared to similar
calculations, such as the minimum commuting distance. For comparison
with survey data from other countries, mpd will naturally be better
suited. More important than the absolute figure is the trend that both
quantities exhibit.
Fig. 4.2. Evolution of the commuting distance (1981-2001) based on
municipality of residence
Chapter 4
136
Fig. 4.3. Evolution of the commuting distance (1981-2001) based on
municipality of workplace
A closer look reveals that over the period 1981-2001 mcd increases in 280
of the 309 municipalities when considering the outbound commute (Fig.
4.2), and increases in 291 of the 309 municipalities when looking at the
incoming commute (Fig. 4.3). Although municipalities with a relative
growth show a certain spatial clustering, the virtually general upward
trend suggests that at least some of the causes of the growth in trip
length are of non-spatial nature.
4.7.2 Evolution of the mean minimum
commuting distance
Given the significant variation in the number of recorded trips between
the three snapshots (1981, 1991 and 2001), we first examine whether this
sample variation affects the calculated mean minimum commuting
distance (mmid). We did this by weighing commuting trips for the year
1981 and 1991 so that the three tables on which the minimization
procedure was then performed all contain 1,907,197 trips (equalling the
number of trips in the origin-destination matrix for 2001, see Table 4.1).
Measuring spatial separation processes through the minimum commute
137
The obtained results are thereupon compared with the results that are
based on the unweighted tables. No significant differences were found,
and Pearson’s correlation coefficient between the two sets of spatially
disaggregated results is 1.00. Next, the global mmid for the entire study
area is calculated for the three different points in time. The obtained
figures are presented in Table 4.2, along with the evolution of mcd, and
the maximum commuting distance (mmad).
Table 4.2. Evolution of the global minimum commuting
distance and excess rate
1981 1991 2001
mmid 9.0 km 8.9 km 9.4 km
evolution mmid (base 1981) -1.4% +4.3%
mcd 12.2 km 12.6 km 14.7 km
evolution mcd (base 1981) +3.4% +20.1%
excess rate (mcd/mmid) 1.36 1.43 1.57
evolution excess rate
(base 1981)
+4.8% +15.1%
mmad 19.5 km 18,6 km 21.6 km
evolution mmad (base 1981) -4.5% +10.8%
A salient element in Table 4.2 is the negative evolution of mmid over the
period 1981-1991 (-1.4%), which is associated with an increase of mcd
(+3.4%). So, the studied residential locations and job locations would
have come 1.4% closer together over this period, while commuting
distances continuously increased. Nevertheless, over the entire period
under investigation (1981-2001) we find, as expected, an increase in
mmid.
The second striking result is the growth rate of mmid: this is much
lower (+4.3%) than the growth rate of mcd (+20.1%). Logically, this
trend is in line with an increase of the excess rate. This result suggests
that over the studied period the average worker became less inclined to
seek a job close to home, or to look for a home close to work. This also
means that only a small part of the increase in commuter traffic can be
attributed to the expansion of the spatial structure.
The evolution of mmad confirms the observed trend of mmid, al-
though the differences are more pronounced. For a theoretical approach
to the possible interpretation of the differences between mmid and mmad,
Chapter 4
138
which is beyond the scope of this paper, we refer to Charron (2007) and
Yang (2008).
Note that in our literature overview only three papers (i.e. Frost et
al., 1998, Horner, 2007 and Yang, 2008) studied time series of minimum
commuting distance. The results of Horner (2007) and Yang (2008) are
analogous with our findings: in Tallahassee, in Boston and in Atlanta
both minimum commuting distances and observed commuting distances
grew, while the former increased more slowly than the latter. Frost et al.
(1998), however, found for all analysed British cities that the growth of
the minimum commuting distance was faster than the growth of the
observed commuting distance, suggesting that changes in the spatial
structure could be held fully responsible for the increase in commuter
traffic volume, along with an improved efficiency in the commute itself
(given the reduction of the excess rate). In contrast to our research,
however, Frost et al. (1998) limited their study area to demarcated cities,
to which only incoming commuting trips were added. So, the spatial
structure of the commuter area around the considered cities was not fully
grasped. Within these demarcated cities urban sprawl hardly occurs, and
also the growth in traffic itself is far more constrained by congestion than
is the case in the suburban areas. While this alternative approach may
explain the difference in results, apparently great caution in interpreting
the results is needed.
4.7.3 Evolution of the spatially disaggregated
minimum commuting distance
A second step in the analysis is the calculation of spatially disaggregated
values of mmid. For each municipality and for each of the three points in
time mmid was calculated twice: once for the outgoing commute (ommd)
(these are the outgoing and internal trips together) and once for the
ingoing commute (immd) (these are the incoming and the internal trips
together). So, for each municipality we obtain two time series.
Then, for each time series the existence of a clear trend for the period
1981-2001 was examined. First, a Student’s t-test was applied to compare
columns and rows of the produced origin-destination matrices. Non-
significant differences (significance level: p < 0.05) are considered as a
status quo. The decision rules that were used to determine the existence
of a trend are presented in Table 4.3. As a general principle it is assumed
that a trend is only acknowledged if the evolution over the period 1981-
Measuring spatial separation processes through the minimum commute
139
1991 is not contradictory to the evolution over the period 1981-2001.
Moreover, the differences have to be statistically significant in at least
one of the two periods.
Table 4.3. Decision table trend
evolution
1981-2001
evolution
1981-1991
evolution
1991-2001
conclusion
+ + + / - / not sig. +
+ - + 0
+ not sig. + +
- - + / - / not sig. -
- + - 0
- not sig. - -
not sig. + / - / not sig. + / - / not sig. 0
For the municipalities where a significant trend was found, the absolute
differences in mmid over the period 1981-2001 were mapped (Fig. 4.4 and
Fig. 4.5). The following applies:
ommd81-01 = ommd01 - ommd81 (4.5)
and immd81-01 = immd01 - immd81 (4.6)
4.7.4 Outbound minimum commuting distance
by municipality
Fig. 4.4 shows the evolution of the minimum commuting distance over
the time frame 1981-2001 for those municipalities that contain more
working residents than jobs. Municipalities with a job surplus are omitted
since the applied method implies that the minimum commuting distance
for workers living in such a municipality remains constant as long as the
job surplus exists. Following issues on the map stand out:
• Most municipalities in the economic core around the triangle Ant-
werp-Brussels-Leuven (A) show a decrease of ommd. Exceptions are
some municipalities with a more rural character that are located on
the edge of the conurbation and have received a large share of hous-
ing suburbanization (Kapellen (B), Nijlen and Berlaar (C), Zemst
(D), Oud-Heverlee (E), Beersel and Sint-Pieters-Leeuw (F)).
Chapter 4
140
• The former mining region in the Kempen (G) shows a sharp increase
of ommd, perhaps due to economic transformation processes that
have led to a decline of job supply in this area.
• In some areas apparently the construction of the motorway has
indeed led to a suburban housing development that entailed only few
jobs. This is clearly the case along the E40 between Ostend and
Ghent (H), and in some municipalities in the Voorkempen (Brecht
and Zoersel (I)) and Limburg (such as Maasmechelen (J), Riemst
(K), Heers (L)).
• In other municipalities also a lot of employment developed in the
proximity of the motorway. This is the case in Leie valley south of
Ghent (M), or in Lummen (N).
Fig. 4.4. Evolution of the minimum commuting distance (1981-2001)
based on municipality of residence
4.7.5 Incoming minimum commuting distance
by municipality
Fig. 4.5 shows the evolution of the minimum commuting distance over
the period 1981-2001 for those municipalities that contain more jobs than
working residents. In this case, municipalities with a surplus of workers
are omitted since the applied method implies that the minimum commut-
Measuring spatial separation processes through the minimum commute
141
ing distance for workers employed in such a municipality remains con-
stant as long as the worker surplus exists. Following issues on the map
stand out:
• The value of immd increases in almost all non-peripheral cities, and
the Brussels region has the strongest growth. This means that these
cities need to cover an expanding recruitment area to have their
available jobs occupied.
• A number of cities in the more remote areas, however, show a
reduction of the minimum commuting distance. Thus, in these places
the concentration of employment is proportionally shrinking. This is
the case in the cities Kortrijk (O), Ostend (P) and Bruges (Q) in the
west, and in the cities of Hasselt and Genk (R), Sint-Truiden (S) and
Tongeren (T) in the east.
Fig. 4.5. Evolution of the minimum commuting distance (1981-2001)
based on municipality of workplace
4.7.6 General interpretation and possible biases
The results of our analysis seem to indicate that at the macro level the
distance between job locations and residential locations of the working
population did not dramatically increase, and that commuter traffic
Chapter 4
142
shows an autonomous growth that can only partially be explained by
changes in the spatial structure.
There are two, possibly complementary, explanations for the rela-
tively limited detected increase of the average distance between home and
work locations. First, jobs have partly followed the suburbanization
process of housing. Sprawl does not simply consist of new residential
allotments, but equally of new business parks. Moreover, many new jobs
engrafted onto the additional residential areas, e.g. new employment that
developed in schools, childcare, local health care, supermarkets and public
services in growing municipalities. At the macro level, suburban areas are
relatively multifunctional: sprawl does not necessarily imply the absence
of a functional mix, even if it is at a lower density in comparison with the
city. Nevertheless, the regional job market remains more spatially
concentrated than the housing market. This is the case in many industrial
activities, but in particular the growth of specialized services (such as
financial services, technology and consultancy) led to an increase in the
number of jobs in the Brussels region. This growth explains a significant
portion of the found increase in the general minimum commuting dis-
tance.
A second, perhaps equally important reason is the large-mesh nature
of the used dataset. Our analysis cannot possibly grasp the sprawl that
occurs within a municipal boundary. Although, given the rather small
area of an average municipality, suburbanization certainly plays a role
too at an intermunicipal scale level, it appears that the intramunicipal
share of these transformation processes cannot be detected from a low
resolution data set. To identify sprawl on a lower scale level time series
data at a much denser spatial aggregation level is needed. Thus, the
discrepancy between our analysis and the morphological analyses men-
tioned above is mainly due to the difference in scale. It is likely that the
increase in the minimum commuting distance, and therefore the expan-
sion of the functional space, would be better reflected in the figures when
examining this micro level. This is due to the modifiable areal unit
problem (MAUP) of which the consequences for the minimum commuting
distance were discussed above. What is certain is that the use of a fine-
mesh data set leads to lower absolute figures for the minimum commuting
distance. On the basis of cross-sectional approximate data for 2007
available for the same study area (Flanders and Brussels) at the level of
(fine-meshed) traffic analysis zones, and using the same algorithm as we
did, Boussauw et al. (2011) found a mean minimum commuting distance
Measuring spatial separation processes through the minimum commute
143
of 6.9 kilometres (to be compared with our calculated 9.4 kilometres
(2001)). Unfortunately, time series data are not available at this micro-
scopic scale level, so it is not possible to verify our assumption that
spatial separation may be mainly taking place at this smaller geographi-
cal scale level.
The inability to outline developments based on fine-mesh data might
lead to an overrepresentation of the impact of regional economic trans-
formations in the results, pushing the effect of the spatial shift in the
housing stock to the background. The decline of some industrial activi-
ties, the development of logistics in the port areas and the general shift
towards service industries play perhaps a more important role at the
studied scale level with respect to spatial proximity between home and
work locations than the suburbanization of the housing stock does.
When we consider the spatially disaggregated evolution of the mini-
mum commuting distance, we notice that the major conurbations in the
economic core of Flanders and Brussels gain importance in terms of
employment, while a reduction of jobs in more peripheral industrial
sectors has led to a local increase of the minimum commuting distance.
The phenomenon of residential suburbanization in the municipalities
along the motorways has led only here and there to spatial separation,
particularly where jobs did not follow the spatial shift in housing.
4.8 Conclusions
The applied method detects spatially disaggregated evolutions in mini-
mum commuting distance, identifying local increases or decreases in
spatial separation between home and work locations at the level of the
municipality. The results show that there is indeed a general loss in
spatial proximity between housing and jobs for the study area in its
entirety, although the pace of this separation process is on average much
lower than the growth in observed commuting distances. It is found that
the minimum commuting distance increased in many municipalities,
especially where population is growing faster than job supply, or where
traditionally high concentrations of employment still increase. Further-
more, there are also municipalities where a decrease is noticed, especially
in suburban areas that are getting a more urban character by acquiring a
considerable functional mix.
Chapter 4
144
The development of the motorway network has undoubtedly contrib-
uted significantly to the growth of actual commuting distances. In those
municipalities where mono-functional planning practice has facilitated
residential development with access to the motorway, the minimum
commuting distance increased, with negative consequences for the spatial
proximity between residential and work locations. In municipalities where
a better balance between different functions was achieved, the develop-
ment of multifunctional sprawl has not necessarily led to an increase of
the minimum commuting distance at the supra-municipal scale level.
Nevertheless, the influence of sprawl on commuting behaviour seems
to be only secondary to the effects of regional-economic transformations,
which for example led to the loss of employment in the Kempen region
and an increase in employment concentration in Brussels and (to a lesser
extent) in Antwerp. Still, in spatial and economic planning it is impor-
tant to ensure the local balance between jobs and inhabitants as good as
possible. Horner and Murray (2003) argue that the most effective way to
do this is raising residential density in areas with a decent job supply:
through the deliberate reallocation of workers’ residences a significant
decrease of the minimum commuting distance can be attained. However,
Yang (2008) shows that job decentralization may also be responsible for
the growth of the excess rate itself, and thus for a weakening land use-
transport connection. From this finding Yang (2008) argues that a policy
of ensuring a good job-housing balance is insufficient: concentrations of
employment both in cities and in suburbs should take the form of com-
pact, yet relatively large, centres and subcentres.
In order to draw valid conclusions, the degree of detail and the con-
sistency of the used data is crucial. We find that the use of municipalities
as a spatial entity is suitable to grasp regional transformations of the
economy, but is far from perfect to detect sprawl in the morphological
sense. If the same analysis could be repeated at a lower scale level,
probably more concrete guidelines for spatial planning could be given.
This is particularly true when mode choice would be taken into account
too. In that case, concentration of activities around the stops of public
transport would remain equally important. For cyclists, at the other
hand, proximity and the availability of infrastructure are also of interest
at the micro level.
Measuring spatial separation processes through the minimum commute
145
References
Aguilera, A. (2005) “Growth in commuting distances in French polycen-
tric metropolitan areas: Paris, Lyon and Marseille.” Urban Studies.
42(9), pp. 1537–1547.
Antrop, M. (2004) “Landscape change and the urbanization process in
Europe.” Landscape and Urban Planning. 67(1-4), pp. 9-26.
Banister, D. (1999) “Planning more to travel less.” Town Planning
Review. 70(3), pp. 313-338.
Bleijenberg, A. (2003) “The driving forces behind transport growth and
their implications for policy”. In: Managing the Fundamental Drivers
of Transport Demand: Proceedings of the International Seminar.
Paris: OECD Publishing.
Boussauw, K., T. Neutens and F. Witlox (2011) “Minimum commuting
distance as a spatial characteristic in a non-monocentric urban sys-
tem: The case of Flanders.” Papers in Regional Science. 90(1), in
press.
Boussauw, K. and F. Witlox (2009) “Introducing a commute-energy
performance index for Flanders.” Transportation Research Part A.
43(5), pp. 580-591.
Charron, M. (2007) “From excess commuting to commuting possibilities:
More extension to the concept of excess commuting.” Environment
and Planning A. 39(5), pp. 1238-1254.
Ewing, R. H. (1994) “Characteristics, causes, and effects of sprawl: A
literature review.” Environmental and Urban Studies. 21(2), pp. 1-15.
Frost, M., B. Linneker and N. Spence (1998) “Excess or wasteful com-
muting in a selection of British cities.” Transportation Research Part
A. 32(7), pp. 529-538.
Gordon, P. and H. W. Richardson (1989) “Gasoline consumption and
cities: A reply.” Journal of the American Planning Association.
55(3), pp. 342-346.
Gulinck, H., S. Meeus, J. Peymen, L. Stalpaert, K. Bomans, V. Dewael-
heyns and S. Heremans (2007) Milieurapport Vlaanderen,
Achtergronddocument Versnippering. Aalst: Flemish Environment
Agency.
Horner, M. W. (2004) “Spatial dimensions of urban commuting: A review
of major issues and their implications for future geographic research.”
The Professional Geographer. 56(2), pp. 160-173.
Chapter 4
146
Horner, M. W. (2007) “A multi-scale analysis of urban form and commut-
ing change in a small metropolitan area (1990-2000).” Annals of
Regional Science. 41(2), pp. 315-332.
Horner, M. W. (2010) “Exploring the sensitivity of jobs - housing statis-
tics to imperfect travel time information.” Environment and Planning
B. 37(2), pp. 367-375.
Horner, M. W. and A. T. Murray (2002) “Excess commuting and the
modifiable areal unit problem.” Urban Studies. 39(1), pp. 131-139.
Horner, M. W. and A. T. Murray (2003) “A multi-objective approach to
improving regional jobs-housing balance.” Regional Studies. 37(2),
pp. 135-146.
Layman, C. and M. W. Horner (2010) Comparing methods for measuring
excess commuting and jobs-housing balance: an empirical analysis of
land use changes. Transportation Research Board Annual Meeting
2010. Washington, DC.
Ma, K. R. and D. Banister (2006) “Excess commuting: A critical review.”
Transport Reviews. 26(6), pp. 749-767.
Ma, K. R. and D. Banister (2007) “Urban spatial change and excess
commuting.” Environment and Planning A. 39(3), pp. 630-646.
Mérenne-Schoumaker, B., H. Van der Haegen and E. Van Hecke (1999)
Algemene Volks- en Woningtelling op 1 maart 1991: Werk- en
Schoolpendel. Brussels: Statistics Belgium.
Murphy, E. (2009) “Excess commuting and modal choice.” Transporta-
tion Research Part A. 43(8), pp. 735-743.
Newman, P. and J. Kenworthy (1989) Cities and Automobile Depend-
ence. A Sourcebook. Aldershot: Gower.
Niedzielski, M. A. (2006) “A spatially disaggregated approach to com-
muting efficiency.” Urban Studies. 43(13), pp. 2485-2502.
O’Kelly, M. E. and W. Lee (2005) “Disaggregate journey-to-work data:
Implications for excess commuting and jobs-housing balance.” Envi-
ronment and Planning A. 37(12), pp. 2233-2252.
O’Kelly, M. E. and M. A. Niedzielski (2008) “Efficient spatial interaction:
Attainable reductions in metropolitan average trip length.” Journal
of Transport Geography. 16(5), pp. 313-323.
Peng, Z. R. (1997) “The jobs-housing balance and urban commuting.”
Urban Studies. 34(8), pp. 1215-1235.
Rietveld, P. and R. Vickerman (2003) “Transport in regional science: The
‘death of distance’ is premature.” Papers in Regional Science. 83(1),
pp. 229-248.
Measuring spatial separation processes through the minimum commute
147
RSV (1997/2004) Ruimtelijk Structuurplan Vlaanderen - Gecoördineerde
Versie. Brussels: Ministry of the Flemish Community.
Schafer, A. (2000) “Regularities in travel demand: An international
perspective.” Journal of Transportation and Statistics. 3(3), pp. 1-31.
Sultana, S. and J. Weber (2007) “Journey-to-work patterns in the age of
sprawl: Evidence from two midsize southern metropolitan areas.” The
Professional Geographer. 59(2), pp. 193-208.
Van Wee, B., P. Rietveld and H. Meurs (2002) A Constant Travel Time
Budget? In Search for Explanations for an Increase in Average
Travel. Amsterdam: Vrije Universiteit Amsterdam.
Vanneste, D., I. Thomas and L. Vanderstraeten (2008) “The spatial
structure(s) of the Belgian housing stock.” Journal of Housing and
the Built Environment. 23(3), pp. 173-198.
Verhetsel, A., I. Thomas and M. Beelen (2010) “Commuting in Belgian
metropolitan areas: The power of the Alonso-Muth model.” Journal
of Transport and Land Use. 2(3), pp. 109-131.
Verhetsel, A., I. Thomas, E. Van Hecke and M. Beelen (2007) Pendel in
België. Deel I: De Woon-Werkverplaatsingen. Brussels: Statistics Bel-
gium.
Witlox, F. (2007) “Evaluating the reliability of reported distance data in
urban travel behaviour analysis.” Journal of Transport Geography.
15(3), pp. 172-183.
Yang, J. (2008) “Policy implications of excess commuting: Examining the
impacts of changes in US metropolitan spatial structure.” Urban
Studies. 45(2), pp. 391-405.
Yang, J. and J. Ferreira (2009) “Informing the public of transportation-
land use connections.” Transportation Research Part C. 17(1), pp.
27-37.
Chapter 4
148
149
Chapter 5:
Excess travel in non-
professional trips: Why
looking for it miles away?
This paper will be published as Boussauw, K., V. Van Acker and F.
Witlox (2011) “Excess travel in non-professional trips: Why looking for it
miles away?” Tijdschrift voor Economische en Sociale Geografie. Copy-
right © The authors. All rights reserved.
Abstract
Based on the spatial distribution of some quasi-daily destination classes
and survey-reported trip distances, regional variation in excess travel in
non-professional trips in Flanders (Belgium) is assessed. To this end,
proximity to various quasi-daily destinations is compared with the
reported distance that is actually travelled to reach similar, but alterna-
tive, facilities.
We note that in rural areas (compared with urban areas) larger dis-
tances are travelled, although the closest facility is chosen more often. In
the most urbanized areas, however, we note that spatial proximity is also
an important aspect in destination choice.
Quantification of these phenomena can support the practice of sus-
tainable spatial planning by distinguishing areas that are too mono-
functional or too remote, and therefore need more functional diversity,
and by identifying areas where densification is useful because the location
is close to most quasi-daily destinations, reducing the need to travel over
large distances.
Keywords: travel behaviour; Flanders; excess commuting; sustainable
spatial development; accessibility
Chapter 5
150
5.1 Introduction
Næss (2003) reduces the relationship between spatial proximity and
mobility to its geometric essence: in an area with a high density of people
and services, distances that are to be covered between potential origins
and destinations are small. As the trip distance is related to the amount
of energy required, empirical research not surprisingly shows that fuel
consumption for transport per capita is actually lower in areas with a
high density than in regions with a low density (Newman and Kenwor-
thy, 1989 and 1999; Næss et al., 1996). This is a logical consequence, as
Næss (2003) states: “The absence of any such influence would also have
been quite sensational.” But reality is obviously more complex than a
geometric problem. All kinds of factors, such as infrastructure configura-
tion, routes of public transport, or the lack of parking space, are
distorting this obvious logic. But also an unbalanced spatial mix, often
caused by functional city planning, may cancel out the positive potential
of high density. Moreover, mode choice plays a role: cities with a high
proportion of pedestrians, cyclists and public transport users will have
less traffic problems. Besides, on the regional level a clear linear relation-
ship exists between fuel consumption and the number of kilometres
travelled per person (Boussauw and Witlox, 2009).
The main deviation between travel behaviour and geometry is due to
the fact that a high degree of spatial proximity, and thus better accessi-
bility, gives rise to new needs (Næss, 2003). Gains, in terms of both time
and money, yielded by a higher level of accessibility are partly offset by
the individual who will make use of the increased choice range. When the
nearest supermarket is located just 100 m from one’s front door, then the
threshold for visiting the second nearest supermarket, at e.g. 500 m, is
particularly low, certainly when the latter offers more products or is a
little cheaper. But if the nearest store is located at 10 km, and the second
nearest is only at 20 km, the same person will for sure go shopping in the
closest store. Although the use of a wider range of accessible destinations,
as well as an increase in the number of trips may offset the potential
efficiency gains of the compact city, the aforementioned empirical studies
suggest that this is only partially the case.
Handy et al. (2005) argue that the reason for the emergence of a trip
can be situated along a choice-necessity continuum. Driving around just
for fun (Ory and Mokhtarian, 2005) is located at the “choice” end of this
continuum, while buying a loaf of bread at the bakery around the corner
Excess travel in non-professional trips: Why looking for it miles away?
151
is at the “need” end of the same spectrum. In these terms, excess travel is
defined as travel beyond what is required for household maintenance,
given choices about residential location, job location, and activity partici-
pation. The required trip length is then defined as the shortest route to
the closest destination possible. The ratio between choice and necessity
determines the excessive nature of any particular trip, meaning the extent
to which the trip distance exceeds the distance to the closest facility that
could possibly satisfy the need of the traveller. The degree of spatial
proximity between the sites that are potential origins or destinations of
trips defines the actual travelled distance. But spatial structure itself is
also one of the factors that influence the decision. In an area with many
options nearby, “choice” will outweigh “need”. Furthermore, in this first
area - with a wide choice range - the total trip length might still be less
than in the second area - where there is little choice.
Vilhelmson (1999) develops a framework that classifies activities
based on the degree of flexibility in terms of physical location and point
in time. Education is an example of an activity where both location and
time are specified exactly. For an activity like jogging the reverse is true.
It are the activities in this last quadrant that are largely determined by
the choice of an individual, and thus may lead to excess travel. Vilhelm-
son (1999) finds that access to a car, not having children and having a
part-time job is associated with an increase in the number of kilometres
travelled during this kind of “free” activities. It is evident that also the
type of trip will play a role in the degree of excess travel. Horner and
O’Kelly (2007) quote the example of the difference between shopping
trips for “comparison goods” (which are only occasionally purchased) and
“convenience goods”. In the latter category, attempts to minimize the
distance travelled will play a greater role than in the first category. Based
on interviews, Næss (2006) shows a willingness to travel longer distances
for work, education and visiting family or friends in comparison with e.g.
schools, kindergartens and grocery shops.
The excess commuting research framework offers opportunities for
studying this phenomenon at a regional scale. In recent decades, excess
commuting has become a major study topic within the discipline of
transport research (Ma and Banister, 2006). The excess commute is that
share of the commute flow (in terms of physical distance or time dis-
tance) that cannot be attributed to the spatial separation between job
locations and residential locations of employees, and is thus rooted in the
travellers’ freedom of choice.
Chapter 5
152
The main goal of our research is examining regional variations in the
relationship between the length of quasi-daily trips and spatial proximity
by applying an excess commuting approach. The paper is structured as
follows. First, we provide a summary of the excess commuting literature
and extend the concept to non-professional travel. Second, we develop a
methodology to define theoretical minimum non-professional trip lengths
and to distinguish spatial categories within reported trip lengths. Subse-
quently, results are obtained by comparing the reported distance travelled
with the theoretical minimum distance travelled, within each spatial
category. Finally, we draw conclusions from our findings and derive
recommendations for sustainable spatial planning practice.
5.2 Excess commuting and excess travel
The concept of “wasteful commuting” or “excess commuting” was first
introduced by Hamilton (1982). Hamilton defined excess commuting as
the difference between the actual commuting distance and the theoretical
minimum commuting distance, suggested by the spatial structure of the
considered city. The attention paid by Hamilton (1982) to minimized
commuting distances stems from the successive oil crises of 1973 and
1979-1980, when the availability, and in particular, the affordability of
fossil oil products was at stake. Daily trips over large distances were
suddenly considered problematic, because of their particularly high energy
consumption and costs.
As transport research progressed, the concept of excess commuting
was extended and applied in different ways. The line of inquiry that was
started by White (1988), compares the spatial structure of different cities
on the basis of the minimum required commute; a method that was later
expanded with the idea of the maximum possible commute (Horner,
2002). Both concepts are measurable properties of spatial structure,
which can be applied not only to compare the morphology of cities, but
also to examine time series and thus measure suburbanization and
evolutions in commuting behaviour (Horner, 2007; Boussauw et al.,
2011b). Further, we can distinguish between the more economically-
inspired research direction that uses travel time as a variable, and the
more environmental approach that focuses on travel distance (Ma and
Banister, 2006). In both cases the minimum commuting distance may be
considered as a measure of proximity, in terms of accessibility (when
Excess travel in non-professional trips: Why looking for it miles away?
153
travel time is studied), or in the sense of spatial proximity (when physical
distance is studied).
An interesting extension of this is the spatially disaggregated ap-
proach where the minimum commuting distance is mapped, considering
this variable as a measure of spatial proximity. Again, we can distinguish
between analyses that are based on time distance (an approach intro-
duced by Giuliano and Small, 1993), and more environmentally-oriented
studies where physical distance is used as a measure (Niedzielski, 2006;
Yang and Ferreira, 2008; Boussauw et al., 2011a). When this type of
research is conducted at a regional scale, it may contribute significantly
to the sustainability of proposed land developments, and to the detection
of regions that are vulnerable because of their extreme remoteness. This
approach is, among other, relevant in the light of the peak oil theory. In
the course of history, the cost of transport has shown a nearly continuous
downward trend, with only a ripple at the time of oil crises. Since today
transport relies almost entirely on finite fossil fuels, we suspect that one
day the cost of transport will evolve in the opposite direction, increasing
systematically as oil supplies decline. The sudden, albeit temporary, surge
in oil prices in 2008 seemed to forecast this hard reality. But even though
there is little point in thinking in doomsday scenarios, it remains a fact
that over time highly car-dependent spatial structures may be particu-
larly vulnerable to oil shortage (Dodson and Sipe, 2008).
To date, the excess commuting research framework (hereafter ex-
tended to excess travel) has to our knowledge only been applied on the
study of the home-to-work commute. There is no doubt about the
primordial economic importance of the commute, which represents a
significant proportion of the number of car kilometres travelled. In
Flanders, the home-to-work commute represents 18.6% of trips. Yet, the
average commuting trip length amounts to 19.0 kilometres, which is
much higher than the average trip length of 12.5 kilometres (for all
purposes combined) (Zwerts and Nuyts, 2004). Moreover, we know that
commuting trips are much less price elastic and thus more inert than
other trips. All these arguments emphasize the importance of studying
commuting behaviour.
In contrast, in the western world the share of commuter traffic in the
overall mobility is decreasing. Leisure travel, and by extension: tourist
trips, are on the rise. In essence, this evolution originates from the ever
growing prosperity and the improved accessibility of high speed travel
modes for an increasingly larger share of the population. Even if the
Chapter 5
154
penetration of the private car and the coverage of the motorway network
would have reached its structural limits, we still continue to use more
and more often aircrafts and high speed trains for recreational and tourist
trips. In the continuum of Handy et al. (2005), these trips are situated
near the “choice” end, and are much less emerging by “necessity”. The
changeable nature of the destinations and the consequent difficulties in
data acquisition represent perhaps the real reason why researchers have
not yet ventured to the study of excess leisure travel, and have thus
confined themselves to the study of the home-to-work commute. Similar
reasons can be found concerning the study of other non-professional trips,
such as shopping, home-to-school travel or visiting public services.
However, non-professional trips did not entirely escape attention in
the excess commuting discourse. Recently, Horner and O’Kelly (2007)
suggested that the study of excess travel for non-professional, but more or
less daily trips could become an interesting extension, possibly shedding
more light on the relationship between non-professional travel behaviour
and spatial structure. Examples are innumerable: bringing children to
day-care or school, doing the groceries, or going to sports or hobby clubs.
The study of non-commuting trips, however, entails considerable meth-
odological problems. Following problems can be identified immediately:
• The capacity of many of the mentioned facilities is deemed elastic,
compared with employment centres that are characterized by a rela-
tively constant number of jobs.
• Many non-professional trips are made frequently, but not daily.
• There are often multiple destinations for one purpose, as is e.g. the
case for multipurpose shopping trips (Handy, 2001).
• Many leisure trips are part of a trip chain that partly includes
commuting, making the distinction between professional and non-
professional travel vague.
• Commuting trip lengths are the result of a “double” selection proce-
dure, being a combination of the preference of an employer for an
employee, and the preference of the worker for the job. In contrast,
non-commuting trips are based on a “single” selection procedure,
comprising only the customer’s preference for the visited facility.
Consequently, excess commuting rates will by definition be higher,
and thus not comparable with values found for non-commuting trips.
• For some destination classes the trade-off between the accessibility of
the widest possible range of customers and the cost of an additional
establishment is inherent in the location of the facility. This is the
Excess travel in non-professional trips: Why looking for it miles away?
155
case with branches of large chain stores, which are often sited based
on a facility location model.
• There is often no area covering sample data available on travel
behaviour of individuals and families, so it is not possible to aggre-
gate data within relatively small traffic analysis zones. Moreover,
available data often contains only information on reported trip
lengths, without mentioning the address of the visited facilities. In
the latter case it is only possible to estimate excess travel that is gen-
erated by residences (an important origin of many trips) in a
particular area, and not e.g. by stores or schools (examples of desti-
nations of non-professional trips).
The conventional calculation of the minimum commuting distance
involves origin-destination matrix optimization through techniques of
linear programming (White, 1988). This method assumes that destina-
tions (i.e. job locations) have a fixed capacity, and do not adapt in case
demand would change. In contrast with jobs, however, daily service and
facility destinations do not fulfil these conditions, an issue that is also
recognized by Fan et al. (2010). Therefore, we need to adapt the defini-
tion of excess travel to the context of non-work-related destinations.
To our knowledge, Fan et al. (2010) are the only authors who have
analyzed a case of excess travel in non-commuting trips. They approach
the concept of excess travel as the difference between the distance
travelled by a household to get through its activity programme (given the
current destinations and travel frequencies), and the distance that would
be travelled in order to achieve the same activity programme in case
home location would be optimized in a geometrical sense (meaning that
the considered family would move house to a location that is more
centrally relative to their activities).
In contrast, our own approach, which we explain into detail in the
next section, compares proximity of destinations, considered as a spatial
characteristic of the studied location, with the travel behaviour of its
residents and users. We assess the ratio between the actual distance
travelled and the distance that would be travelled in case for any trip
purpose an alternative, nearby destination would be chosen. Summariz-
ing, the definition of excess travel in non-commuting trips by Fan et al.
(2010) is household-oriented, while our definition is location-oriented.
Both approaches are therefore complementary.
A final aspect worth mentioning is the comparability of results be-
tween various analyses. It is known that the magnitude of the obtained
Chapter 5
156
values in excess commuting research are largely subject to the modifiable
areal unit problem (MAUP), making attempts to compare results of
research in areas with surveys based on a different zoning system virtu-
ally meaningless (Horner and Murray, 2002). This problem is even
exacerbated in the study of non-commuting excess travel: here, the extent
to which a destination category is representative of a particular destina-
tion is determining the obtained results too, and will additionally bias
comparisons.
5.3 Methodology
Our research takes the Flemish Region and - to some extent - the Brus-
sels Capital Region, together constituting the northern half of Belgium, as
a study area. The study consists of several stages. Given the relatively
scarce data it is impossible to use traffic analysis zones as spatial units,
which would be in line with the spatially disaggregated study of excess
commuting. Therefore we first have to define the spatial classes that
should be distinguished. We do this by mapping the transition between
more and less urbanized areas as accurate as possible.
Then we develop a non-professional equivalent to the minimum com-
muting distance in the form of a proximity map. This is done by defining,
for each statistical ward (corresponding with a neighbourhood) and for
each defined spatial class, the minimum distance that should be covered
in order to reach all facilities that are visited by an average Flemish
household during a week. Obviously, this method implies some simplifica-
tions. The choice of the number of destination categories or travel
purposes is limited by the amount of available data. So we need to
consider those facilities for which data is available as representative for a
certain category of destinations. This choice is to a certain extent arbi-
trary. Furthermore, the activity pattern of a household is also determined
by the spatial context in which this family lives, reducing the assumption
of an average activity pattern that is applicable to every household to a
major simplification. However, assuming an average household travel
pattern, regardless of the location of residence, is a deliberate choice we
make in order to map variations in spatial proximity in an objective way.
So, the average activity pattern of a resident of the Flanders region is
used as a reference in order to quantify proximity as a spatial characteris-
tic.
Excess travel in non-professional trips: Why looking for it miles away?
157
In a third phase, we examine the effective distance travelled by
households during non-professional trips, based on available travel survey
data, while distinguishing our predefined spatial classes and destination
categories. Eventually we compare reported travel distances with calcu-
lated minimum distances, resulting in a ratio that represents a measure
for excess travel. We consider variation of spatial proximity and excess
travel as a spatial characteristic, indicating how sustainable a certain
physical structure is in relation to non-professional trips.
5.4 Determination of spatial classes
Unlike data on commuting, which is based on a census (SEE 2001), data
for non-professional trips is in Flanders only available in the form of a
sample (Travel Behaviour Survey for Flanders 2000-2001 (Onderzoek
Verplaatsingsgedrag Vlaanderen (OVG)) (Zwerts and Nuyts, 2004). This
means that we cannot make a spatially continuous analysis on the basis
of a map for the whole studied region, as opposed to the study of excess
commuting (Boussauw et al., 2011a).
However, we want to relate the observed travel behaviour to different
types of spatial structures. To retain a survey sample that is large enough
for every spatial class, we look for a meaningful spatial classification for
the Flanders region. We base our argument on the existing literature.
Depending on the point of view, two major formats exist. Based on
empirical data (a combination of morphological characteristics and data
on commuting and migration flows) Luyten and Van Hecke (2001) assign
each municipality to one of the following four categories: urban agglom-
eration (“agg”), suburban (“sub”), commuter area (“comm”) and rural
(“rur”) (which is a residual category with very limited urban characteris-
tics). We will call this the “urban region” classification. The urban
agglomeration consists of those municipalities where more than half of the
population lives in an urban core or in the urban fringe, characterized by
a continuously built-up environment. The suburban area is the outer zone
of the city, characterized by an extensive, rural morphology, combined
with an urban functionality. Agglomeration and suburban area together
compose the urban region. The commuter area is attached to the urban
region and relies on this urban region for an important part of its em-
ployment. The demarcation of the different classes follows the municipal
Chapter 5
158
borders, causing some loss of accuracy. An overview is represented in Fig.
5.1.
Fig. 5.1. Spatial classification according to the “urban region” approach
(Luyten and Van Hecke, 2001)
The Spatial Structure Plan for Flanders (Ruimtelijk Structuurplan
Vlaanderen (RSV), 1997/2004) offers a second classification, which is
much more policy-oriented. This means that this format does not only
take into account the current situation, but incorporates also a vision for
future development, which is adopted by the Flemish government. The
main direction of development, favoured by RSV, is indicated by the
dichotomy between urban areas and outlying areas. Urban areas are those
areas that should receive most of the additional housing and businesses,
and are very accurately delineated (on the ward level). The selection
makes a distinction between metropolitan areas (“ma”) (agglomerations
with more than 300,000 inhabitants: Antwerp and Ghent, and the part of
the Brussels agglomeration that is located in Flanders), regional urban
areas (“rua”) (between 50,000 and 150,000 inhabitants) and small urban
areas. Within the latter category a distinction is made between “structure
supporting small urban areas” (“ssua”) (being relatively important
attraction and development poles) and “small urban areas at provincial
level” (“psua”) (being a development pole of minor importance).
Excess travel in non-professional trips: Why looking for it miles away?
159
Fig. 5.2. Spatial classification according to the RSV approach
(Ruimtelijk Structuurplan Vlaanderen, 1997/2004)
The demarcation of urban areas is based on a consultation process and is
consolidated on the basis of cadastral boundaries. For a number of urban
areas this demarcation process is still ongoing. We used a provisional
definition, translated to the ward level. For simplicity, we consider
everything that is not within the definition of any urban area as outlying
area (“oa”). In the outlying area we can still distinguish selected residen-
tial nuclei (“noa”), generally corresponding to villages. In total we
distinguish thus six categories within the framework offered by RSV. A
cartographic overview is shown in Fig. 5.2.
Since the two proposed classifications have their pros and cons, we
decided to include both systems in our analysis.
5.5 Developing a proximity map
5.5.1 Method and selection of destinations
In a first phase of our research we develop a method to quantify the
proximity of non-professional quasi-daily destinations. We use the ward
as a spatial unit, considering the centre of gravity (centroid) of the ward
Chapter 5
160
as a starting point to calculate the network distance to the closest
appropriate facility. The 9708 wards which cover Flanders and Brussels
are therefore regarded as residential locations. Used network data consist
of the seven highest categories of the Streetnet skeleton file, which
contains almost all passable connecting roads.
We select 18 types of facilities for which a location dataset is avail-
able. Each facility type is judiciously assigned to one of the travel
purposes that are applied in the OVG. We only used the purposes that
are non-professional (“work” and “business visit” are thus excluded) and
have a destination for which alternative locations can be found (so the
purposes “walking/driving around” and “visiting someone” were not
considered, just as the indefinite “other purpose”). The remaining
purposes are: shopping (SHP), education (EDU), picking up/taking
something/someone (PCK), leisure/sports/culture (LSC), services (e.g.
medical doctor, commercial bank) (SRV).
The various selected facility types, data sources and links with OVG
purposes can be found in Table 5.1. The purpose “shopping” is repre-
sented by three classes of supermarkets and some more specialized types
of shops. For the interpretation of the purpose “education” the higher
grades of secondary education and higher education (in general: education
for students over 14 years old) were not taken into account, since we
consider these facilities as too specialized and thus rather being part of
the commute. The purpose “leisure/sports/culture” is represented by
cafés, restaurants, sports centres and cinemas, while the category “ser-
vices” is represented by medical doctors and banks. We construct the
purpose “picking up/taking something/someone” by a combination of
education and leisure/sports/culture, supplemented with nursery. This is
of course only an approximation, where we assume that mainly children
are taken to and collected from their activities.
With regard to the quality of the data we mention that the data re-
trieved from Google Maps (2009) is based on commercial information
which is less complete than the other data sets that were used (Federal
Public Service of Economy (2009), Ministry of Education (2009), Child &
Family (2009), Cinebel.be (2009)), which claim to be exhaustive. The
location data from Google Maps include geographic coordinates. The
other data sets used consist of address lists, which were geocoded with
the help of Yahoo! Maps Web Services (2009). Finally, we calculated the
network distance between each ward’s centroid and the nearest location
within each selected type of facility using Dijkstra’s shortest path algo-
Excess travel in non-professional trips: Why looking for it miles away?
161
rithm (implemented in the Closest Facility Tool of ArcGIS Network
Analyst) as follows: Owf
Oiwf
Owf TTfiT min,,min, : ≥∈∀ (5.1)
in which:
Tmin = minimum trip length
w = ward
f = type of facility
i = any possible destination belonging to type of facility f
O = indicates that the spatial unit is always considered
as the base (origin) of the trip
Table 5.1. Selection of facilities and purposes
type of facility n source purpose OVG
baker’s 3747 Google Maps
supermarket class 1
(hypermarket)
54
supermarket class 2
(supermarket)
1484
supermarket class 3
(superette)
869
clothes shop 660
do-it-yourself shop 199
household appliances
(electrical)
180
Federal
Public
Service of
Economy
shopping
kindergarten 2913
primary school 2861
middle school (1st
grade high school)
681
adult education 111
Ministry of
Education
education +
picking up/taking
something/someone
nursery 2844 Child &
Family
picking up/taking
something/someone
café/bar 4746
restaurant 6907
sports centre 1581
Google Maps
cinema 49 Cinebel.be
leisure/sports/culture
medical doctor 9713
commercial bank 3391 Google Maps services
Chapter 5
162
5.5.2 Weighting
The closest facility calculation provides a proximity map for each type of
facility, showing the accessibility, in terms of physical distance, from
every considered ward. However, we want to limit the number of maps to
the five mentioned OVG purposes, and ultimately we want one summary
map. We calculate spatial proximity per spatial class and per travel
purpose as follows:
n
T
T
n
w
Owf
Osf
∑=
=1
min,
min, (5.2)
in which:
s = spatial class
n = number of wards in spatial class
To join the proximity of different purposes into one map it is necessary to
assign weights to the various facility types:
Owf
m
ff
OwH TaT min,
1min, ∑
=
⋅= (5.3)
in which:
H = average weekly haul
af = weight by type of facility
m = total number of facility types
The weighting is determined by the weekly visit frequency to the respec-
tive facilities by an average Flemish household. As a starting point we
take the number of trips per household per purpose, as reported in the
OVG. Overall an average household generates 42.95 trips per week, of
which 23.26 meet our criteria.
Based on a number of other data (demographic statistics and market
research), we estimate the average visit frequency. Visit frequencies of
some destinations are extrapolated to fit in with the OVG data (visit
frequency per OVG purpose). This means that some facility types are
considered as representative for similar destinations: e.g. clothes shops,
do-it-yourself and household appliance shops together are considered
being representative for non-food specialist shops. Estimated visit fre-
quencies are shown in Table 5.2. A trip is seen as a single move, which on
average should be partially attributed to a trip chain. Based on OVG we
expect that visiting a facility generates on average 1.68 trips, since often
more than one facility is visited within one trip chain. The purpose
Excess travel in non-professional trips: Why looking for it miles away?
163
“picking up/taking something/someone” in the table was reduced to the
facility type “nursery”. For comparison with reported trip lengths this
purpose is extended pro rata with the purposes “education” and “lei-
sure/sports/culture” (see below).
Table 5.2. Estimated weekly visit frequency by type of facility,
per household
purpose OVG
• representative facility
# trips/household-
week
shopping 8.99
• bakery 2.11
• supermarket class 1 (hypermarket) 0.69
• supermarket class 2 (supermarket) 3.26
• supermarket class 3 (superette) 0.42
• clothes shop 0.84
• do-it-yourself shop 0.84
• household appliances (electrical) 0.83
education (without higher secondary and
higher education)
3.09
• kindergarten 0.68
• primary school 1.70
• middle school (1st grade high school) 0.52
• adult education 0.19
picking up/taking some-
thing/someone (limited to nursery)
0.34
• nursery 0.34
leisure/sports/culture 9.00
• café/bar 1.84
• restaurant 5.95
• sports centre 1.00
• cinema 0.21
services (e.g. doctor, bank) 1.85
• medical doctor 0.62
• commercial bank 1.23
SUM 23.26
Chapter 5
164
5.5.3 Mapped proximity per spatial class
By mapping the result of equation (5.3) for every ward, we get a
weighted proximity map (Fig. 5.3). This map provides an overview of the
spatial variation in the minimum distance, expressed in kilometres, that
should at least be covered by an average Flemish household to complete
its weekly programme.
Fig. 5.3. Weighted proximity map for Flanders and Brussels (shortest
average weekly haul per household, for selected facilities)
To compare these calculated minimum distances with the distances
reported in the OVG, we calculate the average values for each spatial
class: both the values per purpose (Figs. 5.4 and 5.5) and the weighted
values (Fig. 5.6). Note that the Brussels Capital Region was omitted in
these tables in order to match the study area of OVG.
In general, shorter minimum trip lengths seem to be associated with
higher degrees of urbanization, even if not all distinguished classes are
useful in describing this phenomenon (e.g. the distinction between the
classes “sub” and “comm” is not expressed in our findings). An ANOVA
test, applied to both spatial classifications, indicates that minimum
distances significantly differ among spatial classes (significance level of
0.01) and thus confirms the observed general trend.
Excess travel in non-professional trips: Why looking for it miles away?
165
Fig. 5.4. Estimated minimum distance by purpose (km)
(urban region classification)
Fig. 5.5. Estimated minimum distance by purpose (km)
(RSV classification)
Fig. 5.6. Estimated shortest average weekly haul per household (km)
Chapter 5
166
As expected, agglomerations and metropolitan areas score points in terms
of spatial proximity. The differences between suburban and commuter
areas are minimal, as well as the differences between the structure
supporting small urban areas and small urban areas at provincial level.
The rural area and outlying area (including the residential nuclei in the
outlying area) score poorly, as was also expected. In particular, the
spatial classification according to RSV shows a systematic increase of the
minimum distances to be covered when we move from a more urbanised
to a less urbanised area.
5.6 Reported trip lengths
5.6.1 Data
Data on the effective length of trips made by the inhabitants of the
respective spatial classes are obtained from OVG. This survey reports on
travel behaviour of a sample of 3028 households over two consecutive
days. The sample does not contain households from Brussels (which is
administratively not part of the Flemish Region) and Ghent (for which a
separate survey was conducted). For our study we added a random
selection of data from Ghent to the Flemish data.
We selected only those trips originating from or ending at the resi-
dence of the respondent, with a destination or origin corresponding to one
of the five selected purposes. Thus, we do not only consider tours from
home to the facility and back, but also parts of trip chains between the
house and the facility. For each trip we know the reported distance
travelled, the ward where the respondent resides, and thus the spatial
class in which the residence is located. However, we do not know the
location of the visited facility. This trip selection method is a deliberate
choice, aiming to retain a maximum amount of useful information from
the available data. Nevertheless, this selection entails some specific biases.
Trips of respondents making complex trip chains may be under-
represented, while the distance travelled in retained parts of trip chains
may be less representative. Furthermore, this method assumes that the
most efficient trips have their destination close to home. However, in case
the respondent is a commuter, also a destination that is chosen nearby
the job location may lead to a very efficient trip. In the latter case it
should not be excluded that some of the reported excess travel, e.g. in a
Excess travel in non-professional trips: Why looking for it miles away?
167
shopping trip, is in fact due to the fact that the respondent works far
from home. It is important to keep in mind these possible biases when
interpreting the results of the analyses.
Since we are looking for quasi-daily travel behaviour, trips that cover
extremely large distances are considered as outliers. We eliminated
outliers per purpose, while setting the threshold at three standard
deviations above the mean trip length. Table 5.3 shows the remaining
number of observations after selection.
Table 5.3. Summary of the retained trips and matching OVG purpose
n SHP EDU PCK LSC SRV
agg 1977 777 936 1377 413
sub 816 427 530 483 166
comm 1273 623 669 794 308
rur 1632 835 904 1141 352
ma 941 365 451 636 203
rua 850 334 390 556 188
ssua 411 171 158 222 96
psua 281 119 143 202 61
noa 1990 1022 1104 1291 429
oa 1225 651 793 888 262
5.6.2 Method
To link up the minimum distance to be covered and the reported trav-
elled distances (Witlox, 2007), we follow two different approaches, in
parallel. First we calculate the average reported distance per purpose, per
spatial class:
q
T
T
q
r
Orsp
Oobssp
∑=
=1
,
, (5.4)
in which:
Tobs = average reported distance per purpose
Tr = reported length of trip r
p = purpose
q = total number of reported trips based in spatial class s and with
purpose p
Chapter 5
168
In a next phase we compare this with the minimum distance to be
covered, according to equation (5.2).
This procedure gives an indication of the influence of the degree of
proximity of a certain type of facility on the actual distance travelled to
reach a similar facility. However, the information obtained in this way is
not sufficient if we want to understand the relationship between sustain-
ability of travel patterns and spatial structure. In this second case also
information about trip frequencies is important, since it is conceivable
that people who make relatively short trips will compensate their benefit
- in terms of time and costs - by making more trips, cancelling out gains
in fuel consumption (as an indicator of sustainability).
Therefore we calculate the average distance travelled by a household
during one week by adding up the distances covered by all selected
reported trips per spatial class, and extrapolate this sum to a time frame
of one week:
t
T
T
t
h
q
r
Orsh
OobssH
∑∑= =
⋅=1 1
,
, 5.3 (5.5)
in which:
h = individual household
t = total number of households in spatial class s and included in the
survey
(factor 3.5 extrapolating the two-day survey to a time frame of 7
days)
In a next step we will compare this distance travelled with the weighted
average minimum distance to be covered by an average household to
satisfy its needs (equation (5.3)).
5.6.3 Results
For the observed (reported) values we also calculate averages and display
these by spatial class. Again, the ANOVA test indicates that significant
differences between spatial classes exist (p = 0.00, except for leisure trips
by RSV class, where p = 0.03). The averages for each purpose are shown
in Figs. 5.7 and 5.8, while the total observed distances (extrapolated to a
full week) are shown in Fig. 5.9.
Figs. 5.7 and 5.8 give a rather surprising picture. The differences be-
tween the spatial classes are much smaller than what we might expect
based on the major differences in spatial proximity. For most purposes,
Excess travel in non-professional trips: Why looking for it miles away?
169
the least urbanized classes do not necessarily provide the greatest dis-
tances travelled: the largest trip lengths are rather recorded in the
suburban and commuter areas. Also the minor differences between the
metropolitan and regional urban areas stand out. Despite the result of the
ANOVA test, the often wide confidence intervals also suggest that the
link between observed trip length and spatial class is rather weak.
Pairwise t-tests indeed show non-significant distinctions between some
pairs of classes, confirming the picture provided by Figs. 5.7 and 5.8.
Fig. 5.7. Reported trip length by purpose (km)
(urban region classification)
Fig 5.8. Reported trip length by purpose (km) (RSV classification)
Chapter 5
170
Fig. 5.9. Reported weekly haul length per household (km)
The results that are shown in Fig. 5.9 take into account the spatial
differences in trip frequency. Although, while assessing the results, we
must not forget that our selection method takes no account of trips that
do not have either their origin or their destination at the respondent’s
home. In environments that would be associated with trip chains that are
more complex than average this could lead to an underestimation of the
number of trips per household. Yet, the literature is not unanimous on
this issue. According to Krizek (2003), a high degree of urbanization is
associated with more, but less complex, trip chains, while Maat and
Timmermans (2006) found a higher tour complexity in more urbanized
areas. Moreover, complex trip chains are usually very efficient tours,
conducting a series of activities within a minimum tour distance.
When we consider the “urban regions” classification, it appears that
agglomerations, as expected, yield the shortest average weekly haul. The
commuter area - thus not the rural area - yields the longest weekly haul.
When examining the RSV classification, metropolitan areas constitute the
shortest weekly haul. The weekly hauls are much longer in the regional
urban areas, but still shorter in comparison with the small urban areas
(the structure supporting small urban areas in particular). In the outlying
area, we record the longest weekly haul.
These relations are quite consistent with British research, recording
the shortest travel distances in the major British cities (> 250,000
inhabitants), except London. Small towns and rural areas score poorly
(Banister, 1999). Our study adds a new element: commuter areas which
fit morphologically with the rural area but are still within the sphere of
influence of the agglomeration are scoring worse than the more remote
“real” rural area.
Excess travel in non-professional trips: Why looking for it miles away?
171
Parts of the results are probably explained by the utilization of the
available choice potential due to spatial proximity. This is more often the
case in metropolitan and regional urban areas, as they are both function-
ally and morphologically more urbanized than the rest of Flanders. At a
lower geographical scale the structure supporting small urban areas
generate more mileage than the small urban areas at provincial level.
However the nature of the former class is more urban than the second.
Perhaps the structure supporting small urban areas are functionally more
focused on the larger cities.
5.7 Excess travel
We find that the influence of spatial proximity does not work out in the
same way for each spatial class. We examine this phenomenon in analogy
with excess commuting research methods. We define excess travel as the
difference between the minimum distance that must be covered to visit
the desired type of facility (e.g., the nearest supermarket) and the
observed distance covered. The observed trip length is always larger than
the minimum trip length because of non-spatial factors that are deter-
mined by e.g. personal preferences, transport cost, price differences
between similar facilities, or the organization of trip chains. We choose to
express this difference as the ratio between the minimum distance to be
covered and the observed distance travelled. This ratio is called the
excess rate:
Osp
OobsspO
spE
EE
min,
,= (5.6)
OsH
OobssHO
shE
EE
min,
,= (5.7)
in which:
Ep = excess rate by purpose
EH = excess rate for an average weekly haul
As shown by equations (5.6) and (5.7), we calculate excess rates in two
ways. First we determine per purpose and for each spatial class the
relationship between the reported average distance of a trip with the
considered purpose (as shown in Figs. 5.7 and 5.8), and the minimum
distance to be covered to reach a similar destination (as shown in Figs.
5.4 and 5.5). This excess rate by purpose is shown in Figs. 5.10 and 5.11.
Secondly, we determine the ratio between the total reported distance
travelled, extrapolated to a full week (shown in Fig. 5.9), and the
Chapter 5
172
weighted minimum weekly haul (shown in Table 5.6). This weekly haul
excess rate is shown in Fig. 5.12.
Fig. 5.10. Excess rate by purpose (urban region classification) (ratio)
Fig. 5.11. Excess rate by purpose (RSV classification) (ratio)
Fig. 5.12. Weekly haul excess rate (ratio)
Excess travel in non-professional trips: Why looking for it miles away?
173
In Figs. 5.10 and 5.11 we note for most purposes a systematic downward
trend of the excess rate when we watch the various spatial classes in
order of decreasing urban nature. In short, this means that a high degree
of spatial proximity is only partially reflected in short trip lengths,
because the increased choice of possible destinations generates compara-
tively long journeys. In metropolitan areas the average household goes
shopping almost three times further from home than strictly necessary,
while in the outlying area this rate amounts to one and a half only. Thus,
a higher degree of spatial proximity creates greater choice, compensating
for a significant proportion of the potential gains (in terms of external
costs caused by traffic). Noteworthy is that the differences in excess
travel between the spatial classes are rather small. This is in contrast to
what was found previously in the case of excess commuting (Boussauw et
al., 2011a), with very high values in urban areas, compared to very low
values in rural areas. Regarding differences between purposes, we notice a
very high degree of excess travel in leisure trips, where strong personal
preferences play. This applies to some extent also for trips to services,
although the low visit frequency and the overall high degree of spatial
proximity (there is a doctor in every street, say) play their role in the
obtained excess rate.
When assessing the excess rate of a combined weekly haul, then the
downward trend is not evident anymore. Following the spatial classifica-
tion according to the RSV, regional and small urban areas report the
highest values of excess travel. The rates are much lower in the metro-
politan areas, which is mainly explained by a larger share of chained
trips. Although this puts the low excess rate of metropolitan residents
somewhat into perspective, it also means that the activity pattern of this
group is already very efficient. A household in a regional urban area
covers a weekly distance that is more than 12 times longer than the
minimum distance required by our model. By contrast, a household living
in the outlying area covers only 6 times the minimum required distance.
We can interpret the excess rate as a measure that indicates to what
extent a travel pattern can be made more efficient, given the spatial
context. In this case gaining efficiency means shortening travel distances
by choosing similar destinations closer to home, within the existing
spatial configuration of housing and facilities. Such an adjustment of
households’ travel pattern may happen in case transport would become
more expensive, e.g. by a severe congestion policy or a stringent environ-
mental policy, or by energy scarcity. In the outlying areas, distances are
Chapter 5
174
relatively great, and the excess rate is low. This means that those areas
are most vulnerable to a price increase in transport. In regional urban
and small urban areas, the distances are not only smaller, there is also
more margin leaving the possibility to choose destinations closer to home.
In the metropolitan areas, non-professional trips seem comparatively
efficient, apart from being short anyway.
5.8 Possible biases in the results
The discussion above assesses the number of kilometres travelled per
household. However, there are substantial differences between average
households in the various studied spatial classes. It is generally assumed
that households in urban areas are relatively small, while household
income peaks in the suburban areas near large cities. These factors may
play a role in the sustainability of travel behaviour, even though the
precise effect is often unclear.
A larger number of family members may lead to more kilometres
travelled per household. Yet, within these families carpooling occurs more
often, while children travel only few kilometres independently. Thus,
calculated per person, larger households are expected to produce less
kilometres. When examining the influence of spatial structure, calculation
of the number of kilometres travelled can be justified both per household
and per person, albeit from different viewpoints.
Household income plays a role too. At the macroeconomic level, there
is a linear relationship between income and the number of kilometres
travelled (Schafer and Victor, 1997). If there would exist significant
income differences between the various spatial classes, it would make
sense to control for this variable. Determining income, however, raises
additional methodological problems. It is for example possible that the
effect of a higher income in an urban environment is primarily reflected in
increased tourist travel, and not in longer daily journeys (Holden and
Norland, 2005). Moreover, we are also downplaying car ownership, an
intermediary variable that is influenced both by income, by the surround-
ings (supply of alternative transport means and parking) and by
household size (Van Acker and Witlox, 2010).
These assumptions add much more complexity to the study of the
role of spatial structure in the sustainability of travel behaviour. Should
we measure the distance travelled per household or per person? Is it
Excess travel in non-professional trips: Why looking for it miles away?
175
useful to take income into account, and if so, how do we tackle this issue?
To avoid oversimplification, we did not incorporate these variables in a
statistical analysis. However, below we shed more light on the possible
role of spatial structure in itself in relation to the mentioned issues by
providing a few basic figures.
5.8.1 Household size
Some Flemish policy plans argue that the average family size is smaller in
urban areas than in suburban and rural areas, and consider this phe-
nomenon as a social problem (Boudry et al., 2003, p. 114). It is also
assumed that the phenomenon of shrinking household size occurs more
rapidly in the urban areas, increasing the identified problem. Champion
(2001) paints a more balanced picture. Small households, particularly
one-person households, are only rarely based on the stereotypical young
career maker with a very urban lifestyle. Young singles stay continuously
longer living in the parental home, while more one-person households
than before are the result of a divorce, and are in many cases located in a
suburban area. In contrast, in western city cores especially the immigrant
population is keeping average family size at a relatively high level. For
Flanders (2006) we find the following values (Table 5.4):
Table 5.4. Average household size per spatial class (2006)
class (urban regions) n
agg 2.19
sub 2.55
comm 2.44
rur 2.48
class (RSV) n
ma 2.13
rua 2.23
ssua 2.24
psua 2.33
noa 2.50
oa 2.61
In urban agglomerations and metropolitan areas households are actually
smaller than average. Yet, an exploratory linear regression, trying to
explain weekly distance travelled by variation in family size does not
yield any significant results. When we calculate the average number of
kilometres travelled per person, rather than per household, then differ-
ences between spatial classes (as shown in Fig. 5.9) are somewhat smaller.
When using the spatial classification according to RSV, structure sup-
porting small urban areas stand out more, making travel patterns of the
Chapter 5
176
inhabitants of this class now appearing the least sustainable. The outlying
areas and small urban areas at provincial level follow shortly. The
agglomerations and metropolitan areas still score best.
5.8.2 Income
By way of illustration we examine the average household income, based
on the assessment forms of direct taxation for the year 2006. The avail-
able data are aggregated for each municipality. This level of aggregation
allows us to regroup the data according to the “urban region” classifica-
tion, but not to the RSV classification. It is important to keep in mind
that here too the Brussels Capital Region is not included in the analysis.
Table 5.5. Average household income by spatial class (2006)
class (urban regions) €
agg 28448
sub 28950
comm 26778
rur 24862
There seems to exist a slightly downward trend when we move from more
urbanized into less urbanized areas. This is particularly the case when we
would take into account household size (income per person). Yet, also in
this case an exploratory linear regression does not yield any significant
effect of the level of income per household or per person on the weekly
number of kilometres travelled. We conclude therefore that the macro-
economic theory, arguing that higher income results in more kilometres,
does not apply at the regional scale of our study area. This finding
provides an additional argument for the proposition that spatial structure
indeed plays an important role in the genesis of travel behaviour.
5.9 Conclusions
The excess commuting research framework proves to be very useful in
examining the relationship between non-professional trips and spatial
structure. By mapping the minimum distance that an average household
needs to cover in order to complete its weekly programme, we get an idea
of the variation in spatial proximity between housing and facilities. This
Excess travel in non-professional trips: Why looking for it miles away?
177
combined minimum weekly haul varies from 5 km to 392 km, depending
on the residence location. This wide range of spatial proximity classes
indicates that the distinction between more and less urbanisation is still
reality and remains important in terms of mobility. The proximity map
we obtained in this way (Fig. 5.3) can be used as a guidance for new
developments. Additional densification of areas where the degree of
spatial proximity is already high, or areas that are immediately adjacent
to these, make an excessive increase of newly generated traffic the least
likely. In areas with a relatively low degree of spatial proximity, the
situation could be improved by planning a better functional mix for the
future. Yet, additional housing in areas characterized by a low degree of
spatial proximity will generate more traffic. These findings are in line
with what Banister (1999, p. 318) suggests: “New development should be
of a substantial size and located near to (or within) existing urban areas
so that critical size thresholds can be achieved.”
As stated in the introduction, spatial proximity is only one aspect of
the overall picture. The degree to which choice behaviour is driven by
spatial structure is equally important. Fig. 5.9 shows that the relationship
between spatial proximity and the number of kilometres travelled is not
linear. Residents of agglomerations, metropolitan and regional urban
areas travel over relatively short distances, but the inhabitants of the
suburban and commuter areas and small urban areas appear to have a
less sustainable travel pattern than is suggested by the rather urbanized
spatial structures in which these people live. What is also surprising is
that these variations cannot be explained by differences in family size or
income.
From research on spatially disaggregated excess commuting we know
that less urbanized areas are characterized by a higher minimum com-
muting distance along with a lower excess rate. In other words, residents
of rural areas go to work further from home than urban residents, but
they opt more often for the closest job they can find (Boussauw et al.,
2011a) than city-dwellers do. By analogy, we expected to find a similar
phenomenon in the study of non-professional trips. Based on Fig. 5.12 we
see that this expectation is only partially confirmed. In particular metro-
politan areas, agglomerations and suburban areas are characterized by a
relatively low excess rate, indicating that residents of these areas are still
heavily influenced by spatial proximity when choosing their non-
professional destinations. Possibly, modal choice has to do something
Chapter 5
178
with this: in an urban environment more non-professional trips are made
on foot or by bike, slow transport modes for which trip length is crucial.
We conclude that spatial structure and degree of urbanization is of
great importance to spatial proximity and length of non-professional
trips. Particularly in metropolitan areas, but also in regional urban areas
or the suburban areas that are adjacent to both urban classes, households
look for non-professional activities relatively close to home, especially
when it is possible to walk or bike there in a pleasant way. By attributing
a role to this aspect in spatial planning practice, generation of additional
traffic can be avoided and the vulnerability of spatial developments to
more expensive transport (by rising fuel prices, congestion problems and
congestion policy) can be reduced.
References
Banister, D. (1999) “Planning more to travel less.” Town Planning
Review. 70(3), pp. 313-338.
Boudry, L., P. Cabus, E. Corijn, F. De Rynck, C. Kesteloot and A.
Loeckx (2003) De Eeuw van de Stad - Over Stadsrepublieken en Ras-
tersteden. Brussels: Ministry of the Flemish Community.
Boussauw, K., T. Neutens and F. Witlox (2011a) “Minimum commuting
distance as a spatial characteristic in a non-monocentric urban sys-
tem: The case of Flanders.” Papers in Regional Science. 90(1), in
press.
Boussauw, K., B. Derudder and F. Witlox (2011b) “Measuring spatial
separation processes through the minimum commute: The case of
Flanders.” European Journal of Transport and Infrastructure Re-
search. 11(1), pp. 42-60.
Boussauw, K. and F. Witlox (2009) “Introducing a commute-energy
performance index for Flanders.” Transportation Research Part A.
43(5), pp. 580-591.
Champion, A. G. (2001) “A changing demographic regime and evolving
polycentric urban regions: Consequences for the size, composition and
distribution of city populations.” Urban Studies. 38(4), pp. 657-677.
Dodson, J. and N. Sipe (2008) “Shocking the suburbs: Urban location,
homeownership and oil vulnerability in the Australian city.” Housing
Studies. 23(3), pp. 377-401.
Excess travel in non-professional trips: Why looking for it miles away?
179
Fan, Y., A. Khattak and D. Rodríguez (2010) “Household excess travel
and neighbourhood characteristics: Associations and trade-offs.” Ur-
ban Studies. doi: 10.1177/0042098010369712 pp. 1-19.
Giuliano, G. and K. A. Small (1993) “Is the journey to work explained by
urban structure?” Urban Studies. 30(9), pp. 1485-1500.
Hamilton, B. W. (1982) “Wasteful Commuting.” Journal of Political
Economy. 90(5), pp. 1035-1053.
Handy, S. L. and K. J. Clifton (2001) “Local shopping as a strategy for
reducing automobile travel.” Transportation. 28(4), pp. 317-346.
Handy, S. L., L. Weston and P. L. Mokhtarian (2005) “Driving by choice
or necessity?” Transportation Research Part A. 39(2-3), pp. 183-203.
Holden, E. and I. T. Norland (2005) “Three challenges for the compact
city as a sustainable urban form: Household consumption of energy
and transport in eight residential areas in the greater Oslo region.”
Urban Studies. 42(12), pp. 2145–2166.
Horner, M. W. (2002) “Extensions to the concept of excess commuting.”
Environment and Planning A. 34(3), pp. 543-566.
Horner, M. W. and A. T. Murray (2002) “Excess commuting and the
modifiable areal unit problem.” Urban Studies. 39(1), pp. 131-139.
Horner, M. W. (2007) “A multi-scale analysis of urban form and commut-
ing change in a small metropolitan area (1990-2000).” Annals of
Regional Science. 41(2), pp. 315-332.
Horner, M. W. and M. E. O’Kelly (2007) “Is non-work travel excessive?”
Journal of Transport Geography. 15(6), pp. 411-416.
Krizek, K. J. (2003) “Neighborhood services, trip purpose, and tour-based
travel.” Transportation. 30(4), pp. 387-410.
Luyten, S. and E. Van Hecke (2007) De Belgische Stadsgewesten 2001.
Brussels: Statistics Belgium.
Ma, K.-R. and D. Banister (2006) “Excess commuting: A critical review.”
Transport Reviews. 26(6), pp. 749-767.
Maat, K. and H. Timmermans (2006) “Influence of land use on tour
complexity.” Transportation Research Record. 1977, pp. 234-241.
Mokhtarian, P. L. and I. Salomon (2001) “How derived is the demand for
travel? Some conceptual and measurement considerations.” Transpor-
tation Research Part A. 35(8), pp. 695-719.
Næss, P. (2003) “Urban structures and travel behaviour. Experiences
from empirical research in Norway and Denmark.” European Journal
of Transport and Infrastructure Research. 3(2), pp. 155-178.
Chapter 5
180
Næss, P. (2006) Accessibility, activity participation and location of
activities: Exploring the links between residential location and travel
behaviour.” Urban Studies. 43(3), pp. 627-652.
Næss, P., S. L. Sandberg and P. G. Røe (1996) “Energy use for transpor-
tation in 22 nordic towns.” Scandinavian Housing & Planning
Research. 13(2), pp. 79-97.
Newman, P. and J. Kenworthy (1989) Cities and Automobile Depend-
ence. A Sourcebook. Aldershot: Gower.
Newman, P. and J. Kenworthy (1999) Sustainability and Cities: Over-
coming Automobile Dependence. Washington, DC: Island Press.
Niedzielski, M. A. (2006) “A spatially disaggregated approach to com-
muting efficiency.” Urban Studies. 43(13), pp. 2485-2502.
RSV (1997/2004) Ruimtelijk Structuurplan Vlaanderen - Gecoördineerde
Versie. Brussels: Ministry of the Flemish Community.
Schafer, A. and D. Victor (1997) “The past and future of global mobil-
ity.” Scientific American. 277(4), pp. 58-61.
Van Acker, V. and F. Witlox (2010) “Car ownership as a mediating
variable in car travel behaviour research using a structural equation
modelling approach to identify its dual relationship.” Journal of
Transport Geography. 18(1), pp. 65-74.
Vilhelmson, B. (1999) “Daily mobility and the use of time for different
activities: The case of Sweden.” GeoJournal. 48(3), pp. 177-185.
White, M. J. (1988) “Urban commuting journeys are not ‘wasteful’.”
Journal of Political Economy. 96(5), pp. 1097-1110.
Witlox, F. (2007) “Evaluating the reliability of reported distance data in
urban travel behaviour analysis.” Journal of Transport Geography.
15(3), pp. 172-183.
Yang, J. and J. Ferreira (2008) “Choices versus choice sets: A commuting
spectrum method for representing job-housing possibilities.” Envi-
ronment and Planning B. 35(2), pp. 364-378.
Zwerts, E. and E. Nuyts (2004) Onderzoek Verplaatsingsgedrag
Vlaanderen 2000-2001. Brussels-Diepenbeek: Ministry of the Flemish
Community.
Excess travel in non-professional trips: Why looking for it miles away?
181
Data sources
Cinebel.be (2009), Cinebel.be from http://www.cinebel.be/nl/. Retrieved
24/08/2009.
Google Maps (2009), Google Maps from http://maps.google.be/. Re-
trieved 24/08/2009.
Ministry of Education (2009), Onderwijsaanbod in Vlaanderen from http:
//www.ond.vlaanderen.be/onderwijsaanbod/. Retrieved 24/08/2009.
Child & Family (2009), Opvangadressen from http://www.kindengezin
.be/Themas/Kinderopvang/Opvangadressen/opvangadressen.jsp. Re-
trieved 24/08/2009.
Federal Public Service of Economy (2009), Planologie van de handel from
http://mineco.fgov.be/ministry/publications/publ_planologie_nl.htm
. Retrieved 24/08/2009.
Yahoo! Maps Web Services (2009), Yahoo! Maps Web Services from
http://developer.yahoo.com/maps/. Retrieved 24/08/2009.
Chapter 5
182
183
Chapter 6:
Relationship between
spatial proximity and
travel-to-work distance:
The effect of the
compact city
This paper will be published as Boussauw, K., T. Neutens and F. Witlox
(2011) “Relationship between spatial proximity and travel-to-work
distance: The effect of the compact city.” Regional Studies. Copyright ©
Regional Studies Association - Routledge. All rights reserved.
Abstract
In this paper, an assessment is made of the relationship between selected
aspects of spatial proximity (density, diversity, minimum commuting
distance, jobs-housing balance and job accessibility) and reported com-
muting distances in Flanders (Belgium). Results show that correlations
may depend on the considered trip end. For example, a high residential
density, a high degree of spatial diversity and a high level of job accessi-
bility are all associated with a short commute by residents, while a high
job density is associated with a long commute by employees. A jobs-
housing balance close to 1 is associated with a short commute, by both
residents and by employees. In general, it appears that the alleged
sustainability benefits of the compact city model are still valid in a
context of continuously expanding commuting trip lengths.
Keywords: compact city; spatial proximity; commuting; sustainable
spatial development; Flanders
Chapter 6
184
6.1 Introduction
Although the spatial development model of the ‘compact city’ gained
momentum in the planning literature particularly during the 1980s, there
are still many prevailing spatial policy plans that draw from this concept.
One of those is the Spatial Structure Plan for Flanders, which was
adopted in 1997 (Ministry of the Flemish Community, 1997/2004) and
was inspired, among others, by the European Union’s “Green paper on
the urban environment” (Commission of the European Communities,
1990) and the Netherlands’ “Vierde Nota over de Ruimtelijke Ordening
(Fourth Report on Spatial Planning; Ministry of Housing, Spatial Plan-
ning and the Environment, 1988).
In Europe, the concept of the compact city emerged from a visionary
quest for a model of sustainable urban development, based on a city
tailor-made for pedestrians and cyclists, with a relatively high density, a
high degree of functional mix and efficient public transport (Jenks et al.,
1996, p. 5). In North America, the New Urbanism concept can be consid-
ered the counterpart of the compact city model, offering an express
alternative for the typically American, extensive form of suburbanisation
(Ellis, 2002), although there are differences in terms of scale (New
Urbanism occurs usually in small-scale developments that are strongly
oriented towards walking). Apart from the protection of open space and
economics of scale, the motivation for compact city development is to a
large extent grounded on the sustainability of mobility patterns. Encour-
aging trips over short distances and creating spatial conditions that
stimulate walking, cycling and using public transport are key ingredients
of the compact city model. The model has been criticized because of the
potential of larger social problems in residential neighbourhoods with high
densities, the concentration of pollution in living environments and the
increasing risk of congestion (Burton, 2000). However, the benefits of
enhanced spatial proximity and reduced car dependence are rarely
questioned.
In reality, the distinctly demarcated, quasi-walled, compact city has
gradually disappeared since the nineteenth century. The density gradient
from the Alonso-Muth-Mills model (describing the equilibrium between
the distance to the central business district and real estate demand in a
monocentric urban system) corresponds much better to reality, even
though in the post-war Western world it was overtaken by the develop-
ment of urban sprawl, in which historical centres are usually embedded.
Relationship between spatial proximity and travel-to-work distance
185
However, this does not mean that mobility-influencing characteristics
that are attributed to the compact city are completely absent in subur-
banized and even sprawled areas.
Sustainable urban development implies offering more opportunities
for travel over short distances, and encouraging the use of alternative
transport modes (instead of the private car). These two objectives are not
independent. If the demand for long-distance travel is reduced, slower
transport modes, which are usually less environmentally stressful, are
likely to be chosen more often. Boussauw and Witlox (2009) showed that
at a regional scale the distance travelled per person can be considered a
good approximation for the overall sustainability of the commute, since
positive consequences of modal shifts (especially towards train) are often
counterbalanced by increases in average trip length, while a reduction of
trip lengths may result in a modal shift towards low-impact modes such
as cycling and walking. These patterns result in a correlation between
energy consumption (as a sustainability indicator) and distance travelled
that is even stronger than intuitively expected. Furthermore, many recent
policy plans (as in the study area: the Mobility Plan for Flanders and the
Flemish Climate Policy Plan) overemphasize encouraging a modal shift,
while research into the number of kilometres travelled has faded some-
what into the background. This justifies an approach in which the
distance travelled is considered a sole sustainability indicator, although
the authors are well aware that reality is significantly simplified by not
taking into account modal split, congestion and other factors that
influence the environmental impact of travel.
Within the scope of this paper, it is examined whether and to what
extent there is a discernible link between the alleged qualities of the
compact city and the travel patterns of its users on the basis of commut-
ing data for Flanders and Brussels. To this end, the paper focuses not
only on cities, but also on the presence of characteristics attributed to the
compact city model throughout the suburbanized historically polycentric
spatial structure that characterizes this region. More specifically, an
attempt is made to gain a deeper understanding of the observed spatial
variation in trip lengths on the basis of commuting data available for
Flanders and Brussels. The restriction to home-to-work travel is moti-
vated by the need for accurate data: in Belgium, commuting is the only
category of travel that is surveyed area-wide. Also, the study of the
commute is particularly relevant from an environmental point of view,
Chapter 6
186
given that the longest average trip lengths are recorded in this travel
category (for example, Zwerts and Nuyts, 2004).
Although today a large body of literature describes aspects of the re-
lationship between spatial structure and trip lengths, this paper adds
considerably to the issue. One of the major difficulties regarding the
realistic interpretation of previous studies is the uncertainty about the
spatial scale at which the influence of certain spatial characteristics
manifest. Much research has been done on monocentric urban structures,
in a way neglecting the spatial characteristics that reflect the embedded-
ness in the larger region (for example, Cervero, 1996; Peng, 1997; Wang,
2000; Schwanen and Mokhtarian, 2005a). By studying a region that is
characterized by a large spatial diversity, and by comparing a series of
various spatial variables - which account for the characteristics of the
surrounding area in different ways - this paper adds to the general
research framework on proximity and trip length and thus partly ad-
dresses the mentioned difficulty. In this way, the validity of various
computational methods for spatial proximity is examined explicitly.
Moreover, the analysis is consistently applied to both origins of trips (in
this case, residential locations) and destinations (in this case, work
locations) in order to obtain more insight into the role played by the
deviation in spatial distribution between both location patterns through-
out the region. Although the approach adds to the complexity of the
issue, it contributes to the practice of spatial planning by complementing
the classical triad ’density, diversity and design’ that was introduced by
Cervero en Kockelman (1997).
A second important contribution is to be found in the specific atten-
tion that is paid to methodological problems such as the modifiable areal
unit problem (MAUP), spatial autocorrelation, non-linearity and multi-
collinearity. Although this paper does not aim to solve these issues
directly, it does provide additional insight into the influence of the
analytical methods used on obtained correlations and significances. For
example, it is not inconceivable that in previous studies the apparent
non-significance of assumed relationships between spatial phenomena was
related to a non-deliberate choice of the geographical scale.
While the approach disregards personal variations in travel behaviour
that are not inherently spatial - which is indeed a major simplification - a
purely geographical approach to the concept of proximity remains of
major importance. The present argument for this is grounded in peak oil
theory, which dictates that limited oil production will in the long run
Relationship between spatial proximity and travel-to-work distance
187
lead to an important increase in the mileage-related costs of mobility and
give rise to the importance of spatial proximity (Dodson and Sipe, 2008).
The paper is structured as follows. A summary of the literature on
compact city characteristics and commuting trip length is presented in
the second section. A research methodology is developed, taking into
account statistical issues and data limitations, in the third section. The
results are analysed and confronted with the knowledge of the spatial
structure of the study area in the fourth section. The fifth section dis-
cusses theoretical issues of scale level and origin versus destination.
Finally, the sixth section provides concluding remarks and outlines
possible improvements by further research.
6.2 The relevant literature
6.2.1 Characteristics of the compact city
According to Neuman (2005), the most important mobility-influencing
characteristics of the compact city consist of a high density, a high degree
of functional mix and a fine-grained land-use pattern. The rationale
behind the pursuit of high densities stems from the work of Newman and
Kenworthy (1989, 1999), who pointed out that global cities with more
inhabitants per square kilometre (km2) consume less fuel per capita for
transportation needs. A higher density ensures a critical mass of public
transport patronage, but also enhances the potential for spatial interac-
tion between people based on short distances. While Newman and
Kenworthy’s work has been criticized on many occasions (for example,
Mindali et al., 2004), their thesis is still widely supported in spatial
planning practice.
A sound spatial mix of functions is another important, yet often
equivocal, feature of the compact city. In particular, it is unclear at which
geographical scale this mixture can play its full role. In the most extreme
case, a high degree of spatial mix boils down to self-sufficiency of jobs,
shops, schools and other services per neighbourhood. While several
authors have studied the influence of spatial characteristics on the travel
behaviour of residents at this district level (among them, Frank and Pivo,
1994; Cervero and Kockelman, 1997; Crane and Crepeau, 1998; and
Schwanen and Mokhtarian, 2005a, 2005b), it may also be apposite to
observe this kind of spatial mix at the level of a city or a region, at least
Chapter 6
188
with regard to the more specialized functions. That is because a small
grain size may neglect economic agglomeration benefits at larger scales.
Moreover, some activities (for example, heavy industry) obviously need a
lot of space, especially if desired or legally enforced environmental
buffering is to be taken into account.
Although by the end of the 1990s the debate on the compact city was
transformed into a discourse on sustainable urban development (Williams
et al., 2000), density and diversity were retained as important elements.
Cervero and Kockelman (1997) argued that density, diversity and design
are important spatial factors underlying travel behaviour, and they
focused on the scale of the neighbourhood, attributing an important role
to the design of public space. Stead et al. (2000) found that the size of
the city and the proximity to key infrastructure are additional factors
that determine the relationship between spatial structure and travel;
while Van Acker et al. (2007) focused on the explanatory power of a
range of social characteristics.
Based on these findings, it is argued here that high density, a sound
land-use mix and a small grain size are all spatial elements, amenable to
modification by policy, that are part of an overarching spatial quality
that is called here ‘spatial proximity’.
A less frequently discussed, but nevertheless important, aspect is the
spatial scale at which the influence of certain spatial characteristics is at
its maximum. In the literature on the compact city, often no distinction
is made between small towns and metropolises. In Flanders, the distance
covered by an average trip (all purposes combined) amounts to 12.5
kilometres (one way). The average commuter even covers a distance of
about 19 kilometres per ride between home and work (Zwerts and Nuyts,
2004). For the United Kingdom, Lyons and Chatterjee (2008) found an
average commuter trip length of 13.7 km in 2002-2003. According to
Banister et al. (1997), commuter trip lengths increase year after year in
European and North American metropolitan areas. It is patent that the
average trip length no longer corresponds with the scale of a compact
city, at least in a Belgian context.
The scale problem is acknowledged by many authors. Van Wee
(2002), for example, suggested that the findings of Newman and Kenwor-
thy (1989, 1999), which are based on demarcated cities, cannot be applied
to the Dutch Randstad conurbation, as this urban system is operating at
a higher, regional-scale level. Alberti (1999), on the other hand, referred
to several possible approaches to a sustainable spatial structure, stating
Relationship between spatial proximity and travel-to-work distance
189
that at the regional level the settlement pattern is of interest, while at
the local level neighbourhood design is paramount. In short, the scale
problem is well recognized, but not yet explicitly addressed. As will be
shown in the present paper, this is partly due to the fact that the quanti-
tative analysis of scale effects is hampered by substantial methodological
issues.
Based on these arguments, we hypothesize that the main compact
city variables (density and land-use mix) are insufficient to represent the
differences in the spatial distribution of residential locations and job
locations at the scale of the region, or the embeddedness in the surround-
ing area. Therefore, three additional variables that should grasp the issue
in a more comprehensive way are introduced in the third section.
6.2.2 Commuting trip length
Although it may be intuitively understood that spatial proximity is a
predominant determinant for commuting trip lengths, many studies
indicate that personal, economic and behavioural factors play an impor-
tant role too, making aggregate non-spatial variables even prevalent in
explaining commuting trip lengths. Cervero (1996), Peng (1997), and Van
Acker and Witlox (2011) found a positive, albeit relatively limited, effect
of income level on commuting distance. Wang (2003) was more specific:
above a certain income threshold, employees again tend to live closer to
their work, making the discerned relationship non-linear. Other socio-
economic characteristics that were investigated in the literature are
gender, race, number of workers per household, education and property
status of the residence (Wang, 2000). In the same analyses, urban
characteristics (in contrast to rural characteristics) are considerably
associated with short commuting distances. Irrespective of the type of
model, the construction of the spatial proximity variables and the
available data used, generally low coefficients of determination are
obtained, indicating that a number of unknown or non-quantifiable
‘spurious’ variables are responsible for the largest share of the explained
variance.
However, this is not to say that the spatial aspect is unimportant.
Rather, spatial structure is a rigid constraint that is not able to undergo
rapid change. If transport were to become more expensive, a scenario that
is suggested by peak oil theory, then the role of spatial proximity will
undoubtedly gain importance (Dodson and Sipe, 2008). The present
Chapter 6
190
research focuses on the relationship between spatial proximity and
commuting distance by testing the applicability of several possible
indicators on a regional scale.
6.3 Methods
6.3.1 Research design
The aim is to examine relationships between commuting trip lengths and
a number of spatial characteristics of the zones where the surveyed trips
have their origin and destination. Since the analysis focuses on regional
variations, possible relationships are first explored visually by confronting
maps with knowledge of the study area. A quantitative analysis is then
conducted by calculating correlations between the average trip length and
each of the other selected variables. The variables are aggregated within
geographic zones, introducing effects of spatial autocorrelation and the
modifiable areal unit problem (MAUP). To address this problem, the
choice of an aggregation level that is adapted to the analysis is para-
mount.
The two most obvious variables that are included in this research are
directly obtained from the compact city literature. These are density and
spatial diversity. However, three additional spatial parameters must also
be added: the theoretical minimum commuting distance (as a more
sophisticated measure of proximity), the jobs-housing balance (as a
measure of self-sufficiency of a zone regarding job supply), and the
number of accessible jobs (as a measure of accessibility). In the next
section we will explain how these three additional variables have been
calculated. In our analysis, the dependent variable is trip length both for
trips that depart from and arrive in each considered zone.
Relationships are deduced in two steps. First, all variables are visual-
ized at an intermediate aggregation level to obtain an overview. This map
allows for an initial impression of possibly present links. Subsequently, an
exploratory spatial correlation analysis is applied at several aggregation
levels, with the intention of quantifying a number of relationships and
testing their significance. Based on the initial results, the most appropri-
ate aggregation level for the analysis is selected. Finally, expected
multicollinearity in the variables and the assumed linearity of relation-
Relationship between spatial proximity and travel-to-work distance
191
shipsare asessed, the aim being to detect redundancies and thus improve
the interpretation of the results.
6.3.2 Used data and variable construction
In what follows, the origin and composition of the six used variables:
distance travelled per trip, density, diversity, minimum distance travelled
per trip, jobs-housing balance, and number of potentially accessible jobs,
are discussed.
6.3.2.1 Distance travelled per trip (dpt)
To determine the distance travelled per trip, the origin-destination (OD)
matrices of the Flanders Multimodal Model (MMM) are used. The MMM
is a macro-traffic model that is developed since 1998 and is commissioned
by the Flemish government. The matrices provided for this study simu-
late traffic on an average weekday between 04.00 and 11.00 hours
(morning traffic). The zones corresponding with the matrix are in most
places in line with census wards. The matrices are built on the basis of
the 2001 Census, which is an exhaustive survey of the Belgian population
(excluding children under six years of age), assessing the address of
residence and the address of the workplace (Verhetsel et al., 2007). The
processed data are aggregated by ward and they present a picture of the
daily travelled distances to and from each neighbourhood.
To obtain trip lengths, the shortest distance over the road network is
calculated between the centroids of the connecting zones. Note that the
distances calculated in this way are a slight underestimation of the real
distances travelled, since detour factors associated with faster routes or
the public transport network are not included (Witlox, 2007). Since the
fastest route depends on varying congestion levels, the aim is to avoid
added complexity and thus stick to calculating the shortest path. The
average distance travelled per trip )(obsh to and from each traffic
analysis zone i is calculated as follows:
∑∑
=
i
OiO
iO
Hobsh )( (6.1)
∑
∑=
i
DiD
iD
Hobsh )( (6.2)
where O and D are the number of departing and arriving trips; and H is
the distance covered by each trip.
Chapter 6
192
6.3.2.2 Density (dens)
As an approximation for density, the number of departing commuting
trips, as well as the number of arriving trips per square kilometre, is
calculated based on the MMM. Residential density is approximated by
counting the number of outbound trips per square kilometre in the
morning traffic. Job density is approximated by the density of inbound
trips. The densities C are calculated based on the number of outbound
trips O and the number inbound trips D belonging to traffic analysis zone
i with area A as follows:
COi = Oi/Ai (6.3) CD
i = Di/Ai (6.4)
6.3.2.3 Diversity (div)
As a source, the Strucnet file of the NGI (National Geographical Institute
of Belgium, 2009), which contains all buildings represented on the official
Belgian topographic maps with scale 1:10,000, was used. Different
categories of buildings are distinguished, but the accuracy of categoriza-
tion is limited. All buildings that are morphologically part of a group of
houses are listed as ‘ordinary building’. All other as such recognizable
buildings (those used for industry, schools, hospitals, public services et
cetera) have their own feature class. In practice this means that many
commercial functions, offices and services that are interwoven with
housing are not recognizable. Nevertheless, this inventory can be used to
approximate the diversity of functions in a given zone, and is without
doubt the best currently available area-wide data set in Belgium.
To calculate the spatial-functional diversity per zone, the Shannon
index was applied. This index is used in landscape ecology as a measure
of morphological diversity (Nagendra, 2002), and is in this case also called
spatial entropy (Batty, 1974). An extension to urban diversity is obvious.
The Shannon index is calculated as follows:
∑=
⋅−=
N
nnni ppS
1
ln
(6.5)
where N is the number of features included within the considered aggre-
gation zone i; and pn is the proportion of each function that occurs within
this zone.
Relationship between spatial proximity and travel-to-work distance
193
6.3.2.4 Minimum distance travelled per trip (mdpt)
The theoretical minimum commuting distance is introduced as a proxy
for spatial proximity. The minimum commuting distance stems from the
research on excess commuting (Hamilton, 1982; Ma and Banister, 2006,
2007; Charron, 2007), and is in this case considered as a spatial charac-
teristic (Niedzielski, 2006; Boussauw et al., 2011).
The principle of the method implies linking any observed departure
(in this case, in the morning traffic) to the nearest observed arrival, also
in the morning traffic. Per traffic analysis zone the number of departures,
as well as the number of arrivals, is retained, but the existing relationship
between origins and destinations is cut to minimize the total travelled
distance within the system. This theoretical exercise deliberately disre-
gards non-spatial factors that determine the real-world match between
origin and destination. When commuting trips are considered, this means
that everyone who is part of the active population is considered suitable
to perform any job.
Boussauw et al. (2011) developed an algorithm to calculate the local
values of minimum commuting distance, based on the MMM, for the
commute occurring between 04.00 and 11.00 hours. The algorithm obtains
a general minimization of travel distances via local optimization, simulat-
ing individuals pursuing a job closer to home. For each traffic analysis
zone, the calculation was performed twice: once with the zone considered
as an origin (outbound travel), and again with the zone considered as a
destination (inbound travel). The first computed value is seen as an
approximation for the proximity of housing within the zone in relation to
the Flemish and Brussels labour market, while the second value is
representative of the proximity of the labour market of the zone in
relation to housing in Flanders and Brussels. The average minimum
distance to be covered (min)h per trip from and to a traffic analysis zone
i is calculated as follows:
( )
∑
∑=
i
OiO
iO
Hh
min(min) (6.6)
( )
∑∑
=
i
DiD
iD
Hh
min(min) (6.7)
where min(H) is the minimized commuting distance per zone.
Chapter 6
194
6.3.2.5 Jobs-housing balance (jhb)
The jobs-housing balance is the ratio between the number of jobs local-
ized in a zone and the number of working people who live in the same
zone. Because of its simplicity, this indicator is often used as a spatial
characteristic in commuting research (Cervero, 1989; Peng, 1997; Horner
and Murray, 2003; for an overview, see Horner, 2004).
The jobs-housing balance for every traffic analysis zone i, based on
the MMM, is calculated by dividing the number of arrivals in the morn-
ing traffic by the number of departures:
Bi = Di/Oi (6.8)
6.3.2.6 Number of potentially accessible jobs (potjob)
The accessibility of the job market is also included in the analysis, and is
expressed as the number of jobs located in Belgium that can be reached
from every zone. Accessibility, departing from a particular zone, is
determined by combining a probability curve with the travel time
between the considered zone and all other zones in Belgium, after which
the values for all these other zones are summed. The probability distribu-
tion was deducted from empirical data; while the travel time was
calculated on the basis of a network with attributed impedances. The
basic data on the location of jobs date from 2001 and were provided by
the Statistics Belgium (2009) and is aggregated by borough (that is, a
municipality in the former administrative system).
The number of jobs accessible from a borough i is defined as follows:
( )∑=
⋅=
n
jijji cFJA
1
(6.9)
where Ai is the number of jobs accessible from the considered borough i;
Jj the number of jobs located in each borough j; and F(cij) is the imped-
ance function, based on the modelled probability curve.
The calculated results from Vandenbulcke et al. (2007, 2009) are
adopted, and these two papers are used for the method of exact calcula-
tion. This accessibility index is related to the jobs-housing balance, but
also takes into account the travel time needed to reach the core of the
labour market.
Relationship between spatial proximity and travel-to-work distance
195
6.3.3 Level of aggregation and MAUP
In order to discover correlations, the various parameters must be aggre-
gated within the same zonal classification. The traffic analysis zones of
the MMM are not suitable for this procedure. These zones are small in
high-density regions and vice versa, which apparently results - among
other objections - in a non-normal distribution of the variables linked to
these zones. The large dispersion in area size makes the traffic analysis
zones unsuitable for calculation of the Shannon index, which is area
dependent, and also the calculation of the number of accessible jobs was
initially based on another zoning area (boroughs). For the sake of uni-
formity, we choose to aggregate all data into a grid of square cells.
However, the choice of the size of these cells is not evident. It is a
well-known phenomenon that statistical correlations change whenever a
different level of spatial aggregation is chosen, and that these become
generally stronger when the aggregation level increases, even if accuracy
of the data is in fact lost by increasing the aggregation level (Amrhein,
1995). This mechanism is called the scale effect of the MAUP, and is
inherent in any quantitative analysis of spatially aggregated data.
Openshaw and Taylor (1979) argue that the choice of the aggregation
level should depend on the expected significance for the studied variables.
However, this suggestion does not solve the question. The main variable,
average trip length, shows important dispersion. Even if one chooses the
mesh size of the grid so that the average trip has its destination in a zone
adjacent to the zone of origin, this will often not be the case for trip
lengths deviant from this average. In this context, ‘mesh size’ points to
the length of a side of a cell in a uniform square grid. Regarding the
independent variables it is even less clear what an appropriate level of
aggregation would be.
To support the choice of a suitable aggregation level, four different
grids with mesh sizes of 1, 4, 8 and 16 km were applied in the exploratory
stage. These options are illustrated in Figs 6.1-6.4 for one example (that
is, trip length). The results of a spatial correlation analysis (see below)
indicate that the absolute values of the coefficients increase and that
significances weaken along with an increase of the aggregation level. This
finding confirms the expected influence of the MAUP.
The highest level of aggregation (16 x 16 km) seems inappropriate
because of the large number of non-significant relationships that are
revealed by means of spatial regression, due to the reduced dataset size.
Chapter 6
196
The lowest level of aggregation (1 x 1 km) also seems inapplicable: the
relatively high degree of discontinuity between adjacent cells makes it
sometimes difficult to clearly distinguish local patterns, resulting in often
very low coefficients. Therefore, for the current analysis, an intermediary
level of aggregation with a mesh size of 4 km was chosen (Fig. 6.2). If one
takes into account a general detour factor of 1.40 (Rietveld et al., 1999;
Witlox, 2007), then this choice implies that about 70% of all commuters
have their destination in another zone than the origin zone of their trip
(Zwerts and Nuyts, 2004).
Fig. 6.1. Average commuting distance (dpt), origin zones,
aggregation level = 1 km
Relationship between spatial proximity and travel-to-work distance
197
Fig. 6.2. Average commuting distance (dpt), origin zones,
aggregation level = 4 km
Fig. 6.3. Average commuting distance (dpt), origin zones,
aggregation level = 8 km
Chapter 6
198
Fig. 6.4. Average commuting distance (dpt), origin zones,
aggregation level = 16 km
For the construction of the grid, the border zones of the study area are
joined so that any cell size approaches the area of a square cell. The
aggregation of those data that were originally available at the level of
traffic analysis zones (that is, travelled distance, density, minimum
commuting distance and jobs-housing balance) or borough (access to
jobs) is done proportionally to the geographical overlap between two
geographical divisions. The Shannon index (diversity) is calculated once
for the grid with a mesh size of 1 km, and then aggregated to the applied
scale level.
6.3.4 Spatial regression
Given the nature of the examined spatial characteristics, an important
degree of positive spatial autocorrelation occurs between the zones
themselves. Spatial autocorrelation is the correlation between values of a
variable that has its origin in the vicinity of the locations where these
values are measured. Positive spatial autocorrelation, meaning that
neighbouring areas are similar, is a phenomenon that is typically present
in most empirical spatial datasets, and is explained by Tobler’s (1970)
first law of geography, in that ‘Everything is related to everything else,
Relationship between spatial proximity and travel-to-work distance
199
but near things are more related than distant things.’ (p. 236) The
presence of spatial autocorrelation is a violation of the assumption of the
independence of successive observations, which must be satisfied before
applying classical statistical techniques such as correlation and regression
analysis. Positive spatial autocorrelation leads too often to positive results
of significance tests, and thus to the unjustified rejection of null hypothe-
sises, and to an overestimation of regression and determination
coefficients (Anselin and Griffith, 1988).
Like the MAUP, spatial autocorrelation has often been ignored in
past spatial planning studies. One of the reasons for the neglect of these
issues in research applications is that the theoretical study of useful
spatial alternatives to traditional statistical methods has been debated for
a long time (Cliff and Ord, 2009). An alternative regression method that
was adopted by many authors is the so-called ‘spatial regression’ tech-
nique (Anselin and Bera, 1998). This method starts from a significance
test for spatial autocorrelation based on the calculation of Moran’s I, a
measure of spatial autocorrelation. If this test is not significant, the
application of an ordinary least-squares (OLS) regression is suggested. If
the Moran’s I test is significant (this is usually the case), the occurring
kind of spatial dependence (‘spatial lag’ or ‘spatial error’) should be
detected. The occurrence of spatial lag implies that the dependent
variable in a cell is affected not only by the independent variables in the
cell itself, but also by those of the neighbouring cells. In the case of
spatial error, only the residuals in the regression analysis of neighbouring
cells are correlated. Depending on the occurrence of spatial lag or spatial
error, a regression model is selected that reduces the influence of these
phenomena. In some cases, both the spatial lag and the spatial error
model can be applied. Each of these calculations provides, inter alia, a
regression coefficient, a significance test and a pseudo-coefficient of
determination (pseudo-R2). The relative popularity of this method in the
recent literature is probably due to the user-friendly implementation in
the freely distributed software tool GeoDa (Anselin et al., 2006). In this
study, models were estimated by means of GeoDa 0.9.5-i5, which is a
public domain software package making spatial regression techniques
available in a geographical information system (GIS) environment. The
software was developed by Dr Luc Anselin of the School of Geographical
Sciences and Urban Planning, Arizona State University, Phoenix, Ari-
zona, USA.
Chapter 6
200
6.4 Results
6.4.1 Cartographic analysis
In order to gain a better understanding of the spatial variations of the
variables, a cartographic representation is shown in Figs 6.5-6.13,
aggregated at an intermediate scale in a grid of 4 x 4 km. Fig. 6.14 shows
a schematic reference of the study area (Flanders and Brussels). The
variables distance travelled per trip, minimum commuting distance and
density are each shown twice: once the zones are regarded as origins; and
once the zones are regarded as destinations. The variables diversity, jobs-
housing balance and number of potentially accessible jobs are displayed
once.
Fig. 6.5. Average commuting distance (dpt), origin zones
Relationship between spatial proximity and travel-to-work distance
201
Fig. 6.6. Average commuting distance (dpt), destination zones
Fig. 6.7. Average minimum commuting distance (mdpt), origin zones
Chapter 6
202
Fig. 6.8. Average minimum commuting distance (mdpt),
destination zones
Fig. 6.9. Residential density (dens)
Relationship between spatial proximity and travel-to-work distance
203
Fig. 6.10. Job density (dens)
Fig. 6.11. Shannon index (div)
Chapter 6
204
Fig. 6.12. Jobs-housing balance (jhb)
Fig. 6.13. Number of accessible jobs (potjob)
Relationship between spatial proximity and travel-to-work distance
205
Fig. 6.14. Schematic representation of the spatial vision on Flanders.
Source: RSV (1997/2004)
Before using these data in a statistical spatial analysis, a number of
interesting findings from the maps can already be noted.
The most obvious relationships can be seen in the maps where the
zones are considered origins, or residential zones (Figs 6.5, 6.7 and 6.9).
The spatial variations of the distance travelled per trip (Fig. 6.5) is in
most regions similar to the spatial pattern of the minimum commuting
distance (Fig. 6.7), although values of the second variable are of course
considerably lower. The map showing residential density (Fig. 6.9) is in
many regions nearly the opposite of Figs 6.5 and 6.7. Therefore, there
seems to be a link between high residential densities and short commut-
ing distances. The map of the spatial diversity distribution (Fig. 6.11)
seems to indicate the same kind of relationship, at least as far as the
origin zones are considered. The low measured spatial diversity in the
province of Limburg (in the east) is noticeable.
On the maps where the zones are regarded as a destination or labour
zone (Figs 6.6 and 6.8), the patterns are less consistent. Here the thor-
Chapter 6
206
ough spatial concentration of jobs, in comparison with the more dispersed
structure of housing, plays an important role. In areas with a low job
density, the minimum commuting distance is of little significance.
A comparison of Fig. 6.10 with Figs 6.6 and 6.8 indicates that job
density is correlated with commuting distance, as well as with the
minimum commuting distance to the destination areas. For a better
insight, the jobs-housing balance (Fig. 6.12) should be taken into account.
In theory, a jobs-housing balance fluctuating around a value of 1 would
be most likely to yield short commuting distances (Peng, 1997). Because
of the large differences in spatial concentration between housing and jobs,
Fig. 6.12 does not seem to support this theory at first sight. The jobs-
housing balance therefore requires more specific attention in the following
paragraphs. In the economic core of Flanders, there also seems to be a
correlation between the accessibility of the job market (Fig. 6.13) and the
distance travelled, while this relationship is much less clear in the rest of
Flanders.
6.4.2 Correlation analysis for the average distance
travelled per trip
All the independent variables are considered approximations for spatial
proximity, in this case between the housing market and the job market.
A Pearson’s correlation matrix shows that most coupled variables in the
origin zones exhibit correlations in the range of 20-50%, amounting even
to 72% (between job density and minimum commuting distance) in the
destination zones. The couple job accessibility and jobs-housing balance is
an exception, with a correlation coefficient of only 9.7%.
Because of this inherent multicollinearity, and because of the difficul-
ties regarding interpretation, multiple regression techniques will not be
applied, but the paper confines itself to a bivariate correlation analysis.
All variables are normalized to z-scores and subsequently included in
a bivariate spatial regression using GeoDa. The average distance travelled
is always regarded as the dependent variable. For the spatial lag and
spatial error models, spatial weight matrices are constructed on the basis
of the grids, according to the so-called ‘queen’ method. This means taking
into account the influence of all adjacent cells, including those cells that
touch the studied cell at only one point, such as the adjacent squares in
chess that are covered by the queen.
Relationship between spatial proximity and travel-to-work distance
207
First, an OLS regression is performed, as well as a test on spatial
autocorrelation (through a significance test for Moran’s I), and the
presence of spatial lag and spatial error is detected. In case no significant
spatial autocorrelation is found, the results of the OLS regression are
selected. In the other case, the spatial lag or spatial error model is
applied, depending on which phenomenon is significant. If both phenom-
ena appear significant, both models are applied. Each time the regression
coefficient r, which can be considered as a spatial correlation coefficient,
and the value of the significance test are selected. The zones are first
regarded as origins and then as destinations. The results are presented in
Table 6.1.
Table 6.1. Results of the correlation analysis for the average distance
travelled per trip
Origin dens div mdpt jhb potjob
MI 0.628 0.609 0.643 0.680 0.631
pMI 0.000 0.000 0.000 0.000 0.000
rOLS (-0.318) (-0.448) (0.255) (0.040) (-0.409)
pOLS 0.000 0.000 0.000 0.237 0.000
rSL -0.098 -0.162 0.080 0.019 (-0.093)
pSL 0.000 0.000 0.000 0.322 0.000
rSE (-0.109) -0.229 0.070 0.029 -0.292
pSE 0.000 0.000 0.003 0.223 0.000
Destination dens div mdpt jhb potjob
MI 0.655 0.677 0.624 0.576 0.653
pMI 0.000 0.000 0.000 0.000 0.000
rOLS (0.209) (0.095) (0.426) (0.467) (0.292)
pOLS 0.000 0.005 0.000 0.000 0.000
rSL 0.056 0.020 0.185 0.191 (0.004)
pSL 0.004 0.302 0.000 0.000 0.829
rSE (0.045) 0.012 0.224 0.202 0.171
pSE 0.108 0.678 0.000 0.000 0.014
Note:
origin: each zone is considered as an origin
destination: each zone is considered as a destination
Chapter 6
208
between parentheses: the GeoDa model suggests that one should not
use this method
MI: Moran’s I
pMI: p-value associated with Moran’s I
rOLS: correlation coefficient (ordinary least squares
method)
pOLS: p-value associated with OLS method
rSL: correlation coefficient (‘spatial lag’-method)
(bold = significant at the 0.05-level)
pSL: p-value associated with SL-method
rSE: correlation coefficient (‘spatial error’-method)
(bold = significant at the 0.05-level)
pSE: p-value associated with SE-method
6.4.2.1 Interpretation regarding the origin zones
In this case, each zone is considered as a residential area, acting as a
source of home-to-work trips in the morning traffic. For all variables, a
significant amount of spatial autocorrelation was found, making the
application of a simple linear regression (OLS) inappropriate. The applied
spatial regression models were suggested by GeoDa. Except for the jobs-
housing balance, all variables show a significant relation with the distance
travelled per trip. An overview of the findings now follows.
• A higher population density is associated with shorter commuting
distances.
• A greater diversity is also associated with shorter commuting dis-
tances. The link with spatial diversity is much stronger than the
relationship with population density.
• The minimum commuting distance shows a positive correlation with
the observed commuting distance, the magnitude of r being similar as
for the relationship with population density.
• There is no significant correlation between the observed commuting
distance and the jobs-housing balance. Given the specific characteris-
tics of the job-housing balance, a separate section will be devoted to
this variable below.
• Better access to the job market is associated with shorter commuting
distances. In comparison with the other independent variables, this
relationship can be called strong.
Relationship between spatial proximity and travel-to-work distance
209
6.4.2.2 Interpretation regarding the destination zones
This subsection considers each zone as a labour area, a destination of the
commute during the morning. Again, significant spatial autocorrelation
for all variables is found. The calculation of the spatial regression models
suggested by GeoDa results in following findings:
• In contrast to the population density in the origin zones, a higher
density of jobs in destination zones is associated with longer commut-
ing distances. The relationship with job density is, however, less
strong than the relationship with population density in the origin
zones.
• There is no demonstrable correlation with spatial diversity.
• There is a relatively strong positive correlation between minimum
commuting distance and the observed commuting distance.
• Even if the jobs-housing balance appears not to be relevant in the
origin zones, this variable provides a relatively strong correlation in
the destination zones. The higher is the jobs-housing balance, the lar-
ger the commuting distances. This finding should, however, be
qualified (see the next section).
• The interpretation of the influence of the number of potentially
accessible jobs is not obvious, since this variable represents the num-
ber of jobs accessible departing from the considered zone, and not
from the other zones. This parameter yields a positive r. Better ac-
cess to the job market seen from the considered zone is associated
with a concentration of jobs, and thus with longer commuting dis-
tances to this zone.
6.4.3 Non-linearity in the jobs-housing balance
The use of spatial correlation analysis assumes a linear relationship
between the corresponding variables. This was assessed by the estimation
of a Lowess curve in a scatter plot, showing indeed linearity for most of
the used datasets. Lowess or Loess (locally weighted/estimated scatter-
plot smoothing) is a regression technique that applies linear least squares
regression on segments of the data, yielding a smoothly fitted curve.
However, slopes tend to flatten in the border areas of the sample curve.
This is particularly the case when dealing with extremely high densities
or large minimum commuting distances, or extreme values (both high and
low) of accessibility. Only with regard to the jobs-housing balance is the
relationship clearly not linear, especially when considering origin zones.
Chapter 6
210
On the one hand, it seems logical that a higher jobs-housing balance
leads to a larger share of nearby jobs, reducing average commuting
distances. On the other hand, it is clear that a high jobs-housing balance
in the destination zones is accompanied by an (over)concentration of jobs,
leading to larger commuting distances. Thus, a high jobs-housing balance
would be associated with both long and short commuting distances. This
paradox stems from the assumption of linearity, which appears unjusti-
fied.
When one considers the total amount of travelled distances produced
per zone, adding the distances covered by inbound trips to these covered
by the outbound trips, and plot this summed variable against the jobs-
housing balance, a ‘U’-shaped curve should be obtained (Peng, 1997)
(Fig. 6.15). The minimum distance corresponds to an equilibrium point
where the jobs-housing balance is close to 1, representing a perfectly
equal spatial distribution of housing and jobs.
To verify the validity of this hypothesis, the link between the total
generated distance per zone (4 x 4 km) and the jobs-housing balance is
visualized by means of a scatter plot for which a Lowess curve and a
best-fit quadratic curve are estimated (Fig. 6.16).
The curves appear to correspond to the hypothesis, that is, the total
commuting distance associated with one zone reaches a minimum at an
optimal jobs-housing balance, which is situated around a value of 1.
However, the fit is rather weak. This is partly because there are many
more zones with a low jobs-housing balance than with a high jobs-housing
balance. However, a more important reason is that the natural relation-
ship between spatial segregation and trip length is disturbed and
smoothed by non-spatial influences, as discussed in the Introduction. The
same reason applies to the generally low values of r that were found in
the previous section.
Relationship between spatial proximity and travel-to-work distance
211
Fig. 6.15. Expected distribution of the total amount of commuter travel
and the jobs-housing balance
Fig. 6.16. Observed distribution of the total amount of commuter travel
and the jobs-housing balance
Chapter 6
212
6.5 Discussion
6.5.1 Origin versus destination zones
For most parameters, a positive correlation with the outbound trip length
corresponds to a negative correlation with the inbound trip length. A
higher population density in an origin zone means shorter trips. A higher
job density, by contrast, just means longer trips. A similar pattern is
observed when considering the number of accessible jobs. With respect to
spatial diversity, there seems to be no discernible relationship in the
destination zones, while one should approach the jobs-housing balance in
a non-linear way.
These results derive from the unequal spatial distribution of housing
and jobs. Jobs occur more often than housing in high concentrations in
urban areas, where the density and the spatial diversity are generally
high. In most zones the jobs-housing balance is lower than a value of 1,
making job supply in the other zones, which are much less numerous,
overconcentrated. Over-concentration means that employees more often
come from very far away to take up the available jobs. On the origin side,
however, employees living in an area with urban characteristics (high
density and spatial diversity) more often find a job close to home.
Therefore, Newman and Kenworthy’s (1989, 1999) thesis that fuel
consumption is negatively correlated with density is only valid when the
residential zones are considered. In the case of the labour zones, an
inverse correlation is found. However, modal choice was not taken up in
this study. It can be assumed that zones with a very high job concentra-
tion are accessed more than average by public transport. Consequently,
energy consumption (and other undesired external effects of traffic) is
expected to grow slower than the travelled distance itself. But because
the highest job concentrations entail the most extreme forms of long-
distance commuting, this mitigating effect on fuel consumption will be
rather limited.
Spatial diversity does not explain commuting distance viewed from
the labour zones. This can be explained by the large differences in types
of labour areas: high job concentrations generating long trips occur both
in urban centres with a high spatial diversity and in isolated monofunc-
tional industrial and port areas. This is different with regard to the
residential areas, where diversity is an explaining factor: a high residen-
Relationship between spatial proximity and travel-to-work distance
213
tial density is usually associated with a large supply of jobs, and thus a
sound spatial mix of both functions.
When considering the relationship between the observed commuting
distance and the minimum commuting distance, values of r are signifi-
cantly higher for labour areas compared to residential areas. However, the
relationship with job density (in the labour areas) is weaker if compared
with the relationship with housing density (in residential areas).
When the origin zones are examined, it can be concluded that job
accessibility and spatial diversity should be considered the most appro-
priate indicators influencing commuting trip length. In this case,
residential density and minimum commuting distance have a relatively
low predictive value, while the jobs-housing balance is not a linear
predictor at all. In the destination zones, however, the minimum commut-
ing distance is a relatively good indicator, while the jobs-housing balance
and job density have only conditional predictive value.
6.5.2 Relevant scale levels
The paper has taken into account the fact that the relationships that are
derived from an analysis at the level of zones of 4 x 4 km cannot neces-
sarily be extended to processes that play at a different aggregation level.
The importance of scale may mainly depend on the considered trip-length
classes.
Density, diversity and proximity are spatial characteristics that are
important for traffic generation, but mainly on a scale level that can be
associated with the considered distance class. Influencing the travel
behaviour of motorists should be done especially at higher scale levels. To
influence the trips of pedestrians, cyclists and users of local public
transport, the lowest scale levels are perhaps the most relevant.
The travel behaviour of the twenty-first-century commuter deviates
strongly from that of the historical inhabitants of a medieval or nine-
teenth-century compact city, so that only a limited overall environmental
impact is to be expected from a policy that only focuses on strengthening
these old structures. Potentially more impact could be expected from a
policy that focuses on strengthening density, diversity, the jobs-housing
balance and proximity within the higher levels of aggregation, also
comprising suburban areas. In addition, the quality of the urban structure
at the lowest scale levels may still have an impact on the share of pedes-
Chapter 6
214
trians, cyclists and users of public urban transport, and may in that sense
have a positive impact on the urban environment.
6.6 Conclusions and pathways
for further research
Daily trips on foot, by bicycle or public urban transport, but also short
car trips, fit into the spatial model of the compact city. By strengthening
the characteristics of the compact city, with the emphasis on residential
density and proximity of functions, a spatial framework may be created
that makes trips over long distances less necessary or even unnecessary.
However, the category of short trips has become a niche market, espe-
cially within the commute. In Flanders, the average commuter covers a
distance of about 19 km per trip, a trip length that is well beyond the
compact urban scale. However, a number of qualities that are attributed
to the compact city are still valuable, even if they also influence spatial
processes at the regional scale. Viewed from a sustainability perspective,
an equal jobs-housing balance and a high residential density are para-
mount. A high degree of spatial diversity and accessibility of the labour
market also play a positive role. By contrast, high concentrations of
employment are not desirable. High job densities seem to give rise to
overconcentration easily, so that workers should be recruited from a wide
hinterland. These features are reflected in the minimum commuting
distance, which is a good measure for the proximity of housing and the
job market and clearly evolves parallel with the observed commuting
distances. Since the minimum commuting distance is determined by the
spatial patterns of housing, jobs and infrastructure, it can be concluded
that commuting over short distances can be facilitated by changes in the
spatial structure.
The outlined research design, however, also calls for further elabora-
tion. The results for the commute cannot simply be extrapolated to other
types of travel. Trips to school, to shops or social activities usually are
much shorter than home-to-work trips (Zwerts and Nuyts, 2004). There-
fore, it might be possible that the correlations for this type of travel are
stronger at the lower scale levels in comparison to home-to-work trips.
Another extension concerns the feasibility of an equal jobs-housing
balance: it has to be taken into account that a neighbourhood’s jobs-
housing balance that fluctuates area-wide around 1 may not be compati-
Relationship between spatial proximity and travel-to-work distance
215
ble with agglomeration trends that are important for many economic
sectors.
Another improvement could be incorporating the modal split into the
comparisons. Long commuting distances are often covered by train, which
means that the negative impact of these trips increases at a slower rate
than the number of kilometres travelled. Moreover, it may well be that
increasing spatial proximity at the lowest scale levels will result in a
larger share of pedestrians, cyclists and users of urban public transport in
urban areas, strengthening the positive effect of shorter trips by a modal
shift. The authors hope to address these and related issues in future
research.
References
Alberti, M. (1999) “Urban patterns and environmental performance:
What do we know?” Journal of Planning Education and Research.
19(2), pp. 151-163.
Amrhein, C. G. (1995) “Searching for the elusive aggregation effect:
Evidence from statistical simulations.” Environment and Planning A.
27(1), pp. 105-119.
Anselin, L. and A. Bera (1998) “Spatial dependence in linear regression
models with an introduction to spatial econometrics”. In: A. Ullah
and D. Giles (Eds.) Handbook of Applied Economic Statistics. New
York: Marcel Dekker, pp. 237-290.
Anselin, L. and D. A. Griffith (1988) “Do spatial effects really matter in
regression analysis?” Papers in Regional Science. 65(1), pp. 11-34.
Anselin, L., I. Syabri and Y. Kho (2006) “GeoDa: An introduction to
spatial data analysis.” Geographical Analysis. 38(1), pp. 5-22.
Banister, D., S. Watson and C. Wood (1997) “Sustainable cities, trans-
port, energy and urban form.” Environment and Planning B. 24(1),
pp. 125-143.
Batty, M. (1974) “Spatial entropy.” Geographical Analysis. 6(1), pp. 1-
31.
Boussauw, K., T. Neutens and F. Witlox (2011) “Minimum commuting
distance as a spatial characteristic in a non-monocentric urban sys-
tem: The case of Flanders.” Papers in Regional Science. 90(1), in
press.
Chapter 6
216
Boussauw, K. and F. Witlox (2009) “Introducing a commute-energy
performance index for Flanders.” Transportation Research Part A.
43(5), pp. 580-591.
Burton, E. (2000) “The potential of the compact city for promoting social
equity”. In: K. Williams, E. Burton and M. Jenks (Eds.) Achieving
Sustainable Urban Form. London and New York: E&FN Spon, pp.
19-29.
Cervero, R. (1989) “The jobs-housing balance and regional mobility.”
Journal of the American Planning Association. 55(2), pp. 136-150.
Cervero, R. (1996) “Mixed land-uses and commuting: Evidence from the
American housing survey.” Transportation Research Part A. 30(5),
pp. 361-377.
Cervero, R. and K. Kockelman (1997) “Travel demand and the 3Ds:
Density, diversity, and design.” Transportation Research Part D.
2(3), pp. 199-219.
Charron, M. (2007) “From excess commuting to commuting possibilities:
More extension to the concept of excess commuting.” Environment
and Planning A. 39(5), pp. 1238-1254.
Cliff, A. D. and J. K. Ord (2009) “What were we thinking?” Geographical
Analysis. 41(4), pp. 351-363.
Commission of the European Communities (1990) Green Paper on the
Urban Environment - COM(90)218. Brussels: Commission of the
European Communities.
Crane, R. and R. Crepeau (1998) “Does neighborhood design influence
travel? A behavioral analysis of travel diary and GIS data.” Trans-
portation Research Part D. 3(4), pp. 225-238.
Dodson, J. and N. Sipe (2008) “Shocking the suburbs: Urban location,
homeownership and oil vulnerability in the Australian city.” Housing
Studies. 23(3), pp. 377-401.
Ellis, C. (2002) “The new urbanism: Critiques and rebuttals.” Journal of
Urban Design. 7(3), pp. 261-291.
Frank, L. D. and G. Pivo (1994) “Impacts of mixed use and density on
utilization of three modes of travel: Single-occupant vehicle, transit,
and walking.” Transportation Research Record. 1466, pp. 44-52.
Hamilton, B. W. (1982) “Wasteful commuting.” Journal of Political
Economy. 90(5), pp. 1035-1053.
Horner, M. W. (2004) “Spatial dimensions of urban commuting: A review
of major issues and their implications for future geographic research.”
The Professional Geographer. 56(2), pp. 160-173.
Relationship between spatial proximity and travel-to-work distance
217
Horner, M. W. and A. T. Murray (2003) “A multi-objective approach to
improving regional jobs-housing balance.” Regional Studies. 37(2),
pp. 135-146.
Jenks, M., E. Burton and K. Williams (1996) “Compact cities and
sustainability: An introduction”. In: M. Jenks, E. Burton and K. Wil-
liams (Eds.) The Compact City: A Sustainable Urban Form? London
and New York: E&FN Spon, pp. 3-8.
Lyons, G. and K. Chatterjee (2008) “A human perspective on the daily
commute: Costs, benefits and trade-offs.” Transport Reviews. 28(2),
pp. 181-198.
Ma, K. R. and D. Banister (2006) “Excess commuting: A critical review.”
Transport Reviews. 26(6), pp. 749-767.
Ma, K. R. and D. Banister (2007) “Urban spatial change and excess
commuting.” Environment and Planning A. 39(3), pp. 630-646.
Mindali, O., A. Raveh and I. Salomon (2004) “Urban density and energy
consumption: A new look at old statistics.” Transportation Research
Part A. 38(2), pp. 143-162.
Ministry of Housing, Spatial Planning and the Environment (1988)
Vierde Nota over de Ruimtelijke Ordening. Den Haag: Ministry of
Housing, Spatial Planning and the Environment.
Nagendra, H. (2002) “Opposite trends in response for the Shannon and
Simpson indices of landscape diversity.” Applied Geography. 22(2),
pp. 175-186.
National Geographical Institute of Belgium (2009) Structuur en codering
van de gegevens top10v-gis en top50V-gis. National Geographical In-
stitute of Belgium from
http://www.ngi.be/Common/articles/CA_Top10V-GIS_TOP50V-
GIS/restruct_doc_nl.htm.
Neuman, M. (2005) “The compact city fallacy.” Journal of Planning
Education and Research. 25(1), pp. 11-26.
Newman, P. and J. Kenworthy (1989) Cities and Automobile Depend-
ence. A Sourcebook. Aldershot: Gower.
Newman, P. and J. Kenworthy (1999) Sustainability and Cities: Over-
coming Automobile Dependence. Washington, DC: Island Press.
Niedzielski, M. A. (2006) “A spatially disaggregated approach to com-
muting efficiency.” Urban Studies. 43(13), pp. 2485-2502.
Openshaw, S. and P. Taylor (1979) “A million or so correlation coeffi-
cients: Three experiments on the modifiable areal unit problem”. In:
Chapter 6
218
N. Wrigley (Eds.) Statistical Methods in the Spatial Sciences. Lon-
don: Routledge and Kegan Paul, pp. 127-144.
Peng, Z. R. (1997) “The jobs-housing balance and urban commuting.”
Urban Studies. 34(8), pp. 1215-1235.
Rietveld, P., B. Zwart, B. van Wee and T. van den Hoorn (1999) “On
the relationship between travel time and travel distance of commut-
ers. Reported versus network travel data in the Netherlands.” Annals
of Regional Science. 33(3), pp. 269-287.
RSV (1997/2004) Ruimtelijk Structuurplan Vlaanderen - Gecoördineerde
Versie. Brussels: Ministry of the Flemish Community.
Schwanen, T. and P. L. Mokhtarian (2005a) “What if you live in the
wrong neighborhood? The impact of residential neighborhood type
dissonance on distance traveled.” Transportation Research Part D.
10(2), pp. 127-151.
Schwanen, T. and P. L. Mokhtarian (2005b) “What affects commute
mode choice: Neighborhood physical structure or preferences toward
neighborhoods?” Journal of Transport Geography. 13(1), pp. 83-99.
Stead, D., J. Williams and H. Titheridge (2000) “Land use, transport and
people: Identifying the connections”. In: K. Williams, E. Burton and
M. Jenks (Eds.) Achieving Sustainable Urban Form. London and New
York: E&FN Spon, pp. 174-186.
Tobler, W. R. (1970) “A computer movie simulating urban growth in the
Detroit region.” Economic Geography. 46 (supp.), pp. 234-240.
Van Acker, V. and F. Witlox (2011) “Commuting trips within tours: How
is commuting related to land use?” Transportation. doi:
10.1007/s11116-010-9309-6.
Van Acker, V., F. Witlox and B. van Wee (2007) “The effects of the land
use system on travel behavior: A structural equation modeling ap-
proach.” Transportation Planning and Technology. 30(4), pp. 331-
353.
Van Wee, B. (2002) “Land use and transport: Research and policy
challenges.” Journal of Transport Geography. 10(4), pp. 259-271.
Vandenbulcke, G., T. Steenberghen and I. Thomas (2007) Accessibility
Indicators to Places and Transports - Final Report. Leuven / Lou-
vain-la-Neuve: Federal Public Service of Mobility / Federal Science
Policy.
Vandenbulcke, G., T. Steenberghen and I. Thomas (2009) “Mapping
accessibility in Belgium: A tool for land-use and transport planning?”
Journal of Transport Geography. 17(1), pp. 39-53.
Relationship between spatial proximity and travel-to-work distance
219
Verhetsel, A., I. Thomas, E. Van Hecke and M. Beelen (2007) Pendel in
België. Deel I: De Woon-Werkverplaatsingen. Brussels: Statistics Bel-
gium.
Wang, F. (2000) “Explaining intraurban variations of commuting by job
proximity and workers’ characteristics.” Environment and Planning
B. 28(2), pp. 169-182.
Wang, F. (2003) “Job proximity and accessibility for workers of various
wage groups.” Urban Geography. 24(3), pp. 253-271.
Williams, K., E. Burton and M. Jenks (2000) “Achieving sustainable
urban form: An introduction”. In: K. Williams, E. Burton and M.
Jenks (Eds.) Achieving Sustainable Urban Form. London and New
York: E&FN Spon, pp. 1-5.
Witlox, F. (2007) “Evaluating the reliability of reported distance data in
urban travel behaviour analysis.” Journal of Transport Geography.
15(3), pp. 172-183.
Zwerts, E. and E. Nuyts (2004) Onderzoek Verplaatsingsgedrag Vlaande-
ren 2000-2001. Brussels-Diepenbeek: Ministry of the Flemish
Community.
Chapter 6
220
221
Chapter 7:
Linking expected
mobility production to
sustainable residential
location planning
This paper will be published as Boussauw, K. and F. Witlox (2011)
“Linking expected mobility production to sustainable residential location
planning: Some evidence from Flanders.” Journal of Transport Geogra-
phy. Copyright © Elsevier. All rights reserved.
Abstract
Based on a set of spatial proximity characteristics this paper develops a
model that estimates for every neighbourhood in Flanders (Belgium) the
amount of traffic that would be generated by an additional residential
unit when socio-economic variables are held constant. The results show
that residential density, land use diversity and proximity of facilities
influence daily travelled distances when these variables are measured in
the immediate vicinity of the residential location of the respondent
(within a radius of 1 km). When aggregating these variables at a larger
geographical scale, in most cases the impact proves no longer significant.
Variables based on the spatial distribution of jobs, or on the global
accessibility of the entire population in the study area, do not show any
significant effects on the travel distance.
Despite the statistical significance only a fraction of the observed
variance in reported distances is explained by characteristics of spatial
proximity. However, we can assume that the importance of spatial
structure in the genesis of mobility patterns will increase in case the cost
of transport would rise (cf. peak oil). For this reason, the application of
Chapter 7
222
the mapped results of the proposed model could contribute to the prac-
tice of sustainable spatial planning.
Keywords: spatial proximity; travel behaviour; sustainable spatial
development; Flanders
7.1 Introduction
Research into the relationship between spatial structure and travel
behaviour exists in many forms. Scholars typically focus on the search for
statistical associations between aspects of travel behaviour (such as choice
of mode or destination, travel time or trip length) and spatial characteris-
tics (such as density and degree of mix of homes, jobs and other facilities,
or mere morphological features such as street patterns and neighbourhood
layout). Ewing and Cervero (2010) present an extensive literature review
on this.
An important part of the existing research in this field focuses on the
potential application of the obtained results in the development of a more
sustainable urban and regional spatial structure that can operate on the
basis of minimal energy needs for transport (Ewing et al., 2008). Al-
though many policy plans still refer to Newman and Kenworthy (1989,
1999), who argue that there is a strong inverse relationship between
population density and transport energy consumption per capita, later
research shows that this statement is a serious simplification. Criticisms
of Newman and Kenworthy (1989, 1999) (Mindali et al., 2004; Mees,
2010, pp. 24-26) rely mostly on methodological issues, such as the chosen
demarcation of the studied cities, while quantitative research into the
relationship between spatial characteristics and energy consumption in a
regional network structure finds much more complex interactions (Bous-
sauw et al., 2011a). Energy consumption by transport is partly
determined by the modal split, and partly by the total distance travelled
within the studied system. Previous research shows that in regional
studies (which go beyond the urban scale) in a western context, the daily
distance travelled per person is a good approximation of sustainability of
travel patterns (Boussauw and Witlox, 2009), while the influence of
modal split is only secondary. In the following sections, we will therefore
focus on the relationship between distance travelled and land use charac-
teristics that measure mutual proximity between possible destinations.
Linking expected mobility production to residential location planning
223
In this over the years adequately documented line of research, we can
distinguish two important constants: i.e., (i) the assumed relationships
always appear to be statistically significant, but (ii) explain only a small
share of the observed variance. The first of these two findings is actually
trivial: it would be quite remarkable if the influence of the spatial distri-
bution of different types of destinations, which among others defines
mutual distances that need to be covered, would not pass significance
tests (Næss, 2003). The second finding, however, is a lot less comforting:
the explained variance (in many analyses represented by the coefficient of
determination (R2) of a regression equation) is usually very low (Handy
et al., 2005a; Cervero and Kockelman, 1997; Cervero, 1996; Næss and
Sandberg, 1996). Obviously, this means that spatial characteristics
explain travel behaviour to an only very limited extent.
In socio-geographically inspired research, spatial features are usually
only one of the considered clusters of explanatory variables in the model.
By combining many socio-demographic and economic variables (such as
income, car ownership, family composition, lifestyle or job preference)
with spatial characteristics, a relatively satisfactory fit may be obtained
(Van Acker and Witlox, 2011; Maat and Timmermans, 2006). An
advantage of this approach is the accurate estimation of the model
coefficients since the influence of any potential correlation between spatial
and socio-economic variables is filtered out. An example of such a
correlation is the inverse relationship between income class and residen-
tial density. A major drawback of upgrading a spatial model to a socio-
economic model to explain travel behaviour is that the influence of the
spatial structure, which is present anyway, seems to fade into the back-
ground.
A model built on mere spatial features is nevertheless useful for spa-
tial policy. Although spatial characteristics explain only a small part of
the assessed travel patterns, the built environment is still determining the
physical preconditions for sustainable mobility patterns. Moreover, we
argue that the importance of the spatial component in the genesis of
travel patterns is not constant throughout history, but is linked to the
cost and the speed of mobility. Over the centuries, the absolute cost to
move an individual over a distance of one kilometre has almost continu-
ally been decreasing, if we neglect the slight ripples in the cost curve
during the oil crises in the seventies. Moreover, the average speed of
travel has been continuously increasing world wide (Schafer, 2000), a
phenomenon that is largely explained by the growth in car ownership and
Chapter 7
224
the extension of the road network. Both developments have led to a
systematic decline of the importance of physical distance between poten-
tial destinations (Rietveld and Vickerman, 2004), which in turn resulted
in the weakening of the transport-land use connection (Giuliano, 1995).
In statistical analyses based on spatial characteristics this phenomenon is
reflected in a low coefficient of determination.
The continuing decline in transport costs is only possible through the
abundant availability of energy in the form of fossil fuels and is therefore
finite (Wegener, 2010). According to the peak oil theory (Witze, 2007)
the relative cost of oil products may significantly increase over time,
leading to a reduction in mobility and a growing importance of mutual
spatial proximity of destinations (Dodson and Sipe, 2008). The propor-
tionately small share of the variance in travel patterns that is explained
by spatial structure should not be considered unimportant. It is exactly
physical space that is the most rigid component, and thus the slowest to
adapt to changing economic conditions, in contrast to e.g. behavioural
elements that are more subject to an individual’s choice.
The aim of the current research is the development of a model for the
study area of Flanders (Belgium), based on mere spatial characteristics,
that indicates what level of mobility production (expressed as daily
distance travelled per individual) is associated with the location of an
additional housing unit in a certain area. The use of the residential
location as a reference for the study is taken by the abundance of avail-
able residence-based travel data. We use data from the 2007-2008 Travel
Behaviour Survey for Flanders (Janssens et al., 2009) and a number of
additional data sets containing spatial variables.
In a first phase, the relationship between characteristics of spatial
proximity, as measured in the area of residence, and individually reported
daily travel distances is assessed through a linear regression model. The
model accounts for variability related to the applied aggregation level by
incorporating various geographical scale levels. In a second phase, results
obtained from the regression analysis are used to construct a map which
represents for each statistical ward the expected mobility production by
an inhabitant in this location if only spatial variables are taken into
account. Obviously, the model will explain only a limited share of the
expected variance, an aspect which should be taken into account in the
interpretation. This means that the applicability of the model is largely
relying on the assumption that observed relationships may become
stronger in the future.
Linking expected mobility production to residential location planning
225
Unlike the usual approach taken to the assessment of land use-
transport connections, which aims to detect statistical relationships, our
study expands the findings immediately to an application that is useful in
location policy at a regional scale level.
7.2 Study area
The research focuses on the Flanders region, which is together with the
Brussels Capital Region composing the north of Belgium. The main
borders of the Flanders region are constituted by the North Sea, the
Netherlands and the Walloon region (south of Belgium). The Brussels
Capital Region, which has over one million inhabitants, is the largest
agglomeration in the region, and is in geographical terms centrally
located in Flanders.
In addition to Brussels, the metropolitan areas of Antwerp (400,000
inhabitants) and Ghent (250,000 inhabitants) are located in Flanders, as
are ten regional cities (with a population of around 100,000 inhabitants)
and a series of smaller urban centres and municipalities.
An interesting, typical Belgian, aspect is found in the history of the
institutionalized commute, through government support for the construc-
tion of an extended railway network and cheap commuter tickets, aiming
for the industrialization of the country based on a minimum of urbaniza-
tion (Verhetsel et al., 2010). In the 19th century and early 20th century,
this policy led to a clustering of housing and amenities in the towns and
villages that were connected to the railway network. After World War II,
these structures have fanned out into car-oriented suburban develop-
ments, an evolution that is associated with ever increasing mutual
distances between homes, jobs and daily facilities, and has created a
major source of dispersed traffic (Boussauw et al., 2011b).
7.3 Methodology and data
7.3.1 Analysis and model structure
The objective of the paper is to develop a model that forecasts regional
variations in mobility production based on characteristics of spatial
proximity at the residential location. We use regression analysis, with
daily kilometrage per person as the dependent variable. Explanatory
Chapter 7
226
variables consist of a number of measures of spatial proximity that are
observed at various aggregation levels around the individual residential
locations. In addition, a number of socio-economic variables are used as
control variables. The applied data sets are described below.
We start from a full model that includes all considered variables.
Then, we trim the model and ultimately only retain those variables and
scale levels that show statistically significant. If necessary, transforma-
tions are applied to address potential deviations from the normal
distribution or prominent non-linear relationships.
After building and trimming the model, the obtained equation is used
to estimate the mobility generating character of each neighbourhood (i.e.
census ward) in Flanders. For each ward the relevant spatial variables are
recalculated, from which the expected daily number of generated kilome-
tres per person is regressed. These values are then displayed in the form
of a map. When interpreting the map, it is important to realize that the
extent to which spatial structure explains the mobility of a (new) resident
of any area is indicated by the coefficient of determination (R2) of the
regression equation.
7.3.2 Dependent variable (PKM)
The daily kilometrage per person is used as the dependent variable. The
data source is the Travel Behaviour Survey for Flanders (OVG3)
(Janssens et al., 2009). OVG3 is a mobility survey conducted during
2007-2008 in 8,800 respondents over the age of 6 years and living in the
Flanders region (excluding the Brussels Capital Region). The selection is
based on a sample from the national register. The home address of the
respondents is recorded. Respondents are asked to keep track of all their
trips during a predetermined random day by means of a travel diary. Of
the 8,800 respondents, 7,273 have actually moved on that day, and have
reported the perceived distance covered by their trips. In our analysis we
use the sum of the lengths of all trips reported by the respondent.
Because of the nature of the data possible biases inherent in the use of
travel diaries should be taken into account (Witlox, 2007).
For the sake of calculating the values of the explanatory spatial vari-
ables, address data was geocoded (converted into XY coordinates) using
the Google Maps web service. An overview of the residential locations of
the respondents is given by Fig. 7.1.
Linking expected mobility production to residential location planning
227
Fig. 7.1. Situation of the residential location of the respondents
in the study area
7.3.3 Explanatory variables
A total of six explanatory variables have been selected (in addition to the
control variables, which are discussed subsequently), each of which can be
considered as a measure for the mutual spatial proximity with regard to
potential destinations. The variables are: (i) accessibility, (ii) residential
density, (iii) land use diversity, (iv) job density, (v) minimum commuting
distance, and (vi) proximity of facilities. The construction of these
variables is explained in the following paragraphs.
Since we are using spatially aggregated data, the modifiable areal unit
problem (MAUP) (Openshaw and Taylor, 1979) should be taken into
account. To reduce distortion of the results by the influence of the spatial
scale at which data is aggregated, each variable has been determined at
three different levels of aggregation. To this end, per respondent three
circular zones have been drawn of which the midpoint is the reported
residential location, with a respective radius equalling 1 km, 4 km and 8
km. Although the choice of these three levels may seem rather arbitrary,
the analysis will clearly show that the used variables do only have clear
impact at the lowest scale level. Consequently, the two higher aggrega-
tion levels will be removed from the regression, thus avoiding
multicollinearity problems. At the other hand, given the resolution of the
Chapter 7
228
basic data sets, using an even smaller zone than a circle with r = 1 km
would not be justified.
Within these circles, data is then averaged on the basis of the propor-
tional overlap with the original zones associated with the used data sets
(these are census wards, traffic analysis zones (TAZ’s) and a one kilome-
tre square grid respectively).
7.3.3.1 Accessibility (ACC)
To define regional geographical accessibility, we start from the 2007
population data in Flanders and Brussels, aggregated by census ward. A
distance matrix is calculated between each possible pair of census wards,
based on a shortest path calculation over the road network (Streetnet).
Finally, for each census ward, the total distance that should be covered to
visit each resident of any other census ward in the study area once and
return back home, is summed. This accessibility index thus gives a
measure of the interaction opportunities with all other inhabitants of
Flanders and Brussels, based on physical distance.
A disadvantage of this measure is that neighbouring countries and
regions are not accounted for. However, the nature of a cumulative
accessibility measure requires a clear delineation of the study area. The
applied boundary is justified by rather strong language and cultural
differences making daily travel crossing the outer borders of the Flanders
region relatively rare. For example, in 2007, only 2.0% of Flanders’
employed labour force worked in Wallonia and only 1.5% worked abroad.
(Policy Research Centre on Work and Social Economy, 2010) Also, since
this is a distance based accessibility measure, it does not take into
account variations in travel speed e.g. due to choice options in travel
mode and route, or the presence of congestion.
7.3.3.2 Residential density (POPD)
The residential density is based on government population data for 2007,
aggregated by census ward in Flanders.
7.3.3.3 Land use diversity (DIV)
To approximate the degree of land use mix, the Strucnet file of the
National Geographical Institute of Belgium (2009) was used, containing
all buildings that are represented by the official topographic maps with
Linking expected mobility production to residential location planning
229
scale 1:10,000. The buildings are divided into categories. Although the
accuracy of the categorization is limited, this inventory can be used to
calculate approximate land use diversity in a given area. Since this
dataset contains a functional classification and is available at a high
resolution, it prevails on satellite imagery and is thus the best area
covering dataset currently available.
To calculate spatial-functional diversity, we employ the Shannon in-
dex. This index is used in landscape ecology as a measure of
morphological diversity (Nagendra, 2002), and is sometimes called spatial
entropy (Batty, 1974). The calculation was done for a square grid based
on an area of 1 km2, after which results were proportionally aggregated
within the three described circular zones. In this way, the possible
additional bias caused by the property of the Shannon index to increase
with larger area coverage is avoided.
7.3.3.4 Job density (JOBD)
Job density is based on commuting data as provided by the Multimodal
Model for Flanders (MMM, version 2007). MMM is a simulation of all
personal trips in the Flanders region formatted as an origin-destination
(OD) matrix and is based on a combination of various sources of socio-
economic data. MMM aggregates arrivals of all commuting trips between
4 am and 11 am in the morning traffic within TAZ’s, which are compara-
ble to, but typically slightly larger than, census wards.
7.3.3.5 Minimum commuting distance (MCD)
This variable was constructed based on the OD-matrices for commuting
between 4 am and 11 am, as they were simulated in the MMM. The
principle of the method implies that any departure (in this case in the
morning traffic) is linked to the nearest possible arrival (also in the
morning traffic). Per TAZ, the number of departures as well as the
number of arrivals are retained, but the in reality existing tie between
origins and destinations is cut in order to minimize the total distance
travelled within the system. This theoretical exercise provides a good
measure of the spatial proximity between the housing market and the
labour market. The data are results provided by Boussauw et al. (2011c),
where details on the calculation can be found.
Chapter 7
230
7.3.3.6 Proximity to facilities (SPROX)
This variable was constructed based on the spatial distribution of non-
work related destinations that are often visited by an average Flemish
household, such as schools, shops, cafes, sports clubs, banks, and medical
services. Per census ward the minimum distance was calculated that
needs to be covered by an average Flemish family to get its weekly
programme done when always opting for the closest facility within each
destination class. This weekly programme for an average family was
determined based on data from the second phase of the Travel Behaviour
Survey for Flanders (OVG2) (Zwerts et al., 2004). The data are results
provided by Boussauw et al. (2011d), to which we refer for further
calculation details.
7.3.4 Control variables
The OVG3 (Janssens et al., 2009) contains a number of socio-economic
data that may explain part of the variance in the reported distance.
These variables are: education level (EDU), income level (INC), age
(AGE) and gender (GND). We include these in the model as control
variables. This means that our research does not focus on the explanatory
power of these socio-economic variables, although it is supposed that they
make the regression equation more fitting. The selected control variables
all exhibit a statistically significant relationship with the reported travel
distance and make an important contribution to the model fit.
Education and income levels are included as continuous variables.
Because of the assumed non-linear influence of the respondent’s age, the
age variable is recoded into four dummy variables. Following categories
are considered: 0-19 years, 20-39 years, 40-59 years and 60-79 years, while
80 years or older is used as the reference category. Gender is obviously a
dummy variable; male is considered as the reference group.
7.4 Analysis
Based on the described variables, a multivariate linear regression equa-
tion has been framed. A logarithmic transformation was applied on the
dependent variable PKM, resulting in an adequate approximation of the
normal distribution. After an exploratory test for the presence of non-
linear relationships, a linear regression appeared to provide the best
Linking expected mobility production to residential location planning
231
match with reality if the non-linear effect of the age of the respondent is
modelled by means of dummy variables. As mentioned, the six explana-
tory variables were repeatedly constructed at three separate levels of
aggregation (circles with r = 1 km, r = 4 km and r = 8 km). Ultimately,
the basic equation is composed of eighteen independent variables and four
control variables, and is expressed formally:
εγγγ
γγγγ
βββ
βββα
+⋅+⋅+⋅+
⋅+⋅+⋅+⋅+
⋅+⋅+⋅+
⋅+⋅+⋅+=
−−−−
===
===
∑∑∑
∑∑∑
INCEDUGND
AGEAGEAGEAGE
SPROXMCDJOBD
DIVPOPDACCPKM
rrr
rrr
rrr
rrr
rrr
rrre
765
7960459403392021901
8,4,16
8,4,15
8,4,14
8,4,13
8,4,12
8,4,11)(log
(7.1)
For most spatial variables, no significant effects are yielded. In particular,
those variables that are constructed on the same basis but at a different
level of aggregation (e.g., ACC1, ACC4 and ACC8) appear to be highly
correlated and thus causing effects of multicollinearity. The best results
are achieved by applying only the first level of aggregation (circles with r
= 1 km). Subsequently the variables related to the spatial distribution of
jobs (JOBD1 and MCD1) do not significantly affect the results. Although
this outcome is unexpected, it can be explained by the small proportion
of today’s commuter traffic in the total number of trips (20.6%) and total
distance travelled (34.5%) (Janssens et al., 2009). Also the accessibility
variable ACC1 was excluded from the equation, since no significant
correlation between ACC1 and loge(PKM) was found. The purified
regression equation is as follows:
εγγγ
γγγγ
βββα
+⋅+⋅+⋅+
⋅+⋅+⋅+⋅+
⋅+⋅+⋅+=
−−−−
INCEDUGND
AGEAGEAGEAGE
SPROXDIVPOPDPKMe
765
7960459403392021901
131211)(log
(7.2)
The results of the regression analysis are given in Table 7.1. The results
are consistent with the literature: significances are satisfactory (all results
are within the 0.01 confidence level) at a low coefficient of determination
(R2 = 14.3%). The relationships found meet the expectations. A higher
population density and a higher degree of spatial diversity are associated
with shorter travel distances. Also, a larger minimum distance to reach
Chapter 7
232
daily facilities is associated with shorter real travel distances. The age
group between 20 and 59 years exhibits the most intensive travel pattern,
while women are less mobile than men. Both a higher level of education
and a higher income are associated with increased mobility.
Table 7.1. Coefficients of the regression analysis
R2 = 0.143 coefficient p-value
(constant) 1.502 0.000
POPD1 -3.99 . 10-5 0.000
DIV1 -0.278 0.001
SPROX1 0.004 0.000
AGE0-19 0.847 0.000
AGE20-39 1.066 0.000
AGE40-59 0.969 0.000
AGE60-79 0.624 0.000
GND -0.245 0.000
EDU 0.173 0.000
INC 0.111 0.000
The relatively small share of the observed variance that is explained by
the model, is common for mobility research. Although this phenomenon is
in part due to data deficiencies (including underreporting and randomiza-
tion of reporting days), the truth lies perhaps in the importance of the
many random factors that form the underlying reason for a significant
share of individual trips, but are difficult or even impossible to model. An
example of this is the so-called random taste variation that is accounted
for in many discrete choice modelling techniques (Train, 2003, p. 46). In
Flanders, we find similar difficulties in travel behaviour modelling
attempts in Witlox and Tindemans (2004).
If we redo the regression analysis on the basis of only the control
variables, then we obtain an R2 equalling 11.9%. The same analysis based
on only the spatial variables yields an R2 of 2.1%. This means that within
this last, reduced, model only 2.1% of the observed variance in distance
travelled could be explained by characteristics of spatial proximity of the
residential environment of the respondent. As explained in the introduc-
tion, this is the restrictive but nevertheless relevant context within which
the results should be interpreted.
Linking expected mobility production to residential location planning
233
The sum of the two coefficients of determination of both reduced
models is less than 14.3%, which indicates the occurrence of suppression
and suggests that combining the two sets of explanatory variables (socio-
economic and spatial) indeed yields some added value. However, it is
clear that the explanatory value of spatial variables largely subordinates
to that of the socio-economic variables. This means, for example, that
lowering the average income would be more effective in combating
excessive mobility than increasing housing density.
7.5 Forecasting model for Flanders
In order to develop a forecasting, area covering, model based on the
results of the regression analysis, we isolate the spatial variables. To this
end, the control variables are made constant by equalling these to the
mean value of the considered variable in the dataset. Formally:
133.3
111.0173.0245.0624.0
969.0066.1847.0502.1
7960
59403920190
=
⋅+⋅+⋅−⋅+
⋅+⋅+⋅+=
−
−−−
INCEDUGNDAGE
AGEAGEAGEctrlα
(7.3)
Based on the regression coefficients for the spatial variables the expected
amount of generated kilometres per inhabitant PKMw for each census
ward in Flanders w is determined as follows:
)004.0278.00.0000399133.3exp( wwww SPROXDIVPOPDPKM ⋅+⋅−⋅−=
(7.4)
The mapped result is shown in Fig. 7.2. The expected amount of daily
generated kilometres per inhabitant based on characteristics of spatial
proximity and averaged by census ward is approximately normally
distributed and is characterized by the values that are shown in Table
7.2.
Chapter 7
234
Fig. 7.2. Spatial distribution of the estimated daily generated mobility
per capita based on characteristics of spatial proximity
Table 7.2. Features of the distribution of daily generated mobility per
capita as expected by the model, based on census wards in Flanders
N = 9205 km
km 05% percentile 15.3
mean 23.0 25% percentile 20.2
median 23.0 75% percentile 25.8
standard deviation 5.1 95% percentile 30.1
The 95-percentile value is almost twice as large as the 5-percentile value.
This means that based on characteristics of spatial proximity, the 5%
best-located census wards are estimated to generate only half of the
mobility of the 5% worst-located wards.
As expected, and as shown in Fig. 7.2, urban areas yield the lowest
values, particularly in the historical city centres and a number of 19th-
century neighbourhoods in Ghent and Antwerp. Among the regional
urban areas mainly Leuven, Mechelen, Aalst, Brugge and Oostende score
well. Also the edge of the Brussels conurbation scores quite well, although
the agglomeration effect decays rapidly while moving away from the
centre of the capital. When we examine regions instead of cities, we see
that typically rural areas as well as green and wooded areas with scat-
Linking expected mobility production to residential location planning
235
tered development score badly. Conversely, the immediate vicinity of
large agglomerations score well, just as the highly suburbanized areas
Kortrijk-Leie (in the south-west) and the so-called Flemish Diamond (the
area cornered by Ghent, Antwerp, Leuven and Brussels).
7.6 Discussion
The results can be summarized as follows. Residential density, land use
diversity and proximity of facilities affect the daily distance travelled if
these variables are measured in the immediate vicinity of the residential
location of the respondent (within a radius of 1 km). When these vari-
ables are aggregated at a higher geographical scale, the impact is no
longer significant in most cases. This is also the case for variables that are
based on the overall accessibility of the entire population in the study
area. From this we can deduce that the overall travel pattern of an
average resident of Flanders is to a larger extent determined by local
accessibility of possible destinations than by regional accessibility or the
embeddedness in the wider region.
In addition, also variables that are based on the spatial distribution
of jobs do not show a significant impact on travel distance. This latter
finding is somewhat surprising given the adequate volume of literature
that is specifically focusing on the relationship between commuting and
the spatial distribution of jobs and housing. A comprehensive overview is
given by Horner (2004). Although the commute continues to represent a
significant share of overall travel, it should not be forgotten that a large
part of the population does not commute. Moreover, the share of the
commute in the total traffic volume is systematically decreasing (Pisarski,
2006, p. 2). Also, non-business trips tend to be considerably influenced by
the local supply of potential destinations (Handy et al., 2005b), whereas
commuting is only limitedly influenced by the local supply of jobs. For
the average resident of Flanders, the specific spatial structure of the job
market plays a relatively minor role compared to more generalized
characteristics of overall spatial proximity.
The results of the regression analysis indicate that only a fraction of
the observed variance in distance travelled is explained by spatial fea-
tures. A combination of some very basic socio-economic characteristics
(education, income, age and gender) is already explaining a much larger
Chapter 7
236
share of the variance. Nevertheless, both sets of explanatory variables are
clearly complementary.
The low proportion of variance explained by a model that contains
only spatial variables (R2 = 2.1%) entails in practice the risk of neglect-
ing the importance of spatial structure as a framework for the genesis of
potentially sustainable travel patterns. However, we argue that the
importance of the spatial distribution and thus the mutual distance
between potential destinations has decreased in the course of history as
the cost of transport fell. The cheaper transport is, the larger individual
freedom is in choosing a particular destination from a range of potential
locations where the occurred need can be fulfilled. The more expensive
transport is, the more often the nearest potential destination will be
chosen (Handy et al., 2005b). In the case where transport is expensive,
the distance from the residential location to this nearest facility, which is
derived from the spatial distribution of the whole range of potential
destinations, will largely determine the distance travelled by the consid-
ered individual.
The spatial separation of destinations, a trend which is often desig-
nated as “sprawl” is based on a rapid decline in transport costs combined
with an increase in travel speed (Ewing, 1994). However, based on the
peak oil theory (Witze, 2007), in time, an increase in transport costs is
expected due to oil scarcity. Although there is little discussion on the fact
that future fossil energy supply will have difficulties in matching global
demand, many uncertainties are present. Both the point in time when
peak oil will occur and the severity of the economic implications are
unclear. In addition, technology that reduces the oil dependence of the
transport system, such as the ongoing development and implementation
of the electric car, may mitigate these effects (Van Ruijven and Van
Vuuren, 2009). However, IEA (2008) estimates that the oil dependence of
the global transport system will only drop from 95% (in 2006) to 92% in
2030, mainly through biofuel substitution. So, even after taking into
account the development of alternative fuels, it is very likely that the
energy component of the cost of transport will gradually increase, with
implications for accessibility. Anticipating this by recognizing the princi-
ple of spatial proximity in the practice of spatial planning is thus
important.
Although much literature has yet emphasized the relationship be-
tween spatial characteristics and travel distance, in this paper we have
extrapolated the results of the analysis to a model that calculates and
Linking expected mobility production to residential location planning
237
visualises in which areas an additional residential unit would contribute
the least to mobility growth. Not unexpectedly, the most urbanized areas
turn out to be the most resilient and sustainable locations. This means
that a further increase of residential density and land use mix in urban
areas is the best guarantee for curbing excessive mobility and preparing
for the end of cheap oil. However, this conclusion requires some qualifica-
tion: there are limits to increasing density and land use mix targeted to
sustainable mobility patterns, primarily by environmental standards and
social desirability (Gordon and Richardson, 1997). Moreover, our results
suggest that a policy of compact development - compared to less steered,
dispersed development - will yield an only marginal direct impact on the
distances travelled. This finding is confirmed by numerous policy studies
(e.g. Echenique et al., 2009, p. 81).
7.7 Conclusion and directions
for further research
Although the influence of locally measurable features of spatial structure
on the distances travelled by individuals is statistically significant, the
explanatory power is small. However, in a context of possibly increasing
transport cost, but also of climate policies and congestion problems, this
does not mean that the spatial component is unimportant. Moreover, the
explanation that is provided by spatial variables is complementary to the
variance explained by socio-economic variables.
Using the analysis results in a policy instrument that can be used for
steering additional residential development is therefore useful. In order to
keep undesired mobility growth within certain limits, it is appropriate to
strengthen the urban character of existing cities as much as possible.
However, it is still possible to improve the model based on spatial and
functional disaggregation. This would enable the development of more
accurate sub-models that are optimized for one city or a part of a region,
or for one category of travel (e.g. the commute), or for one section of the
population (e.g. school children). Moreover, a refinement would not only
allow modelling the spatial distribution of residential locations, but also
of destination categories such as schools, jobs, shops or public services.
Chapter 7
238
References
Batty, M. (1974) “Spatial entropy.” Geographical Analysis. 6(1), pp. 1-
31.
Boussauw, K., T. Neutens and F. Witlox (2011a) “Relationship between
spatial proximity and travel-to-work distance: The effect of the com-
pact city.” Regional Studies. in press.
Boussauw, K., B. Derudder and F. Witlox (2011b) “Measuring spatial
separation processes through the minimum commute: The case of
Flanders.” European Journal of Transport and Infrastructure Re-
search. 11(1), pp. 42-60.
Boussauw, K., T. Neutens and F. Witlox (2011c) “Minimum commuting
distance as a spatial characteristic in a non-monocentric urban sys-
tem: The case of Flanders.” Papers in Regional Science. 90(1), in
press.
Boussauw, K., V. Van Acker and F. Witlox (2011d) “Excess travel in
non-professional trips: Why looking for it miles away?” Tijdschrift
voor Economische en Sociale Geografie. in press.
Boussauw, K. and F. Witlox (2009) “Introducing a commute-energy
performance index for Flanders.” Transportation Research Part A.
43(5), pp. 580-591.
Cervero, R. (1996) “Mixed land-uses and commuting: Evidence from the
American housing survey.” Transportation Research Part A. 30(5),
pp. 361-377.
Cervero, R. and K. Kockelman (1997) “Travel demand and the 3Ds:
Density, diversity, and design.” Transportation Research Part D.
2(3), pp. 199-219.
Dodson, J. and N. Sipe (2008) “Shocking the suburbs: Urban location,
homeownership and oil vulnerability in the Australian city.” Housing
Studies. 23(3), pp. 377-401.
Echenique, M., A. Hargreaves, Y. Jin, G. Mitchell and A. Namdeo (2009)
London and the Wider South East Regions Case Study. Cambridge-
London-Bristol: Solutions Project.
Ewing, R. (1994) “Characteristics, causes, and effects of sprawl: A
literature review.” Environmental and Urban Studies. 21(2), pp. 1-15.
Ewing, R., K. Bartholomew, S. Winkelman, J. Walters and D. Chen
(2008) Growing Cooler: The Evidence on Urban Development and
Climate Change. Washington, DC: Urban Land Institute.
Linking expected mobility production to residential location planning
239
Ewing, R. and R. Cervero (2010) “Travel and the built environment: A
meta-analysis.” Journal of the American Planning Association. 76(3),
pp. 265-294.
Giuliano, G. (1995) “The weakening transportation-land use connection.”
Access. 6(1), pp. 3-11.
Gordon, P. and H. W. Richardson (1997) “Are compact cities a desirable
planning goal?” Journal of the American Planning Association.
63(1), pp. 95-106.
Handy, S., X. Cao and P. Mokhtarian (2005a) “Correlation or causality
between the built environment and travel behavior? Evidence from
Northern California.” Transportation Research Part D. 10(6), pp.
427-444.
Handy, S., L. Weston and P. Mokhtarian (2005b) “Driving by choice or
necessity?” Transportation Research Part A. 39(2-3), pp. 183-203.
Horner, M. W. (2004) “Spatial dimensions of urban commuting: A review
of major issues and their implications for future geographic research.”
The Professional Geographer. 56(2), pp. 160-173.
IEA (2008) World Energy Outlook. Paris: International Energy Agency.
Janssens, D., E. Moons, E. Nuyts and G. Wets (2009) Onderzoek Ver-
plaatsingsgedrag Vlaanderen 3 (2007-2008). Brussels-Diepenbeek:
Ministry of the Flemish Community.
Maat, K. and H. Timmermans (2006) “Influence of land use on tour
complexity.” Transportation Research Record. 1977, pp. 234-241.
Mees, P. (2010) Transport for Suburbia: Beyond the Automobile Age.
London: Earthscan.
Mindali, O., A. Raveh and I. Salomon (2004) “Urban density and energy
consumption: A new look at old statistics.” Transportation Research
Part A. 38(2), pp. 143-162.
Næss, P. (2003) “Urban structures and travel behaviour. Experiences
from empirical research in Norway and Denmark.” European Journal
of Transport and Infrastructure Research. 3(2), pp. 155-178.
Næss, P. and S. L. Sandberg (1996) “Workplace location, modal split and
energy use for commuting trips.” Urban Studies. 33(3), pp. 557-580.
Nagendra, H. (2002) “Opposite trends in response for the Shannon and
Simpson indices of landscape diversity.” Applied Geography. 22(2),
pp. 175-186.
National Geographical Institute of Belgium (2009) Structuur en codering
van de gegevens top10v-gis en top50V-gis. National Geographical In-
stitute of Belgium from
Chapter 7
240
http://www.ngi.be/Common/articles/CA_Top10V-GIS_TOP50V-
GIS/restruct_doc_nl.htm.
Newman, P. and J. Kenworthy (1989) Cities and Automobile Depend-
ence. A Sourcebook. Aldershot: Gower.
Newman, P. and J. Kenworthy (1999) Sustainability and Cities: Over-
coming Automobile Dependence. Washington, DC: Island Press.
Openshaw, S. and P. Taylor (1979) “A million or so correlation coeffi-
cients: Three experiments on the modifiable areal unit problem”. In:
N. Wrigley (Eds.) Statistical Methods in the Spatial Sciences. Lon-
don: Routledge and Kegan Paul, pp. 127-144.
Pisarski, A. E. (2006) Commuting in America III. Washington, DC:
Transportation Research Board.
Policy Research Centre on Work and Social Economy (2010) Gewestelijke
uitgaande pendel. Policy Research Centre on Work and Social
Economy from http://www.steunpuntwse.be/view/nl/18767. Re-
trieved 06/08/2010.
Rietveld, P. and R. Vickerman (2004) “Transport in regional science: The
‘death of distance’ is premature.” Papers in Regional Science. 83(1),
pp. 229-248.
Schafer, A. (2000) “Regularities in travel demand: An international
perspective.” Journal of Transportation and Statistics. 3(3), pp. 1-31.
Train, K. (2003) Discrete Choice Methods with Simulation. Cambridge,
UK: University Press.
Van Acker, V. and F. Witlox (2011) “Commuting trips within tours: How
is commuting related to land use?” Transportation. doi:
10.1007/s11116-010-9309-6.
Van Ruijven, B. and D. P. Van Vuuren (2009) “Oil and natural gas
prices and greenhouse gas emission mitigation.” Energy Policy.
37(11), pp. 4797-4808.
Verhetsel, A., I. Thomas and M. Beelen (2010) “Commuting in Belgian
metropolitan areas: The power of the Alonso-Muth model.” Journal
of Transport and Land Use. 2(3), pp. 109-131.
Wegener, M. (2010) The future of mobility in cities: Challenges for urban
modelling. Proceedings of the World Conference on Transport Re-
search. Lisbon.
Witlox, F. (2007) “Evaluating the reliability of reported distance data in
urban travel behaviour analysis.” Journal of Transport Geography.
15(3), pp. 172-183.
Linking expected mobility production to residential location planning
241
Witlox, F. and H. Tindemans (2004) “The application of rough sets
analysis in activity-based modelling. Opportunities and constraints.”
Expert Systems with Applications. 27(4), pp. 585-592.
Witze, A. (2007) “That’s oil, folks….” Nature. 445(4), pp. 14-17.
Zwerts, E. and E. Nuyts (2004) Onderzoek Verplaatsingsgedrag Vlaande-
ren 2000-2001. Brussels-Diepenbeek: Ministry of the Flemish
Community.
Chapter 7
242
243
Chapter 8:
Conclusions and
policy recommendations
8.1 General conclusions
8.1.1 Introduction
For several decades now, analyzing the relationship between travel
behaviour and spatial structure has been a rewarding research topic.
Although its history goes back to Von Thünen (early 19th century) and
Alonso (early sixties) (Arnott and McMillen, 2006), this research line has
only been applied in an environmental framework since the eighties. For
an extensive, recent literature review in this context, we refer to Hickman
et al. (2009), which includes almost 250 studies. In the mean time, the
definition of the environmental problem has also evolved. Tangible
environmental and social effects of traffic such as air pollution, noise and
accidents have been constant concerns since the seventies. In the eighties,
these were complemented with the potential scarcity of fossil fuels; while
issues of global warming completed the debate in the nineties. The peak
oil phenomenon is thus a repackaged form of expected shortage of fossil
fuels that has only recently gained interest.
From the approach that views spatial planning as a means to influ-
ence mobility in a more sustainable direction, both “optimistic” and
“pessimistic” research results have surfaced: some researchers state that
there is a significant influence of spatial structure on travel behaviour,
while others argue the opposite (Van Acker, 2010, p. 273). When we rely
on the conclusions formulated in Chapter 7 and Chapter 8, it seems that
our own research finds only a weak, albeit statistically significant connec-
tion: i.e. the proportion of the variance in the distances travelled that is
explained by spatial structure is small. So, the assumed links are present
indeed, but the processes that underpin these may be too complex to be
steered by simple planning principles. The main reason for the only small
share of observed travel that is explained by spatial characteristics is
Chapter 8
244
however known from the literature on travel behaviour. In Chapter 7 it
has been demonstrated that the role of the spatial distribution of housing
and potential destinations is only secondary to the importance of socio-
economic characteristics. This means that e.g. age, gender, household
composition, disposable income and sector of employment are more
decisive for an individual’s travel pattern than the physical structure of
the residential environment. Recent research additionally involves aspects
of social psychology, and demonstrates that the variance which is still
present within coherent social groups is largely determined by individual
preferences, attitudes and lifestyles (Van Acker, 2010, p. 10). Hence, it is
important to keep these unpredictable human characteristics in mind
when discussing the link between travel behaviour and spatial structure.
A second caveat relates to the reliance on cross-sectional data. Chapter 4,
which is the only chapter that analyses time series, suggests that the link
between mobility and spatial structure weakened between 1981 and 2001.
Although the analysis of Chapter 4 (on commuting) cannot simply be
extrapolated to all travel, observed correlations should not be expected to
remain constant over time.
For the US, Ewing et al. (2008, p. 9) estimate the average difference
between the number of kilometres produced by compact development in
comparison to standard development at about 30%. This estimated
difference would be caused by a combination of a reduction in distance
travelled and a relative increase in the use of public transport. The real
gains that can be achieved also depend on the turnover in the real estate
market and the historical spatial structure. For the US, Ewing et al.
(2008, p. 9) estimate that compact development could lead to a 7 to 10%
reduction of transport-related emissions by 2050. In Europe, though, it is
quite possible that the potential gains are lower than in the US since the
European historical spatial structure is more compact, making opportuni-
ties for infill development rare. In addition, certain forms of the rebound
effect, in particular shifts from daily travel to occasional tourist and
recreational traffic, are usually not included in the analyses on which
these figures are based.
When we compare the expected impact of spatial policy with pricing
policy, the latter appears a much more efficient and controllable way to
curb traffic in the short term (Anas and Rhee, 2006; Rodier, 2009).
Nevertheless, pricing policies seem to bear more fruit within a spatial
structure that allows switching to alternative modes or choosing closer
Conclusions and policy recommendations
245
destinations. In that sense, compact development may contribute to
enhancing the effectiveness of non-spatial measures (Zhang, 2002, p. 3).
A context of rising energy prices, and hence rising transport costs, is
much more compelling than a mere policy objective aimed at reducing
emission levels. A lack of spatial proximity may turn oil price shocks
really problematic and strengthen the difference in accessibility between
metropolitan and rural areas in a way that has serious consequences for
the local economy and quality of life in remote areas. The six sub-studies
that make up this doctoral research can be interpreted from two policy
perspectives: (1) the required reduction in emissions from transport, and
(2) mitigation of accessibility problems caused by rising energy prices.
Below, in this context, we give a systematic overview of the conclusions
that we draw from these sub-studies, both for the commute and for non-
work related quasi-daily mobility.
In the next section we link our findings to the spatial policy that was
devised in the Spatial Structure Plan for Flanders (RSV, 1997/2004). The
last section, finally, gives an overview of research gaps and subsequent
avenues for further research.
8.1.2 Commuting
Chapters 2, 3, 4 and 6 focus on commuter travel. In the literature on
person mobility, commuting is clearly over-represented in respect to other
types of travel. As it is the case in this dissertation, the choice to study
the commute is in most cases data-driven. In many western countries,
and Belgium is no exception, the commute is particularly well docu-
mented. In Belgium, the decennial censuses (which basically cover all
commuters), supplemented with social security data and some special
surveys (Vanoutrive et al., 2010) constitute a very extensive database.
Quasi-daily non-commuting trips are only documented through survey
samples. Besides, in Belgium, these surveys are split up into the Travel
Behaviour Survey for Flanders (Onderzoek Verplaatsingsgedrag
Vlaanderen (OVG)) and the Belgian Mobility Survey (MOBEL) that is
mainly focusing on Wallonia and Brussels.
Chapter 2 can be considered as an exploratory study based on rela-
tively simple data, assumptions and methods. This chapter shows that
there are significant regional variations regarding commuting distance
and commuting energy consumption, viewed from the residential location.
Moreover, on a regional scale, energy consumption for commuting is
Chapter 8
246
highly correlated with commuting distance, which means that the
importance of mode choice is only of second order. These findings are put
in the context of the argument of Newman and Kenworthy (1989, 1999),
who found a strong negative correlation between population density and
per capita energy consumption for transport. Also, in Flanders, this
correlation exists, albeit much weaker.
The relatively narrow focus of Chapter 2 raises the question whether
the spatial variation in commuting distances can be explained by varia-
tions in the spatial proximity between houses and jobs, and if so, how
this proximity can be measured. This question is the basis for Chapter 3,
where a method is developed to calculate the spatial proximity between
houses and jobs based on spatially disaggregated values for the minimum
commuting distance. The minimum commuting distance is a concept
derived from the excess commuting literature. Unlike Chapter 2, the data
set used in Chapter 3 does not only contain the location of residence
(origin), but also the job location (destination) of the commuting flow. It
is therefore possible to develop two spatial proximity maps for the
commute: one that represents the spatial proximity of residential loca-
tions in relation to the job market, and another map showing the spatial
proximity of work sites in relation to the housing market. These maps
demonstrate that inhabitants of relatively remote areas with a low jobs-
housing balance have difficulties to find a job close to home. Moreover,
the inhabitants of these regions would even be more affected when all
employees would collectively look for a job closer to home, or would move
house to live closer to their jobs. Residents of urban areas, in contrast,
have wide margins to adapt their commuting behaviour, for example
under the influence of rising transport costs. The main cause of this
phenomenon is the large difference in spatial distribution between the
housing market and the job market. Jobs occur usually in a more concen-
trated form than houses do, and many agglomerations of jobs (especially
in industrial estates) are located relatively far from the residential
concentrations. Furthermore, the results of Chapter 3 seem to confirm the
assumptions that were raised in Chapter 2: i.e. those regions that are
characterized by a relatively low degree of spatial proximity, viewed from
the residential location, are broadly consistent with the regions where
above average commuting distances are reported. Regarding the job
locations, however, most regions with a high density of jobs score rela-
tively poorly in terms of spatial proximity.
Conclusions and policy recommendations
247
In Chapter 2 a possible link is suggested between the increase in av-
erage commuting distance over the years and the evolution of spatial
proximity between the housing stock and the job market. In Chapter 4,
this hypothesis is assessed through time series of the minimum commut-
ing distance and partly also through the maximum commuting distance.
The spatial resolution of the used commuting matrices is limited to the
municipal level, restricting the utility of the results to the supra-
municipal level. Nevertheless, in many municipalities, a clear trend may
be discerned, indicating an average increase in spatial separation between
housing stock and job market. The reason is that the increase in jobs is
mainly found in the agglomerations, while the housing market is charac-
terized by a trend of further dispersion. More important, however, is the
finding that the growth of the commute is much faster than the observed
processes of spatial separation. We can therefore conclude that the
growth of the commute is a quasi-autonomous, prosperity driven process
that exhibits interaction with spatial separation processes although there
is no unidirectional causal link between these two phenomena.
Chapter 2 suggests an inverse relationship between population density
and per capita energy consumption for transport. However, this relation-
ship is not quantified, nor is there a link to the spatial characteristics of
the job location. Chapter 6 elaborates on this problem by selecting a
number of potential variables for spatial proximity, and by calculating
correlations between commuting distance and these selected variables (on
both ends of the commuting flow). Viewed from the residential location,
residential density, functional mix and overall accessibility are negatively
correlated with commuting distance. The minimum commuting distance,
as expected from Chapter 3, is positively correlated with the reported
commuting distance: this is the case both in residential locations and in
job locations. Job density (in job locations) is positively correlated with
commuting distance, as opposed to residential density in the residential
locations. This means that high density does not lead to a more sustain-
able commuting pattern if it is not accompanied with a high degree of
functional mix, while a skewed jobs-housing balance contributes to an
increase in commuting distance. With respect to the jobs-housing balance
we can say that the average commute is the shortest when the origin or
the destination of the trip is in an area where the jobs-housing balance is
close to or equal to 1.
However, the results of these four chapters should be interpreted
within the limitations of the data available and the methods used. The
Chapter 8
248
main constraint is the lack of differentiation based on sector, income class
or level of education. It is for example quite possible that highly educated
employees, who often perform specialized jobs, live less frequently in the
vicinity of the companies that require their qualifications. This is for
example the case for office jobs in central Brussels that are taken by
workers living in the green suburban belt outside the city. On the other
hand, low-skilled jobs occur relatively common in industrial areas that are
not always near a residential centre or agglomeration. According to Van
Acker and Witlox (2011) higher incomes are associated with longer
commuting distances. Differentiation would clarify this issue and allow
mapping the role of the spatial mismatch between qualifications and
preferences of employees and job requirements.
Moreover, in the sub-studies, too little attention has been paid to the
role of public transport. A high concentration of jobs around a major
railway station will have a smaller impact in terms of energy consumption
than a concentration of jobs on a remote industrial area that is virtually
only accessible by car. Therefore, in many planning strategies for office
development it is argued that specialized jobs which necessarily occur in
strong spatial concentrations because of the required agglomeration
benefits, should preferably be located in the vicinity of a main railway
station (Schwanen et al., 2004). Examples are the North quarter and the
European quarter in Brussels, or the office district of Amsterdam-Zuidas.
However, an analysis of the effectiveness of this planning concept is
beyond the scope of this dissertation.
8.1.3 Daily non-commuting travel
Although the findings of Chapters 2, 3, 4 and 6 clearly demonstrate a link
between the commute and the spatial proximity of housing and jobs,
these conclusions should not simply be extrapolated to other forms of
quasi-daily mobility. Chapters 5 and 7 try to extend the methodology to
travel patterns that are not part of the commute.
In Chapter 5 the concepts of the minimum commuting distance and
excess commuting are applied to quasi-daily trips that are not part of the
commute. The results are displayed in the form of a spatial proximity
map, representing for every census ward the relative proximity to a
bundle of facilities. In addition, for each spatial class an aggregated excess
rate is calculated, which indicates how strong the relationship is between
Conclusions and policy recommendations
249
the degree of spatial proximity to facilities and the actual distance
travelled.
The proximity map indicates which areas are in fact too remote to
possibly achieve sustainable travel patterns in terms of access to everyday
amenities. The comparison of the minimum required travel distance to
the observed travel distance indicates that the relationship between
spatial proximity and travel patterns is stronger in the most remote areas
compared to more urbanized areas, even though the distance travelled is
on average larger in the more remote areas. In other words, residents of
rural areas are more likely to choose the closest possible destination than
residents of more urbanized areas. Metropolitan areas and urban agglom-
erations, which together constitute the most urbanized of all assessed
areas, however, are an exception to this rule: here the actual distances
travelled are short, while the nearest possible destination is chosen
relatively often.
Chapter 7 combines the variables that were developed in Chapter 3, 5
and 6 with the method of analysis developed and explained in Chapter 6.
The spatial correlation analysis of Chapter 6 is extended to a multivariate
regression analysis based on a sample provided by the 2007 phase of OVG
(Janssens et al., 2009). In order to get an overall picture of the impact of
all quasi-daily trips together, commuting and non-commuting trips were
combined and so the total distance travelled per respondent was ana-
lyzed. By extrapolating the obtained regression equation to the whole
study area, a new, more sophisticated, proximity map is generated.
Unlike the analysis in Chapter 6, Chapter 7 only includes the spatial
characteristics of the residential location of the respondents, limiting the
validity of the conclusions to the spatial characteristics of the residential
environment. A major finding of Chapter 7 is that the importance of the
spatial distribution of jobs fades completely into the background when
studying the aggregated travel pattern. So, the impact of the spatial
distribution of jobs on total mobility is relatively unimportant in relation
to the spatial distribution of other quasi-daily destinations. Population
density, functional mix and spatial proximity of facilities, however,
remain intact as determining factors. However, the overall explanatory
value of a model based on spatial properties is rather limited. This means
that attempts to shrink mobility patterns based on spatial policies are
doomed to lead to poor results unless combined with other measures (e.g.,
pricing) or concomitant with (autonomously) rising energy prices.
Chapter 8
250
Chapters 5 and 7 also have their limitations. The main shortcoming
can be summarized as the absence of an analysis based on the spatial
characteristics of the destination. In Chapters 3, 4 and 6 job locations are
also involved in the assessment. In Chapters 5 and 7 we do not gain
insight on the influence of the location of the visited shops, schools,
recreation facilities, et cetera. The reason why such analysis is beyond the
scope of this dissertation is, on the one hand, the lack of a sufficiently
large data set that contains the necessary information, and on the other
hand, the high level of complexity of the issue. Nevertheless, this is an
important track for possible future research. To our knowledge, in
Flanders or Belgium there is no research available that is for example
quantifying the total difference in energy efficiency between suburban
hypermarkets and small urban convenience stores, or between large and
small schools, or between more and less specialized educational institu-
tions.
8.1.4 Non-daily travel
Occasional (i.e. non-daily) trips, including excursions, weekends out, city
breaks, holidays and business trips are not the subject of the six sub-
studies of this dissertation. Although the study of this type of trips is
usually counted as tourism or business travel research, and not as
mobility research in the narrow sense, we still want to give this issue
some thought. As suggested in the introduction, it is quite possible that
the slow-down of the growth of mobility in the segment of quasi-daily
trips is compensated by continued growth within the segment of occa-
sional mobility. This shift would also be speed-driven, mainly based on
the far-reaching democratization of air travel. Moreover, there may also
be a link between the amount of travel (and thus the associated level of
energy consumption) that results from occasional mobility, and the
spatial structure of the residential environment of the traveller (Holden
and Norland, 2005). In order not to leave this hypothesis without any
evidence, in an addendum we present a brief analysis, based on the
methodology used in Chapter 5.
Conclusions and policy recommendations
251
8.2 Options for spatial planning policy
8.2.1 The sustainable mobility paradigm
Banister (2008) describes a new mobility paradigm, in which sustainable
development is paramount and the sense of a further increase in mobility
is questioned. Urry (2002) depicts proximity as a necessary condition for
the occurrence of social interaction, which he calls co-presence, without
automatically being linked with extensive mobility. In policy terms, this
means that improving accessibility should become the objective, as
opposed to intensifying mobility. The ultimate goal is to achieve a high
degree of accessibility based on a minimum of mobility and therefore a
minimum amount of traffic. The above shows that steering spatial
development should be viewed first as a way to enhance resilience to oil
scarcity and more expensive transport, and second as working on a
spatial framework suitable to facilitate climate policies.
Based on the scheme of Banister (1999, pp. 316-319) we make a brief
translation of our research results in recommendations towards spatial
planning policy:
• Wherever possible, developments should be based on a high residen-
tial density, also in areas where, apart from housing, other activities
occur. The desired level of residential density depends on the level of
ambition. Density thresholds mentioned in the literature are usually
linked to the desired role of public transport and were discussed in
Chapter 1. The eco-districts Vauban and Rieselfeld in the German
Freiburg-im-Breisgau, known as prototypes of green contemporary
residential urban extensions, have a gross residential density that
goes up to about 150 inhabitants per hectare (Ryan and Throgmor-
ton, 2003), which is comparable to the most densely populated
neighbourhoods in the 19th century belts of the larger Belgian cities.
However, in Flanders, the aspiration level of the Spatial Structure
Plan for Flanders (RSV, 1997/2004) is limited to 25 dwellings
(equivalent to approximately 55 people) per hectare in urban areas
and 15 dwellings (equivalent to approximately 38 people) per hectare
in the outlying area. Moreover, a progressive shrink of the average
household size threatens to bring down the effective residential den-
sity on the basis of these standards (which are expressed in dwellings
per hectare).
Chapter 8
252
• The size of the nuclei must be large enough to assure a relatively high
degree of self-sufficiency, based on agglomeration economies. Follow-
ing Banister (1999), we could argue that towns or villages with less
than 25,000 residents should in principle not be allowed to grow
anymore, while isolated cities that are not part of an agglomeration
should in time reach a size of at least about 50,000 inhabitants. In
addition, growth should be directed as much as possible in or imme-
diately subsequent to the agglomerations (> 250,000 inhabitants),
where public transport can play an important role.
• New developments should be located within or next to existing urban
areas. Facilities and jobs should be planned and developed simultane-
ously with housing, preferably concentrated in local centres.
Moreover, new developments should have good non-car accessibility.
8.2.2 Possible adjustments to the Spatial
Structure Plan for Flanders
As such, these principles, relying on compact development and economies
of agglomeration, are not new. In developing the RSV (1997/2004), a
number of these items were already taken into account:
• The RSV requires that 60% of additional housing units should be
built in urban areas, and only 40% in the nuclei of the outlying area.
This rule differs from the trend that was observed in the early nine-
ties, when most new homes were built outside the urban areas. To
meet this objective in practice, the urban areas are demarcated and
additional residential land is designated in those urban areas.
• The RSV requires a minimum residential density for new develop-
ments (25 dwellings per hectare in urban areas and 15 dwellings per
hectare in the outlying area).
• Interweaving of functions and activities is a guiding principle in the
development of urban areas.
• Offices are preferably concentrated at nodes of public transport.
• Amenity levels should be attuned to the importance of the respective
urban areas.
Based on our own research results and literature review we suggest some
possible adjustments to the perspective of the RSV regarding the steering
of new developments. We summarize these adjustments in Fig. 8.1. The
main research results we have been drawing from in order to produce Fig.
8.1, are represented in Figs 3.1 and 3.2, in Fig. 5.3, and in Fig. 7.2. The
Conclusions and policy recommendations
253
schematic map that is presented in Fig. 8.1 and is explained below, can
be considered as a possible spatial development perspective that is
motivated from optimizing accessibility based on a minimum amount of
traffic, by maximizing spatial proximity and valorising economies of
agglomeration. This development perspective includes the following
elements:
• The distinction of areas suitable for additional residential develop-
ment and areas where less residential development, or even a status
quo, is desirable, should be based on the existing agglomerations. The
overall degree of spatial proximity of a municipality located near a
metropolitan area is much higher than in a remote small urban area.
It is therefore important to exploit and strengthen existing agglom-
eration effects. The demarcation of the areas that are designated as
“conurbation” in Fig. 8.1 is done on the basis of various indicators of
spatial proximity, and these should be considered as search areas for
additional housing. The areas classified as “potential conurbation”,
are today probably too small or too remote to be considered yet as a
conurbation, but could become a conurbation in the future through
directed growth. Perhaps other areas exist that rather belong to this
category, but were not indicated as such on the map. The term “ur-
ban area” should therefore be extended to a conurbation, with the
intention to grant the existing suburban areas in the conurbations a
more urban character (by increasing density and functional mix)
while stopping the growth of peri-urban developments.
• The proportion of all new dwellings that should be directed into these
conurbations will depend on arguments from sectors other than mo-
bility. From the accessibility perspective, preferably 100% of
additional homes would be located in the conurbations. A variant of
this could allow a minimal amount of new housing outside the conur-
bations to meet the locally occurring shrink of family size, or could
allow for natural growth of the local population. The greater the pro-
portion of additional dwellings that will be located in the
conurbations, the wider the demarcation of the external border of the
conurbation can be. In case there would still be built houses outside
the conurbations, these should be built first in small urban areas, and
only in second order in the nuclei of the outlying area. To ensure the
compactness of the conurbation, the desired housing density should
be adjusted upward: an average gross density of 80 to 120 inhabitants
per hectare appears feasible.
Chapter 8
254
• By concentrating residential development in the conurbations, a
reduction in the total number of vehicle kilometres travelled will be
facilitated. Since it is mainly the mobility which would otherwise
have occurred outside the conurbation that will be suppressed, such a
development will not result in relief of congestion in the conurbations.
The enhanced agglomeration effect, however, provides a stronger ba-
sis for an efficient public transport system. Each demarcated
conurbation should aim to increase the internal accessibility by public
transport (based on restricted tram and (trolley)bus lanes and (light)
rail), and also by bike. By discouraging car traffic that enters the
conurbation, additional space can be created to improve the internal
accessibility. Additional budget for public transport should mainly be
directed at developing these conurbation networks, complemented
with transferia to receive car traffic from the outlying area.
• Within the conurbations, a well-balanced spatial distribution of
facilities should be pursued, preferably concentrated in centres and
sub-centres. Upscaling, in the sense that several smaller establish-
ments such as schools, shops and workshops are replaced by one
campus, one hypermarket or one industrial estate should be discour-
aged because of the consequent reduction of the level of spatial
proximity.
• In small urban areas, as in the outlying area, the amount of addi-
tional houses to be built should be minimized (preferably to zero),
pursuing a status quo of the population. The small urban areas
should serve as subregional centres with the fullest possible range of
amenities and a variety of jobs that meet the needs of the service
area that is surrounding every small urban area. By concentrating
facilities in small urban areas, and also by expanding the supply of
services, the level of spatial proximity in the surrounding service area
will increase, reducing the need for residents to travel to a conurba-
tion. By enlarging the supply of jobs, the average distance between
home and work locations will be reduced. The emphasis should be on
non-specialized jobs (which have a weaker link with the conurbations
and also generate less long-distance commuting), itinerant and home-
based occupations (the latter jobs also fit in the actual countryside).
For commuters from the small urban areas and the surrounding ser-
vice area, rail links to the conurbations should be optimized. Based
on these principles, in Fig. 8.1 the selection of small urban areas is
reproduced from the RSV, unless included in a conurbation. Based on
Conclusions and policy recommendations
255
arguments from other sectors this selection may be changed, or possi-
bly supplemented with certain economic nodes or main villages.
• Since highly specialized office jobs generate long commuting dis-
tances, these should be directed as much as possible to central
locations near major railway stations. In order not to encourage long-
distance commuting, these sites are less suitable for non-specialized
jobs where the aim should be to recruit staff locally wherever possi-
ble. Central locations that have good access by train are limited in
number: we are talking about the major stations in Brussels and
Antwerp, possibly supplemented with Mechelen, Gent and Leuven.
The surroundings of the other major railway stations in Flanders are
also suitable for the establishment of institutions of higher education.
• In an integrated policy plan, the selection and demarcation of urban
development areas should of course not only be determined by mobil-
ity. Economic aspects (such as the siting of major port and industrial
areas that may entail long distances between home and work loca-
tions), the presence of valuable natural areas, forests and landscapes,
agricultural areas and heritage are all factors that should be taken
into account when planning future development. Moreover, the neces-
sary supply of additional housing will be determined on the basis of
demographic trends.
• Spatial policies aimed at strengthening spatial proximity may be
supported by an appropriate fiscal policy. Enhancing the variability
of the transport cost as a function of the distance travelled may help
mitigating differences in the real estate prices between the conurba-
tions and the outlying areas.
Although the basic principles are similar, the spatial development per-
spective outlined in Fig. 8.1 contains a few noticeable differences
compared to existing policy plans and studies. Below, we touch upon the
most striking elements, comparing with both the RSV (1997/2004) and
the Spatial-Economic Main Structure (Cabus et al., 2001), which is a
spatial perspective on the current and future economic development in
Flanders.
Compared with the RSV perspective, which is shown schematically in
Fig. 6.14, we note the following:
• The demarcation of the regional urban areas and metropolitan areas
is expanded, some of which may even grow together.
Chapter 8
256
• The small urban areas are no longer regarded as development areas
for additional homes, but rather as employment and service centres
for the surrounding outlying areas.
• The requirements in terms of residential density and functional mix
in the (re)development of urban neighbourhoods are revised signifi-
cantly upwards.
Fig. 8.1. Spatial development perspective based on maximum access and
minimum mobility
The main difference compared with the vision of the Spatial-Economic
Main Structure (Cabus et al., 2001) is that the economic development
zones, which coincide with the conurbations as suggested in Fig. 8.1,
supplemented with the major ports and industrial areas (which are not
shown in Fig. 8.1) and to a lesser extent the small urban areas, are
viewed much more compact.
While this study has analyzed a number of potential problems and
has issued some possible guidelines based on this analysis, bringing these
suggestions into practice is not obvious. In Flanders and Belgium, the
role of the government is largely limited to the designation of locations
where housing and industry is allowed to develop, along with the plan-
ning of the infrastructure network and public transport. The density of
residential (re)development is hardly centrally controlled, and the legal
and fiscal framework is rather targeted at providing legal certainty than
Conclusions and policy recommendations
257
at creating future-oriented dynamics. Moreover, no efficient economic
planning tools are available that can significantly influence the location of
jobs. Also, the location and size of schools and shops are hardly steered
by a spatial development perspective. So, the main challenge is perhaps
not to be found in research and planning theory, but in the implementa-
tion of the principle of compact urban development.
8.3 Some directions for further research
Results of research in terms of spatial planning and transport policy will
never be entirely satisfactory. Each new analysis raises additional ques-
tions. Moreover, history tells us that planning is not a science. Analyses
and models may underpin policy choices, but remain only one element in
decision making processes. Nevertheless, quantitative research is impor-
tant, and should not fade into the background of planning processes.
Throughout this dissertation, research gaps have been pointed out
continuously, along with the necessary caution that should be observed
when drawing conclusions from the research results. Below we give a list
of possible avenues for further research that build on this dissertation.
Within commuting research, the following directions for further re-
search can be put forward:
• The application of the concept of minimum commuting distance
(based on the method described in Chapter 3) and correlation analy-
sis (as in Chapter 6) based on commuting data that differentiate
between economic sector, level of education and income class.
• The analysis of time series of minimum commuting distance based on
detailed zoning (traffic analysis zones or census wards), with the aim
of measuring processes of spatial separation and sprawl (based on the
method described in Chapter 4).
• Calculating variations in the energy efficiency of the commute to
offices and industrial estates based on the type of location and taking
into account the mode choice, the economic sector, the level of educa-
tion and the income class of staff members.
Within the study of quasi-daily mobility outside the commute, the
following future research tracks may be distinguished:
• Refinement of the concept of the spatial proximity map based on
residential locations (see Chapter 5) and of the underlying methods of
measurement.
Chapter 8
258
• The development of a similar map based on the spatial proximity of
destinations (complementary to the current map which relies only on
residential locations, considered as origins).
• The analysis of time series of the spatial proximity map(s) with the
aim of measuring processes of spatial separation and sprawl (as a
combination of the methods from Chapters 4 and 5).
• Modelling the differences in overall energy efficiency for transport
between small scale and large scale retailing (taking into account
both energy consumption by goods supply and by customer’s trips).
• Modelling the differences in energy efficiency of the school commute
between networks of more numerous, but smaller, schools versus less
numerous, but larger, schools.
Further research based on these suggestions would contribute to a well-
informed practice of spatial planning, which aims to ensure a high level of
accessibility in combination with a sustainable form of mobility.
References
Anas, A. and H.-J. Rhee (2006) “Curbing excess sprawl with congestion
tolls and urban boundaries.” Regional Science and Urban Economics.
36(4), pp. 510-541.
Arnott, R. and D. P. McMillen (2006) A Companion to Urban Econom-
ics. Malden, MA - Oxford - Victoria: Blackwell Publishing.
Banister, D. (1999) “Planning more to travel less.” Town Planning
Review. 70(3), pp. 313-338.
Banister, D. (2008) “The sustainable mobility paradigm.” Transport
Policy. 15(2), pp. 73-80.
Cabus, P., G. Allaert and W. Vanhaverbeke (2001) De Ruimtelijk-
Economische Hoofdstructuur van Vlaanderen: Concept in Hoofdlij-
nen. Gent: Academia Press.
Ewing, R., K. Bartholomew, S. Winkelman, J. Walters and D. Chen
(2008) Growing Cooler: The Evidence on Urban Development and
Climate Change. Washington, DC: Urban Land Institute.
Hickman, R., C. Seaborn, P. Headicar, O. Ashiru, S. Saxena, D. Banister
and T. Pharoah (2009) Land Use and Transport: Settlement Patterns
and the Demand for Travel. Oxford: Commission for Integrated
Transport.
Conclusions and policy recommendations
259
Holden, E. and I. Norland (2005) “Three challenges for the compact city
as a sustainable urban form: Household consumption of energy and
transport in eight residential areas in the greater Oslo region.” Urban
Studies. 42(12), pp. 2145–2166.
Janssens, D., E. Moons, E. Nuyts and G. Wets (2009) Onderzoek Ver-
plaatsingsgedrag Vlaanderen 3 (2007-2008). Brussels-Diepenbeek:
Ministry of the Flemish Community.
Newman, P. and J. Kenworthy (1989) Cities and Automobile Depend-
ence. A Sourcebook. Aldershot: Gower.
Newman, P. and J. Kenworthy (1999) Sustainability and Cities: Over-
coming Automobile Dependence. Washington, DC: Island Press.
Rodier, C. (2009) “Review of international modeling literature: Transit,
land use, and auto pricing strategies to reduce vehicle miles traveled
and greenhouse gas emissions.” Transportation Research Record.
(2132), pp. 1-12.
RSV (1997/2004) Ruimtelijk Structuurplan Vlaanderen - Gecoördineerde
Versie. Brussels: Ministry of the Flemish Community.
Ryan, S. and J. A. Throgmorton (2003) “Sustainable transportation and
land development on the periphery: A case study of Freiburg, Ger-
many and Chula Vista, California.” Transportation Research Part D.
8(1), pp. 37-52.
Schwanen, T., M. Dijst and F. M. Dieleman (2004) “Policies for urban
form and their impact on travel: The Netherlands experience.” Urban
Studies. 41(3), pp. 579-603.
Urry, J. (2002) “Mobility and proximity.” Sociology. 36(2), pp. 255-274.
Van Acker, V. (2010) Spatial and Social Variations in Travel Behaviour.
PhD Thesis. Ghent: Ghent University.
Van Acker, V. and F. Witlox (2011) “Commuting trips within tours: How
is commuting related to land use?” Transportation. doi:
10.1007/s11116-010-9309-6.
Vanoutrive, T., L. Van Malderen, B. Jourquin, I. Thomas, A. Verhetsel
and F. Witlox (2010) “Mobility management measures by employers:
Overview and exploratory analysis for Belgium.” European Journal of
Transport and Infrastructure Research. 10(2), pp. 121-141.
Zhang, M. (2002) Conditions and Effectiveness of Land Use as a Mobility
Tool. PhD Thesis. Cambridge, MA: Massachusetts Institute of Tech-
nology.
Chapter 8
260
261
Addendum:
The spatial component
of air travel behaviour:
An exploration An earlier version of this addendum has been published (in Dutch) as Boussauw (2009a) in the journal Agora and as Boussauw (2009b) on the website Low Tech Magazine.
A.1 The aeroplane: the forgotten
transport mode
While urban and interurban person mobility still seems to increase, fuel consumption and emissions are stagnating. Cleaner cars and improved public transport play a role in this evolution, but it is also true that mobility is slowly but surely clashing with structural capacity limits. Good news for the climate, so it seems. However, data on this, reported by the Flemish Environment Agency (Vlaamse Milieumaatschappij (VMM)) show an important gap (De Vlieger et al., 2007). International aviation is systematically not included in the statistics. However, air travel is the fastest growing segment of the transport sector, which is also responsible for bafflingly large volumes of CO2 emissions from the com-bustion of equally large volumes of fossil fuels. In 2008, the number of passengers in Belgium increased by 5.8% over the previous year, while road traffic, exceptionally, slightly decreased (-1%) (Statistics Belgium, 2009).
But what amount of emissions are we talking about? Since there are no published statistics on fuel consumption by flying inhabitants of the Flanders region, we made a rough estimate based on flight data for 2008 from the Belgian aviation authorities (Brussels Airport, 2009; Statistics Belgium, 2009). According to the demographic composition we assumed that out of the 45% Belgian passengers using Belgian airports, 58% are
Addendum
262
living in Flanders. The travel distances were estimated based on distance classes, provided by Brussels Airport (2009), and complemented with data on the destinations served by the regional airports. Some examples: for flights to Mediterranean resorts, the average length of a flight was estimated at 2,000 km, while for North American destinations this amounted to 7,500 km, and to 1,500 km for Eastern European destina-tions.
Based on a report by the Danish Environmental Protection Agency (2003) we assumed an average of 5 litres of kerosene fuel consumption per 100 km, for each occupied seat. For flights of less than 1,000 km, fuel consumption is generally higher. Moreover, consumption depends strongly on the type of aircraft, and the occupancy rate. No detailed information is available on the used equipment. This is also the case for the occu-pancy rate, which might be much lower since the beginning of the financial crisis, than before. All these factors make our estimate of fuel consumption perhaps rather cautious (underestimate).
Converted to PJ (petajoule = 109 MJ (megajoule)), we find a total consumption by Flemish air travellers of approximately 69.0 PJ. For overland person transport, VMM (for 2006) reports a total consumption of 121.6 PJ (De Vlieger et al., 2007). This means that aviation is respon-sible for more than one third of total energy consumption and hence CO2 emissions for transport of inhabitants of Flanders. And this proportion is increasing rapidly.
It should not be forgotten that one litre of kerosene burned at high altitudes contributes much more to global warming than the consumption of the same amount of energy on the ground level. At altitude also the emission of water vapour and nitrogen oxides has a significant contribu-tion, in addition to the effect of carbon dioxide. This has, among other things, to do with cloud formation, caused by jet engines. According to Åkerman (2005), the total impact of aviation on global warming is 2.7 times greater than the impact of CO2 emissions from aviation alone.
Despite the rapid increase in air traffic, the aircraft remains an excep-tional transport mode for the average inhabitant of Flanders. In the Travel Behaviour Survey for Flanders (OVG) (2000-2001) 52.6% of the respondents indicate not to fly at all (Zwerts and Nuyts, 2004). Less than half of the travellers who used Brussels Airport in 2008 were flying more than four times a year (Brussels Airport, 2009). The vast majority of the emissions from air travel is therefore caused by a small minority of the population.
The spatial component of air travel behaviour: An exploration
263
A natural upper limit to the demand for air travel is not yet in sight. On average, an inhabitant of Flanders flies annually once to, say, Egypt and back. Still, there seems to be a lot of growth opportunity before the saturation level of the average tourist will be reached. The price of air tickets continues to fall, thanks to declining fixed costs. Only the price of fuel, or the introduction of taxes could constitute an impediment to growth. For the time being, however, the oil price remains at an accept-able level, and important eco-taxes are almost nonexistent. Within the current framework a rapid growth in air travel, both in the western world and globally, is inevitable.
Attempts to provide technological solutions have little success. The environmental problem of aviation has in fact only a limited technological dimension: per person kilometre, a plane is about as efficient as a car. The essence of the problem is that the availability of cheap air transport induces long-distance travel, and thus leads to an explosion in the number of kilometres travelled. Given the huge volumes of fuel we are talking about, tentative experiments with aviation on biofuels are of little significance to the market.
Adding aviation figures to fuel consumption and emissions statistics of person transport leads to new insights. On the one hand, in the context of energy and climate issues it appears suddenly very easy to reduce energy consumption and emissions: imposing a high tax on flying would undoubtedly result in a significant reduction in the number of passengers. Besides, the economic downturn would be largely confined to the tourism sector (comprising about two thirds of air passengers), which is much more price sensitive than the business sector and also less critical to the domestic economy.
On the other hand, policy efforts in the field of alternative transpor-tation modes sink away when these are viewed in the light of the growth of aviation. Does it make sense to abandon the car for daily travel, as air traffic continues to grow as fast yet? Moreover, there is no significant societal support to curb air traffic growth, especially not in the non-western world.
Addendum
264
A.2 Urban versus rural lifestyle
Based on Norwegian research, Holden and Norland (2005) suggest that there is not only a link between spatial structure and daily travel pat-terns, but also with air travel. They argue that city dwellers make more unsustainable long distance trips compared to non-city dwellers. They quote several possible reasons for this. As city dwellers have less often a garden, they would need more vacation outdoors. Living in the city would more often be accompanied with an internationally oriented lifestyle. Moreover, city dwellers would have lower daily travel expenses, perhaps because they own less often a car, leaving them with more financial headroom to pay for, among other things, plane tickets.
While in Flanders the contrast between urban and rural areas may be less significant than in Norway, it is still worth to assess the validity of the assertion of Holden and Norland (2005) also in this context. However, the available data is quite limited. We rely on the OVG survey, which distinguishes between respondents who never fly (the reference category) and those who are occasional or frequent flyers. Of each respondent, we know the municipality of residence. Based on Luyten and Van Hecke (2007), we assigned all municipalities into four categories: urban agglom-eration, suburban area, commuter area and rural area (the residual category). This classification was made on an empirical basis, and therefore does not always reflect the policy oriented classification of the Spatial Structure Plan for Flanders (RSV, 1997/2004). In addition, the OVG survey asked a question on the respondent’s perception of his residential environment: we distinguish between people who think they live in a centre (dense built-up area) and those who think they live outside a centre.
We include the variables in two logistic regression models, considering air travel behaviour as the dependent variable. In the first model, the location of the municipality of residence is introduced as an explanatory variable, while in the second model the perception of the residential environment is used. We present the output of these two regressions in Table A.1 and Table A.2 respectively.
The column exp(B) can be interpreted as the odds ratio of the prob-ability that a respondent in this category is flying in comparison to the reference category. Table A.1 shows that the probability that a resident of the urban agglomeration flies is a factor 1.6 higher compared to a rural resident. For residents of suburban and commuter areas, this odds ratio is
The spatial component of air travel behaviour: An exploration
265
1.2 to 1.3. From Table A.2 we learn that the probability that a resident of a centre or a dense built-up area is flying is 1.2-fold higher compared to someone living outside a centre. All results are statistically significant at the 0.05 level. Table A.1. Odds ratios of the probability of flying when not living in the rural area
category exp(B) p-value
intercept - 0.000
urban agglomeration 1.576 0.000
suburban area 1.206 0.014
commuter area 1.262 0.000
rural area (reference category) - -
Table A.2. Odds ratio of the probability of flying when living in a centre
category exp(B) p-value
intercept - 0.000
centre 1.157 0.002
outside centre (reference category) - -
Although the available data is too rough to allow quantifying variations in energy consumption and CO2 emissions, our analysis seems to confirm the argument of Holden and Norland (2005). If we take aviation into account, then the travel pattern of the city dweller may be a lot less sustainable than we thought.
A.3 Rebound effect and policy implications
The above findings fit the theory of the so-called rebound effect (Alcott, 2005). Energy savings in one field are compensated by more consumption in another field, as long as the disposable budget remains the same. A small local ecological footprint, which is made possible by an urban lifestyle, is compensated by a large global ecological footprint which in this case can be written on behalf of air travel.
A macro-economic extension of this thesis is known as the Khazzoom-Brookes postulate (Saunders, 1992). This states that at constant energy prices, increasing energy efficiency does not lead to a decrease, but rather
Addendum
266
to an increase in the volume of energy consumed globally. The reason for this is that greater efficiency leads to more prosperity with relatively low market prices for energy, compared to the baseline scenario. The conse-quent wealth surplus can be spent easily to energy-intensive consumption. Applied to our subject, this means that the money that is saved by an individual through better thermal insulation of his house or by driving a more fuel-efficient car, is easily invested in activities where efficiency seems less important, such as vacations. The rapid growth of air travel indicates that tourist flights play an important role in this mechanism. This observation adds a note of caution to a climate policy that is based solely on encouraging more energy efficiency.
Air traffic increase is conveniently overlooked in many mobility stud-ies and environmental policy plans. For example, the Climate Policy Plan for Flanders (LNE, 2006) includes no measures related to aviation. Nevertheless, in this plan efforts to promote cycling are considered relevant in terms of reducing CO2 emissions, even though the results expected from this are only marginal compared to the emissions from aviation.
Besides the climate issue, also the depletion of fossil energy is a good reason to deal judiciously with the available oil resources. In this light, inducing the demand for city breaks may not be the most sustainable strategy. However, this is exactly what happens today through partner-ships between low-cost airlines and diverse government agencies (Monbiot, 2009).
In addition, we have brought up a surprising aspect of the urban life-style. An urban spatial structure may encourage sustainable local travel behaviour, it has no control over global consumption patterns. Urban dynamics rather stimulate air travel demand. When local mobility clashes with structural capacity limitations, the desire for more interaction with the world results possibly in flying.
References
Åkerman, J. (2005) “Sustainable air transport - On track in 2050.” Transportation Research Part D. 10(2), pp. 111-126.
Alcott, B. (2005) “Jevons’ paradox.” Ecological Economics. 54(1), pp. 9-21.
The spatial component of air travel behaviour: An exploration
267
Boussauw, K. (2009a) “Stadsmens onderweg: Een duurzaamheidspara-dox.” Agora - Magazine voor Sociaalruimtelijke Vraagstukken. 25(5), pp. 7-10.
Boussauw, K. (2009b) Hoe duurzaam leeft de stadsbewoner? Low Tech Magazine from http://www.lowtechmagazine.be/2009/12/hoe-duur zaam-is-de-stadsbewoner.html.
Brussels Airport (2009) Brutrends 2008. Brussels: Brussels Airport. Danish Environmental Protection Agency (2003) Greenhouse Gas
Emissions from International Aviation and Allocation Options. Co-penhagen: Danish Environmental Protection Agency.
De Vlieger, I., E. Cornelis, L. Int Panis, L. Schrooten, L. Govaerts, L. Pelkmans, S. Logghe, K. Vanherle, F. Vanhove, G. De Ceuster, C. Macharis, T. Festraets, E. Pekin, L. Turcksin, K. Van Bladel, M. de Jong, J. Van Mierlo, J. Matheys, J.-M. Timmermans, C. De Geest and E. Van Walsum (2007) MIRA (2007) Milieurapport Vlaanderen, Achtergronddocument 2007, Transport. Aalst: Flemish Environment Agency.
Holden, E. and I. Norland (2005) “Three challenges for the compact city as a sustainable urban form: Household consumption of energy and transport in eight residential areas in the greater Oslo region.” Urban Studies. 42(12), pp. 2145–2166.
LNE (2006) Vlaams Klimaatbeleidsplan 2006-2012. Brussels: Ministry of the Flemish Community.
Luyten, S. and E. Van Hecke (2007) De Belgische Stadsgewesten 2001. Brussels: Statistics Belgium.
Monbiot, G. (2009) “Subsidising the climate crash.” The Guardian. (7th July).
RSV (1997/2004) Ruimtelijk Structuurplan Vlaanderen - Gecoördineerde Versie. Brussels: Ministry of the Flemish Community.
Saunders, H. D. (1992) “The Khazzoom-Brookes postulate and neoclassi-cal growth.” Energy Journal. 13(4), pp. 131-148.
Statistics Belgium (2009) Mobiliteitsportaal from http://www.statbel. fgov.be/port/mob_nl.asp.
Zwerts, E. and E. Nuyts (2004) Onderzoek Verplaatsingsgedrag Vlaanderen 2000-2001. Brussels-Diepenbeek: Ministry of the Flemish Community.
Addendum
268
269
Samenvatting
S.1 Overzicht
Deze verhandeling wil het inzicht in de wederzijdse relatie tussen mobili-
teit en ruimtelijke ontwikkelingen vergroten, en het probleem situeren in
een context van klimaatbeleid en peak oil. Dit wordt gedaan op basis van
literatuurstudie en de toepassing van een aantal, voor een deel nieuw
ontwikkelde, onderzoeksmethoden op de case study Vlaanderen.1
Dit erg ruim geformuleerde onderzoeksveld werd vernauwd tot perso-
nenmobiliteit, waarbij we onze doelstelling richten op het onderzoeken
van de duurzaamheid van de ruimtelijke structuur ten aanzien van
verplaatsingsgedrag, met bijzondere aandacht voor de dagelijks afgelegde
afstanden. Duurzaamheid wordt gedefinieerd in de zin van het robuust
zijn, niet enkel voor een groeiende maar ook voor een in de toekomst
mogelijk krimpende mobiliteit. Een algemeen krimpende mobiliteit is een
scenario dat zich kan ontwikkelen tengevolge van stijgende energiekosten
(het peak-oil scenario) of een stringent klimaatbeleid, terwijl een selectie-
ve krimp van de mobiliteit (slechts van toepassing op een deel van de
bevolking) zich kan voordoen door een toenemende verzadiging van het
verkeerssysteem. Bovendien speelt de ruimtelijke structuur een rol in de
mogelijke sturing van het verplaatsingsgedrag in een meer duurzame
richting.
Deze verhandeling volgt de klassieke wetenschappelijke onderzoekslijn
over het verband tussen ruimtelijke structuur en verplaatsingsgedrag, die
al decennia in ontwikkeling is. Een voorlopige algemene conclusie van
deze onderzoekslijn in een context van duurzaamheid zou als volgt
kunnen worden geformuleerd: “Een duurzaam verplaatsingspatroon kan
slechts tot stand komen binnen een ruimtelijk kader dat daar geschikt
voor is, maar andere maatregelen (financieel en regulerend) zijn nodig om
het verplaatsingsgedrag effectief te wijzigen. Met andere woorden:
ruimtelijke ordening is nodig, maar het is niet voldoende” (naar Zhang,
1 De aanleiding voor dit onderzoek bevindt zich in de taakstelling van het
Steunpunt Ruimte en Wonen (2007-2011), beleidsondersteunend onder-
zoeksconsortium dat aangestuurd wordt door het Departement Ruimtelijke
Ordening, Woonbeleid en Onroerend Erfgoed van de Vlaamse overheid.
Samenvatting
270
2002, p. 3). De ruimtelijke kwaliteit die een verplaatsingspatroon op basis
van korte afstanden faciliteert noemen we “ruimtelijke nabijheid”, zelfs al
is dit begrip in de verkennende fase van dit onderzoek nog niet eenduidig
gedefinieerd.
De verhandeling bestaat uit acht hoofdstukken. Het inleidende hoofd-
stuk schetst de context van het onderzoek en formuleert de
onderzoeksvragen. De hoofdstukken 2 tot en met 7 bestuderen elk een
afzonderlijk aspect van de probleemstelling en kunnen dan ook als
afzonderlijke artikels gelezen worden. Deze zes hoofdstukken werden all
gepubliceerd, of zullen gepubliceerd worden, in internationale peer-
reviewed tijdschriften. Een basisversie van elk van deze hoofdstukken
werd op minstens één internationaal wetenschappelijk congres gepresen-
teerd. In het achtste hoofdstuk worden de bevindingen tenslotte
samengevat, wordt de noodzakelijke kritische nuancering aangebracht, en
worden aanbevelingen voor het beleid geformuleerd.
S.2 Onderzoeksopzet
Het inleidende hoofdstuk geeft een overzicht van de klimaatwijziging en
peak-oil, twee mondiale verschijnselen die rechtstreeks verband houden
met externe effecten van mobiliteit, en legt verbanden met enkele boeien-
de aspecten van tijd- en ruimtebeleving en welvaartstoename die aan de
basis liggen van de groei van de mobiliteit. Ook wordt de mogelijke rol
van de ruimtelijke structuur in een toekomstgerichte benadering van de
mobiliteit onderzocht binnen de beleidscontext in Vlaanderen en Brussel2.
Op basis van de geschetste context worden vervolgens de onderzoeks-
vragen geformuleerd, en wordt een conceptueel kader opgesteld dat de
verwachte verbanden tussen de verschillende concepten visualiseert.
S.2.1 Onderzoeksvragen
Op basis van de beschouwingen uit Hoofdstuk 1 formuleren we de
onderzoeksvraag, die aan de basis ligt van deze verhandeling, als volgt:
• Tot op welke hoogte is de wederzijdse ruimtelijke nabijheid tussen
potentiële bestemmingen bepalend voor de dagelijks afgelegde afstan-
2 Gezien de geografische context werd Brussel in de meeste analyses binnen
deze verhandeling mee opgenomen, tenzij beperkingen in de beschikbare data
dit niet toelieten.
Samenvatting
271
den in Vlaanderen, en wat betekent dit in de context van peak-oil en
klimaatwijziging? (A)
Om de basisvraag te kunnen operationaliseren, splitsen we deze onmiddel-
lijk op in de volgende drie sub-vragen:
• Hoe kan de invloed van de ruimtelijke structuur op de dagelijks
afgelegde afstanden gekwantificeerd worden? (B1)
• Hoe kan ruimtelijke nabijheid gedefinieerd en gemeten worden, en
toegepast worden in de praktijk van duurzame ruimtelijke planning?
(B2)
• Welke locaties zijn het meest geschikt om bijkomende woningen en
jobs te realiseren, als we het bijkomend gegenereerd verkeer tot het
minimum willen beperken? (B3)
S.2.2 Conceptueel kader
Fig. S.1. Conceptueel kader
De gestelde onderzoeksvragen vormen de gebalde verwoording van een
onderzoekskader dat voortspruit uit de ruimere geschetste context.
Hierboven geven we een schematische weergave van dit conceptueel
kader, dat de verbanden tussen de verschillende aspecten van het onder-
zoek verduidelijkt (Fig. S.1).
Samenvatting
272
S.3 Bevindingen
S.3.1 Pendel
Hoofdstukken 2, 3, 4 en 6 behandelen enkel het pendelverkeer. Hoofdstuk
2 kan beschouwd worden als een verkennend onderzoek op basis van
relatief eenvoudige basisgegevens, assumpties en methodes. Dit hoofdstuk
toont aan dat er belangrijke regionale variaties bestaan inzake pendelaf-
standen en energieverbruik, bekeken vanuit de woonlocatie. Bovendien
blijkt het energieverbruik voor de pendel op regionale schaal sterk
gecorreleerd te zijn met de pendelafstand, wat betekent dat het belang
van de moduskeuze slechts van de tweede orde is. Deze bevindingen
worden in de context geplaatst van de stelling van Newman en Ken-
worthy (1989, 1999), die een sterk negatief verband vinden tussen
bevolkingsdichtheid en energieverbruik per capita voor transport. Ook in
Vlaanderen lijkt er een dergelijk, zij het vrij zwak, verband te bestaan.
De relatief beperkte focus van Hoofdstuk 2 roept de vraag op of de
ruimtelijke variatie in pendelafstanden verklaard kan worden door
variaties in de ruimtelijke nabijheid tussen woningen en jobs, en zo ja,
hoe deze nabijheid gemeten kan worden. Deze vraag is de basis voor
Hoofdstuk 3, waarin een methode ontwikkeld wordt om de ruimtelijke
nabijheid tussen woningen en jobs te berekenen op basis van ruimtelijk
gedesaggregeerde waarden voor de minimale pendelafstand. De minimale
pendelafstand is een concept dat afkomstig is uit de literatuur rond
bovenmatige pendel (“excess commuting”). Anders dan in Hoofdstuk 2
bevat de dataset die in deze paper werd gebruikt niet enkel de woonloca-
tie (oorsprong) maar ook de werklocatie (bestemming) van de
pendelstromen. Het is dan ook mogelijk om twee ruimtelijke nabijheids-
kaarten voor het pendelverkeer te ontwikkelen: één die de ruimtelijke
nabijheid van een woonlocatie ten opzichte van de jobmarkt weergeeft, en
één die de ruimtelijke nabijheid van een werklocatie ten opzichte van de
woonmarkt toont. Deze kaarten geven aan dat inwoners van relatief
afgelegen gebieden met een lage arbeidsbalans3 erg moeilijk een job
dichtbij huis kunnen vinden. Bovendien zijn de inwoners van deze regio’s
extra benadeeld wanneer alle werknemers collectief een job dichter bij
huis zouden gaan zoeken of dichter bij hun werk zouden gaan wonen.
3 De arbeidsbalans is de verhouding tussen het aantal jobs en het aantal
werkende inwoners in een gebied.
Samenvatting
273
Inwoners van de agglomeraties daarentegen hebben ruime marges om hun
pendelgedrag aan te passen, bijvoorbeeld onder invloed van stijgende
vervoerskosten. De belangrijkste oorzaak van dit fenomeen is het grote
verschil in ruimtelijke spreiding tussen de woonmarkt en de jobmarkt.
Jobs zijn doorgaans sterker geconcentreerd dan woningen, en heel wat
concentraties van jobs (in het bijzonder in industriegebieden) zijn relatief
ver verwijderd van de woonconcentraties. Verder lijken de resultaten van
Hoofdstuk 3 de vermoedens die Hoofdstuk 2 opriep te bevestigen: de
regio’s die gekenmerkt worden door een relatief lage mate van ruimtelijke
nabijheid, bekeken vanuit de woonlocatie, komen in grote lijnen overeen
met de regio’s waar we bovengemiddelde pendelafstanden waarnemen.
Met betrekking tot de werklocaties zijn het echter voornamelijk de regio’s
met een hoge dichtheid aan jobs die relatief slecht scoren op het vlak van
ruimtelijke nabijheid.
In Hoofdstuk 2 wordt een mogelijk verband gesuggereerd tussen de
toename van de gemiddelde pendelafstand in de loop der jaren en de
evolutie van de ruimtelijke nabijheid tussen woonmarkt en jobmarkt. In
Hoofdstuk 4 wordt deze hypothese onderzocht op basis van tijdsreeksen
voor de minimale pendelafstand, en deels ook voor een variant hierop (de
maximale pendelafstand). De ruimtelijke resolutie van de gebruikte
pendelmatrices is beperkt tot het gemeenteniveau, wat de bruikbaarheid
van de resultaten beperkt tot het bovengemeentelijke schaalniveau.
Niettemin is er in heel wat gemeenten een duidelijk waarneembare trend,
die gemiddeld wijst op een toename van de ruimtelijke scheiding tussen
woonmarkt en jobmarkt. De reden daarvoor is dat de toename van jobs
vooral te vinden is in de agglomeraties, terwijl de woningmarkt geken-
merkt wordt door een vorm van verdere ruimtelijke uitspreiding.
Belangrijker is echter de vaststelling dat de groei van het pendelverkeer
een stuk sneller verloopt dan de processen van ruimtelijke scheiding. We
kunnen dan ook besluiten dat de groei van het pendelverkeer een quasi-
autonoom, welvaartsgestuurd proces is dat een wisselwerking vertoont
met ruimtelijke scheidingsprocessen zonder dat er een causaal éénrich-
tingsverband bestaat.
Hoofdstuk 2 suggereert een omgekeerd verband tussen bevolkings-
dichtheid en energieverbruik per capita voor transport. Dit verband
wordt echter niet gekwantificeerd, noch wordt er een verband gelegd met
de ruimtelijke kenmerken van de werklocatie. Hoofdstuk 6 gaat dieper in
op dit probleem door eerst een aantal potentiële grootheden voor ruimte-
lijke nabijheid te selecteren, en vervolgens correlaties te berekenen tussen
Samenvatting
274
de geobserveerde pendelafstand en deze geselecteerde variabelen (aan
beide uiteinden van de pendelstroom). Bekeken vanuit de woonlocatie is
de woondichtheid, de ruimtelijke mix en de algemene bereikbaarheid
negatief gecorreleerd met de pendelafstand. De minimale pendelafstand is,
zoals verwacht op basis van Hoofdstuk 3, steeds positief gecorreleerd met
de pendelafstand, zowel in de woonlocaties als in de joblocaties. Jobdicht-
heid (in de joblocaties) is echter positief gecorreleerd met de
pendelafstand, in tegenstelling tot woondichtheid in de woonlocaties. Dit
betekent dat een hoge dichtheid niet leidt tot een duurzamer pendelpa-
troon als deze niet gepaard gaat met een hoge graad van functionele mix,
en dat een scheve arbeidsbalans bijdraagt tot een toename van de
pendelafstand. Met betrekking tot de arbeidsbalans kunnen we stellen dat
de pendelafstanden gemiddeld het kortste zijn als de herkomst of de
bestemming van de verplaatsing zich in een gebied bevindt waar de
arbeidsbalans nagenoeg in evenwicht is (d.w.z. gelijk is aan 1).
De resultaten van deze vier hoofdstukken moeten echter geïnterpre-
teerd worden binnen de beperkingen van de gebruikte data en de
gehanteerde methode. De belangrijkste randvoorwaarde is het ontbreken
van een differentiatie op basis van sector, inkomen of opleidingsniveau.
Het is namelijk best mogelijk dat hoogopgeleiden, die vaker gespeciali-
seerde jobs uitvoeren, minder vaak in de buurt van de voor hen geschikte
jobs wonen. Dit is bijvoorbeeld het geval voor kantoorjobs in het centrum
van Brussel die door werknemers uit de villawijken in de groene rand
worden ingevuld. Anderzijds komen jobs voor laagopgeleiden relatief vaak
voor in industriegebieden die niet steeds in de buurt van een woonkern of
agglomeratie liggen. Volgens Van Acker en Witlox (2011) is een hoger
inkomen geassocieerd met grotere pendelafstanden. Een differentiatie zou
deze problematiek scherper kunnen stellen en de rol van de ruimtelijke
mismatch tussen de kwalificatie en voorkeuren van de werknemers en de
vereisten die een bepaalde job stelt beter in kaart kunnen brengen.
Bovendien is de rol van het openbaar vervoer in deze deelonderzoeken
onderbelicht. Een hoge concentratie van jobs rondom een belangrijk
station zal in termen van energieverbruik een minder grote impact
hebben dan een concentratie van jobs op een afgelegen industrieterrein
dat nagenoeg enkel per auto bereikbaar is.
Samenvatting
275
S.3.2 Dagelijkse niet-pendel
Hoewel de bevindingen van Hoofdstukken 2, 3, 4 en 6 duidelijk aantonen
dat er een verband bestaat tussen de ruimtelijke nabijheid van woningen
en jobs en het pendelverkeer, mogen deze conclusies niet zomaar geëxtra-
poleerd worden naar andere vormen van quasi-dagelijkse mobiliteit.
Hoofdstukken 5 en 7 proberen de gehanteerde methodiek uit te breiden
met verplaatsingspatronen die niet tot de pendel behoren.
In Hoofdstuk 5 wordt het concept van de minimale pendelafstand en
bovenmatige pendel toegepast op quasi-dagelijkse verplaatsingen die niet
tot de pendel behoren. De resultaten worden weergegeven in de vorm van
een nabijheidskaart, die voor elke statistische sector de relatieve nabijheid
tot een korf van voorzieningen weergeeft. Daarnaast wordt voor elke
ruimtelijke categorie een geaggregeerde excesfactor berekend, die aangeeft
hoe sterk het verband is tussen de ruimtelijke nabijheid van de voorzie-
ningen en de effectief afgelegde afstanden.
De nabijheidskaart geeft een indicatie van welke gebieden in feite te
afgelegen zijn om een duurzaam verplaatsingspatroon in functie van het
bereiken van dagelijkse voorzieningen mogelijk te maken. De vergelijking
van de minimaal af te leggen afstand met de geobserveerde afgelegde
afstand geeft aan dat het verband tussen ruimtelijke nabijheid en ver-
plaatsingspatronen sterker is in de meest afgelegen gebieden dan in de
meer verstedelijkte gebieden, ook al is de afgelegde afstand gemiddeld
groter in de meer afgelegen gebieden. Met andere woorden: inwoners van
het platteland zullen vaker de dichtstbijzijnde mogelijke bestemming
kiezen dan inwoners van meer verstedelijkte gebieden. De grootstedelijke
gebieden en de agglomeraties, die samen de meest verstedelijkte van alle
onderzochte gebieden zijn, vormen echter een uitzondering: hier zijn de
effectief afgelegde afstanden kort, terwijl er toch relatief vaak voor de
dichtstbijzijnde mogelijke bestemming geopteerd wordt.
Hoofdstuk 7, tenslotte, combineert de variabelen die ontwikkeld wer-
den in Hoofdstuk 3, 5 en 6 met de analysemethode van Hoofdstuk 6. De
ruimtelijke correlatieanalyse van Hoofdstuk 6 wordt uitgebreid tot een
multivariate regressieanalyse op basis van een steekproef die geleverd
wordt door het Onderzoek Verplaatsingsgedrag van 2007 (Janssens et al.,
2009). Om een globaal beeld te geven van de impact van alle quasi-dage-
lijkse verplaatsingen samen werden pendel- en niet-pendelverplaatsingen
samengevoegd en werd dus de totale afgelegde afstand per respondent
geanalyseerd. Door een extrapolatie op basis van de bekomen regressie-
Samenvatting
276
vergelijking wordt een nieuwe, verfijnde, nabijheidskaart gegenereerd. In
tegenstelling tot de analyse in Hoofdstuk 6, worden in de analyse van
Hoofdstuk 7 enkel de ruimtelijke kenmerken van de woonlocatie van de
respondenten opgenomen, zodat de conclusies enkel geldig zijn met
betrekking tot de ruimtelijke kenmerken van de woonomgeving. Een
belangrijke conclusie van Hoofdstuk 7 is dat het belang van de ruimtelijke
distributie van jobs helemaal naar de achtergrond verdwijnt wanneer we
het volledige verplaatsingspatroon gaan bestuderen. De impact van de
ruimtelijke distributie van jobs op de totale mobiliteit is dus relatief
onbelangrijk in verhouding tot de ruimtelijke distributie van andere
quasi-dagelijkse bestemmingen. Bevolkingsdichtheid, functionele mix en
ruimtelijke nabijheid van voorzieningen blijven echter wel overeind als
bepalende factoren. Globaal blijkt de verklarende waarde van een model
op basis van ruimtelijke eigenschappen echter zeer beperkt te zijn. Dat
betekent dat pogingen om mobiliteitspatronen te doen krimpen op basis
van een ruimtelijk beleid gedoemd zijn om tot povere resultaten te leiden,
tenzij in combinatie met andere maatregelen (zoals bv. prijsbeleid) of
samengaand met (autonoom) stijgende energieprijzen.
Ook Hoofdstukken 5 en 7 hebben echter hun beperkingen. De belang-
rijkste lacune kan samengevat worden als het ontbreken van een analyse
op basis van de ruimtelijke kenmerken van de bestemming. In Hoofdstuk-
ken 3, 4 en 6 wordt ook de joblocatie in het onderzoek betrokken. In
Hoofdstukken 5 en 7 hebben we echter geen zicht op de invloed van de
locatie van de bezochte winkels, scholen, ontspanningscentra, ... De reden
waarom een dergelijke analyse buiten het bestek van deze verhandeling
valt, is enerzijds het gebrek aan een voldoende grote dataset die de
benodigde informatie bevat, en anderzijds de erg grote complexiteit van
dit soort onderzoek. Niettemin ligt hier een belangrijke piste voor moge-
lijk vervolgonderzoek. Er is voor België of Vlaanderen geen onderzoek
beschikbaar dat bijvoorbeeld het totale verschil in energetische efficiëntie
tussen hypermarkten en buurtwinkels kwantificeert, of ook tussen grote
en kleine scholen, of tussen meer en minder gespecialiseerde onderwijsin-
stellingen.
S.3.3 Niet-dagelijkse verplaatsingen
Occasionele (dus niet-dagelijkse) verplaatsingen, zoals daguitstapjes,
weekendjes, citytrips, vakanties of zakenreizen vormen geen voorwerp van
de zes deelstudies van deze verhandeling. Hoewel dit soort verplaatsingen
Samenvatting
277
slechts zelden wordt bestudeerd binnen het mobiliteitsonderzoek, is het
best mogelijk dat de afvlakking van de groei van de mobiliteit binnen het
segment van de quasi-dagelijkse verplaatsingen gecompenseerd wordt
door een voortdurende groei binnen het segment van de occasionele
mobiliteit. Deze verschuiving zou dan opnieuw snelheidsgedreven zijn,
voornamelijk op basis van de verregaande democratisering van het
luchtverkeer. Bovendien is er mogelijk ook een verband tussen het
verkeer, en dus het energieverbruik, dat het gevolg is van occasionele
mobiliteit, en de ruimtelijke structuur van de woonlocatie van de reiziger
(Holden en Norland, 2005).
Deze hypothese wordt verkennend getoetst in een addendum. We stel-
len vast dat inwoners van meer verstedelijkte gebieden vaker het vliegtuig
nemen dan inwoners van minder verstedelijkte gebieden. Hoewel de
beschikbare gegevens niet van die aard zijn dat verschillen in energiever-
bruik en CO2-uitstoot kunnen worden gekwantificeerd, lijkt de analyse te
suggereren dat, als we vliegverkeer in de analyse zouden opnemen, het
verplaatsingspatroon van de stadsbewoner wellicht een stuk minder
duurzaam is dan algemeen gedacht.
S.4 Aanbevelingen voor het ruimtelijk beleid
Banister (2008) volgend, stellen we een ontwikkelingsperspectief voor dat
gemotiveerd wordt vanuit het optimaliseren van de bereikbaarheid op
basis van een minimaal verkeersvolume, door het maximaliseren van de
ruimtelijke nabijheid en het valoriseren van agglomeratie-effecten. Uit het
voorgaande blijkt dat het sturen van ruimtelijke ontwikkelingen ten
eerste moet gezien worden als een manier om het systeem robuuster te
maken voor olieschaarste en duurder wordend transport, en ten tweede
als het werken aan een ruimtelijk kader dat geschikt is om klimaatbeleid
te faciliteren. Aan de hand van het stramien van Banister (1999, p. 316-
319) kunnen we een beknopte vertaling maken van onze onderzoeksresul-
taten in aanbevelingen voor het ruimtelijk beleid:
• Ontwikkelingen moeten zoveel mogelijk gebeuren op basis van een
hoge woondichtheid, ook in gebieden waar andere activiteiten dan
wonen voorkomen. Hoe hoog de woondichtheid moet zijn hangt af
van het ambitieniveau. Dichtheidsdrempels die in de literatuur wor-
den genoemd zijn meestal gekoppeld aan de gewenste rol van het
openbaar vervoer en worden besproken in Hoofdstuk 1. Een aantal
Samenvatting
278
nieuwe eco-wijken in het buitenland hebben een bruto-woondichtheid
die gaat tot zo’n 150 inwoners per hectare, wat vergelijkbaar is met
de dichtstbevolkte wijken in de 19e-eeuwse gordels van de Belgische
steden. Het ambitieniveau van het RSV is echter beperkt tot 25 wo-
ningen (equivalent met ongeveer 55 inwoners) per hectare in
stedelijke gebieden en 15 woningen (equivalent met ongeveer 38 in-
woners) per hectare in het buitengebied. Een voortschrijdende
gezinsverdunning dreigt de effectieve woondichtheid op basis van deze
normen (die uitgedrukt zijn in woningen per hectare) zelfs nog verder
naar beneden te halen.
• De omvang van de kernen moet voldoende groot zijn om op basis van
agglomeratievoordelen een relatief hoge mate van zelfvoorziening te
kunnen verzekeren. Banister (1999) volgend zouden we kunnen stellen
dat kernen die minder dan 25000 inwoners huisvesten in principe niet
meer zouden mogen groeien, en zouden geïsoleerde steden, die geen
deel uitmaken van een agglomeratie, op termijn een omvang moeten
bereiken van tenminste zo’n 50000 inwoners. Daarnaast zou groei zo-
veel mogelijk in of onmiddellijk aansluitend bij de agglomeraties
(> 250000 inwoners) moeten plaatsvinden, waar het openbaar vervoer
een belangrijke rol kan spelen.
• De locatie van nieuwe ontwikkelingen zou moeten gebeuren aanslui-
tend bij of in bestaande stedelijke gebieden. Daarbij moeten
faciliteiten en jobs tegelijkertijd met woningen gepland en ontwikkeld
worden, bij voorkeur geconcentreerd in lokale centra. Bovendien moe-
ten nieuwe ontwikkelingen zoveel mogelijk bereikbaar zijn zonder
auto.
Op basis van deze principes wordt in Hoofdstuk 8 een vertaling gemaakt
naar mogelijke aanpassingen aan het Ruimtelijk Structuurplan Vlaande-
ren (RSV, 1997/2004), waarvan we de belangrijkste hieronder opsommen:
• Bijkomende woningen zouden zoveel mogelijk in of onmiddellijk
aansluitend bij de agglomeraties moeten worden gebouwd, met een
hoge woondichtheid en een goede mix van functies. De agglomeraties
bestaan in grote lijnen uit de grootstedelijke en regionaalstedelijke
gebieden (zoals geselecteerd in het RSV), inclusief de suburbane gor-
dels die onmiddellijk bij deze steden aansluit.
• Binnen deze agglomeraties moet een uitgebalanceerde ruimtelijke
distributie van voorzieningen worden nagestreefd, bij voorkeur gecon-
centreerd in centra en subcentra. Schaalvergroting waarbij
verschillende vestigingen van bijvoorbeeld scholen, winkels of ateliers
Samenvatting
279
vervangen worden door één campus, hypermarkt of industrieterrein
moet ontmoedigd worden wegens de reductie van de ruimtelijke na-
bijheid die hier het gevolg van kan zijn. Op het vlak van openbaar
vervoer zou de verbetering van het interne openbaar-vervoernetwerk
in de agglomeraties prioriteit moeten krijgen.
• In de kleinstedelijke gebieden moet een volwaardig aanbod aan jobs
en voorzieningen geboden worden voor de inwoners van het omlig-
gende buitengebied, zodat de noodzaak voor verplaatsingen naar de
agglomeraties beperkt wordt. Bijkomende woningen zijn echter niet
gewenst in de kleinstedelijke gebieden en het buitengebied, aangezien
deze bijkomend lange-afstandsverkeer genereren.
• Zeer gespecialiseerde kantoorjobs zouden zoveel mogelijk op centrale
locaties bij belangrijke spoorstations moeten gesitueerd worden (Brus-
sel, Antwerpen, eventueel ook Mechelen, Gent en Leuven). Om lange-
afstandspendel niet aan te moedigen, zijn deze locaties echter minder
geschikt voor niet-gespecialiseerde jobs waarbij het de bedoeling is om
het personeel zoveel mogelijk lokaal te recruteren.
S.5 Verder onderzoek
Tot slot wordt een overzicht gegeven van mogelijke richtingen voor
verder onderzoek dat een bijdrage zou kunnen leveren tot een goed
geïnformeerde praktijk van ruimtelijke ordening, gericht op de ontwikke-
ling van een goede bereikbaarheid in combinatie met een duurzame vorm
van mobiliteit.
Referenties
Banister, D. (1999) “Planning more to travel less.” Town Planning
Review. 70(3), pp. 313-338.
Banister, D. (2008) “The sustainable mobility paradigm.” Transport
Policy. 15(2), pp. 73-80.
Holden, E. en I. Norland (2005) “Three challenges for the compact city as
a sustainable urban form: Household consumption of energy and
transport in eight residential areas in the greater Oslo region.” Urban
Studies. 42(12), pp. 2145–2166.
Samenvatting
280
Janssens, D., E. Moons, E. Nuyts en G. Wets (2009) Onderzoek Ver-
plaatsingsgedrag Vlaanderen 3 (2007-2008). Brussel-Diepenbeek:
Ministerie van de Vlaamse Gemeenschap.
Newman, P. en J. Kenworthy (1989) Cities and Automobile Dependence.
A Sourcebook. Aldershot: Gower.
Newman, P. en J. Kenworthy (1999) Sustainability and Cities: Overcom-
ing Automobile Dependence. Washington, DC: Island Press.
RSV (1997/2004) Ruimtelijk Structuurplan Vlaanderen - Gecoördineerde
Versie. Brussel: Ministerie van de Vlaamse Gemeenschap.
Van Acker, V. en F. Witlox (2011) “Commuting trips within tours: How
is commuting related to land use?” Transportation. doi: 10.1007
/s11116-010-9309-6.
Zhang, M. (2002) Conditions and Effectiveness of Land Use as a Mobility
Tool. PhD Thesis. Cambridge, MA: Massachusetts Institute of Tech-
nology.
281
Curriculum vitae Kobe Boussauw (°1978, Bruges) is a researcher
at the Geography Department of Ghent Univer-
sity (since November 2007). He graduated as a
civil engineer-architect (Ghent University, 2001)
and as a spatial planner (Ghent University,
2002), and obtained the certificate of post
academic traffic and mobility studies (University
of Antwerp, 2005). During his PhD research,
Kobe also finished the complete doctoral training
programme (Ghent University, 2011).
Kobe has worked as a consultant in a private company (iris consult-
ing, 2001-2003), as a civil servant for the Flemish Government (Planning
Department, 2003-2006), and as an advisor for the UN-Habitat pro-
gramme in Kosovo (2006-2007). Kobe’s PhD research was funded by the
Policy Research Centre on Regional Planning and Housing - Flanders
(Steunpunt Ruimte en Wonen 2007-2011).
Kobe is the author of several publications in international peer-
reviewed academic journals, and published also in various regional
journals. He acted as a referee for the journals Papers in Regional
Science, Regional Studies and Tijdschrift voor Economische en Sociale
Geografie, and for the Transportation Research Board’s annual meetings
(2010 and 2011). He presented his work at many national and interna-
tional conferences in Europe and the US. At the Transport Research
Arena conference in Brussels (2010), he was awarded a bronze medal for
his paper “Spatial variations in destination proximity: A regional case
study”. Kobe was also an occasional guest lecturer at Ghent University,
at the Artesis University College of Antwerp and at the University of
Prishtina.
Articles in international peer-reviewed journals included in the
ISI / Thomson Reuters Web of Science index
Boussauw, K. and F. Witlox (2009) “Introducing a commute-energy
performance index for Flanders.” Transportation Research Part A.
43(5), pp. 580-591.
Boussauw, K., T. Neutens and F. Witlox (2011) “Minimum commuting
distance as a spatial characteristic in a non-monocentric urban sys-
Curriculum vitae
282
tem: The case of Flanders.” Papers in Regional Science. 90(1), in
press.
Boussauw, K., T. Neutens and F. Witlox (2011) “Relationship between
spatial proximity and travel-to-work distance: The effect of the com-
pact city.” Regional Studies. in press.
Boussauw, K., B. Derudder and F. Witlox (2011) “Measuring spatial
separation processes through the minimum commute: The case of
Flanders.” European Journal of Transport and Infrastructure Re-
search. 11(1), pp. 42-60.
Boussauw, K., V. Van Acker and F. Witlox (2011) “Excess travel in non-
professional trips: Why looking for it miles away?” Tijdschrift voor
Economische en Sociale Geografie. in press.
Boussauw, K. and F. Witlox (2011) “Linking expected mobility produc-
tion to sustainable residential location planning: Some evidence from
Flanders.” Journal of Transport Geography. in press.
Articles in regional journals
Boussauw, K. (2006) “Sprawl of duurzaamheid? Leren van utopische
concepten voor stedelijke mobiliteit.” Ruimte en Planning. 26(1), pp.
22-33.
Boussauw, K., D. Lauwers and F. Witlox (2008) “Ruimtelijke structuur
en energieverbruik voor vervoer: Een eerste verkenning voor Vlaande-
ren.” Ruimte en Planning. 28(3), pp. 35-48.
Boussauw, K. (2008) “Stedenbouw in Kosovo: Visievorming voor transi-
tie.” Agora - Magazine voor sociaalruimtelijke vraagstukken. 24(4),
pp. 12-14.
Boussauw, K. (2009) “Stadsmens onderweg: een duurzaamheidsparadox.”
Agora - Magazine voor sociaalruimtelijke vraagstukken. 25(5), pp. 7-
10.
Boussauw, K. and F. Witlox (2009) “Theoretische minimale pendelaf-
stand als ruimtelijke karakteristiek: Een denkoefening.” Tijdschrift
Vervoerswetenschap. 45(4), pp. 150-158.
Boussauw, K., T. Neutens and F. Witlox (2010) “Pendel in en om de
compacte stad: Een ruimtelijke analyse van de afstand tot het werk.”
Ruimte & Maatschappij. 2(2), pp. 5-22.
Curriculum vitae
283
International conference papers
Boussauw, K. and F. Witlox (2008) Introducing a commute-energy
performance index for Flanders. 48th Congress of the European Re-
gional Science Association. Liverpool.
Boussauw, K. and F. Witlox (2009) Variation and evolution of minimum
commuting distances in Flanders. 4th Kuhmo-Nectar Conference and
Summer School. Copenhagen.
Boussauw, K., T. Neutens and F. Witlox (2010) Spatial variations of the
minimum home-to-work distance in the north of Belgium. Transpor-
tation Research Board 2010 Annual Meeting. Washington, DC.
Boussauw, K., T. Neutens and F. Witlox (2010) Does spatial proximity
influence commuting trip length? An approach based on evidence
from Flanders and Brussels. Transportation Research Board 2010
Annual Meeting. Washington, DC.
Boussauw, K. and F. Witlox (2010) Spatial variations in destination
proximity: A regional case study. Transport Research Arena 2010.
Brussels.
Boussauw, K. (2010) Aspects of spatial proximity and sustainable travel
behavior in Flanders: A quantitative approach. Association of Colle-
giate Schools of Planning 2010 PhD Workshop. Atlanta, GA.
Boussauw, K. and F. Witlox (2010) Excess travel in non-commuting
trips: A regional case study. World Conference on Transport Re-
search. Lisbon.
Boussauw, K. and F. Witlox (2011) Regional variations in travel energy
consumption: Some evidence from Flanders. Mobil.TUM 2011 Con-
ference. Munich.
Boussauw, K. and F. Witlox (2011) The role of spatial proximity in daily
mobility production: A case study in the North of Belgium. Annual
Meeting of the Association of American Geographers. Seattle.
National and regional conference papers
Boussauw, K., D. Lauwers and F. Witlox (2008) Ruimtelijke structuur en
energieverbruik voor vervoer: Een eerste verkenning voor Vlaanderen.
Uitdagingen voor de ruimtelijke ordening in Vlaanderen. Brussels.
Boussauw, K. and F. Witlox (2008) L’introduction d’un indice de per-
formance énergétique pour la navette en Flandre et Bruxelles. Derde
Belgische dagen van de geografie - Troisièmes journées belges de la
géographie. Brussels.
Curriculum vitae
284
Boussauw, K., D. Lauwers and F. Witlox (2008) En wat als de olie op is?
De relatie tussen ruimte en energieverbruik voor vervoer. Colloquium
Vervoersplanologisch Speurwerk 2008. Santpoort (The Netherlands).
Boussauw, K. and F. Witlox (2008) Kilometers malen: Maatschappelijke
dwangneurose of ruimtelijk probleem? Duurzame mobiliteit Vlaande-
ren: De leefbare stad. Ghent, 205-228.
Boussauw, K. and F. Witlox (2009) De wereld om de hoek: Is ruimtelijke
nabijheid maakbaar? PlanDag 2009. G. Bouma, F. Filius, H. Lein-
felder and B. Waterhout. Brussels, pp. 425-434.
Boussauw, K., T. Neutens and F. Witlox (2009) “Excess commuting in
Flanders and Brussels”. In: C. Macharis and L. Turcksin (Eds.)
BIVEC-GIBET Transport Research Day 2009. Brussels: VUBPress,
pp. 157-171.
Boussauw, K., T. Neutens and F. Witlox (2009) Pendelgedrag en nabij-
heid: Speelt de compacte stad haar rol? Colloquium Vervoers-
planologisch Speurwerk 2009. Antwerp.
Boussauw, K. and F. Witlox (2010) Travel energy consumption and the
built environment: Evidence from Flanders. Colloque de la Con-
férence Permanente du Développement Territorial 2010. Liège.
Boussauw, K. and F. Witlox (2010) A residential location model based on
characteristics of spatial proximity: A case study in the North of Bel-
gium. 11th TRAIL Congress. Rotterdam.
Book chapters
Boussauw, K., Zwerts, E. and F. Witlox (2009) “Mobiel Vlaanderen”. In:
L. Vanderleyden, M. Callens and J. Noppe (Eds.) De Sociale Staat
van Vlaanderen 2009. Brussel: Studiedienst van de Vlaamse Reger-
ing, pp. 279-312.
Boussauw, K., D. Lauwers and F. Witlox (2009) “Ruimtelijke structuur
en energieverbruik voor vervoer: Een eerste verkenning voor Vlaande-
ren”. In: Re-Marc-able Landscapes - Marc-ante Landschappen: Liber
Amicorum Marc Antrop. Gent: Academia Press, pp. 235-246.
Witlox, F., K. Boussauw, W. Debauche, B. Derudder, C. Macharis and S.
Verlinde (2009) “Vandaag besteld, vannacht geleverd. Over de moge-
lijkheden van nacht- en daldistributie als oplossing voor het probleem
van het stedelijk goederenvervoer in België”. In: C. Kesteloot, M.
Goossens, H. Van der Haegen et al. (Eds.) Bas-Congo tot Dadizele.
Veelzijdigheid in de Geografie. Liber Amicorum Etienne Van Hecke.
Curriculum vitae
285
Leuven: KULeuven, Instituut voor Sociale en Economische Geografie,
pp. 227-242.
Others
Boussauw, K., D. Lauwers, E. Zwerts and F. Witlox (2009) Visie Ruim-
tegebruik en Ruimtebeslag 2020-2050: Sectornota Mobiliteit. Ghent:
Steunpunt Ruimte en Wonen.
Boussauw, K. (2009) Hoe duurzaam leeft de stadsbewoner? Low Tech
Magazine from http://www.lowtechmagazine.be/2009/12/hoe-duur
zaam-is-de-stadsbewoner.html.
Boussauw, K., R. Simoen and F. Witlox (2010) Focusnota Ruimte,
Logistiek en Multimodaliteit. Ghent: Steunpunt Ruimte en Wonen.
Zwerts, E., K. Boussauw, L. Bral, P. De Maeyer, B. Derudder, V. Van
Acker, L. Verdonck and F. Witlox (2010) Algemeen profiel Oost-
Vlaamse O&O-bedrijven. Ghent: POM Oost-Vlaanderen.
284
285
286
This dissertation aims to contribute to the insight in the reciprocal relationship
between person mobility and spatial development, taking into account the societal
context of climate targets and imminent peak oil. This is done through the
development of a number of quantitative research methods, which are embedded
in a literature review and is applied to the case study of Flanders (Belgium).
The research focuses on exploring the sustainability of spatial structure with
respect to travel behaviour, with particular attention to the daily distances
travelled. Sustainability is defined in terms of resilience, not only for growing
mobility but also for a possible declining future mobility, a scenario that is
suggested by peak oil theory or may be the consequence of a stringent climate
policy. Moreover, spatial structure plays a role in the potential steering of travel
behaviour in a more sustainable direction.
From this point of view, the dissertation assesses to what extent mutual spa-
tial proximity between potential destinations is determining the daily distances
covered in Flanders, and how spatial development can play a role in pursuing a
high degree of accessibility based on a minimum amount of traffic.
Kobe Boussauw (°1978, Bruges) is a researcher at the Geography Department of
Ghent University. He is a civil engineer-architect and a spatial planner. Before,
Kobe has worked as a consultant in a private company, as a civil servant for the
Flemish Government, and as an advisor for the UN-Habitat programme in
Kosovo. Kobe’s PhD research was funded by the Policy Research Centre on
Regional Planning and Housing - Flanders (Steunpunt Ruimte en Wonen 2007-
2011).