Socio Economic Applications in Geographical

Socio Economic Applications in Geographical
Information Science
Socio-Economic
Applications of Geographic
Information Science
Editors David Kidner, Gary Higgs and Sean White

First published 2003 by Taylor & Francis
11 New Fetter Lane, London EC4P 4EE
Simultaneously published in the USA and Canada
By Taylor & Francis Inc,
29 West 35th Street, New York, NY 10001
Taylor & Francis is an imprint of the Taylor & Francis Group
This edition published in the Taylor & Francis e-Library, 2005.
“To purchase your own copy of this or any of Taylor & Francis or Routledge’s
collection of thousands of eBooks please go to www.eBookstore.tandf.co.uk.”
© 2002 Taylor & Francis
Publisher’s Note
This book has been prepared from camera-ready copy provided by the editors.
All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or
by any electronic, mechanical or other means, now known or hereafter invented, including
photocopying and recording, or in any information storage or retrieval system, without permission
in writing from the publishers.
Every effort has been made to ensure that the advice and information in this book is true and
accurate at the time of going to press. However, neither the publisher nor the authors can accept
any legal responsibility or liability for any errors or omissions that may be made. In the case of drug
administration, any medical procedure of the use of technical equipment mentioned within this
book, you are strongly advised to consult the manufacturer’s guidelines.
British Library Cataloguing in Publication Data
A catalogue record for this book is available from the British Library
Library of Congress Cataloging in Publication Data
A catalog record for this book has been requested
ISBN 0-203-30107-2 Master e-book ISBN
ISBN 0-203-34562-2 (Adobe eReader Format)

ISBN 0-415-27910-0 (Print Edition)
Contents
Preface vi
Contributors vii
GISRUK Committees and Sponsors xii
1. Introduction 1
Gary Higgs, David Kidner and Sean White
2. Revisiting fundamental principles of GIS 10
Nicholas Chrisman
PART I: GIS AND CRIME 21

3. When is a hotspot a hotspot? A procedure for creating statistically robust 22
hotspot maps of crime
Spencer Chainey, Svein Reid and Neil Stuart
4. Evaluating situational crime prevention: the Merseyside ‘alleygating’ 37
schemes
Chris Young, Alex Hirschfield, Kate Bowers and Shane Johnson
5. Crime hot spot prediction: a framework for progress 51
Jonathan Corcoran and Andrew Ware
PART II: GIS AND PLANNING 67

6. E-community participation: communicating spatial planning and design 68
using web-based maps
Kheir Al-Kodmany
7. A collaborative three dimensional GIS for London: Phase 1 Woodberry 92
Down
Andrew Hudson-Smith and Steve Evans
8. Historic time horizons in GIS: East of England historic landscape 106
assessment
Lynn Dyson-Bruce
9. Using GIS to research low and changing demand for housing 118
Peter Lee and Brendan Nevin
v
PART III: GIS AND URBAN APPLICATIONS 133

10. Geographical visual information system (GVIS) to support urban 134
regeneration: design issues
Xiaonan Zhang, Nigel Trodd and Andy Hamilton
11. Using measures of spatial autocorrelation to describe socio-economic and 146
racial residential patterns in US urban areas
Andrea Frank
12. Georeferencing social spatial data and intra-urban property price 162
modelling in a data-poor context: a case study for Shanghai
Fulong Wu
PART IV: GIS AND RURAL APPLICATIONS 178

13. Accessibility to GP surgeries in South Norfolk: a GIS-based assessment of 179
the changing situation 1997–2000
Andrew Lovett, Gisela Sünnenberg and Robin Haynes
14. Measuring accessibility for remote rural populations 197
Mandy Kelly, Robin Flowerdew, Brian Francis and Juliet Harman
15. Assessing the transport implications of housing and facility provision in 210
Gloucestershire
Helena Titheridge
PART V: GIS IN SOCIO-ECONOMIC POLICY 226

16. Using GIS for sub-ward measures of urban deprivation in Brent, England 227
Richard Harris and Martin Frost
17. The spatial analysis of UK local electoral behaviour: turnout in a Bristol 239
ward
Scott Orford and Andrew Schuman
18. Towards a European peripherality index 255
Carsten Schürmann and Ahmed Talaat
19. Using a mixed-method approach to investigate the use of GIS within the 267
UK National Health Service
Darren Smith, Gary Higgs and Myles Gould
Index 279
Preface
This volume contains papers presented at the 9th Annual GIS Research UK conference held at
the University of Glamorgan and co-hosted by the University of Wales, Cardiff. We are proud
to declare that this was the most well attended GISRUK conference to date and had the highest
number of paper submissions. This stemmed from an increasing interest in GIS Research from
both within, and outside, the academic community and reflects a growing maturity in the use
of GIS in a number of different sectors. The Local Organising Committee for GISRUK 2001 made
a conscious effort to target a broader range of papers to appeal to a wider audience and this led
to sessions containing papers arranged according to different policy sectors which together
occupied the middle day of the conference. This is reflected in the diversity of papers included
in this volume which, to a certain extent, complements those included in last years volume
edited by Peter Halls which was largely concerned with the innovative use of GIS in
environmental applications. The aim here is to reflect the significant body of innovative
research that is being conducted largely in the social sciences. At the same time, we have not
neglected important research initiatives in the physical sciences—papers arranged around these
themes will be included in special issues edited by the co-chairs of the meeting. Of course,
many of the papers presented here transcend such arbitrary, and increasingly fuzzy, boundaries
and a key message from the papers presented here and in the other conference outputs, is that
many of the techniques developed in either ‘domain’ are indeed transferable. This, in turn,
suggests that the oft-repeated claim that GIS is helping to break down barriers between the
physical and social world is being realised through a host of interdisciplinary initiatives.
At this point, it is appropriate to remind readers of the aims of the GISRUK conference,
which are:
• to act as a focus for GIS Research in the UK;

• to provide a mechanism for the announcement and publication of GIS Research;
• to act as an interdisciplinary forum for the discussion of research ideas;
• to promote active collaboration amongst UK researchers from diverse parent disciplines;
and
• to provide a framework in which postgraduate students can see their work in a national
context.
This year’s programme, and attendance list, suggests that GISRUK has evolved into something
more than just a British forum for GIS research. Approximately one third of the presentations
were made up from International contributions, which is very encouraging. We have made a
vii
deliberate effort to include contributions from a range of social and physical environments in
our conference outputs which, we hope, will add to the international appeal of the volume.
The conference included plenary keynote addresses from Vanessa Lawrence (Ordnance
Survey), David Maguire (ESRI), Martien Molenaar (ITC) and Nick Chrisman (University of
Washington). We are delighted to include Nick Chrisman’s contribution as the opening
chapter in this volume, which explores a number of important research challenges addressed
by many of the chapter contributors. As in previous years, a prize (sponsored by the
Association for Geographic Information) was awarded to the best paper presented at GISRUK
2001 and this year the prize was given to Anna Symington of the University of Newcastle (to
our knowledge, the youngest presenter at the meeting) for her research exploring the use of
statistical techniques to trace errors when assessing map lineage. In addition a prize, sponsored
by CADCORP Ltd, was given to the best poster presentation and, after much discussion
amongst the Steering Committee, this hotly contested award was given to Oliver Duke-
Williams and John Stillwell of the University of Leeds for their poster illustrating their
research on developing web-based interfaces to access migration and travel to work statistics
from the 1981 and 1991 UK censuses.
As in previous years, GISRUK actively encourages full involvement from postgraduate and
young researchers with the help of substantial registration discounts and Bursary awards, both
of which are made possible by our sponsors (AGI, RRL.net and Ordnance Survey). The Young
Researchers Forum (YRF) which precedes the main conference is a unique component of
GISRUK and we would like to thank those members of the national and local organising
committee, as well as our keynote speakers, who attended this year’s forum. Particular thanks
should go to Peter Halls and Scott Orford who helped put an extremely interesting and varied
timetable together for this year’s forum. Feedback from this and previous years’ YRFs has been
very encouraging and it is hoped that they continue as a regular feature of the GISRUK
meetings.
The Steering Committee are enormously grateful to the sponsors (listed on page xiv) who
generously supported GISRUK 2001. We would particularly like to thank Taylor and Francis
and the AGI for sponsoring the keynote presentations this year and to Wileys and Elsevier for
their sponsorship of the wine receptions, always a key feature of GISRUK! We would also like
to thank the staff at the Glamorgan Business Centre, and the Local Organising Committee
(listed on page xiv), for their support during the planning of the meeting. The continued
support of members of the National Organising Committee is much appreciated. A particular
thanks should go to Jan Cross and Caroline Bowen for all their help in the day-to-day running
of the conference. To all: Diolch yn fawr.
Lastly, to keep informed of future GISRUK conferences, please bookmark: http://
www.geo.ed.ac.uk/gisruk/gisruk.html.
Gary Higgs, David Kidner and Sean White
Trefforest
January 2002
Contributors
Kheir Al-Kodmany
University of Illinois at Chicago, UIC, United States of America
E-mail: Kheir@uic.edu
Kate Bowers
URPERRL, Department of Civic Design, University of Liverpool, Liverpool, L69 7ZQ, United Kingdom
E-mail: kjb@liverpool.ac.uk
Spencer Chainey
InfoTech Enterprises Europe, 40–42 Parker Street, London, WC2B 5PQ, United Kingdom
E-mail: schainey@infotech-europe.com
Nicholas Chrisman
Department of Geography, University of Washington, Seattle, WA 98195–3550, United States of
America
E-mail: chrisman@u.washington.edu
Jonathan Corcoran
GIS Research Centre, School of Computing, University of Glamorgan, Pontypridd, CF37 1DL, United
Kingdom
E-mail: jcorcora@glam.ac.uk
Lynn Dyson-Bruce
Planning Division, Essex County Council, County Hall, Chelmsford, Essex, CM1 1QH, United Kingdom
E-mail: lynn.dyson-bruce@essexcc.gov.uk
Steve Evans
Centre for Advanced Spatial Analysis, University College London, 1–19 Torrington Place, WC1E 6BT,
United Kingdom
E-mail: sevans@geog.ucl.ac.uk
Robin Flowerdew
School of Geography and GeoSciences, University of St Andrews, KY16 9AL, United Kingdom
E-mail: r.flowerdew@st-andrews.ac.uk
Brian Francis
Lancaster University, Centre for Applied Statistics, Lancaster, LA1 4YB, United Kingdom
E-mail: b.francis@lancaster.ac.uk
Andrea Frank
Department of City and Regional Planning, University of Wales, Cardiff, CF10 3WA, United Kingdom
E-mail: FrankA@cardiff.ac.uk
Martin Frost
School of Geography, Birkbeck College, University of London, 7–15 Gresse Street, London, W1P 2LL,
United Kingdom
E-mail: m.frost@bbk.ac.uk
Myles Gould
School of Geography, University of Leeds, Leeds, LS2 9JT, United Kingdom
ix
E-mail: M.Gould@geog.leeds.ac.uk
Andy Hamilton
School of Construction and Property Management, Salford University, Salford, M7 9NU, United Kingdom
E-mail: a.hamilton@salford.ac.uk
Juliet Harman
Lancaster University, Centre for Applied Statistics, Lancaster, LAl 4YB, United Kingdom
E-mail: j harman@lancaster.ac.uk
Richard Harris
School of Geography, Birkbeck College, University of London, 7–15 Gresse Street, London, W1P 2LL,
United Kingdom
E-mail: r.harris@bbk.ac.uk
Robin Haynes
School of Environmental Sciences and School of Health Policy and Practice, University of East Anglia,
Norwich, NR4 7TJ, United Kingdom
E-mail: R.Haynes@uea.ac.uk
Gary Higgs
Kingdom
E-mail: ghiggs@glam.ac.uk
Alex Hirschfleld
URPERRL, Department of Civic Design, University of Liverpool, Liverpool, L69 3BX, United Kingdom
E-mail: hirsch@liv.ac.uk
Andrew Hudson-Smith
Centre for Advanced Spatial Analysis, University College London, 1–19 Torrington Place, WC1E 6BT,
United Kingdom
E-mail: asmith@geog.ucl.ac.uk
Shane Johnson
E-mail: s.d.johnson@liverpool.ac.uk
Mandy Kelly
Lancaster University, North-West Regional Research Laboratory, Department of Geography, Lancaster,
LA1 4YB, United Kingdom
E-mail: Mandy.Kelly@lancaster.ac.uk
David Kidner
Kingdom
E-mail: dbkidner@glam.ac.uk
Peter Lee
Centre for Urban and Regional Studies, University of Birmingham, JG Smith Building, Pritchatts Road,
Edgbaston, Birmingham, B15 2TT, United Kingdom
E-mail: p.w.lee@bham.ac.uk
x
Andrew Lovett
School of Environmental Sciences, University of East Anglia, Norwich, NR4 7TJ, United Kingdom
E-mail: A.Lovett@uea.ac.uk
Brendan Nevin
Centre for Urban and Regional Studies, University of Birmingham, JG Smith Building, Pritchatts Road,
Edgbaston, Birmingham, B15 2TT, United Kingdom
E-mail: b.nevin@bham.ac.uk
Scott Orford
Department of City and Regional Planning, University of Wales, Cardiff, CF10 3WA, United Kingdom
E-mail: OrfordS@cardiff.ac.uk
Svein Reid
Department of Geography, University of Edinburgh, Drummond Street, Edinburgh, EH8 9XP, United
Kingdom
Andrew Schuman
Local Government Chronicle Elections Centre, University of Plymouth, Plymouth, PL4 8AA, United
Kingdom
E-mail: Andrew.Schumann@bris.ac.uk
Carsten Schürmann
Institute of Spatial Planning, University of Dortmund, August-Schmidt-Str-6, D-44221, Dortmund,
Germany
E-mail: cs@irpud.rp.uni-dortmund-de
Darren Smith
Department of Geography, School of the Built Environment, University of Brighton, Brighton, BN2 4GJ,
United Kingdom
E-mail: D.Smith@brighton.ac.uk
Neil Stuart
Department of Geography, University of Edinburgh, Drummond Street, Edinburgh, EH8 9XP, United
Kingdom
E-mail: ns@geo.ed.ac.uk
Gisela Sünnenberg
School of Environmental Sciences and School of Health Policy and Practice, University of East Anglia,
Norwich, NR4 7TJ, United Kingdom
E-mail: G.Sunnenberg@uea.ac.uk
Ahmed Talaat
Institute of Spatial Planning, University of Dortmund, August-Schmidt-Str-6, D-44221, Dortmund,
Germany
E-mail: at@irpud.rp.uni-dortmund-de
Helena Titheridge
The Bartlett School of Planning, UCL, 22 Gordon Street, London, WC1H OQB, United Kingdom
E-mail: h.titheridge@ucl.ac.uk
Nigel Trodd
xi
Telford Institute of Environmental Systems (TIES), School of Environment and Life Science, Salford
University, M5 4WT, United Kingdom
E-mail: n.m.trodd@salford.ac.uk
Andrew Ware
School of Computing, University of Glamorgan, Pontypridd, CF37 IDL, United Kingdom
E-mail: jaware@glam.ac.uk
Scan White
Department of City and Regional Planning, University of Wales, Cardiff Cardiff CF10 3WA, United
Kingdom
E-mail: WhiteSD@cardiff.ac.uk
Fulong Wu
Department of Geography, University of Southampton, Southampton, S017 1BJ, United Kingdom
E-mail: F.Wu@soton.ac.uk
Chris Young
E-mail: youngca@liv.ac.uk
Xiaonan Zhang
Telford Institute of Environmental Systems (TIES), School of Environment and Life Science, Salford
University, M5 4WT, United Kingdom
E-mail: X.Zhang1@pgr.salford.ac.uk
GISRUK Committees and Sponsors
GISRUK National Steering Committee (as of GISRUK 2001)
Steve Carver University of Leeds steve@geog.leeds.ac.uk

Jane Drummond University of Glasgow jdrummond@geog.gla.ac.uk
Dave Fairbairn University of Newcastle dave.fairbairn@newcastle.ac.uk
Bruce Gittings (Chair) University of Edinburgh bruce@geo.ed.ac.uk
Peter Halls University of York pjhl@york.ac.uk
Gary Higgs University of Glamorgan ghiggs@glam.ac.uk
Zarine Kemp University of Kent zk@ukc.ac.uk
David Kidner University of Glamorgan dbkidner@glam.ac.uk
Andrew Lovett University of East Anglia a.lovett@uea.ac.uk
David Miller The Macaulay Institute mi008@mluri.sari.ac.uk
George Taylor University of Glamorgan getaylor@glam.ac.uk
Steve Wise University of Sheffield s.wise@sheffield.ac.uk
Jo Wood City University jwo@soi.city.ac.uk
GISRUK 2001 Local Organising Committee
David Kidner (Co-Chair) Linus Mofor Nathan Thomas

Gary Higgs (Co-Chair) Andrew Morris Andrew Ware
Jonathan Corcoran Scott Orford Mark Ware
Chris Jones Alun Rogers Chris Webster
Marc Le Blanc George Taylor Scan White
Mike Lonergan Malcolm Thomas
GISRUK 2001 Sponsors

The GISRUK Steering Committee is enormously grateful to the following organisations who
generously supported GISRUK 2001:
The Association for Geographic Information (AGI) Blackwell Publishers

Cadcorp Edina
Elsevier Science ESRI (UK) Ltd
xiii
Geo-Europe Ordnance Survey

Oxford University Press Pearson Education
The Quantitative Methods Research Group of the RGS/IBG RRL.Net
Taylor & Francis Ltd John Wiley and Sons
1
Introduction
Gary Higgs, David B.Kidner and Sean D.White
1.1
SOCIO-ECONOMIC APPLICATIONS IN GEOGRAPHICAL
INFORMATION SCIENCE
In a previous volume in this series papers were included that demonstrated the use of
innovative research techniques in environmental studies (Halls, 2000). The aim of this volume
is to include papers presented at GISRUK 2001 that illustrate the state-of-the-art in
researching socio-economic applications with GIS. In so-doing, we recognise that the
intertwined relationships between the physical and human world mean that such arbitrary
divisions are not straightforward. There are clearly environmental aspects to the majority of
papers that follow, but at the same time, given the quality of papers presented at the
conference, we have taken this opportunity to demonstrate the dynamic nature of GIS research
in what would traditionally be called ‘human geography’. A number of texts have been
concerned with focusing on the use of GIS in socio-economic applications (e.g. Martin, 1996)
and a number of others contain extensive overviews of the use of GIS in different policy
sectors (e.g. Longley et al., 1999). There have also been texts that have examined the use of
GIS in application areas such as crime (Goldsmith et al., 2000; Hirschfield and Bowers, 2001),
health (Gatrell and Loytonen, 1998; Hay et al., 2000) and transport (Thill, 2000). Previous
volumes in the Innovations in GIS series have variously contained papers related to such topics
and readers are advised to consult these to get a flavour of the significant research
developments that have taken place since the first GIS Research UK conference held in Keele
in 1993. However, to date, no one volume has been given over to solely highlighting the use
of up-to-date GIS-based techniques in a range of socio-economic applications. In this volume
we redress this ‘gap’.
In the following chapter, Nicholas Chrisman examines how social issues are embedded in
the different stages of GIS development. In particular, he examines the limitations of
technological determinism from a GIS perspective and draws on wider, more theoretical
literatures, to call for a greater understanding of the interactions between people,
organisations and such technologies. This is illustrated with a number of examples that draw
attention to the role of software vendors in influencing GIS users in their work from a number
of different, but inter-connected, perspectives. This, in turn, points to the potential benefits to
be gained from greater participation of users in software development processes and their
continued independence and objectivity in GIS research. This provides a valuable introduction
2 SOCIO-ECONOMIC APPLICATIONS OF GEOGRAPHIC INFORMATION SCIENCE
for many of the chapters in this volume which are largely concerned with enhancing existing
software tools in innovative ways to address social applications.
The rest of the book is composed of seventeen chapters, divided into five parts, which focus
on the use of GIS in different sectors. In the first section, we focus on the integration of
analytical techniques with GIS in order to investigate spatial patterns of crime incidence. GIS is
increasingly being recognised as a vital tool in the exploration of spatial trends, in the
prediction of crime events and in the evaluation of crime prevention measures. In the UK,
much of the impetus for the use of GIS in crime analysis has resulted from legislation such as
the Crime and Disorder Act (1998) which made annual crime and disorder audits a statutory
requirement. As a result, GIS has been used extensively in the preparation of maps and in
preliminary analyses of spatial data patterns in the majority of police authorities. In addition,
the Home Office document “Guidance on Statutory Crime and Disorder Partnerships” (Stationary
Office, 1998) has emphasised the potential for GIS in extracting patterns from audits. One of
the recommendations in the guidance documents was that agencies such as local authorities,
police and social services need to work together to produce such audits. Together with their
use in community safety strategies for local areas, GIS is seen as an ideal tool to permit such
‘joined-up’ working. However, many of these authorities are in early stages in their GIS
development and the 3 chapters in this section, illustrate how GIS can be used in a more
advanced way to address a number of current strands in researching crime patterns.
The second section of the book contains 4 chapters that focus on the use of GIS in planning
tasks. Planning departments in the UK tend to be the lead department in the majority of local
authority GIS implementations having traditionally been major users of spatial information and
paper maps (Campbell and Masser, 1995). However, despite having a relatively long history of
use within the profession, GIS continues to be utilised predominantly to address relatively low
level, operational and routine tasks, contrasting to their use in more strategic contexts which has
traditionally been very limited (Gill et al., 1999). The reasons for this are complex and varied,
but the contributions contained in this section, point the way towards the more innovative use
of GIS in areas such as using web-based tools in public participation and collaborative visual
assessment. One sector where GIS has made less in-roads is in the study of housing issues.
Once again, GIS has been used as an operational tool in a significant number of agencies in
their housing management functions but less research has been conducted, certainly in the UK
context, in using GIS to examine changing demand-supply relationships through the
integration of spatial analytical techniques. Clearly, there are many other application areas
which are of potential concern to planners, and arguably many of the contributions in the other
sections of the book could have been included here, but these chapters begin to examine the
rich contribution GIS can make in more strategic contexts.
In the third section, contributions focus on urban dimensions and the three chapters
contained herein variously describe the use of GIS in urban regeneration, in examining spatial
trends in socio-economic patterns and in exploring intra-urban variations in house prices. GIS
has been used extensively in the latter to examine the impacts, for example, of access to urban
services and facilities on house prices (e.g. Orford, 1999). Again, the visualisation aspects of
GIS technologies come to the fore here. One of the current research strands pre-occupying
many research groups relates to the use of GIS in measuring access to facilities in urban and
rural contexts and in section four, we include three chapters which although focusing on such
issues from a rural perspective, highlight methodologies that could easily be adapted for urban
scenarios. In particular, these chapters draw attention to the limitations of simplistic measures
based on ‘crow-fly’ distances by demonstrating the potential to incorporate travel times into
such indicators. Further research is on-going to include information from public transport
INTRODUCTION 3
timetables into such analyses and this continues to be a fruitful area for farther research. The
increasing importance of Global Positioning Systems (GPS) in such studies is also likely to form
a vital component of future papers given at the annual conference.
In the last section of the volume, we focus on studies of the use of GIS in various aspects of
socio-economic research. GIS has been used extensively to develop indices of deprivation in
both urban and rural contexts (see Higgs and White, 2000 for examples of the latter).
However, here we are concerned with the derivation of measures using non-census based
sources, or more accurately lifestyle databases, that tend not to be made available at detailed
spatial scales. Such datasets have real potential for researchers concerned with investigating socio-
economic patterns in inter-censal years as well as those concerned with a wider range of urban
issues (Longley and Harris, 1999). Other chapters in this section are concerned with using GIS
to investigate aspects of electoral geography and in the development of indices of
peripherality. The final chapter provides a detailed account of how we can begin to investigate
current uses of GIS in researching health issues within the UK National Health Service using
multi-method approaches. Although we acknowledge there are many other application areas
that are not covered in this volume, and that many of the chapters could easily have been
placed in a number of sections, we believe that these contributions are both illustrative of the
types of research initiatives underway in the UK and beyond, and that many of the techniques
described are transferable to other application areas which have not been included this time
round.
1.2
GIS AND CRIME
In Chapter 3, Spencer Chainey, Svein Reid and Neil Stuart highlight techniques for
describing clusters of crime patterns or so-called ‘hotspot maps’. They contrast a number of
approaches to creating continuous surface maps according to different criteria including ease of
use and interpretation, relevance for particular crime data sets and visual appearance. Often
despite the use of sophisticated techniques to create such surfaces, it is ultimately up to the
user to decide when a cluster can be identified as a hotspot. To address such issues, the authors
apply point pattern analysis, and in particular local tests of spatial association as well as spatial
autocorrelation, to create surfaces for, in this case, four different categories of crime over a 3
month time period. They contrast different techniques for defining thresholds when applied to
these surfaces and compare the outcomes for different types of crime. As Chainey et al.
suggest, the use of space-time methods to explore temporal clusters could prove a useful
addition to the range of techniques available to agencies and there are a number of research
initiatives underway to apply such techniques (e.g. Ratcliffe, 1998).
Chris Young, Alex Hirschfleld, Kate Bowers and Shane Johnson explore another
application area for GIS in relation to crime analysis in Chapter 4, namely that of evaluating the
success, or otherwise, of crime prevention measures using the case study of an area of
Liverpool, Mersey side. Although, based on a relatively short run of data post-intervention,
they examine the impacts of alleygating (installing gates at the entrances of small streets at the
rear of premises) as part of a series of burglary reduction projects. GIS has significant potential
here to investigate the type, nature, and extent of crimes in relation to interventions such as
CCTV and Neighbourhood Watch schemes but can also present some interesting challenges
for researchers faced with using data sets at a range of temporal and spatial scales. This is
further complicated by the use of incompatible spatial units such as police beat areas which do
not neatly correspond to other administrative geographies. Further concerns relate to data
confidentiality and in particular, the use of spatially disaggregate data on victim incidences.
New methods of analysing crime data are addressed in Chapter 5 by Jonathan Corcoran
and Andrew Ware who describe a project currently underway with a police force in Wales
which is concerned with exploring the use of Artificial Neural Networks (ANN) in examining
crime incidence. Their paper provides a preliminary analysis of the use of such techniques in
predicting crime patterns in Cardiff, South Wales. A significant amount of work is needed to pre-
prepare the data for such an analysis and they describe the issues that arise from the use of such
detailed data sources before outlining the nature of the techniques used in these contexts.
Finally, the authors re-iterate the policy benefits of such an approach to crime analysis.
1.3
GIS AND PLANNING
In the second section of the book, we are concerned with showing the potential for GIS in a
range of planning tasks. There is increasing interest in using GIS within group decision making
processes to address community concerns and preferences, and in Chapter 6, Kheir Al-
Kodmany describes the potential for web-based GIS tools in such a public participation
exercise. Previous research has demonstrated the use of such software tools in, for example,
‘Planning for Real’ collaborative exercises in UK contexts (e.g. Kingston et al., 2000). Such
studies have highlighted the importance of designing appropriate interfaces in community-
based GIS studies to take into account the findings from research into cognitive aspects of using
screen-based images and maps. This study has been concerned with experimenting with
prototype interfaces as part of a wider study involving the use of web-based tools for
community based initiatives for low-income neighbourhoods in Chicago. Residents are asked
to comment on the visual appearance of their communities through various types of interface.
Further research is needed to compare the different approaches to interface design in a user
environment but this study has highlighted the types of advances that have been made in
designing web-based tools in public participation exercises.
The use of Internet based GIS tools in collaborative planning is also demonstrated in
Chapter 7 by Andrew Hudson-Smith and Steve Evans who are concerned with
developing 3D models of urban form in London. Research at the Centre for Advanced Spatial
Analysis (CASA) has used basic building blocks and other locational features in 2D maps to
create photo-realistic 3D block models in networked environments for pilot areas of the city.
They critically review some of the data sets needed to create such models including
comparatively new data sets such as LIDAR (Light Detection and Ranging) images. They also
highlight some of the drawbacks in creating such 3D models within existing GIS software
before advocating an approach based on linking web enabled GIS with 3D modelling systems.
Users of such systems can query attributes for objects within such models, hyperlink to other
web-sites or conduct fly- or walk- throughs and the authors conclude by demonstrating a
prototype in one London borough. Further research is planned to refine this model and to test
its usefulness in practical planning tasks such as in regeneration initiatives but the approach
adopted has significant potential in addressing community concerns.
The visual aspects of GIS are also highlighted in Chapter 8 by Lynn Dyson-Bruce who
describes the use of Historic Landscape Assessment (HLA) to examine landscape diversity in a
study area in the East of England. GIS has shown great potential to aid the planning process
where development may impinge on land which may have significant historical and cultural
INTRODUCTION 5
significance. This study has also drawn attention to the need for consistent meta-data relating
to data holdings and the not insignificant amount of work that is involved in collating such data
from a range of agencies. Finally Dyson-Bruce provides examples of how this approach has
been applied in the study region and reiterates the advantages of GIS vis-à-vis traditional
approaches to handling such data.
The last paper in this section by Peter Lee and Brendan Nevin (Chapter 9) is concerned
with the use of GIS in assessing the demand for housing in local authorities in the UK and, in
particular, with the potential for such approaches in targeting resources. To date, much of the
use of GIS in housing departments has been to manage local authority or social housing stock.
This research points the way to a more proactive use of GIS in helping to combat social
exclusion in a number of urban authority areas. A key element has been the derivation of
suitable indices of poverty in such contexts and the use of administrative data in identifying
areas at risk of low demand. This, in turn, has been tested against data relating to housing
turnover, transfer requests and housing stock condition to examine the appropriateness of such
indices.
1.4
GIS AND URBAN APPLICATIONS
In the third section of the book, we are primarily concerned with studies that have specifically
involved the use of innovative GIS-based techniques in urban contexts. Clearly, there is
overlap here with some of the issues arising from the papers in Section 2, but the aim has been
to include papers which have a broader focus. In Chapter 10, Xiaonan Zhang, Nigel Trodd
and Andy Hamilton outline a theoretical framework relating to the application of
Geographical Visual Information Systems (GVIS) in urban regeneration. They describe the use
of such frameworks in relation to stages in the public participation process and document the
advantages of such an integrated approach in relation to a pilot area in Salford in the north of
England.
In Chapter 11, Andrea Frank describes how spatial autocorrelation can be used to identify
distinctive areas within US cities. Using census data to identify clusters of areas based on socio-
demographic characteristics, the aim has been to examine the hypothesis that local political
forces influence residential patterns through land use, taxation and planning policies and that
these can be identified through the use of spatial analytical approaches. This has involved a
loose coupling approach in order to integrate autocorrelation techniques with a commercially
available GIS and applied to cities with contrasting political structures and population
characteristics. This, in turn, has drawn attention to key methodological concerns facing
researchers in this field, for example, in relation to the definitions of neighbourhoods. Frank
reiterates the policy implications of such research and the applicability of such techniques in
other environments.
The subject of the importance of detailed, spatially disaggregate data sources in studies of
socio-economic variations, is returned to by Fulong Wu in Chapter 12. Using the case study
of Shanghai, contrasts are presented to the situation in the developing world where such data is
generally not available. A high resolution spatial data set is created with which to study intra-
urban spatial variations in property prices. Recent approaches have investigated the
incorporation of locational attributes into hedonic regression models to analyse the
determinants of local house prices. In this study, residuals from regression models (having
controlled for structural variables of properties) are interpolated from sample points in order
to examine the spatial distribution across the city and to gauge the importance of
environmental factors on house prices. Such techniques offer significant benefits in the data-
poor contexts of some developing countries.
1.5
GIS AND RURAL APPLICATIONS
The use of GIS research in rural contexts is explored in the fourth section, which includes
three chapters on analysing various aspects of rural service provision. In Chapter 13, Andrew
Lovett, Gisela Sünnenberg and Robin Haynes, describe the use of GIS-based techniques
to examine the accessibility of health services, in this case General Practitioner (GP) surgeries
in Norfolk, in the east of England. Traditionally such studies have used straight-line distances
to investigate such access issues. This is fundamentally flawed both because of the state of the
road network and the reliance of some sections of the community on public transport, which
in recent decades has been in decline in some rural areas. In this study, the researchers are
concerned with taking into account the availability of such services when examining access for
those on patient registers. In particular they demonstrate the potential for GIS-based
techniques when examining the implications of recent investments in public transport
provision within communities. Their study also points the way to identifying communities that
are still ill-served by existing provision and those that should be targeted for more investment
in public or community transport schemes.
In Chapter 14, Mandy Kelly, Robin Flowerdew, Brian Francis and Juliet Harman
continue the theme of using GIS to measure access for rural populations. Their prime concern
in this paper, however, is to demonstrate how GIS can be used in resource allocation measures
such as the Standard Spending Assessments (SSAs). There is concern that rural areas are losing
out under existing funding mechanisms because of the failure to take into account the unique
set of circumstances facing service providers in such areas. This research proposes a
methodology whereby the remoteness of such areas can be taken into account through the use
of GIS-based measures of travel times which considers the nature of the road network and the
time taken to access such areas. The researchers draw attention to the financial implications of
incorporating such measures into existing sparsity measures for authorities in England in terms
of the ‘winners’ and ‘losers’ in resource allocation. The results are not as clear-cut as first
thought which prompts the researchers to call for more research into deriving measures which
reflect the costs of providing services for dispersed populations in remote areas.
The theme of examining the potential for GIS in the area of service provision in rural areas
is also addressed in Chapter 15 by Helena Titheridge. This research, funded by the EPSRC
Sustainable Cities programme, is concerned with examining the potential for using GIS-based
models to reduce travel to services and hence energy consumption. The ESTEEM (Estimation
of Travel, Energy, and Emissions Model) has been applied to Gloucestershire in the west of
England in order to compare travel patterns under a number of policy scenarios relating to
service and housing provision. This is run as an extension to an existing commercially available
GIS package. This is based on a gravity model which is used to examine travel patterns based
on variables such as car ownership and existing transport networks and the impacts of new
developments on fuel consumption and emissions are modelled. In this way, the policies for a
typical local authority can be monitored both for the authority as a whole or for individual
settlements. This, in turn, has drawn attention to the limitations of existing data sources used
to calibrate these models. Many authorities do not routinely collect data on changes in the
INTRODUCTION 7
quantity (and quality) of facilities and a significant amount of effort is needed to collate data for
such modelling efforts. This study has drawn attention to the limitations of those data sets that
do exist before modelling the transport implications of different policy scenarios involving the
location of housing developments and facilities. This, in turn, has drawn attention to the role of
public transport in rural areas.
1.6
GIS IN SOCIO-ECONOMIC POLICY
The final section of the book contains four chapters that are concerned with demonstrating the
potential for GIS in addressing a variety of socio-economic issues. In Chapter 16, Richard
Harris and Martin Frost highlight the limitations of using existing (largely) census based
measures at relatively coarse scales for measuring intra-urban variations in socio-demographic
characteristics. At the same time, it is recognised that existing area based measures need to be
refined to take on board the fact that many deprived households are located in otherwise
affluent areas and vice versa. Many private sector organisations are developing rich data sets on
individuals/households based on consumer patterns that have great potential for describing
such patterns at finer, and more flexible, spatial scales (Longley and Clarke, 1995). The use of
one such spatially disaggregated data set, for the London borough of Brent, is described in this
contribution in order to map the social fragmentation that exists within a ‘typical’ local
authority area. The authors conclude by drawing attention to the potential use of such data sets
in relation to recent research concerned with deriving indices of deprivation based on a wider
range of measures across the UK and in relation to Government initiatives based around the
concept of neighbourhoods.
There has been concern expressed regarding the apathy of voters in recent European and
General elections in the UK. In Chapter 17, Scott Orford and Andrew Schuman
investigate the use of a combined GIS-spatial analytical approach to examine the factors
determining voter turn-out using the example of voting patterns at a local Council election in a
Bristol ward. Specifically they examine the relative influences of geographical factors such as
access to polling booths as well as household characteristics using electoral registers showing
those individuals who had voted. These were geo-coded via the postcode and a lifestyles
database used to examine intra-ward socio-economic patterns. It is then possible to examine
the importance of distance to polling stations in relation to such patterns and make some
preliminary attempts to suggest ways in which turn-out could potentially be improved, e.g. re-
locating stations or by altering their catchment areas.
In Chapter 18, Carsten Schürmann and Ahmed Talaat broaden the discussion to an all-
Europe level by describing a project which has been concerned with developing a European
Peripherality Index using GIS-based methodologies. As well as distance factors such indices
need to take account of the economic potential of the regions and the paper includes some
preliminary examples of the types of measures that can be developed. This, in turn, has
brought to the attention of policy makers the need for standardised data sets across the EU
with which to examine spatial patterns. In order to assess the peripherality of regions travel
times by different modes of transport are used in the calculations and combined with economic
measures of the various regions (such as Gross Domestic Product) and applied to the NUTS
Level 3 spatial units. These measures are compared visually and the policy implications of using
various indicators of peripherality discussed. Finally the authors suggest some refinements to
the methodology, such as the incorporation of measures based on other transport modes (e.g.
rail, air), in order to support the assessment of EU policies with respect to peripherality and
cohesion.
The final chapter in the volume by Barren Smith, Gary Higgs and Myles Could
describes an on-going project which is concerned with examining the factors influencing the
take-up of GIS by health organisations in the UK. This involves the use of mixed method
methodologies and the researchers discuss the rationale for such an approach, the expected
outputs from the study and the wider implications of the research.
1.7
CONCLUSIONS
In this introductory chapter, we have attempted to outline the main themes to be drawn from
the contributions included in this volume. Although we have focused here on the use of
innovative GIS-based research techniques in a variety of socio-economic applications, many of
the techniques highlighted have relevance for those concerned with addressing wider
environmental concerns. We contend that many of the issues discussed in this summary chapter
clearly have resonance for those working in other sectors that we have not been able to include
in the book. In addition, although we have tended to focus on technical issues of
implementation many overarching organisational issues will be common across all sectors.
Convincing senior managers and politicians of the important contributions GIS can make in
addressing crucial social and environmental issues at a time of restricted research budgets is of
prime concern to many researchers. It is to be hoped, that by demonstrating the policy
benefits of using GIS in conjunction with other spatial analytical approaches, this volume has made
a contribution to such laudable aims.
1.8
REFERENCES
Campbell, H. and Masser, I., 1995, GIS and Organizations: How Effective are GIS in Practice? (London:
Taylor and Francis).
Gatrell, A. and Loytonen, M. (editors), 1998, GIS and Health. (London: Taylor and Francis).
Gill, S., Higgs, G. and Nevitt, P., 1999, GIS in planning departments: preliminary results from a
survey of local authorities in Wales. Planning Policy and Practice, 14 (3), pp. 341–361.
Goldsmith, V., McGuire, P.G., Mollenkopf, J.H., Ross, T.A. (editors), 2000, Analyzing Crime
Patterns: Frontiers of Practice. (London: Sage).
Halls, P.J., (editor), 2000, Spatial Information and the Environment: Innovations in GIS 8. (London:
Taylor and Francis).
Hay, S.I., Randolph, S.E., and Rogers, D.J. (editors), 2000, Advances in Parasitology: Remote Sensing
and Geographical Information Systems in Epidemiology. (San Diego: Academic Press).
Higgs, G. and White, S.D., 2000, Alternative indicators of social disadvantage in rural
communities: the example of rural Wales. Progress in Planning, 53 (1), pp. 1–81.
Hirschfield, A. and Bowers, K. (editors), 2001, Mapping and Analysing Crime Data: Lessons from
Research and Practice. (London: Taylor and Francis).
Kingston, R., Carver, S., Evans, A. and Turton, I., 2000, Web-based public participation
geographical information systems: an aid to local environmental decision-making. Computers,
Environment and Urban Systems, 24 (2), pp. 109–125.
INTRODUCTION 9
Longley, P. and Clarke, G. (editors), 1995, GIS for Business and Service Planning. (London:
GeoInformation International).
Longley, P.A., Goodchild, M.F., Maguire, D.J. and Rhind, D.W. (editors), 1999, Geographical
Information Systems: Principles, Techniques, Management Issues and Applications (2nd edition).
(Chichester: Wiley).
Longley, P.A. and Harris, R.J., 1999, Towards a new digital data infrastructure for urban analysis
and modelling. Environment and Planning B, 26, pp. 855–878.
Martin, D., 1996, Geographic Information Systems: Socio-Economic Applications. (London: Routledge).
Orford, S., 1999, Valuing the Built Environment: GIS and House Price Analysis. (Aldershot: Ashgate).
Ratcliffe, J.H., 1998, Aoristic crime analysis. International Journal of Geographical Information Science,
12 (7), pp. 751–764.
Stationary Office, 1998, Guidance on Statutory Crime and Disorder Partnerships, (London: Home Office
Communication Directorate).
Thill, J-C. (editor), 2000, Geographic Information Systems in Transportation Research. (New York:
Permagon).
2
Revisiting fundamental principles of GIS
Nicholas Chrisman
2.1
INTRODUCTION
In 1987, I presented ‘Fundamental Principles of GIS’ at Auto-Carto 8 (Chrisman, 1987a). I want
to review the content of this paper, and update it with more recent research. First, the paper
demonstrates that the GIS community has always been embedded in social issues. We have had
our own little version of the Science Wars, and like most wars, both sides have lost more than
they have gained. There are good reasons to ask probing questions about the human values of
any technical system, and I hope we can ask probing questions about GIS.
On balance, the views in the 1987 article might be asking good questions, but they seem a
bit formulaic in the kinds of remedies presented. Research on ‘Society and GIS’ has progressed.
I will present some of the exciting new directions, some of them developed from a closer
reading of the literature on the social studies of technology and science (STS). The most
exciting one for me are the ones that are directly linked to the most seemingly technical
details. I want to develop in particular the way that the current software industry seems intent
on ‘configuring the user’—rather than developing new modes of interaction.
2.2
RETURNING TO 1987
In 1987, I presented a paper at Auto-Carto 8 under the title Fundamental Principles of GIS (later
published in Photogrammetric Engineering and Remote Sensing under the more descriptive but less
punchy title: Design of GIS based on social and cultural goals (Chrisman, 1987b). I do not intend to
indulge in some form of nostalgia, but rather to use the paper I wrote fourteen years ago as a
lens to examine the current state of geographic information systems research.
First, let us return to that time period to set some of the scene. As before and after, the GIS
community was marvelling at rapid expansion. In the autumn of 1986, the British hosted an
unexpectedly overflowing Auto-Carto London (the surplus revenues from that event may have
played a role in many graduate student bursaries ever since, and perhaps also played a role in
founding GISRUK). The first GIS textbook (Burrough, 1986) appeared at that conference. As
Director of the next Auto-Carto, I adjusted my deadline for abstracts to attract good papers
from authors at the London event.
The state of the art in GIS in 1986/87 was (retrospectively) at a high point in the acceptance
of topological principles to organize software. For example, Intergraph had developed TIGRIS
REVISITING FUNDAMENTAL PRINCIPLES OF GIS 11
(Herring, 1987), an object-oriented system using a seamless topological data structure indexed
using R-trees. Since this had been my original agenda at earlier Auto-Cartos, you might expect
me to have been pleased. Actually, I had already turned towards a more decentralized model
that I articulated more carefully in the Fundamental Principles paper.
As I sat down to write my own paper for Auto-Carto 8, I knew I did not want to write a
narrow technical contribution. It was an opportunity to make a more comprehensive
statement. It is important to remember how important these conferences were to the
community at that time. The International Journal of GIS had just announced that it would begin
in 1987, but we mostly relied on conference proceedings as the medium to express ideas
publicly. While it may seem academically retrograde, a well-organized meeting can often
accomplish much more than the most rigorous peer-reviewed journal can ever pull together.
The basic point of the paper (and the sentence most cited since) stated that GIS technology
“must be accountable economically, but also politically, socially, and even ethically” (Chrisman,
1987b, p. 1367). Such a noble sentiment might seem somewhat non-controversial in the
current epoch in which all undertakings are subject to careful questioning, but at the time this
was somewhat novel, and even slightly shocking to those committed to the technical agenda. The
paper proceeded to spend most of its energy on the design of databases. It started with an
allusion to the earlier era of raster/vector debate as a dead issue, without much of an apology
for the role I played in it. It tried to critique the reliance on user-need studies because they
tended to ratify the current status quo. It then tried to implicate the classical communications
model as a source of the problem, and to present a culturally and historically embedded
alternative. Probably nobody ever figured out the diagram that I offered to explain how
institutions manage their data holdings are the historical result of the interactions of people and
their environment. When I reviewed this diagram in 1992, I found it overly structuralist
(Chrisman 1992). Any situation involving long-term organizations and the people acting within
them will have to deal with the tension between structure and agency, though I did not
articulate the connection to the social science literature on that subject originally. The article
makes what would now appear to be a quaint foray into the lack of objectivity of geographic
information. The framing of these ideas still needs to be worked out, it has only been in the
past few years that I have begun to locate some ways to address these problems (see below).
If the paper had a positive suggestion, it was in the form of a fairly legalistic approach to
requirements analysis through the analysis of mandates assigned to custodians. To some extent,
the GIS coordination efforts of the past fourteen years have confirmed this approach. It is not
too surprising that Wisconsin adopted this strategy nearly verbatim, but it also influenced the
US Federal Geographic Data Committee and a number of other similar efforts. In most cases,
the process was somewhat reversed. Instead of rethinking the content of databases from the
start using the mandates as a guide, it seems that the existing agencies declared themselves
custodians through the political process of turf battles, then designed databases that ratified the
division of labour. In either case, the result is a more decentralized and collaborative
framework of cooperation than was being advanced in that period (National Research Council
1983, for example). My voice was just one of many, and the idea of cooperation among
contributors had many origins and many supporters. In any case, it took years to see these
movements take root in the community. During these years many other forces interacted, so it
is pointless to argue which elements started anything.
The final point of the paper was to argue that equity was a more important goal than the
measures of efficiency that had dominated the technical arguments. “Geographic information
systems should be developed on the primary principle that they will ensure fairer treatment of
those affected by the use of the information” (Chrisman, 1987b, p. 1370). I turned this
argument into a slight dig at the raster world by arguing that the system should retain the units
of interest, not impose some external set of arbitrary units. I ended on a reflexive note,
arguing that my earlier work (Chrisman, 1975) proposing an integrated topological database
“was flawed because it centralizes definitions” (Chrisman, 1987b, p. 1370). The argument was
based on displacing authority to a technical elite that should be more accessible to the political
and administrative process. The final sentences demonstrate the ringing polemic of the paper:
“…the search for technical efficiency must not be allowed to overturn political choices
without careful examination through the political process. The true challenge is to use
the increased sophistication of our automated systems to promote equity and other
social ends which will never fit into a benefit/cost reckoning. I am convinced that the
future of geographic information systems will lie in placing our technical concerns in
their proper place, as serious issues worthy of careful attention. These technical
concerns must remain secondary to the social goals that they serve.” (Chrisman, 1987b,
p. 1370.)
2.3
SO WHAT?
Perhaps it is somewhat soothing to know that conferences fifteen years ago, back in the dark
ages of computing, could include such stirring sentiment. No doubt conference speakers have
proclaimed equally lofty goals before and since. If my goal were simply to provide a bit of
moral uplift after a few days of technical detail, I could simply repeat the main points I
delivered in 1987. But, fair listeners, you will not be so easily rid of me. As I took apart my
own work from 1974 and 1975 in the 1987 talk, I will now consider how my research
direction has changed, and how the prospects for the future seem somewhat different. I will
embark on this project with a lot more assistance than I had in the prior enterprise, in part due
to an expanded group of GIS researchers who study ‘Society and GIS’ in various forms, and in
part because I have spent the last seven years reading heavily in the interdisciplinary field of
Science and Technology Studies (STS). This work leads me to quite a different set of
prescriptions from those advanced at Auto-Carto 8. In particular, I want to demonstrate one
theme begun in the earlier paper: there are traces of the social in the deepest and most
technical parts of a GIS. But first, I need to deal with the discovery of GIS by the rest of the
geography research community and the Science Wars that spilled into our isolated world.
2.4
SCIENCE WARS: SHIPS PASSING IN THE NIGHT IN K-D SPACE
Part of the value of returning to 1987 is that it is prior to the wave of criticism that followed.
It is not that my paper was the first to connect technical details to social concerns (e.g. see Bie,
1984), but that the nature of discourse changed radically.
At first, there was a kind of nervous criticism, most clearly articulated in the intemperate
newsletter column of the President of the AAG (Jordan 1988). Professor Jordan, an historical
geographer from the ‘exceptionalist school’, worried that GIS might ‘swamp’ the discipline
and displace focus from the ‘theoretical core’. Brian Harley (1989) brought in a post-modernist
critique, mostly of maps, but with some mention of the GIS movement. GIS became the
sticking point for geographers who wanted to complain about relationships of power and
representation. Perhaps the most heated rhetoric (Taylor & Overton, 1991; Smith, 1992) was
enflamed by the ‘best defense is a good offense’ strategy of Stan Openshaw (1991). When
Jerome Dobson had published his original ‘automated geography’ (Dobson, 1983), the
commentary had been almost exclusively from cartographers and GIS insiders. When he
renewed his vision after ten years (Dobson, 1993), the commentary included a lot sharper
criticism (Pickles, 1993). Viewed from inside, it might seem that the GIS movement was being
attacked for being successful, but the criticism was not simply a reaction to GIS, it is a small
part of a much larger intellectual movement that swept across the humanities and social
sciences. Just as GIS folk felt attacked, the science community reacted, in even more
flamboyant ways than Openshaw (most notably the Sokal affair).
The publication of Ground Truth (Pickles, 1995b) moved beyond an era of pure polemic and
showed some attempt to include authors from inside GIS alongside the critics. Despite the effort,
the insiders stuck to their scientism (Goodchild, 1992; Goodchild, 1995, in particular), and
the critics to theirs (Pickles, 1995a, talks about totalitarian tendencies). While this book was
quite important, it did not serve to construct a new common ground, but more to
demonstrate how divergent the views were. As the literature continued (Sheppard, 1995;
Curry, 1998), the focus concentrated on the impacts (potential, imagined, observed) of GIS on
society. Only a few talked about the reverse, that our current GIS might be conditioned by
societal pressures, cultural presuppositions, and political choices (Chrisman, 1992, 1996).
Arguments about GIS technology often slip into a discourse of technological determinacy. GIS-
proponents and critics alike assert, consciously or unconsciously, that technology is
intrinsically independent from the social world. This perpetuates the two major tenets of
technological determinism: (1) technology engages unilinear progress from less to more
advanced systems; and (2) technology is an imperative to which social institutions and people
must adapt (Bijker et al., 1987; Woolgar, 1987; Bijker and Law, 1992; Feenberg, 1995).
Technological determinism leads to the belief that the technology can be studied solely by
itself, outside of the context of its construction or use. As a consequence, ‘implications’
remain as the sole issue.
Proponents often acclaim geographic information technology as the means to make more
efficient and socially equitable decisions. These proponents hope to clear away subjective
issues and rationalize the process of establishing consensus, so that decisions can be made
objectively (Dobson, 1983; Openshaw, 1991, 1993). Most of this literature aligns itself with a
‘March of Progress’ metaphor, an attitude about history with limited utility to detect the
choices and inconsistencies involved in technological change (Chrisman, 1993). The idea of an
automated geography implies that the technology is somehow independent of the people,
operating on its own internal logic. Critics of GIS are quite justified in calling attention to
flaws in the proponents’ claims.
The heralds of progress create the impression that improvement is inexorable and assured.
The GIS bandwagon suggests that jumping aboard is the way to success; technology can fulfill
every demand, and bring you the world. Dobson places GIS technology on a clear rational path
towards a better tomorrow, arguing that “GIS has become a sine qua non for geographic analysis
and research…the beginning stage of a technological, scientific, and intellectual revolution”
(Dobson, 1993, p. 431). The authors of Ground Truth made much of the claims of GIS
proponents (Pickles, 1995c) as well as the advertising of GIS vendors (Goss, 1995; Roberts
and Schein, 1995). The more arrogant the claim, the better it seems to serve the critics.
The critics (e.g. Smith, 1992; Curry, 1998; Pickles, 1995b; Sheppard, 1995) have also
focused on the impacts of technology. They often portray the technology as a force out of
social control, something external to the social discourse. They use a somewhat sophisticated
form of C.P.Snow’s (1959) ‘two cultures’ argument, saying that technologists are not
connected to the same literature and not engaged in the same bases of theory. The gap
between two discourses does not mean that technology and technologists do not respond to
their own versions of social forces. Both proponent and critic alike need to see where exactly
the social comes into GIS. It may not be in the places they are watching.
Technological determinism, proclaimed by proponents or implied by critics, obscures the
relationships between GIS technology and society largely by neglecting linkages. The
contention between progress-believing technologists and humanistic-orientated social theorists
omits the people involved with the technology and the complex interactions required to
maintain it. GIS technology serves to extend human capabilities by other means, not a
superorganic force in itself. The people who use GIS are not mere instruments of progress
towards better information systems nor are they simply victims of its social consequences. The
systems now in place reflect many layers of negotiation between social goals and technical
capacity to respond. The simplistic metaphors must be replaced with more nuanced
understanding of interactions between people and technology.
Rather than a vast superhuman realm, GIS technology is the result of localized social
construction. This construction occurs when the technology is created, and continues as it is
configured for each application. The march of progress myth must be replaced with a careful
examination of the social divisions created and maintained by geographic information
technologies. This paper will consider one example of these social divisions, after it presents
some approaches to technology and society that move beyond technological determinism.
2.5
STUDIES OF SCIENCE AND TECHNOLOGY (STS)
In place of the technological determinism common in treating GIS, this paper draws
specifically on recent theoretical insights from a number of interlocking literatures including
the sociology of scientific knowledge (SSK), studies of technology and science (STS), history of
technology and of science, philosophy of science and related fields. The twentieth century
began with a fairly coherent expectation of the cumulative development of scientific
knowledge (Carnap, 1966). By mid-century, the logical positivists seem to have conquered all
opposition, broadcasting a message of method as a path of coherent science. Kuhn (1970)
introduced an observation that science in this period was by no means as linear as it was meant
to have been. The development of relativity in physics, for example, required replacing the
whole ‘paradigm’, not just the incremental accumulation of adjustments to earlier schemes.
Kuhn’s approach left science (and thus technology) fairly independent from social concerns.
Kuhn’s work was so pervasive that the quantifiers in geography adopted the terminology of
paradigms (e.g. Berry, 1973), a basically anti-positivist theory of knowledge. Some recent
studies in the history of science (Galison, 1997) demonstrate further refinements in
understanding how science operates, extending the concept of paradigms to allow for greater
ambiguity in the negotiations between theorists and instrumentalists. The assurance that a
particular scientific method always works has been strongly questioned (Feyerabend, 1993).
Thus, the history and philosophy of science no longer provide support for the old mythology
of inexorable progress.
Studies of science and technology (e.g. Barnes, 1974; Bloor, 1976; Latour & Woolgar, 1986)
provide strong documentation of complex networks linking social organization, political
structure, economic interaction, and cultural foundations to the development of a technology.
The sociology of scientific knowledge developed a ‘strong program’ of researchers (Bloor,

1976; Collins, 1981) who argued that social relationships underpin the development of science
and technology. This strong program argues against the study of ‘impacts’ from technology to
society. The constructivist literature (Latour & Woolgar, 1986; Bijker et al. 1987; Latour,
1987, 1988; Bijker & Law, 1992, 1993), though inherently quite diverse and far from
unambiguous, modified the unidirectional direction providing a more complex dynamic of
mutual constitution. Latour (1993) argues that the division between ‘nature’—a realm of
scientific enquiry—and ‘society’—a realm for human creation—obscures intricate
interactions required to sustain the hybrid networks of current technology.
This literature argues that science and technology are constructed from a multiplicity of
viewpoints, and that this construction is distinctly local, not universal. Multiple social forces
interact in the process of developing a complex technology such as GIS. Implementation of any
technology depends on the specific local environment that strongly constrains how actors
interact with the artifacts they construct. This literature digs deeper than the argument of
‘inherent logic’. Any logic in a technology was put there by developers and adopted by users
each group acting for their own reasons.
Social constructivist approaches provide a theoretical framework for examining and
understanding the tight linkages between the actions of people and the technology they create
and use. The web of technology and society consists of many complex relationships between
artifacts and people, institutions and data, software and researchers. Martin (2000) has applied
this actor-network approach to demonstrate how different GIS organizations interact in
Ecuador. In his analysis, there are ‘social’ institutional logics tied in with ‘technical’ logics in a
way that defies any simple explanation of one element overruling the other.
The STS literature offers some additional analytical approaches. Star, Greisemer (1989) and
Fujimura (1992) develop the relationships between multiple actors and artifacts through what
they call boundary objects. Boundary objects mediate between different groups; they serve a
dual function: at the same time they serve to distinguish differences, they also supply common
points of reference (Harvey & Chrisman, 1998). Institutions and disciplines play a crucial role
in formulating boundary objects that allow for stable translations between different
perspectives on the same phenomenon. Galison (1997) provides a further development of
these boundary concepts that may apply more directly to the interdisciplinary nature of GIS
practice. He argues that translation implies too much mutual comprehension; he uses the
linguistic metaphor of a pidgin dialect operating in a ‘trading zone’. This concept offers an
important insight for the design of GIS technology, particularly in tempering the rush for a
universal ontology through ‘common sense’.
2.6
CONFIGURING THE USER
As a specific resource for this paper, I am borrowing the phrase ‘configure the user’ from a
paper written by Steven Woolgar (1991), a British sociologist of technology and science. In the
days when the IBM XT was the dominant PC, he observed how a microcomputer
manufacturing organization decided how to design their next model. He argues that the group
did not configure a machine to suit a specific body of users, but rather that they built the
machine that they could and attempted to configure the users to suit the machine. He was
contributing to a literature about the role of technical artifacts (e.g. Latour & Woolgar, 1986,
1987). This theme has recently been extended (in a more interactive form) to the study of
software developments (Mackay et al., 2000). This concept provides the most direct linkage to
the critique of user-needs assessment in the 1987 paper. I only have space to introduce a few
examples here.
Lets start out with the concept of ‘user’ in the first place. This term pollutes our understanding
of the interaction between software, computer, data, organizations and the multiple skills of
the varied people involved in any real-life GIS. Software vendors thrive by making these
people dependent on the software, dependent on never switching packages. The annual user
conference for dominant software vendors is lavish, and thoroughly suffused with a division of
labor between the few software designers and the many users. The next version—just around
the corner, real-soon-now—will clear up all problems, bring peace on Earth, and other minor
side effects. The user is being configured in a social network of dependency, rather than
treating the software as just one component, and often the least important compared to the
data and the analytical models required to do any real work.
In an earlier paper written for a conference on error analysis (Chrisman, 1999), I examined
a very specific GIS operation to register a digitized source to a coordinate system. This GIS
operation adopts the mathematical formulation from photogrammetry that divides error into
three parts: systematic, random and blunders. An appropriate numerical procedure, typically
least-squares regression, can minimize the random error, and estimate the parameters of the
systematic error (which can be totaly removed). This depends on having no ‘blunders’ to
contaminate the estimation process. The choice of least-squares is not as straight-forward as it
might seem. It is only optimal under some very special conditions, ones that are unlikely in
routine use (Unwin & Mather, 1998). The technical details are important, but my point here is
that the software designer isn’t just choosing what mathematical model to apply, but who does
what kind of work. The model handles two parts of the error, but not the other. The user is
meant to weed out blunders, even though the software provides very limited tools to detect
and remove them. If this analysis of errors were universal, then there would be little point in
complaining. However, there are alternatives (from the realm of robust statistics) that divide
error differently, and thus divide the labor along different lines. Such a change may not seem
like much of a difference, but it uncovers a set of expectations about who is meant to know
how much, which disciplines are best prepared to address certain issues, and so on. These are
important elements in the present and future of GIS. If we continue to use least-squares, we
are requiring users to perform a certain kind of blunder detection to protect the software’s
mathematics. Yes, we have software that perfomrs the ‘optimal’ solution, but only under
unrealistic conditions. With the normal error-prone data entry of real-life users, least squares
solutions are much worse than they need to be. If we convert to robust technology, the
balance shifts. A small change in software technology configures the user in the sense of
designing what a user has to know.
In a broader sense, the organization of data in a GIS serves to decide what kind of
administrative arrangements belong in a GIS. The layer-cake design (that I did much to
promote at Harvard and at Wisconsin) reflects a particular expectation of cooperating agencies
each with their independent coverage. There were alternatives in our past, such as the
integrated terrain unit techniques promoted by the Australians (Mabbutt, 1968) and adopted
by the Food and Agriculture Organization of the UN (FAO, 1976). The layer based design is
not neutral, it makes certain kinds of arms-length custodian approaches easier. While I can
contend that these are more likely to succeed and perhaps more likely to contribute to equity,
such a decision should not be furtive. Rather than configuring the user, we do need to ensure
that the software is not out of control.
One possible path towards such change in control is set forth as the choice between the
paradigms of the Cathedral and the Bazaar (Raymond, 1998). Raymond sets out a distinction
between centralized software development (the cathedral) and a democratized, decentralized
scheme (the bazaar). Certainly, most commercial GIS suffers from the divisions of labor implicit
in the cathedral model. Broader participation and freer dissemination might unblock some of
these barriers, also changing the roles in configuring the user. I do accept some of Bezroukov’s
(1999) critique of Raymond; the distinctions are not quite as easy. In particular, the role of social
status in open code networks limits the pure dissemination of a bazaar. In any even, this calls
for closer examination of the social roles in GIS software.
2.7
CONCLUSION
The 1987 paper argued that the fundamental principles of GIS were to be found in the social
goals that a GIS serves. While this may still be a grand objective, the situation seems a bit more
complex in the light of understanding the criticism of GIS from inside and outside. Society and
GIS are no longer simple categories that remain distinct. Software and databases are integrally
implicated in social relationships at all levels. This should be no surprise, rather it should focus
each of us on communicating all the choices to all affected parties. The social goals may not
need consideration just at the outset but in each step and each detail of GIS operation.
2.8
ACKNOWLEDGEMENTS
Partially supported by a grant from National Science Foundation (USA) Grant SBR 98–10075.
AGI provided funding to attend and present this paper at GISRUK 2001.
2.9
REFERENCES
Barnes, B., 1974, Scientific Knowledge and Sociological Theory. (London: Routledge & Kegan Paul).
Berry, B.J.L., 1973, Paradigms for modern geography. In Directions in Geography, edited by
Chorley, R.J., (London: Methuen), pp. 3–22.
Bezroukov, N., 1999, A second look at the cathedral and the bazaar, http://first monday.dk/issues/
issue4_12/bezroukov/index html.
Bie, S., 1984, Organizational needs for technological advancement. Cartographica 21 (3),
pp. 44–50.
Bijker, W.E., Hughes, T.P. and Finch, T.J. (editors), 1987, The Social Construction of Technological
Systems: New Directions in the Sociology and History of Technology. (Cambridge: MIT Press).
Bijker, W.E. and Law, J. (editors), 1992, Shaping Technology/Building Society. (Cambridge: MIT
Press).
Bloor, D., 1976, Knowledge and Social Imagery. (London: Routledge & Kegan Paul).
Burrough, P.A., 1986, Principles of Geographical Information Systems for Land Resource Assessment . (Oxford:
Clarendon Press).
Carnap, R., 1966, Philosophical foundations of physics; an introduction to the philosophy of science. (New
York: Basic Books).
Chrisman, N.R., 1975, Topological data structures for geographic representation. Proc. of the 2nd
Int. Symp. on Computer-Assisted Cartography (Auto-Carto II), pp. 346–351.
Chrisman, N.R., 1987a, Fundamental principles of geographic information systems. In Proc. of the
8th Int. Symp. on Computer-Assisted Cartography (Auto Carto 8), Baltimore, edited by Chrisman, N.R.,
pp. 32–41.
Chrisman, N.R., 1987b, Design of information systems based on social and cultural
goals. Photogrammetric Engineering and Remote Sensing 53, pp. 1367–1370.
Chrisman, N.R., 1992, Ethics for the practitioners of Geographic Information Systems embedded in
‘Real World’ constraints of guilds, professions and institutional sponsorship. Proceedings GIS/LIS
92 1, pp. 129–137.
Chrisman, N.R., 1993, Beyond spatio-temporal data models: a model of GIS as a technology
embedded in historical context. Proceedings Auto-Carto II, pp. 23–32.
Chrisman, N.R., 1996, GIS as social practice. In Technical Report 96–7, edited by Harris, T. and
Weiner, D., pp. D9–D11. (Santa Barbara: National Center for Geographic Information and
Analysis ).
Chrisman, N.R., 1999, Speaking truth to power: An agenda for change. In Proc. 3rd Int. Symp. on
Accuracy of Spatial Databases Applied to Natural Resource Issues, edited by Lowell, K. and Jaton, A.
(Ann Arbor: Ann Arbor Press).
Collins, H.M., 1981, Understanding science. Fundamentals of Science 2, pp. 367–380.
Curry, M.R., 1998, Digital Places: Living with Geographic Information Technologies. (New York:
Routledge).
Dobson, J., 1983, Automated geography. Professional Geographer 35, pp. 135–143.
Dobson, J., 1993, The geographic revolution: A retrospective on the age of automated geography.
Professional Geographer 45, pp. 431–439.
FAO, 1976, A Framework for Land Evaluation. (Rome: Food & Agriculture Organization).
Feenberg, A., 1995, Subversive Rationalization: Technology, Power, and Democracy. In Technology
and the Politics of Knowledge, edited by Feenberg, A. and Hannay, A. pp. 3–22. (Bloomington:
Indiana University Press).
Feyerabend, P.A., 1993, Against method: outline of an anarchistic theory of knowledge. (London: Verso)
third edition.
Fujimura, J.H., 1992, Crafting science: standardized packages, boundary objects, and ‘translation’.
In Science as Practice and Culture, edited by Pickering, A. pp. 168–211. (Chicago: University of
Chicago Press).
Galison, P., 1997, Image and Logic. (Chicago: Chicago University Press).
Goodchild, M.F., 1992, Geographical information science. International Journal of Geographical
Information Systems 6 (1), pp. 31–45.
Goodchild, M.F., 1995, GIS and geographic research. In Ground Truth: The Social Implications of
Geographic Information Systems, edited by Pickles, J. pp. 31–50. (New York: Guilford).
Goss, J., 1995, Marketing the new marketing: the strategic discourse of geodemographic
information systems. In Ground Truth: Social Implications of Geographic Information Systems, edoted
by Pickles, J. pp. 130–170. (New York: Guilford).
Harley, J.B., 1989, Deconstructing the map. Cartographica 26 (2), pp. 1–20.
Harvey, F. and Chrisman, N.R., 1998, Boundary objects and the social construction of GIS
technology. Environment and Planning A 30, pp. 1683–1694.
Herring, J., 1987, TIGRIS: topologically integrated geographic information system. In Proc. of the
8th Int. Symp. on Computer-Assisted Cartography (Auto Carto 8), Baltimore, edited by Chrisman, N.R.,
pp. 282–291.
Jordan, T., 1988, President’s column: the intellectual core. Newsletter of the Association of American
Geographers 23 (5), pp. 1.
Kuhn, T.S., 1970, The Structure of Scientific Revolutions. (Chicago: University of Chicago Press)
second edition.
Latour, B., 1987, Science in Action. (Cambridge: Harvard University Press).
Latour, B., 1988, The Pasteurization of France. (Cambridge: Harvard University Press).
Latour, B., 1993, We Never Were Modern. (Cambridge: Harvard University Press).
Latour, B. and Woolgar, S., 1986, Laboratory Life: The Construction of Scientific Facts. (Princeton:
Princeton University Press) 2nd edition.
Mabbutt, J.A., 1968, Review of concepts of land classification. In Land Evaluation, edited by
Stewart, G.A. pp. 11–28. (Melbourne: Macmillan).
Mackay, H., Carne, C., Beynon-Davies, P. and Tudhope, D., 2000, Reconfiguring the user: using
rapid application development filing cabinet. Social Studies of Science 30, pp. 737–757.
Martin, E.W., 2000, Actor-networks and implementation: examples from conservation GIS in
Ecuador. International Journal of Geographical Information Science, 15 (8), pp. 715–738.
National Research Council, 1983, Procedures and Standards for a Multipurpose Cadastre. Panel on a
Multipurpose Cadastre. (Washington DC: National Academy Press).
Openshaw, S., 1991, A view on the crisis in geography, or using GIS to put Humpty-Dumpty back
together again. Environment and Planning A 23, pp. 621–628.
Pickles, J., 1993, Discourse on method and the history of a discipline: reflections on Jerome
Dobson’s 1993 ‘Automated Geography’. Professional Geographer 45, pp. 451–455.
Pickles, J., 1995a, Conclusion: Toward an ecomony of electronic representation and the virtual
sign. In Ground Truth: Social Implications of Geographic Information Systems, edited by Pickles, J.,
pp. 223–240. (New York: Guilford).
Pickles, J. (editor), 1995b, Ground Truth: Social Implications of Geographic Information Systems. (New
York: Guilford).
Pickles, J., 1995c, Representations in an electronic age: geography, GIS and democracy. In Ground
Truth: Social Implications of Geographic Information Systems, edited by Pickles, J., pp. 1–30. (New
York: Guilford).
Raymond, E.S., 1998, The Cathedral and the Bazaar, http://firstmonday.dk/issues/issue3_3/
raymond/index html.
Roberts, S.M. and Schein, R.H., 1995, Earth shattering: global imagery and GIS. In Ground Truth:
Social Implications of Geographic Information Systems, edited by Pickles, J., pp. 171–195. (New
York: Guilford).
Sheppard, E., 1995, GIS and society: towards a research agenda. Cartography and Geographic
Information Systems 22, pp. 5–16.
Smith, N., 1992, History and philosophy of geography: real wars, theory wars. Progress in Human
Geography 16, pp. 257–271.
Snow, C.P., 1959, The Two Cultures and the Scientific Revolution. (Cambridge: Cambridge University
Press).
Star, S.L. and Greisemer, J.R., 1989, Institutional ecology, ‘translations’, and boundary objects:
amateurs and professionals in Berkeley’s Museum of Vertebrate Zoology. Social Studies of Science
19, pp. 387–420.
Taylor, P. and Overton, M., 1991, Commentary: further thoughts on geography and GIS.
Environment and Planning A 23, pp. 1087–1094.
Unwin, D.J. and Mather, P.M., 1998, Selecting and using ground control points in image
rectification and registration. Geographical Systems 5 (3), pp. 239–260.
Woolgar, S., 1987, Reconstructing man and machine: a note on sociological critiques of cognitivism.
In The Social Construction of Technological Systems: New Directions in the Sociology and History of
Technology, edited by Bijker, W.E., Hughes T.P. and Finch, T.J., pp. 311–328. (Cambridge:
MIT Press).
Woolgar, S., 1991, Configuring the user: the case of usability trials. In A Sociology of Monsters: Essays
on Power, Technology and Domination, edited by Law, J., pp. 58—97. (London: Routledge).
PART I
GIS and Crime

3
When is a hotspot a hotspot? A procedure for
creating statistically robust hotspot maps of
crime
Spencer Chainey, Svein Reid and Neil Stuart
3.1
INTRODUCTION
Several techniques and algorithms are used in practice for the generation of continuous surface
hotspot maps, all of which have different merits. These mainly relate to their ease of use,
application to different types of events, visual results and interpretation. Few of these methods
help to distinguish a consistent defining threshold that helps the analyst decide when a cluster of
crimes can be defined as a ‘hotspot’. This paper will present results of some partnered research
that explores a procedure for creating statistically robust hotspot maps. Our methods include
the application of point pattern analysis techniques to identify for spatial clustering, spatial
dispersion, spatial autocorrelation, and Local Indicators of Spatial Association (LISA statistics),
plus review the use of spatial analysis tools to visualise and assist in formatting the design of the
crime hotspot map and the definition of hotspot thematic thresholds.
Maps showing the distribution of crime as a digital continuous surface are increasingly
replacing point (or pin) maps, thematic boundary maps and spatial ellipses as ways to visualise
and understand patterns of crime events. Each of these techniques has their application but
suffer from problems that relate to their operation and interpretation of crime event patterns
(see Table 3.1).
The increased application of continuous surface smoothing methods has been in partial
recognition to these problems, but is also linked to their more common availability, their
appealing visual allure, and the improved precision and accuracy of geocoded crime records.
The creation of these continuous surface or ‘hotspot maps’ allow for easier interpretations of
clusters of crime, reflect more accurately the spatial distribution of the crime hotspot, and
often have fewer parameters to set. Continuous surface smoothing methods also consider
concentrations of crime at all event levels, rather than cluster grouping some and discounting
others. However, as their appeal has increased few questions are being made of the outputs
generated. Instead, many agencies that use these methods are becoming easily caught in the
‘false lure’ of the sophisticated looking geo-graphic they have produced, being reluctant to
question its validity or its statistical robustness. Where this is particularly the case is when little
regard is given to the legend thresholds that are set that help the analyst decide when a cluster
of crimes can be defined as a hotspot. This visual definition of a hotspot being very much left to
the ‘whims and fancies’ of the map designer. For example, a map showing the distribution of
crime as a continuous surface can have as little or as many hotspots on it depending on the
WHEN IS A HOTSPOT A HOTSPOT 23
Table 3.1 Crime mapping applications and problems associated with point patterns, thematic
boundary maps, and spatial ellipse methods.
ranges selected by the map designer to show spatial concentrations of these point events.
Different methods also produce different surface results (Jefferis, 1999) and can often further
mislead.
This paper suggests a pragmatic procedure for creating statistically robust hotspot maps of
crime. The procedure includes provisional steps that test for clustering and dispersion, reviews
three methods (kernel density estimation, inverse distance weighting, and a location profiler)
that have previously been applied to convert point maps of crime into crime pattern surfaces,
explores the ways to standardise parameter settings of these continuous surface smoothing
methods, and suggests an easy to apply and interpret method for defining crime hotspot
thresholds. We also demonstrate options to extend the procedure to a more advanced level of
analysis using local indicators of spatial association (LISA statistics).
The data we use in our tests are property/location precise allegations of crime from the
Metropolitan Police Crime Report Information System for the London Borough of Hackney.
We experiment with four different categories of crime—all crime, robbery, residential
burglary, and vehicle crime—for the three-month period, June to August 1999.
3.2
TESTING FOR CLUSTERING AND DISPERSION
A primary step that is often avoided but is fundamental to the detection of clusters of point
events are global tests that indicate if clustering and dispersion exist in the original point
distribution. This first step can provide insight into what types of patterns will be expected
when the mapped crime points are converted into a continuous surface. For example, whether
the points show evidence of clustering or are randomly distributed, and how dispersed the
distribution is relative to other crime types. Crime analysts often assume crime distributions to
be clustered in some form or other, and whether clusters do or do not exist, some can be
identified from random crime distributions.
There are several approaches for analysing a point distribution for spatial randomness. Most
of them incorporate the basic principles of hypothesis testing and classical statistics, where the
initial assumption is that the point distribution is one of complete spatial randomness (CSR).
By setting the CSR assumption as the null hypothesis the point distribution can be compared
against a set significance level to accept or reject the null hypothesis. Spatial autocorrelation
tests have been used previously to test for evidence of crime event clustering (Chakravorty,
1995). Spatial autocorrelation techniques do though require an intensity value, beit a
weighting linked to the event or a count of crimes where the crime point relates to the
centroid of an area to which crime events have been aggregated. From our point of view, if the
original crime event data exists as accurate point georeferenced data, aggregating this data to a
common point will lose spatial detail. With the increased availability of accurate and precise
geocoded records of crime it would seem more important to use methods that do not require
an intensity value but that retain and perform tests on the original crime event point data. Point
retaining methods only require x (eastings) and y (northings) coordinates which are used to test
for evidence of point clustering and dispersion.
Where the user has access to data where each point relates to individual crime events
(irrespective of whether some of these events are mapped to exactly the same location) we
suggest the point retaining methods of nearest neighbour analysis (NNA) to test for clustering
and standard distance for dispersion. Both these tests are familiar to many crime analysts, being
available in STAC (Illinois Criminal Justice Information Authority, 1996) and CrimeStat
(United States Institute of Justice, 1999). They are also simple and quick to apply and provide
a good global measure that can help lead the analyst in creating more confident mapping
output.
If analysts only have access to crime point data that are aggregate counts (representing the
number of crime events within a certain geographic area, e.g. Census tracts) we suggest the use
of the spatial autocorrelation technique, Moran’s I, to test for clustering. The statistic works
by comparing the value at any one location with the value at all other locations (Levine, 2002,
Bailey and Gatrell, 1995, Anselin, 1992, Ebdon, 1985). The significance of the result can then
be tested against a theoretical distribution (one that is normally distributed) by dividing by its
theoretical standard deviation (for more details on this test see Levine, 2002).
Table 3.2 Standard distance and nearest neighbour analysis results for four categories of crime in
the London Borough of Hackney (June 1999–August 1999). All categories show evidence of crime
point event clustering, and describe relative differences in dispersion.
NNA and standard distance tests were performed on the crime data available for the London
Borough of Hackney. Table 3.2 shows that all crime data point sets showed evidence of
clustering in their distribution. The degree of difference in the observed clustering between
the two different bounding areas was noted with additional tests suggesting the need to base all
calculations on the true boundary area (Reid, 1999).
To test for dispersion we calculated each crime category distributions’ standard distance.
This measure could then be used on a relative scale to describe different levels of dispersion
between crime types. The lower the relative standard distance, the less dispersed the crime
distribution. Robbery was seen to be the least dispersed crime type, suggesting a distribution
that is concentrated (clustered) at a few locations. Residential burglary showed evidence of
dispersion similar to all crime, whilst vehicle crime, although being clustered, tended to be
more dispersed.
We now have evidence of clustering for all crime types and an indicator of their relative
levels of dispersal. A final preparatory step before creating hotspot maps of crime was to carry
out nearest neighbour tests for each point’s 20 nearest neighbours (i.e. a K order of 20). This
would return mean nearest neighbour distances for 1, 2, 3…20 nearest neighbours that could
be used to help create search radii for the proposed continuous surface smoothing mapping
methods (Williamson et al., 1999). By gathering these simple global tests for clustering and
dispersal we can more confidently proceed to creating hotspot maps of crime.
3.3
CONTINUOUS SURFACE SMOOTHING METHODS
The methods we reviewed were those that were available through commonly used
geographical information systems software packages. This review allowed us to re-evaluate
techniques that the London Borough of Hackney had employed in creating hotspot maps of
crime and decide on a method that we could focus on for further analysis. The methods we
considered were kernel density estimation, location profiler (unique to the Maplnfo add-on,
Vertical Mapper) and inverse distance weighting.
Figure 3.1 Step 1 kernel density estimation—a fine grid is placed over the area covered by the
crime points. In most cases the user has the option to specify the grid cell size. (Source:
Ratcliffe, 1999a.)
Figure 3.2 A search radius (or bandwidth) is then selected, within which intensity values for
each point are calculated. Points are weighted, where incidents closer to the centre contribute a
higher value to the cells intensity value. (Source: Ratcliffe, 1999a.)
3.3.1
Kernel Density Estimation
The kernel density method creates a smooth surface of the variation in the density of point
events across an area. The method is explained in the following steps:
• a fine grid is generated over the point distribution (see Figure 3.1);
• a moving three-dimensional function of a specified radius visits each cell and calculates
weights for each point within the kernel’s radius. Points closer to the centre will receive a
higher weight, and therefore contribute more to the cells total density value (see
Figure 3.2);
• final grid cell values are calculated by summing the values of all circle surfaces for each
location (see Figure 3.3).
Figure 3.3 Each grid cell value is the sum of values of all circle surfaces for each location.
(Source: Levine, 2002.)
The quartic kernel estimation method we applied requires two parameters to be set prior to
running. These are the grid cell size and bandwidth (search radius). Of these, bandwidth is the
parameter that will lead to most differences in output when varied. The method for choosing
bandwidth that we have adopted follows from that of Williamson et al. (1999) where
bandwidth relates to the mean nearest neighbour distance for different orders of K. This still
requires the user to make a choice over which K order to apply. We regard this as an
advantage of the technique where users are prompted to think about the data they are mapping
and apply the K order’s mean nearest neighbour distance as the bandwidth relevant to their
application. For example, if the application is one that requires output for focused police
patrolling a smaller bandwidth would be used than one that requires a more strategic view of
crime hotspots. Having this flexibility will be demonstrated in Section 3.4.
Users are also required to specify a grid cell size. We again regard this as a positive feature
in kernel estimation whereupon users have the flexibility to set sizes relevant to the scale at which
the output will be viewed. Large cell sizes will result in coarser or blocky looking maps but are
fine for large scale output, whilst smaller cell sizes add to the visual appeal of the continuous
surface produced but may create large file sizes. Where the user is unsure over the cell size to
use, we suggest following the methodology of Ratcliffe (1999b) where cell size resolution is
the result of dividing the shorter side of the minimum bounding rectangle by 150. Cell size for
the London Borough of Hackney was set to 25m. The cell values generated from quartic
kernel estimation are also in meaningful units for describing crime distributions (e.g. the
number of crimes per square kilometre).
3.3.2
The Location Profiler
The Location Profiler method uses a technique where the distance from an overlaid grid cell to
a specified number of nearest points within a specified radius are measured, these distances are
then summed and the resulting value used to indicate local intensities (Northwood
Geoscience, 1998). The first step therefore requires a grid of user defined size to be overlaid
across the point distribution. Size of the grid cell can follow the flexible methodology to set
this parameter as previously described. Distances from the centre of each cell to a specified
number of points, or points within a specified search radius are then measured. Choice of the
number of points to measure distances to is somewhat arbitrary and as Eck et al. (1998) note,
there exists no underpinning theory to aid its choice. Search radius can follow from the
methodology described for choosing kernel bandwidth, where the radius entered relates to a
mean nearest neighbour distance for a particular K order.
A problem with the Location Profiler is that cell values generated do not fit logically as
outputs expected to describe crime hotspots. The cell values refer to the average distance to
points within the specified search radius. Cells with low values represent areas of high crime
density, and cells with high values are areas of low density. These values can be visualised
thematically on a map, but the cell unit values can add confusion when attempting to explain
the differences between local patterns of crime.
By choosing a search radius based on the mean nearest neighbour distances approach, the
effect was to highlight isolated crime events rather than highlighting areas of obvious clustering.
This occurred because there were no other points within the search radius to measure
distances to, the average distance to all other points being zero. Average distances in areas of
crime clustering would be low, but greater than for those areas where there are these isolated
points (see Figure 3.4).
A better approach in using Location Profiler would be to set the search radius to cover the
whole point distribution, set the display radius as a K order mean nearest neighbour distance
and choose an appropriate number of points to measure distances between. Choosing an
appropriate number of points is still an arbitrary process. If we were to base the number of
points on the K order of the mean nearest neighbour distance used as the display radius (e.g. if
a K order of 2 returned a mean nearest neighbour distance of 216m, 216m would be the
display radius and 2 would be the number of points), in cases where our point number was
small, areas of isolated points would often appear as being just as significant crime hotspots
than areas where there was an obvious cluster of crimes (see Figure 3.5).
3.3.3
Inverse Distance Weighting
The inverse distance weighting method (IDW) uses a floating circle to visit each new node of a
grid created across the distribution of points, assigns weights to the data points lying within a
prescribed distance, which are then averaged to return a value for each grid cell. The weights
assigned to points are based on their distance from the grid cell node. The IDW method is
most commonly applied to data which can be extremely variable over short distances
(Northwood Geoscience, 1998). For this reason the method has been used as a technique to help
visualise spatial patterns of crime. However, in its strictest sense IDW is an interpolation
technique that requires an intensity value for each crime event. Interpolation techniques are
most often used to generate estimates of an attribute at unsampled locations. For example,
Figure 3.4 Location Profiler returns cell output values based on the average distance from a number
of points within a specified radius. In the figure above, the darker the shading the lower the average
distance. By choosing a radius based on descriptions of the crime points spatial distribution (i.e. K
order mean nearest neighbour distances), the effect was to highlight isolated crime events where
there were no other points within the search radius, rather than identifying areas of obvious
clustering.
Figure 3.5 Location Profiler parameters were set as the whole study area for the search radius,
216m for the display radius (the mean nearest neighbour distance for 2 orders based on the point
distribution) and the number of points to visit to measure distances as 2 (the K order). The figure
shows that if we use small numbers of points, areas that are relatively isolated but that have two
points at the same location (area B), will be highlighted at least (and often more) significantly than
areas of obvious greater clustering (area A).
interpolation methods can be applied to create continuous surface maps of rainfall distribution,
where the point locations are rain gauges and the attribute values attached to them is the
volume of rainfall collected in the gauge. When we create continuous surface maps of crime
we are not estimating the number of crimes at ‘unsampled’ locations, but instead using the
point distribution to perform an area data aggregate which can be smoothed through
application of a suitable spatial function to describe crime pattern distribution.
To apply IDW an intensity value is required. For each crime point this intensity value can be
set to ‘1’, such that each point has equal intensity. Parameters required for IDW include a
search radius, display radius, cell size and exponent. There is little guide as to what exponent
to set, but experience in Hackney has always left this setting as the default. Grid cell size again
is best set in relation to the scale of viewed output, whilst search and display radii can follow
the mean nearest neighbour distance settings previously described. IDW does though require
an initial point aggregation technique to be applied. The resulting output of this required
technique will mean the loss of some spatial detail, therefore questioning already the detailed
accuracy of the final continuous surface output. A coincident point distance parameter setting is
also required to apply data aggregation. This coincident point distance is a search radius applied
to each new grid node across a grid network covering the distribution of points. In this sense it
is similar to the location from which a search radius or bandwidth is applied in the Location
Profiler technique and the quartic kernel density estimation technique.
Through its nature, the IDW method does not honour local high or local low values. As
Figure 3.6 shows, the averaging method will create output values that are ‘cooled’ at local high
points, and that could be over emphasised in low value areas.
The cell values returned also relate to the values of nearby aggregated points and not to the
original spatial point detail (see Figure 3.7). The moving window average technique can also lead
to visual inaccuracies relating to crime distribution. An example of this is shown in Figure 3.7
where the region to the west of the main hotspot shows a small increase in crime intensity,
even though there are no crimes in that area to suggest this increased intensity. Description of
the cell values returned can be best made by rounding up the values to integers and using these
integers with the search radius to describe local densities. For example, at the hotspot
indicated by the darkest area, the distribution of crime relates to the number of points (i.e. the
cell value, 5) within 216m (the search radius) of the point of interest. This description can
though lead to confusion where there is high variability across short distances.
From this review it was decided that the quartic kernel density estimation method provided
the best opportunity to proceed with creating statistical robust crime hotspot mapping output.
The method creates understandable grid cell value outputs that relate to crime, plus requires
fewer parameters to be set than those required for many other methods and where the
parameters entered can relate to the spatial distribution of the points being analysed. The
method also has the advantage of deriving crime density estimates based on calculations
performed at all locations (Levine, 2002), and retains some practical flexibility in map design.
3.4
DEFINING CRIME HOTSPOT LEGEND THRESHOLDS
In our aims to define crime hotspot legend thresholds we set out criteria that we felt were
important in helping to decide on suitable method implementation. These criteria were:
• the method needed to be practical and simple to apply;

• the thresholds generated were meaningful and could be linked to a value that could be
easily understood as a unit describing crime concentration;
Figure 3.6 The surface created using the moving average technique in the IDW method is less than
the local max. value and greater than the local min. value. (Source: modified from Northwood
Geoscience, 1998.)
Figure 3.7 Inverse distance weighting output using a search radius of 216m (this equals the K order
2 mean nearest neighbour distance of the point distribution). To apply IDW, a data aggregation
routine initially needs to be carried out. Data aggregation was applied with a coincident point
distance search radius of 216m. The data aggregation output is displayed above as the lighter points.
The moving average technique applied in the inverse distance weighting method has the tendency to
‘cool’ values at local high points, and can lead to visual inaccuracies of crime distribution.
• the separation of thematic thresholds followed a consistent methodology and where the
upper most categories consistently defined when a crime concentration reached hotspot
status;
• the method was more robust in taking into account the statistical spatial distribution of the
point data;
• but still retained flexibility in map design, appropriate to output required at different scales
and for different applications.
There is still much debate around the theoretical definition of understanding when a crime
hotspot becomes a hotspot. By ‘hotspot’, we refer to areas of distinctly high concentrations of
crime, relative to the overall distribution of crime across the borough (see also Home Office,
2001). This would mean that if clusters were identified from the nearest neighbour analysis
tests, a consistent methodology would identify where these hotspots were located and indicate
a level of concentration relative to the full crime distribution. By investigating crime hotspots
on a relative scale means that we are not restricted to a static setting that defines levels in
crime concentration. Implementing this relative approach would also enable analysts operating
at different geographic scales (e.g. local, regional and national levels) to apply the same
methodology to identify and prioritise the tackling of crime hotspots. We present results from
two different techniques we have investigated for defining thematic hotspot legend thresholds.
3.4.1
Incremental Standard Deviation Approach
Our first approach for defining crime hotspot thresholds was to explore the use of incremental
standard deviations above the grid cell mean. Standard deviations are useful in defining where
in a distribution the cut off is between values of different statistical significance. The procedure
we explored applied calculations on grid cells that had a value of greater than 0 and were
within the study area boundary. From this grid cell set, the mean and standard deviation was
calculated, with grid cell thematic thresholds set at:
• 0 to mean
• Mean to +1 standard deviation (SD)
• +1 SD to +2 SD
• +2 SD to +3 SD
• +3 SD to +4 SD
• Greater than +4 SD
Plate 1 shows the application of this incremental standard deviation threshold approach for
robbery (June to August 1999). Visually, the method is appealing, clearly identifying in a
structured manner areas of increasing crime concentration significance. The method is easy to
apply, makes use of K order mean nearest neighbour distances to define bandwidths, retains
flexibility in the map design by allowing the user to apply different K order bandwidths and
grid cell sizes to the suited application, and uses a consistent methodology to separate thematic
thresholds. However, standard deviations are not the easiest of statistics to describe to the
novice crime hotspot map reader. If there is a need to explain the concept of standard
deviations for identifying thresholds of significance before the crime hotspot map can be
interpreted then the threshold method is not acceptable as it itself creates a barrier to map
interpretation. Also, grid cell values of the quartic kernel density output are not normally
distributed. As our aim is to develop robustness in our crime hotspot mapping output, it
would seem inappropriate to apply a standard deviation approach to defining legend
thresholds.
3.4.2
Incremental Mean Approach
Our second method applied incremental multiples of the grid cells’ mean to define crime
hotspot map thresholds. Again, calculations were only applied to grid cells that had a value of
greater than 0 and were within the study area boundary. From this grid cell set, the mean was
calculated, with grid cell thematic thresholds set at:
• 0 to mean
• Mean to 2 mean
• 2 mean to 3 mean
• 3 mean to 4 mean
• mean to 5 mean
• Greater than 5 mean
Plate 2 shows the application of this incremental mean threshold approach for robbery (June to
August 1999). Comparisons in the map can be immediately drawn with the incremental
standard deviation approach where the method is visually appealing and structures the thematic
thresholds to clearly identify areas of highest crime concentration. The method is simple to
apply, requiring only the calculation of the grid cell mean, makes use of K order mean nearest
neighbour distances to define bandwidths, retains flexibility in the map design by allowing the
user to apply different K order bandwidths and grid cell sizes to the suited application, and
uses a consistent methodology to separate thematic thresholds.
As a statistic, the mean is an easier value for the novice map reader to grasp. Increments of
the mean would be more obviously linked to increasing values, and their relative significance.
This makes this approach immediately more appealing as a method to define crime hotspot
legend thresholds.
The incremental mean approach therefore shows potential in being a suitable method for
defining crime hotspot legend thresholds. At this stage we applied further tests using this
method but on different crime types and for different time periods. This would allow us to
investigate consistency in the technique, and explore the flexibility in map design, described as
an advantage in the quartic kernel density estimation method.
The maps in Plate 4 show the results of applying this incremental mean crime hotspot
threshold approach to the four different types of crime and for different bandwidths. The
incremental mean legend threshold approach consistently defines mapped crime hotspots. By
using different bandwidths, different scales in output can be generated to suite different
applications. Maps with small bandwidths demonstrate the spiky and more focused crime
mapping output returned, whilst larger bandwidths tend to smooth the data and indicate, for
example, general areas in need of strategic priority. As bandwidth is linked to the underlying
spatial distribution of the crime point data, all the maps can be described as statistically robust,
and use a consistent, easy to apply and easy to understand method for defining crime hotspot
thresholds. In addition, we can now observe the global descriptions calculated in Section 3.2
and see that they match the crime hotspot mapping output. All maps display areas of crime
clustering, and when compared against each other show relative levels of dispersal. For
example, the hotspot maps of vehicle crime show a greater degree of dispersion compared to
that of robbery.
3.5
ADVANCED HOTSPOT ANALYSIS: APPLICATION OF LOCAL
INDICATORS OF SPATIAL ASSOCIATION
Local indicators of spatial association (LISA statistics) have been described as being particularly
suited to identifying crime hotspots (Anselin, 1995; Getis and Ord, 1996). LISA statistics
assess the local association between data by comparing local averages to global averages. For
this reason they are useful in adding definition to crime hotspots, and placing a spatial limit on
these areas of highest crime event concentration (Ratcliffe and McCullagh, 1998). One of the
more applied LISA statistics on crime point events is the Gi* statistic (see Ratcliffe and
McCullagh, 1998 for more details). The Gi* statistic is applied to a grid cell output, such as a
quartic kernel density estimation map, from which local associations are compared against the
global average. The user is typically required to enter a search distance within which local
associations are explored, and a significance level related to the confidence in the final output.
The search distance is usually set to three times the grid cell size of the original kernel
estimation surface (i.e. for our data the search distance was set to 75m) and whilst levels in
significance can often range between 99.9%, 99% and 95%, this range has far less effect over
eventual hotspot areas than parameters set for grid cell size or bandwidth (Ratcliffe, 1999b).
The most common significance level to apply is 99.9%.
Plate 3 shows the result of the Gi* LISA statistic for robbery. This LISA output has been
mapped with the quartic kernel density estimation surface generated using a bandwidth of
118m (K2). The map shows the matching between this LISA output and kernel value results
above the three mean threshold. This LISA output adds definition to our quartic kernel density
estimation, indicating the level at which hotspots can be more clearly distinguished against
other levels of crime concentration.
3.6
DISCUSSION
We can now suggest a procedure for creating statistically robust hotspot maps of crime:
• Apply nearest neighbour analysis to test for clustering

• Calculate standard distance to return a relative measure of dispersion
• Perform a nearest neighbour test for each point’s 20 nearest neighbours (i.e. a K order of
20). Performing this test for at least each point’s 20 nearest neighbours will return
sufficient information to help set parameters for continuous surface smoothing
• Create a continuous surface map by applying the quartic kernel density estimation method.
Parameters should be set with respect to different K order mean nearest neighbour
distances, where the choice of K order is dependent on the crime mapping application
• Use an incremental mean approach to set grid cell thematic legend thresholds
• Advanced option—perform a Gi* LISA statistic analysis to quartic kernel surface. This
mapped LISA output will add definition to our current quartic kernel surface, indicating
the level at which hotspots can be more clearly distinguished against other levels of crime
concentration.
But when is a hotspot a hotspot? We argue that there is the need for flexibility in defining
when a crime concentration reaches hotspot status, depending largely on the level of
significance required in the result. However, from our results we suggest that hotspot status is
reached at 3 multiples of the mean grid cells’ value and above. This defining threshold visually
fits with a LISA statistic output of 99.9% significance.
Our methodology has yet to experiment with secondary variables that describe
environmental heterogeneity. This is where we would use a secondary variable to describe the
underlying population (e.g. distribution of residential properties), from which a continuous
surface hotspot map would be an estimate of crime risk. Theoretically, the incremental mean
approach for defining hotspot thresholds could still be successfully applied.
The crime hotspot maps we generate are also only snapshots of a particular period in time.
Future areas of work we aim to explore include space-time interaction, using test statistics
devised by Knox and Mantel (see Bailey and Gatrell, 1995). This would reveal whether certain
types of crime display temporal hotspots in particular areas (for example, whether there are
hotspots of crime that emerge, and only emerge on certain days of the week). A step that we
have already taken in this direction is the creation of crime hotspot animations to visualise
space and time interaction. Each frame produced for the animation is created following a
similar procedure to that described above, where the first frame’s parameters and incremental
mean thematic thresholds are used as the control for all Other subsequent frames (see http://
www.agi.org.uk/cdsig/hotspots-space-time-animation.htm).
3.7
CONCLUSION
The procedure we suggest creates statistically robust hotspot maps of crime. Preliminary
analysis tests for clustering and dispersion in our crime point distribution. The continuous
surface smoothing method (quartic kernel density estimation) we applied creates understandable
grid cell value outputs that relate to crime, uses parameters linked to the spatial distribution of
the points being analysed, and derives crime density estimates based on calculations performed
at all locations (Levine, 2002), whilst retaining some practical flexibility in map design. An
incremental mean approach helps us then to consistently define crime hotspot legend
thresholds, producing mapping output that describes relative levels of crime concentration.
From the results on data for the London Borough of Hackney, we suggest a crime density value
above 3 multiples of the grid cells’ mean to be a useful defining threshold that describes when
a concentration reaches hotspot status. This definition of crime concentration can then be
supported and added to with an advanced option that applies the Gi* local indicator of spatial
association.
Any form of map creation and design requires flexibility. As mentioned, the procedure we
suggest retains that flexibility but introduces some simple to apply operations. By including
these operations into continuous surface crime hotspot map creation could avoid mistakes in
their interpretation and add confidence to the final output.
3.8
REFERENCES
Anselin, L., 1992, SpaceStat: A Program for the Statistical Analysis of Spatial Data. (Santa Barbara:
National Center for Geographic Information and Analysis, University of California).
Anselin, L., 1995, Local indicators of spatial association—LISA. Geographical Analysis. 27 (2),
pp. 93–115.
Bailey, T.C. and Gatrell, A.C., 1995, Interactive Spatial Data Analysis. (Harlow: Longman).
Chakravorty, S., 1995, Identifying crime clusters: the spatial principles. Middle States Geographer, 28,
pp. 53–58.
Eck, J., Drapkin, D. and Gersh, J., 1998, A multi-method exploration of crime hot-spots: An
assessment of Vertical Mapper for examining crime concentration. Crime Mapping Research Centre
Intramural Project, (Washington D.C.: The National Institute of Justice).
Ebdon, D., 1985, Statistics in Geography. (Oxford: Blackwell).
Getis, A. and Ord, J.K., 1996, Local spatial statistics: An overview. In Spatial Analysis: Modelling in a
GIS Environment, edited by Longley, P. and Batty, M. (Cambridge: Geolnformation
International), pp. 261–277.
Home Office, 2001, Focus Areas (Hotspots) Toolkit. Crime Reduction Unit. http://
www.crimereduction.gov.uk/toolkits/fa00.htm.
Illinois Criminal Justice Information Authority, 1996, Spatial and Temporal Analysis of Crime. State of
Illinois.
Jefferis, E., 1999, A multi-method exploration of crime hot-spots: a summary of findings. Crime
Mapping Research Centre Intramural Project, (Washington D.C.: The National Institute of Justice).
Levine, N., 2002, CrimeStat II: A Spatial Statistics Program for the Analysis of Crime Incident Locations
(version 2.0). Ned Levine & Associates, Houston, Texas, and the National Institute of Justice,
Washington, DC., April, http://www.icpsr.umich.edu/NACJD/crimestat.html.
Northwood Geoscience, 1998, Vertical Mapper Version 2.0 Manual. (Ontario: Nepean).
Openshaw, S., 1984, The Modifiable Areal Unit Problem. Concepts and Techniques in Modern Geography,
38, (Norwich: GeoBooks).
Ratcliffe, J., 1999a, Spatial Pattern Analysis Machine Version 1.2 Users Guide, http://www.bigfoot.com/
~jerry.ratcliffe.
Ratcliffe, J., 1999b, Hotspot Detective for Maplnfo Helpfile Version 1.0. http://www. bigfoot.com/
~jerry.ratcliffe.
Ratcliffe, J. and McCullagh, M.J., 1998, Hotbeds of crime and the search for spatial accuracy.
Paper presented to the 2nd Crime Mapping Research Center Conference: Mapping Out Crime, Arlington,
Virginia, USA. December 10–12, 1998. http://www.bigfoot.com/~jerry.ratcliffe.
Rayment, M.R., 1995, Spatial and temporal crime analysis techniques. Crime Analysis Unit paper,
Vancouver Police Department.
Reid, S., 1999, Crime: When is a hotspot a hotspot? MSc dissertation for Master of Science Degree in
Geographical Information Systems. University of Edinburgh, Scotland.
U.S. Institute of Justice, 1999, CrimeStat: A Spatial Statistics Program for the Analysis of Crime Incident
Records. http://www.ojp.usdoj.gov/cmrc/tools/welcome.html #crimestat.
Williamson, D., McLafferty, S., Goldsmith, V., Mallenkopf, J. and McGuire, P., 1999, A better
method to smooth crime incident data. ESRI ArcUser Magazine, January–March, pp. 1–5. http://
www.esri.com/news/arcuser/0199/crimedata html.
4
Evaluating situational crime prevention: the
Merseyside ‘alleygating’ schemes
Chris Young, Alex Hirschfield, Kate Bowers and Shane Johnson
4.1
INTRODUCTION
It is encouraging to note that examples of mapping and GIS research in the field of crime
analysis are becoming increasingly widespread and extensive (see Hirschfield and Bowers,
2001), but to date there has been little reference to the use of GIS for the monitoring and
evaluation of Alley gating—an intervention designed to both reduce and prevent crime, reduce
fear of crime, improve the urban environment, (re)build communities, and improve health.
Such schemes on Merseyside attempt to be truly holistic because the gates are also co-designed
by the police and constructed by recovering drugs offenders.
This chapter focuses specifically upon the crime reduction aspect of Alleygating, and the use
of GIS and police recorded crime data to measure the impact of this intervention on Merseyside
—a metropolitan region in the NW of England, (principle city Liverpool situated on the River
Mersey). The evaluation is set against the backdrop of a generally downward trend of crime in
Britain—the main exception being robbery and theft from the person as reported in the 2000
British Crime Survey—BCS (Kershaw et al., 2000). According to the BCS, burglary was down
21% from 1997 to its lowest level since 1991, the number of vehicles stolen was down by 11%
from its 1997 level and theft from vehicles was down by 16%. However the BCS, where
possible, also compares the results from their survey with recorded offences and from 1997–
99, burglary was down by 13%, and theft from vehicles was down by 9%. Possible
explanations for these differences include precision of the BCS estimates and trends in
reporting and recording of crimes to the police. However, the overall downward trend still
provides a benchmark against which the Merseyside Alleygating evaluation can be compared.
4.2
BACKGROUND TO THE EVALUATION
A ‘Northern Evaluation Consortium’ from the Universities of Liverpool, Hull and
Huddersfield, lead by the Environmental Criminology Research Unit (ECRU) in the
Department of Civic Design at Liverpool, was commissioned by the Home Office in September
1999 to evaluate 21 burglary reduction projects situated in the north of England. These 21
projects were part of the first phase of 63 country-wide Strategic Development Projects
(SDPs) of the Burglary Reduction Initiative, an arm of the government’s wider £250 million
national Crime Reduction Programme.
Figure 4.1 Alleygate in-situ.
The 63 SDPs were funded, after an application process, on the basis that they were
attempting to tackle domestic burglary in ways that adopted a mixture of interventions, that
were either new, or at least applied existing techniques in a novel manner or, for instance, in
combination with other interventions. The evaluation is still ongoing and final outcome and
process reports will not be submitted to the Home Office until 2002.
Across such a large number of projects, it was perhaps inevitable that some of the 21 SDPs
would end up trying to reduce domestic burglary by adopting interventions that were not
unique. It was therefore an exciting prospect for the evaluation consortium to be able to
compare the application of an intervention (or at least an intervention within 21 various
‘packages’ of interventions) in at least 2 or more different geographical and demographic
areas. However, an important aspect and first step of the evaluation is to start with individual
SDPs. Final results of the 21 SDPs by which their performance can be assessed, will be the sum
total of each individual project, which in turn is composed of several individual interventions.
A single intervention type is the subject of this chapter. Some example interventions include
target hardening, offender supervision and surveillance, diversion schemes and market
disruption, but it is ‘Alleygating’ which offers a particularly attractive option for study here.
Alleygating has been chosen because it is a novel domestic burglary reduction method
involving the installation of secure gates at the entrances/exits to Alleyways at the rear of
domestic, usually older Edwardian terraced, properties. An example of one such gate is shown
in Figure 4.1. All residents (as well as principle utility services and emergency services) are
provided with a registered key to the gate that provides access to their own Alley. An Alley can
be of various sizes in order to allow either pedestrians or vehicles to gain easy access and exit
to and from the rear of properties for both legitimate and unfortunately, illegal purposes.
Data indicates that a high percentage of domestic burglaries are perpetrated via the rear of
properties; the British Crime Survey reports 55%, but research specific to mostly terraced
EVALUATING SITUATIONAL CRIME PREVENTION 39
streets finds that the figure increases to 72% for this type of housing (Johnson & Loxley—
2001). This is likely to be because of the more concealed aspect of the Alleyway compared to
the front of a property in urban areas, where the offender is at greater risk of being disturbed
and/or observed. Alleyways also provide a convenient and again, relatively hidden, network
by which an escape can be made after an offence has been committed. In addition Alleys are
often poorly lit (although ‘Alleygate’ schemes often attempt to improve lighting at the same
time as the gates are installed) and create the perfect environment for other illicit activities
such as prostitution and disorder offences including ‘youths causing annoyance’. They also
become an area for fly-tipping and rubbish dumping allowing domestic pets to rummage
through bin bags, leading to obvious unwanted environmental issues.
The mechanics and practicalities of Alleygating from initial plans through to eventual
installation is a long and complex process, involving procedures such as highway closure,
public rights of way issues and resident consultation and consent, that in itself merits the attention
of a dedicated study. Unpublished results from a ‘pilot’ Alleygate scheme covering some 200
households and four streets in an area of terraced housing in Liverpool on Merseyside indicated
that domestic burglary had decreased by around 50% after the installation of the Alley-gates. The
pilot area demonstrated other positive outcomes such as reductions in fear of crime,
environmental improvements and the formation of an active residents association. This chapter
extends the spatial analysis of recorded domestic burglary patterns within Alley-gated zones
because it is not limited to just one single scheme. It benefits from a before and after gate
installation period in 20 gated areas (including the pilot area) in one region of the UK. By
concentrating upon the range of data and GIS techniques required to undertake a robust
evaluation of the Alleygating intervention the validity of any results extolling the virtues of
Alleygating as a crime reduction technique are also given added weight.
The use of GIS is particularly appropriate in the evaluation of Alleygating because this
intervention protects individual properties within geographically defined Alley-gated zones.
The area of Merseyside is also appropriate because it is a ‘pioneering’ Alleygate region within
the UK and many of the agencies concerned with process management, and the raw data
required for evaluation were already familiar to ECRU. The original ‘pilot’ scheme was lead
by the Safer Merseyside Partnership (SMP) and Merseyside Police but because of the
spectacular expansion and popularity of Alleygating throughout Merseyside, an agency (Safer
Terraces Alleygating Project—STAP) was established to manage the entire Alleygating
process, but with continued input and support from the SMP, Police and other relevant
players such as Local Authorities, Fire Service and residents groups.
In summary, this chapter will describe aspects of the evaluation of one particular type of
crime reduction intervention in one geographical region; Alleygating in Merseyside. The
techniques can be used by the Northern Evaluation Consortium in some of the other SDPs that
are attempting to install gates, but importantly they form one component within a toolbox of
techniques that can be applied in the evaluation of crime reduction programmes.
4.3
METHODOLOGY
STAP provided ECRU with an Ordnance Survey map for each of the 20 currently gated areas.
Each map displayed the location of every gate within each Alleygated area, and the position of
these gates in relation to the network of Alleys and streets allowed a clearly defined
‘Alleygated’ zone to be identified. Each zone was digitised to create a GIS base map containing
20 polygons—each polygon therefore representing an area protected from burglary by

Alleygates. Instead of having just one Alleygated area the author was now free to choose
combinations of 2 or more areas up to a maximum of 20 areas for analysis. This fact is
potentially useful because each scheme had a different start date and not all were complete,
that is, some areas were only partly ‘protected’ because at the time of writing they were
waiting for remaining gates to be installed. However for completeness and in order to
maximise the impact of the data, all 20 gated areas were analysed as one ‘block’. A larger
number of schemes increases the validity of statistical analysis and although just 6 areas had
been fully gated as of 1st April 2000 it is not unreasonable to assume that work in progress in
the remaining 14 areas will have positive effects upon the frequency of domestic burglary.
Appropriately the police recorded crime data provided to the author was for financial year
periods (i.e. April 1st to March 31st inclusive). Also because, at this stage of the research, the
authors are concerned with the medium to long-term effects of Alleygating, and not the
month-by-month or day-to-day effects, the precise date tag attached to each burglary record
within the police data was ignored. Therefore GIS analysis within this chapter will investigate
higher level data ‘patterns’ in the Alleygated areas. The ‘before’ Alleygating period is
straightforward since crime data is available from 1995/6, and the first gates were not installed
until April 1998. The next phase of installation was between January and October 2000, with
6 fully gated at 1st April 2000 as stated above. The crime data for the period April 1999 to
March 2000 and April 2000 to March 2001 inclusive would therefore represent at best a
‘transition’ period rather than an ‘after’ (in the true sense of the word) Alleygating period.
That is, this study had 20 areas at its disposal that were either fully or partially gated as of the
31st March 2001. One issue regarding the treatment of all 20 areas as one single block of
protected properties was that standardised buffer zones could potentially overlap, but as will
be described below, crime rates could be calculated for any resultant shape of buffer zone since
the number of properties in each buffer zone would be known based upon property counts
within each buffer zone.
4.3.1
Crime Counts
In order to calculate crime counts and rates for each of the Alleygated areas, data relating to
domestic burglary and properties were obtained from Merseyside Police and Ordnance Survey
(via the local planning authority) respectively.
The Merseyside Police Integrated Criminal Justice System (ICJS) has been ‘live’ since the
early 1990s. However the algorithms used to generate grid references from domestic address
information, incorporated within ICJS, were generated during the late 1980s and therefore
pre-date national standard data sets such as Ordnance Survey’s Address Point. Nevertheless,
the algorithms developed were quite pioneering and it is worth dwelling on the process of data
capture here, to understand how a grid reference for a recorded domestic burglary is obtained.
When a crime occurs a police officer enters the address into ICJS and the algorithm calls upon
a database called MARS (Merseyside Address Referencing System). MARS is a geo-referenced
street and address system made up of line and point segments. A long straight road can be
made of several segments and curved sections can be made up of many more. Each segment
has a known length and a street or road name attribute. MARS also contains a ‘per segment’
addressed property count. To allocate the crime to a one-metre grid reference along the
segment the algorithm first matches the address and then divides the known length of the segment
by the known number of properties. Thus a 100 metre line segment with 20 properties
numbered 1 to 20 will place house number 5, 25metres along from the start point of the
segment. It is clear that this may be suitable for a row of identical terraced properties but is
not entirely appropriate for a line segment representing a road consisting of different sized
properties such as a mix of semi-detached/detached and terraced houses; the grid reference
allocated could fall just outside the boundary of the property it is supposed to represent.
Visual inspections by the authors revealed that most of the burglary crime record grid
references were at least allocated to the correct Alleygated zone, if not always the correct
property. A problem might arise if one was to research, using GIS tools, crime reduction
interventions at the individual property level (for example target hardening) because an error
of only a few metres could place an incident in the wrong property. If this was the case,
address cleaning processes that deal with data quality issues are available and are discussed by
Johnson et al. (1997). Cleaned addresses can be matched with a database such as Ordnance
Survey’s Address Point; a procedure which would then include the allocation of a grid
reference from the database, that falls within the property boundary, close to the property
centroid.
This research could not use aggregate crime data (that is, at the police beat level) because
although Alleygating itself is areal, it is protecting new groups of individual properties that do
not conform to sets of existing boundaries such as police beats1, Census Enumeration Districts
(Eds), or local authority wards. Best-fit look-up tables are an option—matching police beat
crime data to approximate boundaries of Alleygated zones (based on wards or EDs) but this is
clearly not ideal. Disaggregate level, individually geo-referenced crime records are therefore
the best option for analysis.
ICJS data provided by the Merseyside Police Force was for complete financial years from
1995/6 to 2000/01 inclusive. Thus a substantial ‘before’, and a reasonable ‘after’ Alleygating
period (given that the first Alleygate scheme went live in April 1998, the rest beginning to
come on-stream during early 2000) of domestic burglary data was available for this research.
To comply with data protection agreements and victim confidentiality, data security measures
(including secure storage and the use of non-networked computers) were in place at all times
when dealing with and analysing the individual-level data derived from ICJS.
4.3.2
Property Counts
To calculate a burglary rate for the Alleygated areas it was necessary to produce property
counts for the areas protected by Alley-gates. 1991 Census household data has been used in the
past because it takes account of Houses in Multiple Occupancy (HMOs). One property could
contain several separate flats or apartments and a burglary rate would effectively be increased
if calculations were based on properties. For example, 10 burglaries recorded in 100
properties during a year equates to a rate of 100 burglaries per 1000 properties. However if
each of those 100 properties contained 2 separate flats (i.e. 200 flats or households) the rate
would reduce to 50 burglaries per 1000 households over the same period. A decision was take
not to use 1991 census household counts firstly because it is becoming out of date given that
Census night 2001 has already taken place. However, it could be argued that household—and
1 Police beats define operational boundaries within which police officers patrol.
property counts—especially when measured at the regional level, do not change dramatically,
even over a 10 year period. A more relevant consideration is that household counts are only
available down to ED level and it was felt that an approximation to the digitised Alleygated
areas, although possible using GIS, would be less appropriate than using property counts based
upon Ordnance Survey’s Address Point. Address Point data for 1999 for Merseyside was
available for this research from the local planning authority, and the value of the dataset would
be highlighted when property counts were required for the buffer zones to be used in the
analysis of displacement. Rates based on properties would at least be consistent (despite being
artificially inflated in reality if it were possible to compare them with rates based upon
household counts) since they would be applied to all polygons under investigation, including
both, Alleygated areas and all buffer zones. In essence, a denominator is required in order to
ascertain changes in the burglary rate year-on-year.
The ICJS recorded crime data was imported into Arcview GIS as a .dbf file with the
National Grid Eastings and Northings of each crime incident providing X and Y coordinates for
the conversion to shape files for each of the financial years from 1995/6 to 2000/2001.
Ordnance Survey’s Address Point tiles2 in national transfer format (NTF) were imported
using a programme called the ‘Classic NTF Translator’ produced by ‘ByDesign’ for Mapinfo,
and supplied by TMS of Harrogate. This programme reads the Address Point data into a Mapinfo
table. The table was then translated from Mapinfo into an Arcview shape file for an area large
enough both to cover all of the 20 Alleygated zones and a region extending beyond all 20
Alleygated zones to be used as appropriate buffer zones. There was no scientific reason for
choosing Arcview over Mapinfo; the researcher simply had more experience of data
manipulation and analysis with the former, but found that the NTF conversion programme
available for Mapinfo was particularly user-friendly.
4.3.3
Buffer Zone Analysis
A concentric ring buffer zone was chosen for several reasons, but primarily because it allows
researchers to make full use of individual level crime data. With aggregate data representing
police beats for example, the researcher would have to look at all of the police beats
surrounding the selected beat for evidence of displacement. Figure 4.2 shows an example of
Alleygated areas in relation to police beat topography, and it also demonstrates that the crime
reduction intervention will not always occupy an entire beat. Concentric buffer zones also
provide the researcher with a standard systematic tool to analyse any potential displacement/
diffusion of benefit from Alleygated zones. A fuller discussion of the benefits of both
concentric buffer zones and disaggregate level data can be found in Bowers & Johnson (2001).
Ten equally spaced concentric rings up to a distance of 2000 metres (i.e. under one and a
half imperial miles) were chosen because of examples (Wiles and Costello, 2000) in the
literature indicating that most crimes of domestic burglary are committed within one mile of
the offender’s home address. Ten individual rings of 200 metres were chosen so that the area of
each ring approximated the areal sizes of the protected Alleygated zones (although this is
problematic as they range from just 17 up to 397 properties, but with an average of 173
properties). Figure 4.3 represents these buffer zones geographically and the extent to which
2 A tile is simply a supply unit—for example a 1km×1km digital map sheet.

Figure 4.2 Police beat and Alleygated zone geography.

they cover a large proportion of the Liverpool district, including the River Mersey which
would not, in reality of course, suffer from displacement of domestic burglary! Principal
transport routes are also shown to give a representation of scale.
By using concentric ring buffer zones it would also be interesting to see if any single ring
suffered more displacement than any other ring. Indeed it might be that ‘diffusion of benefits’,
instead of displacement, into areas surrounding Alleygated zones takes place as was the case for
a scheme analysed by Johnson et al. (2001). In this case displacement was found to be worst in
the 600 metre buffer zone, with diffusion of benefit occurring in the 400 metre buffer zone.
Street and building topography within the buffer zones will probably affect diffusion/
displacement—for example domestic burglary is unlikely to displace into a public park that
does not contain any households. Likewise a nearby affluent area containing many large
detached houses with adequate security measures might not be the preferred new target of a
career criminal more used to breaking and entering via the rear of a poorly defended terraced
property in a less affluent area. A diffusion of benefit can occur if domestic burglars perceive
Figure 4.3 Concentric ring buffer zones around Alleygated areas in Mersey side.
that a wider area has been Alleygated than is actually the case. On the other hand even if they
know exactly where all of the gates are they may think (mistakenly or not) that other measures
have been introduced in conjunction with the gates such as more police patrols for instance.
The buffer zones cover quite a large region of Merseyside but data available for the entire
Police Force Area (PFA) of Merseyside (i.e. the rest of Merseyside excluding the Alleygated
areas and the buffer zones) could be used for comparison (or control) purposes. Data for the
PFA would therefore help provide some understanding of the general pattern of domestic
burglary in the whole region, within which the Alleygated areas and buffer zones are
operating, throughout the six-year period for which burglary data was available.
Table 4.1 Results in all 20 Alleygated zones (3,442 properties)—plus PFA rates.
4.4
RESULTS
Table 4.1 shows the raw counts and crime rates for all of the burglary data available for the 20
combined Alleygated areas at the time of writing. It is based on 3,442 properties in 20-gated
areas protected by a total of 208 individual gates, which were erected between January and
October 2000 except for one pilot area, which was gated in April 1998. Data provided by
STAP indicated that a lesser total of 3,160 properties were protected. This discrepancy is mainly
due to the fact that STAP exclude void properties from the protected property count but a
contributory factor might be the fact that Ordnance Surveys’ Address Point includes non-
domestic properties that may not have been included in the STAP ‘domestic’ property total.
Nevertheless the Address Point total of 3,442 properties was used so that crime rates in
Alleygated areas would be consistent with those calculated in the buffer zones. Table 4.1 also
contains the crime rates for the PFA which shows trends in domestic burglary in the whole of
Merseyside between 1995/6 and 2000/01.
Table 4.1 was produced using 12 month slices of data only (individual crime record dates
were not used). It is worth mentioning here that any seasonal fluctuations within the data—for
example an increase in domestic burglary during the Christmas period—would therefore be
masked. Also, some of the Alleygated areas did not receive gates until October 2000.
Therefore despite work to fit Alleygates being in progress (April to October 2000) this fact was
likely to have a detrimental effect on the burglary rate and therefore ‘contaminate’ the
burglary data for the period April 2000 to March 2001, which was supposed to represent an
‘after’ Alley gating period. Nevertheless Alley gating appears to be having a positive effect on
the burglary rate within all of the aggregated Alley-gate zones because the incident rate of 29.
05 in 2000/01 is substantially lower than the previous lowest rate of 43.58 burglaries in 1997/
8. The total number of 8 repeat burglary incidents is also significantly better than the previous
best of 21, also in 1997/8. Put another way repeats make up 8% of incidents in 2000/01 in
Alleygated areas—in other years between 1995/6 and 1999/00 this figure varied between 12.
1% and 17.3%.
Observing the results in all of the Alleygated zones and the PFA shown in Table 4.1, against
the results for all ten buffer zones combined—Table 4.2 and Figure 4.4 below—it can be seen
Table 4.2 Results in all ten buffer zones (109,654 properties).
that national trends of falling burglary rates (Kershaw et al., 2000) are echoed in Merseyside.
There are year-on-year reductions since 1998/9 in Alleygated areas and buffer zones, but since
1995/6 in the PFA. In the entire buffer zone, repeat incidents account for between 8.4%
(1997/8) and 10.9% (1995/6) of burglary incidents taking place during any one year, which
compares favourably with the figure of 8% quoted above for the Alleygated areas in 2000/01.
Figure 4.4 is a graphical representation of the incident rates per 1000 properties shown in
Tables 4.1 and 4.2 respectively for the Alleygated areas, buffer zones and PFA. It can be seen
that 2000/2001 is the first time that the rate in the newly Alleygated zones has fallen below
the rate in the surrounding buffer zones. This implies that Alleygating does have a positive and
significant impact upon burglaries within protected zones. It is also evident that during the six
year period between 1995/6 and 2000/01 burglary rates overall have fallen by 37.5% in the
PFA, 32.8% in the buffer zones and 53.3% in the Alleygated zones.
Figure 4.5 shows the difference between the burglary rate in each of the 200 metre buffer
zones (these figures are not shown but were used to produce the totals in column 2 of
Table 4.2) and the burglary rate in the Alleygated areas shown in Table 4.1. The crime rate in
the Alleygated zones was subtracted from the crime rate of each individual buffer zone for
every year of the six years that burglary data is available.
In Figure 4.5 if a bar is greater than zero it indicates that the rate recorded in the individual
buffer zone was higher than the rate recorded in the Alleygated area during that particular year
within the period 1995–2001. For the period that preceded the start of the bulk of the
schemes (i.e. 1995/6 to 1999/00) this situation arises only on three occasions—in the 200m
buffer zone in 1995/6 and the 200m and 1000m buffer zones in 1997/8. Following the
inception of Alleygating in 2000/01 the first five of the 200 metre buffer zones, from zero to
1000 metres, experience a burglary rate which is higher than that inside the Alleygated zones.
Obviously any randomly selected area can have a rate greater than another randomly selected
area, but the difference here is that an intervention—Alleygating—has been introduced.
Therefore, the data for 2000/2001 favours an argument that the correct areas were selected
for Alleygating because there has been an impressive reduction in the burglary rate.
Unfortunately it appears that some displacement is occurring from the 20 Alleygated areas into
the buffer zones.
It is also apparent from Figure 4.5 that, in descending order, the four zones suffering the
most from displacement effects are the 200, 400, 1000 and 600 metre buffer zones. Johnson et
Figure 4.4 Comparison of domestic burglary rate between all Alley-gated areas, buffer zones and
the PFA from 1995/6 to 2000/01.
al. (2001) found that the 600 metre buffer zone was worse off; but this study examined just one
SDP with Alleygating, and a known series of further interventions.
4.5
CONCLUSIONS
There are perhaps three main points to be drawn from this research:
• Alleygating appears to be effective in reducing the recorded burglary rate—by more than
50% compared to years when gates were not installed. This reduction appears to exceed
the current national decline of 13% since 1997 in this crime category (Kershaw et al.
2000).
• Displacement from newly Alleygated areas is an issue that requires attention by
practitioners, especially if surrounding areas consist of similar housing types and
demographies.
• GIS can be an effective tool for the planning and management of Alley-gating schemes. This
chapter has demonstrated the usefulness of GIS for the monitoring and evaluation of Alley-
gate interventions particularly because this crime reduction/prevention initiative, despite
being clearly defined and geographically concentrated, does not conform to any pre-
existing boundary or land parcel.
The burglary rates for the Alleygated areas shown in Table 4.1 vary between 29.05 and 62.17
—a reduction of almost 54% during the period 1995/6 to 2000/01. Even the previous ‘best’
of 43.58 burglaries per 1000 properties in 1997/8 has been improved by some 33%. At the
same time the domestic burglary rate in the buffer zones (Table 4.2) ranges between 31.79 and
Figure 4.5 Difference between individual buffer zone rates and the Alley-gated area rate for each
year between 1995/6 and 2000/01.
47.29—a reduction of almost 33% during the period 1995/6 to 2000/01. However the
previous ‘best’ of 34.23 burglaries per 1000 properties in 1997/8 has only been improved by
7.1%.
Displacement is an issue that cannot be ignored but other studies are optimistic that
diffusion of benefits can also result from the installation of Alley-gates (Johnson et al., 2001).
Using GIS gives the researcher freedom to explore different avenues in an evaluation, but
does require the availability of detailed and accurate data. The only drawback might be that
results published outside of the police service have to be re-aggregated for data protection and
confidentiality purposes, but at least the GIS means that a researcher can choose the size and
type of area and is not restricted to statutory boundaries.
4.6
FURTHER RESEARCH
It would also be interesting from both a research and a social policy effectiveness point of
view, to analyse a further 12 or even 24 months of data to March 2003 for example, to
ascertain the true sustainability of Alleygating as a crime reduction intervention. However
consideration will have to be given to the fact that more Alleygated areas will become ‘live’ during
that time and some may be sited inside the buffer zones of Figure 4.3, therefore affecting the
results. Repeat burglary figures were shown in Tables 4.1 and 4.2 and although they are also
encouraging in terms of a reduction, they merit a more indepth analysis in their own right.
A fuller evaluation, now that more and more areas are being Alleygated should include a
more detailed look at the crime data. The modus operandi (MO) of domestic burglary in both
new and established Alleygated areas requires investigation. Do burglars simply switch to
another mode of entry other than the rear of the property or two or three years down the line
do residents simply become complacent and leave the gates unlocked allowing burglars to gain
easy access again?
An analysis of the temporal pattern of domestic burglary (i.e. year-on-year, month-to-
month, weekly or even daily) in Alleygated areas might also prove to be a useful exercise. For
example, criminals might simply begin to take advantage of the periods when gates are left
open to allow refuse collectors to gain access.
Displacement to other types of acquisitive crime (crime switch) should also be considered.
With access to domestic properties restricted will offenders attempt to carry out more street
robberies or thefts of/from motor vehicles for instance, or will the figures for attempted
burglary/criminal damage simply go up if it’s harder to get into a house? Burglars may even up
the stakes by being prepared to use violence against a resident during a burglary, or use
confrontational techniques such as distraction burglary—therefore aggravated burglary or
assault figures might increase? Alleygating may well have to be implemented at the same time
as complementary vehicle crime reduction campaigns for example, in order to be truly
successful. Conversely researchers may have to try and isolate Alleygating as a crime reduction
method when trying to evaluate its specific successes or failures. In this chapter no account has
been taken of any other crime reduction campaign that may have been operating concurrently
(either in the Alleygated areas or the buffer zones)—perhaps a target hardening scheme or
intensive offender supervision programme had a larger impact upon the decrease in domestic
burglary?
To support the work discussed above, a cost effectiveness study of Alleygating both as a
process and crime reduction tool would be appropriate but this should not ignore any human
benefits—what price for improved peace of mind and maybe even better health for the
residents of these areas? Resident surveys, health impact questionnaires and data from other
agencies such as the Fire Service should help complete the bigger picture of the Alleygating
phenomenon.
4.7
ACKNOWLEDGEMENTS
This research is based upon the initiatives of the Safer Merseyside Partnership and the authors
are grateful for their co-operation and support. Photograph of Alley-gate courtesy of Safer
Terraces Alleygating Project.
4.8
REFERENCES
Bowers, K.J. and Johnson, S.D., 2001, Measuring the geographical displacement of Crime.
Submitted to the Journal of Quantitative Criminology.
Hirschfield, A. and Bowers, K. (editors), 2001, Mapping and Analysing Crime Data: Lessons from
Research and Practice. (London: Taylor and Francis).
Johnson, S.D., Bowers, K., and Hirschfield, A., 1997, New insights into the spatial and temporal
distribution of repeat victimisation. The British Journal of Criminology, 37 (2), pp. 224–244.
Johnson, S.D., Bowers, K.J., Young, C.A. and Hirschfield, A.F.G., 2001, Uncovering the true
picture: Evaluating crime reduction initiatives using disaggregate crime data. Crime Prevention and
Community Safety: An International Journal, in press.
Johnson, S.D., and Loxley, C, 2001, Installing Alley-Gates: Practical Lessons from Burglary Prevention
Projects. Home Office Policing & Reducing Crime Unit, Briefing Note 2/01. (http://
www.homeoffice.gov.uk/rds/prgpdfs/brf201.pdf).
Kershaw, C. et al., 2000, The 2000 British Crime Survey. Home Office Research Development and
Statistics Directorate.
Wiles, P. and Costello, A., 2000, The Road to Nowhere: The Evidence for Travelling Criminals. Home
Office Research Study No 207. London: Policing and Reducing Crime Unit, Home Office
Research, Development and Statistics Directorate.
5
Crime hot spot prediction: a framework for
progress
Jonathan Corcoran and Andrew Ware
5.1
INTRODUCTION
Crime rates differ between types of urban district, and these disparities are best explained by
the variation in use of urban sites by differing populations. A database of violent incidents is
rich in spatial information and studies have to date, provided a statistical analysis of the
variables within this data. However, a much richer survey can be undertaken by linking this
database with other spatial databases, such as the Census of Population, weather and police
databases. Coupling Geographical Information Systems (GIS) with Artificial Neural Networks
(ANNs) offers a means of uncovering hidden relationships and trends within these disparate
databases. Therefore, this paper outlines the first stage in the development of such a system,
designed to facilitate the prediction of crime hot spots. For this stage, a series of Kohonen Self-
Organising Maps (KSOMs) will be used to cluster the data in a way that should allow common
features to be extracted.
5.2
CRIME PREDICTION
The advent of computers and the availability of desktop mapping software have advanced the
analytical process, allowing efficient generation of virtual pin maps depicting crime incidents.
A logical step beyond visualisation and analysis of trends and patterns is the development of a
predictive system capable of forecasting changes to existing hot spots and the evolution of new
ones.
Crime prediction in criminological research has been established for a number of decades
(Ohlin and Duncan, 1949; Glueck, 1960; Francis, 1971), although its foundation within a
geographic and GIS context is still in its infancy.
As predictive crime analysis is a new research area, very little literature currently exists.
Olligschlaeger (1997) provides an overview of existing forecasting techniques, concluding the
time, level of user interaction and the expertise that each demands would be unrealistic for
implementation in an operational policing environment (Olligschlaeger and Gorr, 1997). In
addition, the inherent inflexibility and inability to dynamically adapt to change would
compromise their viability in policing. ANNs are presented as one technique that offers
minimal user interaction in addition to dynamic adaptability, and thus a potential operational
forecasting solution.
5.2.1
Potential for Crime Prediction by the Police
A recent survey (Corcoran and Ware, 2001) has highlighted the uptake and use of computer
based crime-mapping technologies by British Police Forces. Computerised mapping
technologies are rapidly becoming a vital prerequisite for visualisation of incident distributions
and assisting in both the identification/allocation of resources and production/evaluation of
policing strategies. The ability to efficiently generate simple maps to depict crime location and
densities can be used directly to inform police officers and policing strategies, therefore
maximising effectiveness and potential. A recent report published by the Home Office (Home
Office, 2000) has underlined the importance of geographic data for analysis and interpretation
of crime at the local level. In addition, the Chief Constable of Kent County Constabulary notes
that “over the last few years, police activity has shifted its centre of balance away from the reactive
investigation after events, towards targeting active criminals on the balance of intelligence” (Phillips,
2000).
A natural step beyond visualisation and analysis of past and current incidents is the
prediction of future occurrences, thus providing an early warning system for the Police
(Olligschlaeger and Gorr, 1997). Prediction can help prevent crime in that it facilitates the
optimal allocation of limited resources. Prevention might be better than cure, but in the real
world, very often, this is under financial constraints. The development of tools for prediction
can thus help prevent crime and maintain optimal operational costs.
Prediction requirements for the police have been classified into three main categories
according to the period of time involved (Gorr, Olligschlaeger et al., 2000):
• Short-term (tactical deployment);

• Medium-term (resource allocation);
• Long-term (strategic planning).
The focus for crime forecasting lies with short-term models as police tend to respond to
existing and emerging crime patterns on relatively short time-scales for example on the basis
of daily, weekly and monthly figures (Gorr and Olligschlager, 1998). This paper details a
potential prediction framework for short-term, tactical deployment of police resources. The
framework will allow identification of risk factors from which probabilities of criminal activity
(specifically emergence of hot spots) can be derived and the necessary resources deployed.
5.2.2
COPDAT
The Crime and Offender Pattern Detection Analysis Technique (COPDAT), outlined in this
paper, offers a potential framework that can be applied to geographic prediction. COPDAT
entails the implementation of a GIS, integrating various spatial databases to analyse and map
the identified trends.
CRIME HOT SPOT PREDICTION: A FRAMEWORK FOR PROGRESS 53
5.3
METHODOLOGY
The volume of crime is insufficient to accurately predict an offence (in terms of location and
time) when extrapolating from past offences. Therefore, in the proposed system, the type of
crime predicted to take place within a particular time-window is supplemented by a separate
prediction of the likely vulnerable areas for the same epoch. In addition, it would seem prudent
to have the facility in a finished system to allow information based on police intelligence and
experience to be built into the predictive model. The idea is to enhance current police
predictive capabilities not replace them.
The spatial framework for the prototype COPDAT conforms to police sector boundaries
for Cardiff (the capital city of Wales in the United Kingdom), whereby the volume of crime is
sufficient to derive accurate predictions.
5.3.1
Data Sets
The accurate forecasting of the temporal-geography of crime (where and when a crime is likely
to take place) would be of immense benefit, for accurate prediction if acted upon should lead
to effective prevention. However, crime prediction frequently relies on the use of data
appertaining to past perpetrators and/or past victims. Such data is therefore subject to legal
and ethical restriction on its use, resulting in an ethical conundrum (Ware and Corcoran
2001). Therefore, the prototype COPDAT involves the use of only two central data sets—one
pertaining to crime and the other providing contextual information.
5.3.2
GIS Techniques
Visual inspection of the various spatio-temporal data sets is a vital pre-requisite in assimilating
an understanding of fundamental relationships and trends. The GIS is used as a tool to conduct
this basic pattern analysis, including cluster and hot spot techniques.
5.3.3
Artificial Neural Network Techniques
ANN models provide a mechanism through which the various spatial, non-spatial and temporal
data sets can be analysed to identify patterns and trends previously undiscovered. Identification
and consolidation of such trends and patterns between the various data sets allows
generalisation and subsequent prediction of future events and scenarios based upon that
generality. For example, identifying that a specific kind of crime is likely to occur in a certain
location type under a given set of ambient conditions allows future incidents to be predicted.
The result of this scenario can be to produce a spatial probability matrix encapsulating a risk
assessment of the study area. The spatial probability matrix can be directed to the GIS for
visualisation in the form of a thematic contour map differentiating different areas with respect
to their potential risk. In essence, the ultimate goal of the COPDAT is to learn decision
criteria for assigning risk levels to new and future situations. This, for example, may involve
identifying and predicting areas most at risk during a hot summer bank holiday football match
Figure 5.1 Data preparation process in relation to ANN processing.
in the City. Provision of such information is of obvious interest and of operational benefit to
the police in terms of both resource allocation and policing strategies.
5.3.4
Data Preparation and Processing
Data representation and structuring is of key importance in the production of a robust
predictive model. It has been shown in previous neural network studies that a certain level of
pre-processing of the raw data is advantageous to model accuracy, efficiency and stability. This
approach subsequently requires a certain level of post-processing in order to generate the
required output values (illustrated in Figure 5.1).
5.3.5
Data Pre-Process
First, the data will undergo systematic and comprehensive analysis. This is followed, where
necessary, by converting the data into an alternative representation more suited for input into
an ANN. This sequential process can be broken down to a series of discrete stages:
• Format conversion and integration;

• Error detection and editing;
• Data reduction, transformation and generalisation.
The final stage in the pre-processing is of critical consequence in terms of a successful ANN
implementation. The process of feature extraction and encoding of such characteristics impacts
upon the ANN’s ability to learn and assimilate relationships between salient variables and
hence its accuracy in prediction. This process can be further decomposed into three distinct
procedures:
1. Transformation and scaling may include:
• Applying a mathematical function (e.g. logarithm or square) to an input;

• Scaling the different inputs so that they are all within a fixed range can greatly effect the
reliability of an ANN system.
2. Reduction of relevant data includes simple operations such as filtering combinations of

inputs to optimise the information content. This is particularly important when the data
is noisy or contains irrelevant and potentially erroneous information.
3. Encoding of identified features for input to the ANN. The data types include a range of
data categories (discrete, continuous, categorical and symbolic), each to be handled and
represented in an explicit manner.
5.3.6
Clustering Using a Kohonen Self Organising Map
Temporal, spatial and incident data will be clustered using a series of KSOMs. These clusters,
whose data share the same characteristics, will form the basis for rule abduction. (Note,
deduction is where the effect is deduced from the cause—for example, ‘the burglary was
committed because the house was left unlocked.’ Abduction is where the cause is gleaned from
analysing the effect—for example, ‘sun and alcohol often causes socially unacceptable
behaviour’.)
An unsupervised network, such as the Kohonen Self Organising Map (KSOM), organises
itself in such a way as to represent classes within a data set. The 2D KSOM allows classes to be
visualised on a feature map, in which similar inputs are spatially clustered. Figure 5.2 shows a
typical 2D KSOM along with an abridged algorithm (N.B., the number of nodes are arbitrarily
selected for example purposes).
Each output node on the KSOM contains a vector of length j, where j is equal to the
number of input attributes. Before training begins, the network is placed into an initialised
state, i.e. the directions of the vectors in each node are randomised. Training involves passing
an input vector into the network through the input nodes. Each node on the KSOM is then
compared with the input vector, and the closest node is changed to be more like the input
vector. Neighbouring nodes also become more like the input vector. Iterating this process
achieves clustering of similar input vectors in Euclidean space.
Figure 5.2 A Kohonen Self Organising Feature Map.

5.3.7
An Overview of the Methodology
The methodology uses a KSOM to find clusters in the input vectors and then the data from
each cluster is used to train a separate MLP network. The advantage of using the KSOM for
this application is that it can identify clusters within the parent dataset that are difficult to
identify using simple sort procedures. Figure 5.3 gives an overview of the method. A dataset
containing the required elements of the vector is passed through the KSOM during the training
stage and allowed to develop into clusters. After training, the clusters are inspected and the
primary clustered features used to describe the sub-datasets. These sub-datasets are then used
as the basis for rule abduction.
However, two fundamental problems need to be resolved before this method can be of any
use. First, the problem of separating adjacent clusters, and second, the desire to proceed to the
abduction phase only using ‘good’ clusters (see Figure 5.6).
The first problem has been recognised in other studies and some guidelines have been
provided (James 1994). In essence, the problem lies in the attribution of boundary nodes to a
specific cluster. Figure 5.4 provides an example of a KSOM output with adjacent clusters.
There appear to be four classes within the dataset, but there are regions of uncertainty relating
to the boundaries of each cluster (the Digits in each node show the number of vectors mapped
to that node).
Figure 5.3 Using a KSOM to partition a dataset.

To overcome this problem a simple method of identifying class boundaries or discriminants
can be used, which relies on the fact that a KSOM clusters primarily on binary features. For
example, if the type of crime is represented using binary inputs, the KSOM will tend to cluster
records according to this attribute. Boundaries between adjacent clusters on a 2D map can
then be found by inspecting the records mapped to each node and grouping together nodes
that contain the same classification values. However, this level of clustering can be achieved
using a multi-level sort procedure. In essence, the binary representation of the data will dictate
the make-up of the resulting clusters and more importantly the homogeneity of the data sets.
If the data are represented using continuous inputs, the clusters formed by the KSOM
would provide more generalised classes which would be difficult to achieve using a sort
procedure. However, the inspection method would no longer identify class boundaries, as the
similarities between records would not always be apparent. Clearly, before meaningful
training data sets can be formed, the problem of discerning effective class boundaries in a
KSOM must be addressed. Ideally, the network adaption rule should cluster similar inputs and
clearly distance individual clusters. Zurada (1992) explains “One possible network adaption rule is: A
pattern added to the cluster has to be closer to the centre of the cluster than to the centre of any other
cluster”. Using this rule, each node can be examined and the distance from the surrounding
centroids can be calculated (a centroid is taken to be a node, outside any known cluster
boundaries, that has the largest number of input vectors mapped to it). The subject node can
then be added to the nearest cluster. Figure 5.5 illustrates a hypothetical situation where it is
unclear where to draw the boundaries around clusters on a KSOM.
By simply calculating the Euclidean distance of the subject node from the two centroids, the
subject node can be assigned to the closest cluster. However, in this application, which aims to
generate clusters with latent but meaningful information that can be subsequently extracted
using abduction techniques, the formation of a class boundary for Cluster 1 (including the
Figure 5.4 An Example Trained Kohonen Self Organising Feature Map.
subject node) may dramatically increase the variance of the training data. This increase will
reduce the potential accuracy of the model. In the example, it may have been better to exclude
the subject node from either of the clusters, and deem the vectors mapped to the subject node
as either being outliers or a separate cluster.
In addition to identifying boundaries around input clusters, it is also important for this
application to match input clusters to appropriate output clusters. In terms of criminal activity,
if, for example, the KSOM has clustered crimes from two different locational areas, it is
reasonable to expect these crimes to have similar characteristics.
Figure 5.6a illustrates a cluster of similar input vectors. When the corresponding data in
output space is examined, all the examples describe similar output values. Conversely,
Figure 5.6b shows a situation where the data can only be modelled using two or more functions.
The problem now is to estimate the ‘usefulness’ of a given cluster. There are a number of
options available of which the following are the most useful:
Figure 5.5 An Example KSOM.
Figure 5.6 Example cluster mappings from input to output space.
The problem now is to estimate the ‘usefulness’ of a given cluster. There are a number of
options available of which the following are the most useful:
• Multi-Layered Perceptron (MLP) Model (Chen 1997)

• Class Entropy (Quinlan 1986)
• R2 Almy (Almy 1998)
• Gamma Test (Stefánsson 1997)
For classification problems, Class Entropy can be used to decide if input clusters are
homogenous with respect to output clusters. For example, Quinlan’s C4.5 and C5.0 (Quinlan
1993) uses Class Entropy to segment the input space until each segment points toward a single
class in output space. However, this approach is not applicable for regression problems such as
this one and this rules out the use of class entropy.
Figure 5.7 Interpreting the output from the Gamma Test.
A quick and easy estimate of the susceptibility of a dataset for function induction can be
achieved by executing a multiple regression analysis on the data and use the R2 value to discern
trainable clusters. This technique is useful for data where the function is known to be linear.
However, this is not known to be true for crime analysis data.
The Gamma Test attempts to estimate the best mean square error that can be achieved by
any smooth modelling technique using the data. If y is the output of a function then the
Gamma test estimates the variance of the part of y that cannot be accounted for by a smooth
(differentiable) functional transformation. Thus if y=f(x)+r, where the function f is unknown
and r is statistical noise, the Gamma test estimates Var(r). Var(r) provides a lower bound for
the mean squared error of the output y, beyond which additional training is of no significant
use. Therefore, knowing Var(r) for a data set allows prediction beforehand of what the MSE of
the best possible neural network trained on that data would be. The Gamma test provides a
method of determining the quality of the data stratification—a good stratification technique
will result in a low value of Var(r) for each subset. Interpreting the output from the Gamma
test requires considerable care and attention.
The least squares regression line provides two pieces of information. First, the intercept on
the Gamma axis is an estimate of the best MSE achievable by any smooth modelling technique.
Second, the gradient gives an indication of the complexity of the underlying smooth function
running through the data. The Gamma test may estimate a very low MSE but unfortunately
show a high level of complexity that could be difficult to accurately model. It is easier to see
this situation when the output from the Gamma test is presented graphically.
A hypothetical example with high noise content and high complexity is shown in
Figure 5.7a; high noise and low complexity Figure 5.7b; low noise and high complexity in
Figure 5.7c; and, finally, low noise and low complexity (the desired outcome) in Figure 5.7d.
In summary, for this methodology to be successful, the following is required:
• class boundaries must be identified around clusters formed by the KSOM over the input
space that exclude outliers and nodes from neighbouring clusters, and;
• only ‘good’ clusters (illustrated in Figure 5.6) should go on to form training data sets for
subsequent back propagation models.
5.3.8
A Detailed Look at the Methodology
The Gamma test can be used at a number of abstraction levels within the KSOM stratification
method:
• Cluster level;
• Node Level;
• Record Level.
Data stratification is achieved at cluster level or at node level, depending on the ease at which
cluster boundaries can be determined. The record level gives an indication of outliers.
5.3.8.1
Cluster Level Analysis
This can be achieved thus:
Identify Cluster boundaries in KSOM

For every cluster
Place records mapped to cluster into a file
Apply Gamma test to data in the file
lf Var(r)<=some Threshold then
Use data file as the training set for a MLP
else
Process at Node Level
Endif
Next Cluster
This level of abstraction is the least computationally intensive as it only requires one pass of the
Gamma test for each cluster. The disadvantage with this method is that it is often difficult to
identify boundaries manually. In this case the Gamma test should be applied at the Node level.
5.3.8.2
Node Level Analysis
At this abstraction level, the methodology attempts to identify useful clusters by selecting a
centroid and adding neighbouring nodes—where the addition of a node increases the variance
significantly it is subsequently removed. This process iterates until the cluster size is maximised
within a specified variance threshold. This algorithm identifies useful clusters on a 2D KSOM.
This is achieved thus:
number_of_clusters:=0
While there are nodes to cluster
number_of_clusters:=number_of_clusters+1
Select the unclustered node with the largest record count
Apply Gamma test to the data in the selected node
If Var(r)<=Threshold then
nodes_of_interest:=None (cluster includes only the data from selected node)
For each unclustered node next to selected node
Add data from unclustered node to the cluster
Run gamma test on cluster
Add node number to nodes_of_interest
else
Remove data from the unclustered node from the cluster
Endif
NextNode
While nodes_of_interest <> None
Select c_nodefrom nodes_of_interest
Remove c_nodefrom nodes_of_interest
For each unclustered node immediately surrounding c_node
Add data from unclustered node to the cluster
Run gamma test on cluster
Add node to nodes_of_interest
else
Remove data from the node from cluster
Record the boundaries of this cluster
Endif
NextNode
EndWhile
else
Process at Record Level
Endif
EndWhile
The boundary detection algorithm for a 1D KSOM is very similar except neighbouring nodes
are selected progressively further away from the left and the right of the centroid node. This
level of analysis is more computationally intensive than the cluster level analysis, as it requires
m*Σni passes of the Gamma test, where i is the number of nodes investigated for cluster n for a
KSOM containing m clusters. If using either the cluster level analysis or the node level analysis,
useful clusters have been identified, it is then possible to train an independent MLP on each
subset. The KSOM is then used to select the appropriate MLP on which to predict the value of
a previously unseen example. The resulting system is closely related to a panel judgement
system. However, if both methods have still resulted in poor training sets (useless clusters)
then the analysis is taken to the most detailed abstraction level, that is the record level.
5.3.8.3
Record Level Analysis
The record level analysis is the most computationally intensive. The purpose of this level of the
methodology is to identify data subsets from examples that have mapped to the same node on
the KSOM. This facilitates extraction of outliers from a data set as well as giving some
indication as to the examples that require additional features. The algorithm developed for this
level of analysis is very similar to that shown for the node level analysis. However now, sets of
records are iteratively analysed using the Gamma test. This is achieved thus:
For each node in the KSOM

Apply Gamma test to estimate the variance for the data in node
If Var(r)>Threshold then
For each record at node
Remove record from data set
Apply Gamma test to estimate the variance for the data in node
If New Var(r)<Previous Var(r) then
Mark record as outlier
else
Add record back into data set
Endif
NextRecord
else
Proceed at Node Level
Endif
NextNode
This level of analysis will identify the need for additional features and highlight records that
may be classed as outliers.
5.3.9
Post-Processing
Natural language rules will be derived from each ‘good’ cluster found by the KSOM. Each rule
will represent a broad and generic definition (that with time can be fine tuned) of a specific
sub-model that can be applied to best predict crime, for example:
if Centre of City
and Weather includes Wet
and Day is Friday
and Time is Night
then Problems will include inside pubs (probability 0.9)
if Centre of City
and Weather includes Dry
and Day is Friday

and Time is Night
then Problems will include inside pubs (probability 0.4)
if Centre of City
and Weather includes Dry
and Weather includes Warm
and Day is Friday
and Time is Night
then Problems areas will include outside pubs (probability 0.9)
These rules together with other determining factors (including temporal information such as
the day, time and prevailing weather) are then directed to the GIS for production of a thematic
crime risk contour map. When used in a predictive manner, the GIS can provide an important
visual reference for analysing the relative impact of multiple factors on crime levels in a given
area.
5.4
IMPLEMENTATION DETAILS
A series of KSOMs will provide the mechanism for classifying the heterogeneous nature of
crime and criminal activity in a spatial, temporal and contextual framework. The assumption
that there is a single simple linear relationship between the various factors is unrealistic and
misplaced. In the light of this, a valid approach to assimilate an understanding of crime and
criminal activity is to dissemble it into its homogeneous components for independent analysis.
This analysis can be used to generate a series of rules and generalisations that can be utilised in
a predictive capacity at a broader heterogeneous level.
The KSOM provides an unsupervised technique, through which this stratification process
can be automated, resulting in the formulation of homogeneous clusters (illustrated in
Figure 5.8). Where clusters are not clearly delineated it maybe necessary to calculate a
function such as the Euclidean distance or Gamma test (Lewis, Ware et al. 1997b, 1997a) in
order to assign output to specific clusters.
The user will be able to present the trained suite of KSOMs, in the COPDAT, with
structured scenarios from which probabilities and vulnerabilities of criminal activity will be
derived and presented to the user via spatial thematic representations. Using standard GIS
functionality, the user could subsequently overlay additional information, which may for
example involve identification of specific localities, (street names, building etc) for
deployment of resources.
5.5
SUMMARY AND BENEFITS
Coupling the predictive capabilities provided by the proposed COPDAT model with existing
GIS techniques would enable law enforcement agencies to more effectively evaluate resource
and tactical policies. Crime mapping and analysis using COPDAT will potentially facilitate
research, assist with offender management and monitoring, and enable community planners to
develop policy and forecast future needs for public safety resources. Provision of such facilities
will rely upon use of multiple data sources and be dependent upon the use of pre-defined data
Figure 5.8 Proposed COPDAT Hybrid System.
processing and encoding techniques that will require new standards and techniques to be
integrated into current practices. The Chief Constable of Kent County Constabulary raises this
point in respect to police intelligence noting that “this process is only possible if we can mix and
match information across the board. To this end, it is essential that common standards and discipline
attach to the intelligence process” (Phillips 2000).
This paper outlines a potential framework that can be expanded to model and predict a
broader variety of crimes based upon a larger mixture of criminal and contextual data sets and
a common schema for crime prediction.
5.6
REFERENCES
Almy, R., Horbas J., Cusack, M., Gloudemans, R., (editors), 1998, The Valuation of Residential
Property Using Regression Analysis. Computer Assisted Mass Appraisal: An International Review,
(England: Ashgate Publishing Company).
Chen, K., Xiang,Y., Huisheng, C., 1997, Combining linear discriminant functions with neural
networks for supervised learning. Journal of Neural Computing and Applications 6.
Corcoran, J. and Ware, J.A., 2001, Application of Computer Crime Mapping Technologies by British Police
Forces. Available from: jcorcora@giam.ac.uk.
Francis, S.H., 1971, Prediction Methods in Criminology, Home Office Research Studies.
Glueck, S.E., 1960, Predicting Delinquency and Crime. (Cambridge, Massachusetts: Harvard
University Press).
Gorr, W. and Olligschlager, A., 1998, Crime Hot Spot Forecasting: Modelling and Comparative
Evaluation, http://www.ojp.usdoj.gov/cmrc/grants/welcome.html.
Gorr, W., Olligschlaeger, A. and Thompson, Y., 2000, Assessment of crime forecasting accuracy
for deployment of police, International Journal of Forecasting.
Home Office, 2000, Review of Crime Statistics: A Discussion Document, http://
www.homeoffice.gov.uk/crimprev/crimstco htm.
James, H., 1994, An ‘Automatic Pilot’ for Surveyors, RICS Cutting Edge.
Lewis, O.M., Ware, J.A. and Jenkins, D.H., 1997a, A novel neural network technique for
modelling data containing multiple functions. Computational Intelligence—Theory and Applications,
Springer Verlag, pp. 141–149.
Lewis, O.M., Ware, J.A. and Jenkins, D.H., 1997b, A novel neural network technique for the
valuation of residential properties, Journal of Neural Computing and Applications 5, pp. 224–229.
Ohlin, L.E. and Duncan, O.D., 1949, The efficiency of prediction in criminology, American Journal
of Sociology: LIV 5.
Olligschlaeger, A. and Gorr, W., 1997, Spatio-Temporal Forecasting of Crime: Application of Classical
and Neural Network Methods, http://www.heinz.emu.edu/wpapers/active/wp00162.html.
Phillips, D., 2000, National Intelligence Model, National Criminal Intelligence Service.
Quinlan, J.R., 1986, Induction of decision trees, Machine Learning 1, pp. 81–106.
Quinlan, J.R., 1993, C4.5: Programs for Machine Learning, Morgan Kaufmann.
Stefánsson, A., Koncar, N., Jones, A.J., 1997, A note of the Gamma test, Neural Computing and
Applications 5, pp. 131–133.
Ware, J.A. and Corcoran, J., 2001, Forecasting Crime: An Ethical Conundrum. Available from:
jaware@glam.ac.uk.
Zurada, J. M., 1992, Introduction to Artificial Neural Network Systems, West Publishing Company.
PART II
GIS and Planning

6
E-community participation: communicating
spatial planning and design using web-based
maps
Kheir Al-Kodmany
6.1
INTRODUCTION
This chapter addresses innovative methods in using Web-based mapping and Geographic
Information Systems (GIS) in public participation and community planning. The ultimate goal
is to explore how Web mapping can be advanced to facilitate two-way communication
between planners and the public. Presently, most Web-based mapping systems use one-way
communication to provide static information to the public. Current GIS technologies lend
themselves well to one-way communication. While this type of system can be very useful, it is
important to consider new technologies that offer opportunities to use the Web in ways that
have not previously been recognised. Using examples from a university-community
partnership between the University of Illinois at Chicago (UIC) and community groups in
Chicago’s Pilsen and North Lawndale neighbourhoods, several such new technologies are
introduced and examined. These examples show the progression from one-way
communication to two-way and even three-way communication using Web-based maps. The
challenges of creating interactive screen-based maps for public participation include navigating
large geographic area maps on a small screen, selecting specific map features, geo-referencing
public input and transferring this input to conventional GIS systems. These challenges as well
as many successes of our Web mapping research are explored in detail. As Web-based
mapping capabilities are refined and improved, new avenues are opened for public
participation in the planning process.
The research in this chapter reflects recent views about the importance of incorporating
citizen participation into computer-based planning efforts. The phrase ‘public participation
GIS’, or PPGIS, describes recent research from the planning profession that is rooted in the
concern that all voices should be heard in a democracy. In particular, it aims at improving
access to GIS among non-governmental organizations and individuals, especially those who
have been historically under-represented in public policy making. Individuals and citizens’
groups without access to GIS may find it difficult to gain entry into a public policy-making
process that relies on GIS data and difficult to challenge policies that were created through the
use of ‘expert’ GIS systems. Advocates of PPGIS claim that GIS technology does not
adequately represent many societal groups. Some researchers and practitioners instead seek to
develop alternative GIS systems (called GIS/2) that are more adaptable to input from citizens
and non-official sources (Obermeyer, 1998b). In this model, the role of participants in
E-COMMUNITY PARTICIPATION 69
creating and evaluating data is primary. GIS/2 systems seek to accommodate an equitable
representation of diverse views (Aitken and Michel, 1995; Rundstrom, 1995; Curry, 1995;
Obermeyer and Pinto, 1994; Obermeyer, 1995; Al-Kodmany, 2000).
This chapter also follows the framework first developed by Lynch (1960), Nasar (1998),
and Sanoff (1991). These scholars advocate users’ need-driven design and planning or a user-
oriented approach. They argue that current practices do not adequately address users’ needs
and preferences. The research described here is propelled by the belief that public needs are
underrepresented in today’s planning and design practices and that Web technology offers a
medium of communication that can potentially address this problem. The goal is to determine
a strategy that allows residents to voice their views on existing conditions or on proposed plans
and development schemes and to share these views amongst themselves and with community
organizations and authorities. While it has become common for local governments,
community planners and community agencies to utilize the Web to offer information to
community members, this project takes the next step by using the Web as a medium for two-
way communication; the tool allows people to become both receivers and providers of
information. This project lays out some of the potential contributions and limitations related to
enabling two-way communication and participation over the Web.
The University of Illinois at Chicago, where this project originated, aims at building
mutually beneficial relationships among and between faculty, scholars, students, community
leaders, institutions, and researchers. It promotes engaged urban research that relies on
collaboration and partnership between the university and individuals and organizations in the
public and private sectors. The communication among the partner groups is a model for the
type of two-way communication that we strive for with Web-based technology in planning.
Two-way interaction is critical for supporting the university-community partnership. In two-
way communication any partner may originate and/or receive communication, and as a result,
a cycle of communication, or a flow of knowledge is generated. Two-way communication
enables each partner to become both a consumer and a producer of knowledge and opens the
possibility for sharing that knowledge with neighbouring communities throughout Chicago, the
region and the world.
The project described in this chapter consists of a number of prototypical Web sites that
represent a variety of attempts to expand upon present Web-based and GIS technology to a
new kind of interactive mapping. Each application we have developed is intended to provide a
structured process to gather information from multiple participants and then to present these
views pictorially. The goal is to help the University and communities think more effectively as
a group focused on a common endeavor without losing their individuality, and to help
Community Based Organizations (CBO)s manage the complexity of their ideas without
trivializing them or losing detail.
In developing the one-way, two-way and three-way spatial communications, we address
several challenges in creating Web-based participatory maps. One challenge has been to create
a way to access user spatial feedback without using a GIS system. Since GIS software such as
ArcIMS, demands long download times and substantial bandwidth, and its interface design is
not intuitive, in our view, we have developed interactive websites using other technologies.
Other challenges involve efforts to replicate traditional, paper-based participation methods
on the computer. For instance, we have experimented with various ways of representing and
navigating a large geographic area on a small computer screen. Unlike paper maps, computer
screens limit the amount of geographic area that can be viewed at one time. Scrolling across a
large map online is usually impossible because of the slow speed and large amount of computer
memory that is required to reload the images. We have developed two interfaces that present
alternative ways of presenting a map of a large community area without excess complication or
computer resource use.
Another challenge is developing a flexible, accurate way for users to select and comment on
a specific neighbourhood, block, or building. In one Web interface design, we used pre-
defined square units on a map (using a grid system) so that it would be possible to easily
analyse the location selection data from a large group of participants. However, we
subsequently found that this method could excessively limit user choices and so we developed
a new interface that features freeform drawing as the method of selection. The development of
Web-based freehand drawing tools, where users can use an online ‘pen’ to select their own
locations, is a particularly exciting innovation. In both of these research areas we have
experimented with the use of aerial photographs, structure base maps and GIS maps of the
community as the background upon which users can select locations of concern. After
presenting several options for navigation and selection, we discuss one interface where we
attempted to incorporate the best features into one ‘complete’ or integrated interface. It will
be most helpful to read this chapter in front of a computer in order to try out the interfaces. It
will be most helpful to read this chapter in front of a computer in order to try out the
interfaces. The Internet Explorer browser is recommended for best results.
6.2
THEORETICAL BACKGROUND: IMAGEABILITY AND PUBLIC
PARTICIPATION IN PLANNING
The theoretical foundations for these Web interface designs can be found in the work of Lynch
(1960) and Nasar (1998), who argue for the importance of discovering how city design affects
citizens. By studying what kinds of evaluative images residents have of their community,
planners, researchers and community leaders can derive valuable information about how to
improve the physical form of their communities.
In his seminal book, The Image of the City (1960), Lynch uses the concept of ‘imageability’ as
a theoretical framework for studying cognitive maps, urban form, and the spatial relationships
of cities. Imageability “is that quality in a physical object which gives it a high probability of
evoking a strong image in any given observer. It is that shape, colour, or arrangement that
facilitates the making of vividly identified, powerfully structured, highly useful mental images
of the environment. It might also be called legibility” (Lynch, 1960, 9). If most people like the
imageable elements of a city, then it will probably convey a positive evaluative image. If they
dislike them, the city will convey a negative evaluative image, suggesting a need for changes in
the city’s appearance. This aspect of city image is what Nasar calls the likability of the
cityscape. Likability refers to the probability that an environment will evoke a positive
evaluative response among the groups of people experiencing it. Inhabitants of a city with a
good evaluative image find pleasure in the appearance of its memorable and visible parts.
The city landscape may be a source of pleasure and delight to its residents and visitors, and
can potentially counteract the stresses of daily life. Thus, the development of a city’s form
should be guided by a visual plan that is concerned with visual form on the urban scale (Lynch,
1960). However, to devise such a plan, we need to know how the public evaluates the
cityscape and what meanings they find in it.
Nasar suggests that it is possible to learn the public’s preferences by empirically measuring
them. Just as we weigh objects to find how light or heavy they are, Nasar says that we can
measure preferences to determine the degree to which people like or dislike various areas of a
city. Nasar developed and implemented a method of surveying residents (using traditional
phone surveys and manual map-making) to determine which areas they like and dislike in their
community, with the goal of creating a single ‘evaluative image’ of the community that could
guide future design and development.
In the Web-based mapping projects described in this chapter, we have adopted and
advanced Nasar’s original method by using the Web and GIS to survey and map resident
preferences. Most of the interfaces we have developed are based on the concept of surveying a
group of community residents to discover their preferences and opinions about the
imageability of their community. GIS is in the background, in the sense that it functions as the
method for analysing the data that is collected via the Web. The Web is used as the primary
platform for collecting spatial information.
6.3
CURRENT WORK ON PUBLIC PARTICIPATION AND THE
INTERNET
The Internet has already proven to be valuable on its own as a low-cost mode of
communication for participatory planning though Web sites, email, surveys and on-line
conferencing (Craig 1998, Al-Kodmany, 2000). The Neighbourhood Knowledge Los Angeles
project (http://www.nkla.sppsr.ucla.edu) provides a strong example of how the Internet is
being used as a communication tool for empowering the public. The project, a collaboration of
the municipality of Los Angeles and UCLA, knits together 6 municipal databases and
inspection records, looks for indicators of urban decay, and plots the information on city maps
posted on its Web site. The project utilizes off-the-shelf software for database management
(Microsoft Access and Internet Information Server) and for Internet mapping (ESRI’s
MapObjects and Internet Map Server). Their aim is to provide public access to government
records and electronic mapping through home computers and at ‘touch-screen’ information
kiosks.
The Internet is now able to support interactive programs in a manner similar to stand-alone
GIS and stand-alone hypermedia systems. Peng (1999) concludes that the speed of
technological development provides an opportunity to expand GIS technology and spatial
information to the general public. While the technology is not quite there yet, “it is likely that
we will see an increasing number of distributed GIS with multimedia components which are
organized around a spatial data infrastructure and delivered through wide-area networks such
as the WWW” (Shiffer 1998 p. 731). Krygier notes that the WWW has great potential for
implementing Public Participation Visualization (PPV) and Public Participation GIS (PPGIS)
(1999). Recent developments in Web-based programming languages are making highly
interactive advanced GIS applications available to anyone with a modem and Internet browser.
Even novices can access geographic information, amend and add information, and interactively
explore ‘what if’ scenarios.
The East St. Louis Action Research Project (ESLARP) is one of many examples of
community organizations utilizing this new technology. ESLARP (http://www.imlab.
uiuc.edu/eslarp/) is a reciprocal learning effort between members of the East St. Louis
community and students and faculty at the University of Illinois. Their primary goal is to help
community-based development organizations increase their planning, design and development
capabilities while educating planning and design students. One result of the partnership is
EGRETS, a geographic information retrieval system designed to be used by community
residents. This allows the public to interact with GIS data through a web browser without ever
owning GIS software on their machines. In EGRETS, residents can either search for maps
already made or create their own maps.
The Internet as a medium of communication will be increasingly utilized in all aspects of
planning. It is valuable on its own as a low-cost mode of communication for participatory
planning but it becomes particularly powerful when it is used to distribute and disseminate
other visualization technologies. While there is great excitement about future possibilities for
Internet-based public participation, concerns generally centre around access to the technology.
First, access must be ensured in terms of making sure the pool of participants has Internet
access so that there will be wide representation in public participation forums, and second, in
terms of creating a critical mass of users of a particular Website to sustain meaningful
interactions.
Shiffer and others at MIT have researched a variety of ways of using the Web for urban
planning and design. They have explored how emerging information technologies and Web
technologies can improve the processing and communication of planning-related information
in metropolitan planning organisations. They provide case studies illustrating how to deliver
spatially-referenced multimedia material for site planning and reviews using projects in
Washington DC, the South Boston Seaport and Boston’s waterfront development. Such
multimedia interfaces, coupled with the accessibility of the Web, have the possibility of
opening up a new paradigm within urban design, one which helps to communicate ideas and
developments to other agencies and the general public (Shiffer, 1995).
As Shiffer’s work has demonstrated, one result of this move toward digital visualisation of
urban form and distribution of information on the Web is that there are new possibilities for
involving the lay public in design decision-making. As the number of households with Web
access increases, and the demographic profile of Web users diversifies, the potential for using
the Web for public participation planning increases exponentially. Thus, the exploration of which
types of digital tools and interfaces can best engage the public in planning activities is a
promising avenue for research.
A survey is one important tool for public participation on the Web. Citizens can use a Web-
based survey to become information creators, rather than passive recipients of information. This
is an important leap forward in using the Web as a communication tool. Most of these
applications utilise simple feedback forms where users type in comments in response to
questions and then click ‘submit’ to send their responses to the Web server. There are fewer
examples of Web-based surveys that utilise graphics, maps or other kinds of visualisation to
inquire about the public’s locational preferences. One example is the Landscape Scenic
Preference survey developed by the Macaulay Land Use Research Institute, which aims to
quantify the landscape preferences of the general public. Participants are shown pairs of
landscape images and asked to choose which they prefer (http://www.mluri.sari.ac.uk:80/
~mi550/landscape/). The Same survey was conducted in a traditional manner where people
were shown photocopied photographs and then asked to give each landscape an evaluative
score. Early indications have shown that the results from the paper-based questionnaire were
not significantly different from the on-line version. The researchers identified areas that
needed improvement but the initial reaction was that the on-line survey worked well.
In another example, Kingston (1998) has developed several projects that use the Web to
facilitate public input on several environmental problems in Britain (http://
www.ccg.leeds.ac.uk/democracy/). In one project, a Web-based decision-making
environment was developed which allows the public to model a number of possible planting
scenarios in locating areas for regeneration of native woodland (Kingston, 1998).
The drawbacks of using the Web for communication between planners, designers and the
public centre around the broad issue of access. First, though access to the Web is increasing, it
is still difficult to draw a random, representative sample since Web users do not yet accurately
reflect the real demographic makeup of the general public. Also, users must be quite motivated
to log-on and find the Web site in order to participate. This fact underlies a concern about
creating a critical mass of users to sustain meaningful Web interactions. This prompts the
question of what factors are needed to achieve a critical level of activity (Shiffer, 1995). The
examples given above suggest that Web-based surveys need to be further refined but have now
become quite feasible as public planning tools.
6.4
BACKGROUND TO OUR PROJECTS
6.4.1
The University of Illinois Chicago
In 1993 the University of Illinois at Chicago (UIC) established the Great Cities Institute
(http://www.uic.edu/cuppa/gci/about/index.htm) to respond to urban problems facing
American cities. ‘Great Cities’ refers to the university’s commitment to use its teaching,
research, and service programs to improve the quality of life in metropolitan Chicago. Under
the Great Cities program, UIC has worked with mainly Chicago communities on
approximately 220 different projects and programs.
One of the newest urban outreach initiatives under Great Cities is the Great Cities Urban
Data Visualisation program (GCUDV), established in 1998. GCUDV’s mission is to explore
how advanced visualisation capabilities can be used in community planning to create innovative
computer applications and to test them in actual urban projects (http://www.uic.edu/cuppa/
udv/). Researchers in the GCUDV program work to link databases, GIS, and 3D visualisation
into a medium suited for urban and regional planning and policy. Projects focus on social and
ecological data as well as urban image visualisation. The program is staffed by a
multidisciplinary team that includes faculty, students and recent graduates from UIC’s
Electronic Visualisation Laboratory and from several other disciplines such as Architecture,
Art History, Information Technology, and Urban Planning.
6.4.2
The Communities
Our research focuses on two Chicago communities: Pilsen (Lower West Side) and North
Lawndale. Pilsen, the first of the three communities we work with, has long served as a port
of entry for many of Chicago’s immigrants. By the end of the 19th century,
rapid industrialisation and urbanisation had transformed the largely Bohemian (Czech and
German) working-class neighbourhood into a national centre of labour activism. This activity
drew Poles, Croatians, Lithuanians, Italians and other immigrants to the community. After a
few decades, these residents moved on, giving way to newer Mexican immigrants. In 1990, 88
per cent of the area population was Hispanic, and Pilsen had the second highest concentration
of Mexicans of all community areas in the city. Interestingly, each of the successive groups has,
in turn, left its unique imprint on the architecture of the community, creating a cultural
mosaic in the built form of Pilsen.
Despite being a welcoming home for new immigrants, Pilsen presently struggles to retain
residents; many people start out in the community but eventually move on to other
neighbourhoods or the suburbs as they assimilate into American culture and become financially
secure. The number of housing units in the community has declined and little new residential
construction has taken place since 1930. In 1980, 27 per cent of the housing units were
overcrowded, having more than one person per room (CFBC, 1990). Over the years, several
strong community organisations have formed to help Pilsen’s disenfranchised residents address
issues like housing and economic development.
The second community, North Lawndale, is located northwest of the university, four miles
from downtown and near the United Centre sports stadium. This community, like Pilsen,
experienced successive waves of immigrants, beginning with Bohemians and Polish, and followed
by a wave of Russian Jews. However, the population decreased by approximately 30,000
people per decade between 1960 and 1980. The 1990 total population was 47,000, a decline of
23 per cent from 1980’s total of 61,000. Ninety-six per cent of its 1990 population was
African American (CFBC, 1990).
The number of housing units in North Lawndale continues to decline, causing a housing
shortage. The community has experienced a net loss of almost half of its housing units since
1960. North Lawndale has an abundance of redbrick and graystone homes, mostly built in the
1910s and 1920s, and though the community has suffered from urban blight, the housing stock
has proved to be surprisingly solid for renovation. Recently, the area has begun to see some
revival, as one development corporation has turned the old Sears catalog complex into homes
at Homan Square, which brings infrastructure and housing improvements to 55 acres. New
businesses and families are moving into Lawndale and the economy is starting to turn around.
However, the community still has many vacant lots and abandoned buildings.
6.5
METHODOLOGICAL APPROACH
Most of the Web-based GIS sites that are currently available on the Web provide one-way
information delivery. Although providing proper spatial information is an important aspect of
participatory planning; however, it lacks engaging the public in the planning process. In order
to become a robust tool for use in participatory planning, Web-based GIS should go one step
further: it should be a medium for two-way spatially based information exchange. Interactive
Web-based GIS could become a critical and widely used tool to gain important feedback in
community planning.
The general intent of most of the interfaces described below is to discover how residents of
Pilsen evaluate the appearance of their community by asking what particular places they like
and dislike. The survey instructions, while not included in all of the interface prototypes,
emphasise the issue of community appearance and encourage respondents to think of which places
they find to be pleasant and which places they find to be blighted and unattractive. The results
are intended to be used as guidelines for decisions on community appearance and visual form.
The map tools have been developed to work on average computers and do not require the loading
of geographic information systems (GIS) or Internet map-server (IMS) applications. Local
residents simply interact with Web-maps by clicking map locations and typing text to give us
their opinions.
Conceptually, the prototypes discussed here are organised into three categories: one-way,
two-way, and three-way spatial communication. We present the development of the tool from
simple Web sites that display spatial information to more complex and interactive sites that
actually allow participants to draw and comment on maps of their community. The increasing
levels of sophistication can be described as follows: 1) simply displaying a map (such as
disseminating the results of a spatial analysis) 2) viewing a map with different types of data
included, and some ability to navigate the map, 3) maps that include some GIS functions so
that users can retrieve attribute data along with the map and 4) map surveys that allow users
provide feedback directly on the maps (Kim, 2001). The degree of user interaction is
associated with the type of tools and software that are used, such as HTML, plug-ins, Java
applets, Java Servlets, and Internet Map Server technologies.
6.5.1
One-way Spatial Communication
One-way spatial communication describes a paradigm by which information is presented to an
audience without any possibility for interactive response. There are several approaches to one-
way communication that help to make a large geographic area navigable online, including map
view, map overlay, nesting maps with increasing resolution and the use of thumbnail
photographs.
6.5.1.1
Viewing Maps Online
Viewing maps online is a basic function and the simplest to deliver. It is a one-way
communication method in which information is transferred in only one pre-assigned direction.
Examples of maps that we used for the purpose of neighbourhood planning include site and
neighbourhood plans, survey maps, topographical maps, census block and tract maps,
transportation and land use maps, utilities map, and aerial photographs.
By using simple HTML or Web authoring tools, we are able to publish these maps in a one-
way static format. A very minimal level of interaction is provided by the Web browser, which
allows scrolling right-left and up-down to view the different portions of the map. Maps are
published directly on the Web page after preparing them in an HTML compatible image format,
such as GIF or JPEG. These maps can be snapshots from a from digital map, exported using
electronic software or scanned in from a paper map. The Web Site, http://www.uic.edu/
~kheir/community_maps/main.html, shows examples of GIS maps of two neighbouring
communities, Pilsen and North Lawndale. These maps include locational, ethnicity,
population, median income and median rent information for these communities. These maps
were produced using U.S. HUD (Housing Urban Development) Community 2020 Maptitude
GIS software. Later, maps were exported as GIF image from the application and embedded
and hyperlinked in an HTML document. The client and server architecture adopted here is
exactly same as the basic framework of the client and server model. When the client requests
maps embedded in regular HTML documents, the Web server pass them to the browser in the
client computer (Figure 6.1).
Figure 6.1 Viewing maps on the Web requires a simple client-server architecture.
6.5.1.2
Map Overlay
Overlay analysis is a classic technique that has been discussed extensively by numerous
geographers and planners (McHarg 1963, Hopkins 1975, and Edwards, 1984). It is an excellent
method for analysing spatial and component relationships. Various facets of a developing
concept can be synthesized using successive layers. Physical and non-physical considerations
can be fused and viewed graphically, as this procedure illuminates more complex relationships.
The map layers act as transparent acetate sheets that, together, form a composite image. By
overlaying individual images, in which each details different pieces of information for the same
geographic location, one can better understand complex relationships between different types
of spatial data within an area. That level of analysis may only be available by examining the
overlaps between the various images. Moreover, computerized image manipulation on the
Web can be done in ways that have never been possible by traditional graphic means.
By using Java Applet layers, we simulate the GIS overlay function on the Web. The user can
interactively turn layers on and off to visualize spatial information and relationships. With this
method, there is no need to use ESRI GIS ArcIMS. The Web enables users to display and
interact with thematic layers intuitively. The Web site, http://www.uic.edu/~kheir/layer/
p1/, provides examples of using a Java applet for overlay function. Layers include the
background of the Lakefront community as well as maps showing census tract, median rent,
median value, and median income information.
6.5.1.3
Nested Maps
This Web Survey (http://g015.cuppa.uic.edu/gridFeedbak/final/index.html) attempts to
evaluate participants’ likes and dislikes for the four square-mile community of Pilsen. In earlier
versions of this Web map/survey tool, we had the problem of not being able to zoom on the
map. We wanted to publish the map at the highest possible resolution, yet this took up too much
space on the screen; users had to scroll up and down and right and left to see the map. In this
new interface, where we are trying to cover a much larger geographic area, this was a serious
problem. We found a solution in the concept of a geographic hierarchy with three zooming
levels. This Web interface utilises a grid and an aerial map of the community. Users are able to
navigate the map by utilising three zooming levels, each with increasing resolution.
When the user selects one of the 16 squares on the initial map (the first level), the program
zooms in, and the selected square appears as a new full-screen aerial map that is again broken
down into a grid of 16 squares (the second level). When one of the 16 cells is selected from
the second level, an aerial close-up of the selected area (the third level) then loads into the
window. The actual geographic size of the final selection is four city blocks. The highest
resolution is on this third level with six inches per pixel. At all times, the user has the ability to
zoom in and out between the three levels. At the third level, users can see the fine details and
distinguish buildings and streets. In a later stage, we further developed the above Web site to
allow two-way communication by adding a selection and textbox tool for participants to type
in comments about the location.
6.5.1.4
Thumbnails
Another option for navigating a large community area is to show a number of very small
‘thumbnail’ photos, or tiles, which, upon clicking, bring up a larger map of the area. A
demonstration of this technique can be found at http://g015.cuppa.uic.edu/feng/project.html
and http://g015.cuppa.uic.edu/ramki/image.html. Like the other interfaces, this one is
designed using an aerial map of the Pilsen community. Cells (miniature photographs) are
displayed across the top of the screen with some street names visible to help orient
participants. The instructions read, “Select one of the thumbnail photos by clicking on it. In the text
box below the map, describe any physical changes to the community that have occurred since the map
photo was taken 10 years ago” This is only one possible use of thumbnails; other uses could also
be envisioned. Also, more tiles could be used to cover a greater area of the community
(however, this will slow down the interface).
6.5.1.5
Zooming
The two prototypes described above utilise different methods of ‘zooming’ on a Web-based
map, and each has pros and cons. The first method, using nested maps, uses Java. In the Java
interface we write the scripts for zooming. There are four Java layers of different scales
(zoom). Once you click the zooming function on the map you call one layer after the other,
simulating a ‘zoom’ on the photograph. When the user clicks to get the next map (zoomed
version) the new map is called from the remote server. The disadvantage of this method is that
it can be extremely slow, particularly for users with modem connections. The advantages are
that Java provides great flexibility in creating and adding new features.
The second method (the thumbnail interface) uses Flash Shockwave software. Flash is Web
design tool to create animated Web pages. The strength of this tool is that it handles images in
vector format. Thus, users do not lose resolution no matter how closely they zoom in on the
map. This is a very fast method of zooming since the function is available within the program
itself. Zooming is done via a small movie file that makes the zooming easy and in real time.
Once a user enters the Web site, a message appears asking the user to install Flash software. If
the user clicks yes, it loads in just a few seconds. The Web site contains a small movie file that
automatically gets loaded on the user’s hard drive and then the zooming is done from the
user’s own machine and it is very fast (Figure 6.2). While speed is a significant advantage of
this method, the present Flash software package does not allow the incorporation of features
from other programs. Thus, we cannot mesh this site with other functions available with the
Java programming language (Figure 6.3). For instance, at this time there is no way to include a
selection feature so that users can select a building within the photo.
Figure 6.2 A map constructed using GIS and published on the Web with Macromedia Flash
Software to provide navigational capabilities.
Figure 6.3 The relationship between the server, the client computer, and plug-ins in a zooming
environment.
Figure 6.4 ESRI ArcView IMS Information System for North Lawndale community. The view
shows the manufacturers and minority business thematic layers.
6.5.1.6
Map and Data Retrieval from GIS
Map and data retrieval refers to enabling users not only to navigate maps but also to examine
attribute data and perform spatial queries. The purpose is to bring simple GIS functionality on
the Web without the need for GIS software at the end user machine. However, this technology
continues to provide limited spatial function on the Web. The user cannot perform advanced
analytical functions but can perform simple ones such as turn on or off layer, set active layer,
examine attribute data related to map feature and basic spatial queries. Commercial Internet
mapping software from major vendors, such as ESRI ArcIMS, Intergraph GeoMedia Web
Map, AutoDesk MapGuide, and MapInfo MapXtreme, are appropriate for performing simple
GIS functions online. The following Web site (http://e036.cuppa.uic.edu/arcviewims/north-
lawndale/index.html) shows an example of using ArcView IMS software for delivering spatial
information to the North Lawndale community (Figure 6.4).
Figure 6.5 The interaction between client, server, interpreter, and GIS software for map and data
retrieval.
Map retrieval enables users to retrieve maps and related attribute data from the server at
runtime. The Web server, interpreter, and GIS software work together for this. Whenever
users change the view or perform spatial query, the Web server pass it to the GIS software via
the interpreter. The GIS software performs the tasks and passes the outcomes, via the
interpreter again, to the Web server. Then, Web server sends final outcome in HTML format
to the client. A key part in this framework is GIS software. What implements user request
spatially is the GIS software. The Web server only receives user requests and sends final
outputs. However, since the GIS software cannot directly communicate with the Web server,
a middleware is required. The middleware acts as an interpreter for the Web server and GIS
software. It passes parameters received by Web server to the GIS software and the results of
the task done by the GIS software to the Web server. The Common Gateway Interface (CGI)
has often been used for this, but Java Servlet technology is rapidly replacing CGI (Figure 6.5).
6.5.2
Two-way Spatial Communication
Getting user feedback on the Web is becoming increasing popular. Sending text input or
polling choices is widely used to get users’ opinions. However, when working with spatial
information, it is important to be able to visualize such feedback. In two-way communication,
we want to allow users to annotate maps and to delineate their concerns on maps. One option
is to allow users to draw various features, such as points, lines, circles, and polygons on the
map. It is also possible to add text along with drawn features. When users send their input to
the server, the server detects the x, y coordinates of such drawings and saves them. This makes
it possible to define a particular location. Users can indicate exactly which location they are
talking about.
Such a spatial feedback system may be designed with or without GIS software. However, at
the time of this writing, the only solution that receives spatial user feedback on the map is GIS
Figure 6.6 ESRI ArcIMS Property Information System for North Lawndale Community. In the
middle is a Web-site window that is hyperlinked to the system. It provides additional information
about particular properties, such as historic sites in the community.
Figure 6.7 Two-way spatial communication. ESRI ArcIMS Property Information System for North
Lawndale Community. System enables participants to type comments on map and submit them to
the University server.
software, specifically, ArcIMS. The workflow on the server side is the same as in the data
retrieval method. However, the function of drawing on the map requires Java applets and Java
and ArcIMS plug-ins in the client side. The server and client must work together heavily for
this method. The following Web page (http://e036.cuppa.uic.edu/arcims/north-lawndale)
shows an example of two-way communication using ArcIMS (Figures 6.6 & 6.7).
However, ArcIMS has substantial drawbacks, particularly at the client side. It requires a
thick client and a heavy server. A ‘thin’ client means a client computer with just a Web
browser, while a ‘thick’ client implies a computer with a Web browser with other add-ins,
Figure 6.8 Complicated client-server configuration required for ESRI ArcIMS.
such as Java applets, ActiveX controls, and plug-ins to evoke special effects. The thick client
needs to download such add-ins at runtime or beforehand. A ‘light’ server is a Web server
computer having only HTML documents and related files, but a ‘heavy’ server has other
components, such as database applications and GIS software working together with the Web
server. For ArcIMS, the server requires heavy processing with GIS software and a Java Servlet,
and the client also needs a Java plug-in installed beforehand and Java applets at runtime to enable
the drawing functions (Figure 6.8).
In ArcIMS, the download time for data, map and attribute data retrieval is quite long and it
requires substantial bandwidth. Most importantly, the interface design and interaction
behaviour is not intuitive. Because of these drawbacks of ArcIMS, we have invested research in
developing alternative methods for two-way communication. We divide them into four
categories: grid-based, freehand, a combination of grid and freehand or a ‘complete interface’,
and a compositional method.
6.5.2.1
Grid Based
We have created several different prototypes that utilise a grid as the underlying selection
structure behind a given map. Below we describe three different interfaces that utilise the grid-
based selection method.
Our earliest project (http://g015.cuppa.uic.edu/gridfeedbak/xarial18St/arial_18st.html)
was a survey consisting of one exercise titled, Urban Likability and Dislikability (ULD) for
Pilsen. Simply speaking, participants logged on to the project Web site where they could view
a high-resolution aerial photograph of Pilsen with a grid overlaid on top of it. We limited the
geographic area to the vicinity of the 18th Street commercial district, since this area was the
primary focus of revitalisation efforts. Participants were asked to point out areas on the map that
they liked and disliked and to provide the reasons for their responses. Each square of the grid
was identified by its centroid (the centre of the square). This centroid was coded as the actual
longitude and latitude of the centre of the square.
Participants were asked to identify the areas of their community that they most liked and
disliked by clicking on the appropriate square of the grid. The only visual assistance on the map
was the name of streets. They were to use a GREEN pointer to indicate liked areas and a RED
pointer to indicate disliked areas. By clicking on one of two radio buttons located on the side
of the map, they could ‘load’ the pointer with either colour. When participants clicked on the
square, a small window with a question mark appeared, asking them to state their reasons for
liking or disliking that area of the community. When finished, the participant clicked on a
button labelled ‘submit’ and their input was transferred to the UIC server. Each of the
participants’ selections could then be stored in the Web access logs for analysis and feedback into
the planning process.
Since the server was linked to an Oracle database and a GIS application, we had the
capability of taking all the points that were selected by the participants, sorting them by
longitude and latitude, and plotting them on a map automatically. The Oracle database could
also group the associated comments. In this manner, a community-input database was created
that contained the range of views about areas liked and disliked with the associated reasons.
We then created a number of GIS choroplethic maps to illustrate the intensity of likes and
dislikes (urban likability and urban dislikability). We used dots to represent intensity: the
number of dots in each cell of the map was proportionate to the number of times that area was
selected by the residents in the survey exercise. In addition, these GIS maps were interactive;
clicking on an area (or cell) of the map opened a window of text that listed the residents’
reasons for liking or disliking that area. Since the maps provided written evaluations for each
point, they were extremely useful in supplying specific directions for improvement and could
easily be incorporated into the next stages of the Pilsen community planning process.
In a later prototype (http://g015.cuppa.uic.edu/gridFeedbak/final/index.html)
(Figure 6.9) we used nested maps. We also used a grid but it advanced the initial project by
dealing with a larger geographic scale. Not only does the user navigate using the grid, but he or
she also selects a square of the grid to comment upon. As mentioned above, the primary
disadvantage of this method is that users cannot select the specific buildings or combination of
buildings that they wish to comment upon, since the size of the final selection square is
predefined. The advantages are clearly evident in the above example where the uniform
selection areas allowed us to easily create sort and analyse the users’ feedback.
In evaluating the grid as a selection mechanism, it has clear advantages and disadvantages. As
in the above example, the grid enables very fast analysis and compilation of spatial data that can
be easily compared among participants. On the other hand, we found that users did not have
enough discretion in selecting the particular areas of the community and buildings that they
wished to comment on. In addition, they were constrained to the square shape of the grid:
even if they only wanted to comment on one building in a corner of this large square, they had
to select the entire square.
6.5.2.2
Freehand
This project introduces an entirely new technology for enabling user feedback using Web-
based maps. In this new prototype (http://g015.cuppa.uic.edu/zhaoxia/blankcanvas/
Figure 6.9 Two-way spatial communication: the nested map method and the three zooming levels.
sketch html) (Figure 6.10), participants can go to the online survey Web site and use a
drawing tool to select the areas of the community that they wish to comment upon; their
locational choices are not limited by the pre-defined geographic areas of the square grid. On
the initial screen, the participants view a structure base map of the community along with two
buttons labelled ‘Click to select area with drawing tool’ and ‘Click to type in comments’.
When a participant clicks on the drawing button, the cursor turns into an arrow and upon
pressing and holding down the left button of the mouse, it starts to draw. The user may draw
any shape on the map and when the mouse is released, the lines close on themselves to form a
polygon. If the user does not draw an enclosed shape, the program approximates the line that
closes the shape into a polygon.
Figure 6.10 Two-way spatial communication: freehand sketch on the Web. Colour gradation
indicates intensity of concern.
Once the participant draws a shape and releases the mouse button, the shape is immediately
filled with a light colour. As the participant continues to select additional areas with the
drawing tool, other shapes are filled in as well. When areas overlap, the program indicates this
by increasing the density of colour in these areas. Areas that participants delineate most
frequently will be the darkest, while the areas chosen the least would be lightest. This works
through the placement of a very fine grid at the back of the map. When a participant delineates
an area, cells underneath will be activated and will create one tone. As participants highlight
additional areas, the overlapping areas would be activated twice, generating ‘double’ tones and
hence increasing the density of the colour. This shading technology is beneficial when compiling
multiple users’ responses, so that the darker shades would represent the most frequently
chosen locations.
Once the user has drawn and selected an area with the drawing tool, the participant can
then click on the text button to type in comments about the delineated area. In order to avoid
clutter on the map generated by multiple annotations when a participant selects several areas,
we added a check box that would allow a participant to turn the text boxes on or off. The
participant may add a text box for each section of the community that is selected. It is possible
to make two separate interfaces: one for drawing and writing about liked locations and one for
disliked locations. This may assist in sorting positive and negative opinions in the database.
Another possibility would be to simply use one interface and allow participants to record all
opinions, positive or negative.
The most important feature of this project is that we replace the Java applet grid with
freehand sketching capabilities using Java to create the Web interfaces so that the user is not
restricted to pre-determined geographic areas within a grid. This new capability of doing
freehand sketching on the Web may have numerous applications. People can make their own
maps and then share these maps with other participants. Another possibility is that the map
could be a photographic image and we could ask participants to comment on the scene by drawing
directly on the picture, indicating elements that they like or dislike. Also, we hypothesise that
community residents will appreciate this feature since it enhances the ability of the computer
to mimic traditional public participation tools that people generally enjoy, such as drawing and
writing comments on a paper map.
6.5.2.3
Toward a Complete Interactive Web Map Design
The prototypes discussed above each represent a particular feature; some are designed to
sketch, others to type in comments. Some are grid based and others are based on freeform
drawing. Some have zooming others do not. We are in the process of developing a Web site
(http://g015.cuppa.uic.edu/mapTool3/mt.html) that will combine the best design features of
the interfaces described above. Users will have options to work with or without the grid, to
zoom in and out, to add layers, to type in comments, to sketch, and so on. This has apparent
advantages (more choices) and disadvantages (confusion, technical difficulties to create the
interface and to work out the database). This site is currently undergoing further
development.
6.5.2.4
Compositional Methods
The final challenge we attempt to address using Web maps was how to allow participants to
essentially create their own maps. This prototype (http://g015.cuppa.uic.edu/zx/sketch2/
project3.htm) (Figure 6.11) was developed for a slightly different use than the ones described
above. While the others use maps of the community to help residents share their opinion and
knowledge about the community by selecting locations and typing comments, this prototype
aims at allowing community participants to draw alternate site plans for a specific location,
such as to show a preferred arrangement of houses in a new subdivision. Called the
Collaborative Decision Support System, it utilises the freehand drawing technology described
above. Users actually draw their own boxes (to represent houses, other buildings, or land use
zones) and then move them around into a desired configuration. The purpose of this project is
to allow users to create and compose maps online.
In the current prototype, the boxes are drawn on a structure base map of Pilsen. The
primary buttons available are ‘draw’, ‘move’, and ‘copy’. Users may correct their work with
‘delete’ and ‘clear’ buttons. When the drawing is complete, the participant clicks ‘submit’ to
send the drawing to the server database. The user is then able to view all the drawings
previously submitted by other users. While at present the prototype simply demonstrates the
capability of creating and moving the boxes, there are many possible applications for the tool.
It could be used for planning work on a small scale; for instance, the survey site could open
with a site map and a description of the particular site along with a set of development plans. Users
would be given a set of instructions, such as asking them how they would prefer to site the
Figure 6.11 Two-way spatial communication: compositional method.
structures in a development that calls for 8 single family homes, 16 town homes, a community
centre, and a playground. Users would then create the boxes and move them around on the
screen until they developed a satisfactory plan. Some of the fixed details of the site could
already be indicated on the site map, such as the direction of traffic, existing driveways, etc.
Alternatively, the proposed issue could be planning work at a larger scale, such as a land use
and zoning plan. In this case, participants could help deal with the issue of urban sprawl by
drawing shapes of different colours to indicate their preferences for placement of various types
of land use zones. As the Web site is further developed, it will need to incorporate a legend to
explain what each drawn shape is representing so that various users could interpret one
another’s sketch plans.
6.5.3
Three-way Spatial Communication
Three-way communication enables users not only to view and input data but also to view input
data of all participants. The user is able to tap into the database and view all entries. The user
may receive a return map in real-time showing the accumulative responses of all participants.
In the following example, (http://g015.cuppa.uic.edu/gridFeedbak/xdbgrid 18St/
grid_18St.html) we created a ‘real-time feedback’ interactive map. Participants could view a
structure base map of Pilsen and choose areas of ‘like’ and ‘dislike’. They could also type in
comments about the selected areas. Once their selections were submitted, they could
immediately receive a return map showing the accumulative evaluative image of the examined
area. This worked in the following way: the update servlet on the database is contacted and the
data sent to the servlet translates the visual information into numeric, string data that is stored
into the database. Once the database is updated the data servlet is called which queries the
database for all the relevant information and then translates the data into visual information and
sends it back to the user. The visual information includes the cumulative ‘like’ and ‘dislike’
opinions of the area, represented by increasing intensity of colour along with the user comments
for each block in the grid. The data obtained from different surveys can be co-related as it is
stored in the database, sorted by latitudes and longitude. The grid was a good structure that
was easy to code and to be read by the Oracle database that created the return map.
Several of the other Web design interfaces can easily be upgraded to function as three-way
communication mediums. For instance, in the Collaborative Decision Support System
described above, we have added a feature whereby when users submit their design, they are
then able to view all the drawings previously submitted by other users.
6.6
IMPLICATIONS FOR PARTICIPATORY PLANNING
Public participation in a community planning process is important in democratic society;
however it is a complex undertaking. Our research explores state-of-the-art information
technology (IT) to facilitate the process. Web-based mapping opens up the potential for
involving a wider range of people by bridging time and space. At the same, IT has the potential
to automatically collate participants’ responses and ideas in a cohesive manner as described in
the previous examples. Traditional public participation in planning usually relies on same-place
and same-time meetings, which restricts involvement. Web-based mapping can be utilized for
widening and diversifying channels of communication among the public, planners, and
politicians.
Our research has asked the question, how can Web-based GIS aid participatory planning? It
would be useful to place this question in the framework provided by Weidemann and Femers
(1993). They presented a public participation ladder model that arranged the tasks of public
participation in a vertical dimension. The order, from top to bottom, is as follows:
1. public participation in final decision;

2. public participation in assessing risks and recommending solutions;
3. public participation in defining interest, actors and determining agenda;
4. public right to object;
5. informing the public; and
6. public right to know.
Kingston (1998) argues that most Web-based GIS models are confined within the bottom two
rungs of the public participation ladder; the ‘public right to know’ and ‘informing the public’.
Our research aims at expanding the role of Web-based GIS from being limited to the last two
tasks to including as many as all six tasks. We envision that a loop of communication facilitated
by two-way spatial communication may enable the public to ‘object’, to ‘define interests and
agenda’, and to ‘assess risks and recommend solutions’. This could be possible by further
developing Web-based mapping that employs a wider range of increasingly powerful Web
technology.
Present commercial Web-GIS software mainly enables one-way communication. The

method is appropriate to allow the public to access spatial information from remote places
such as home or office. Two-way communication on Web-based maps is intended to not only
provide spatial information about a particular planning problem, but also to provide a forum
for the public and planners to express their perspectives and concerns in a spatial format.
Planners as well as government officials can learn about local knowledge, which is necessary
for sound planning. The three-way communication concept further supports democratic
decision-making by allowing the public to view the opinions of all participants.
The tools we have developed in these prototypical interfaces have the potential to be used
by a variety of agencies for multiple purposes. For example, it has been observed that
transportation planning and local ‘comprehensive’ planning tend to occur separately, resulting
in some cases in policies that work at cross purposes. A partial solution could be that
comprehensive planning and transportation planning leaders could incorporate these kinds of
Web-based tools to both inform the public and also learn about public concerns and views. By
sharing future planning ideas and learning about public concerns, costly mistakes could be
avoided. This kind of communication could be used to address other issues such as urban sprawl,
creating subdivisions, transportation planning, landscaping, and identifying advantageous
options for development and environmental protection.
6.7
CONCLUSION
It is increasingly important to direct research to discovering the most effective methods and
tools for interacting with the public using maps. As there has been a dramatic increase in the
number of people exposed to and using screen-based geographic information products for
general use, two-way communication of spatial information must become more efficient and
more easily comprehended. Researchers in the field have estimated that ‘up to 90% of all
business data has a geographic component’ (ARC News, 1997), while an estimated 80 per cent
of all government information is spatially referenced (Huxhold, 1991). Governments
themselves publish geographic information on-line. For example, the San Diego, California
Police Department publishes data on crimes on a public Web site just 24 hours after the
incidents occur. Local planning agencies use the Web as an adjunct to the traditional public
meeting format. A number of U.S. government agencies, including the U.S. Census Bureau
(http://www.census.gov), the Geological Survey (http://www.usgs.gov), and the National
Cancer Institute (http://web.nci.nih.gov/atlas), provide maps via the Web. Without direct
research into which designs, tools and methods of presenting and receiving spatial information
are most easily used and comprehended by the public, we may be misinformed in estimating
the level of communication that is actually occurring.
In this chapter we have described a variety of digital map designs and tools created for a dual
purpose: to communicate spatial information to average map users, and to allow those users to
navigate and make selections on the maps in order to give feedback into a community planning
process. We have provided examples of solutions to the basic problems of creating interactive
screen-based maps, which include navigating large geographic area maps and making selections
on maps. New Web technology has made it possible to create map-based surveys to receive
feedback from the general public, but it is not yet clear what kind of graphic design
alternatives and digital map designs and tools are the most useful for novice map-readers. This
research attempts to explore different interface designs with the purpose of finding which
combinations of online tools and maps are most productive in soliciting feedback in a
community planning process. In the future, we plan to empirically compare these interfaces to
learn more about how people comprehend screen-based maps vs. paper-based maps. More
research is needed in order to understand the unique challenges and important advantages of
Web-based maps in general, and the usefulness of each graphic design and tool in particular.
Our project has explored how Web-based maps can be advanced beyond mere information
provision to actual two-way interaction with the public. Tools such as freehand drawing may
provide new avenues for people to take a greater role in public decision making. By beginning
to examine the graphic design alternatives possible on the Web, we are taking one step further
toward understanding how people comprehend and utilize screen-based maps.
6.8
REFERENCES
Aitken, S. and Michel, S., 1995, Who contrives the “Real” in GIS? Geographic information,
planning and critical theory. Cartography and Geographic Information Systems, 22 (1).
Al-Kodmany, K., 2000, Extending geographic information systems to meet neighbourhood
planning needs: The case of three Chicago communities. The Journal of Urban and Regional
Information Systems Association, 12 (3).
ARCNews, 1997, Visio Corporation adopts MapObjects. ARCNews, Environmental Systems
Research Institute, Inc. 19 (2), p. 1.
Chicago Fact Book Consortium (CFBC), 1990, Local Community Fact Book Chicago Metropolitan Area.
(Chicago: Academy Chicago Publishers).
Craig, W.J., 1998, The Internet aids community participation in the planning process. Computers,
Curry, M., 1995, Rethinking rights and responsibilities in geographic information systems: Beyond
the power of the image. Cartography and Geographic Information Systems, 22 (1).
Edwards, C., 1984, Overlay Drafting Systems. (New York: McGraw Hill).
Hopkins, L., 1975, Optimum-Seeking Models for Design of Suburban Landuse Plans. Unpublished
Dissertation.
Huxhold, W.E., 1991, An Introduction to Urban Geographic Information Systems. (New York: Oxford
University Press).
Kim, D., 2001, Design and Assessment of Web-based Geographic Information Systems. Masters project,
University of Illinois at Chicago, College of Urban Planning and Public Affairs.
Kingston, R., 1998, Web based GIS for public participation decision making in the UK. A paper
presented at the Empowerment, Marginalisation, and Public Participation GIS Meeting in Santa Barbara,
California October 14–17th, 1998.
Krygier, J., 1999, Public Participation Visualization: Conceptual and Applied Research Issues, http://
www.geog.buffalo.edu/~jkrygier/krygier_html/
Lynch, K., 1960, The Image of the City. (Cambridge: MIT Press).
McHarg, I., 1969, Design with Nature. (New York: Natural History Press).
Nasar, J., 1998, The Evaluative Image of the City. (London: SAGE Publications).
Obermeyer, N., 1995, The hidden GIS technocracy. Cartography and Geographic Information Systems,
22 (1).
Obermeyer, N. and Pinto, J., 1994, Managing Geographic Information Systems. (New York: Guildford
Press).
Peng, Z-R., 1999, An assessment framework for the development of Internet GIS. Environment and
Planning B: Planning and Design 22, pp. 117–132.
Rundstrom, R., 1995, GIS, indigenous peoples and epistemological diversity. Cartography and
Geographic Information Systems, 22 (1).
Sanoff, H. 1991, Visual Research Methods in Design. (New York: Van Nostrand Reinhold).
Shiffer, M., 1995, Interactive multimedia planning support: Moving from stand-alone systems to
the world wide web, Environment and Planning B: Planning and Design 22.
Shiffer, M., 1998, Multimedia GIS for planning support and public discourse. Cartography and
Geographic Information Systems, 25 pp. 89–94.
Weidemann, P. and Susanne, F., 1993, Public participation in waste management decision making:
Analysis and management of conflicts. Journal of Hazardous Materials, 33.
6.8.1
URLs: Examples and References
http://www.mluri.sari.ac.uk:80/-mi550/landscape/, Macaulay Land Use Research
Institute (Web survey using images).
http://www.ccg.leeds.ac.uk/vdmisp/, Kingston’s Web-based decision-making systems in Britain
http://www.uic.edu/cuppa/gci/about/index.htm, University of Illinois at Chicago’s (UIC) Great
Cities Institute
http://www.uic.edu/cuppa/udv/, UIC’s Great Cities Urban Data Visualisation program (GCUDV)
http://www.census.gov, The U.S. Census Bureau
http://www.usgs.gov, The Geological Survey and
http://www.nci.nih.gov/atlas, The National Cancer Institute
6.8.2
URLs: Project Interfaces
http://www.uic.edu/~kheir/layer/p1, Map Overlay.
http://e036.cuppa.uic.edu/arcims/north-lawndale, ArcIMS for North Lawndale.
http://e036.cuppa.uic.edu/arcviewims/north-lawndale/index.html, ArcView IMS for North
Lawndale.
http://g015.cuppa.uic.edu/gridFeedbak/final/index.html, Nested Maps.
http://g015.cuppa.uic.edu/feng/project html, Thumbnails.
http://g015.cuppa.uic.edu/gridfeedbak/xarial18St/arial_18St.html, Grid-Based Map-Survey
Interface.
http://g015.cuppa.uic.edu/gridFeedbak/drawMap3/dm htm, Real-Time Feedback Interactive Map-
Survey.
http://g015.cuppa.uic.edu/zhaoxia/blankcanvas/sketch.html, Freehand Drawing and Selection Tools.
http://g015.cuppa.uic.edu/mapTool3/mt.html, Integrated Interface with Combined Features.
http://g015.cuppa.uic.edu/zx/sketch2/project3.htm, Collaborative Decision Support System
(composing site plans online).
7
A collaborative three dimensional GIS for
London: Phase 1 Woodberry Down
Andrew Hudson-Smith and Steve Evans
7.1
INTRODUCTION
In this chapter we outline the various issues involved in developing a three dimensional
geographic information system (3dGIS) model for a large urban area, such as London. There
are now many techniques which can be used to construct two-dimensional (2d) and three-
dimensional (3d) models of cities and their characteristics, which are useful for various types of
visualisation. Until quite recently the main approach was based on rendering wire frame
models of buildings based on Computer-Aided Architectural Design (CAAD/CAD).
However, with the widespread development of Geographic Information Systems (GIS), there
has been a move from the 2d map to the 3d block model which is based on extruding building
plots, street line data, and basic topography. At the same time there has been a move towards
the distribution of information relating to urban form and its representation via the Internet.
Research at CASA into Internet based urban planning techniques has resulted in techniques
whereby the common CAD based models of urban form can be adapted into a networked
collaborative design and information system. A system which can be queried, explored and
indeed inhabited through standard desktop machines linked to the Internet via standard dial up
connections is currently being piloted.
This chapter addresses two issues, which involve our proposal to construct a 3dGIS for
London, firstly the construction of an Internet distributed three-dimensional model of London
and secondly its integration with web enabled GIS. It explores the current ‘state of the art’ in
3d GIS and compares these with other photogrammetric techniques which produce 3d models.
Through the use of multi-way interaction, it demonstrates how photorealistic models of the built
environment can be distributed and queried online through networked GIS. The paper also
demonstrates how such models can be populated, creating distributed virtual environments
underpinned by a GIS. In short it details methods to communicate, distribute and query built
form for the purpose of Online Planning. We will elaborate our proposed paper in three sections:
spatial data technologies, model construction and dynamic 3dGIS for Online Planning.
7.2
SPATIAL DATA TECHNOLOGIES
For a model to be useful it must tie in with a range of consistent spatial data. We make a
distinction between content and geometric data. Content data is now widely available at different
A COLLABORATIVE THREE DIMENSIONAL GIS FOR LONDON 93
levels from the Census, the Office of National Statistics, the Greater London Authority, the
Valuation Office, and a variety of property companies and related concerns. There are four
sources which are being built in as layers and which users can query and display:
• Public domain data ranging from population to employment, deprivation indicators, and
related socio-economic attributes;
• Detailed data sets that have been compiled for one-off projects such as the Kings Cross
Development etc;
• Unconventional content such as visual data, advertising data etc, which is available and can
be added in stages;
• Private data sets which we are soliciting as part of the wider support for the project will be
sought on a case-by-case basis.
Geometric data for the basic visual content is a different matter. There are a number of data
providers and there is the possibility that we might produce our own data, the cost
effectiveness of these options is crucial to a successful model. However, this spatial data is of
little use if it is only available to those with the technical expertise or the relevant software.
Access to the GIS must be easy and widespread. Even though the concepts underlying GIS have
existed for over 30 years, it is only in the last 10 years that technological developments in
computing have enabled planners to exploit the potential (Coppock and Rhind, 1991).
Consequently virtually all of the GIS systems implemented by the planning community have
been since 1990 (Masser and Craglia, 1996). Even more recently GIS, like many related
technologies, has become ‘web-enabled’. Whilst the Internet does not change the fundamental
nature of GIS (Harder, 1998), it does open the technology to an infinitely wider audience.
GIS is no longer only reserved for the technical experts or those with the relevant software.
Through the use of the Internet, home users are increasingly exploiting GIS to plan a journey
or locate the nearest cash-point. In most cases people are exploiting this software without even
realising they are using GIS, yet it provides a perfect introduction to the technology which can
be exploited by the planning community. As the use of GIS technology becomes more
widespread and more accessible to home users, it is inevitable that its role in planning will
increase. If this spatial data and associated information can be taken one stage further and
visualised in a 3d environment by home users, then we are able to offer a system that may be
even more appealing to architects and urban planners alike.
7.3
CONSTRUCTING THE MODEL: PHOTOREALISM AND
DISTRIBUTION
The aim is to provide a detailed geometric mapping and three-dimensional visualisation,
rendered to a sufficiently high level of accuracy to provide users with a definite feel that
‘virtual’ London is ‘real’ London. This sense of location and place needs to be achieved while
bearing in mind Brutzman’s (1997) six components of Internet distributed three-dimensional
graphics. They are as follows: connectivity, content, interaction, economics, applications and
personal impacts. The combination of these six components is central to our understanding
into how networked three-dimensional graphics, and the resulting virtual environments, are
produced by the modeller, distributed over the Internet and browsed by the end user.
Figure 7.1 illustrates the interlinking features of Brutzmans’ six components with an additional
Figure 7.1 Interpretation of Brutzmans’ six components of graphics internetworking with the
addition of file size.
Table 7.1 Download times according to connection, adapted from Telegraphy, 2000.
seventh, that of file size. Each of these features and components are examined in turn in
relation to creating and distributing a three-dimensional model of London.
The first, and perhaps most important component is connectivity. Connectivity refers to the
capacity, bandwidth, protocols and multicasting capabilities on the networking infrastructure
(Rhyne, 2000). Although networking is considered to be ‘different’ than computer graphics,
network considerations are integral to the production and distribution of large-scale interactive
three-dimensional graphics. Graphics and networks are two interlocking halves of a greater
whole: distributed virtual environments (Brutzman, 1997). Connectivity, in terms of available
bandwidth, is a decreasing problem. At the start of research, in 1997, the average home users
connection speed was 28.8bps with many users connecting via slower 14.4bps analogue
modems. At present average connection speeds for home users are 56Kps (Graphics,
Visualisation and Usability, 1998). The increase in available bandwidth has allowed an increase
in file size for the virtual environments produced during the research into Online Planning.
Table 1, lists theoretical download times, based on a perfect connection, of a typical 1.5mb file
for a virtual environment. Download time is listed in relation to connection speeds and thus
increasing bandwidth availability.
Download times are representative of distribution mediums using a client/server model,
whereby the entire file is required to be downloaded before the virtual environment can be
rendered. Alternatives to the client/server model are discussed later. There is a direct
relationship between connectivity and file size in the client/server model. This relationship
represents a bottleneck in the ability to distribute three-dimensional models online.
Figure 7.2 Bottlenecks in access speed, adapted from Halabi, 1997.
Figure 7.3 Level of required interaction with application.
As Figure 7.2 illustrates, even if a model is hosted via a T3 link the fastest connection the
client will obtain is 56Kbps using a 56Kbps standard modem. Online three-dimensional
environments therefore need to be tailored to the average connection rate, currently 56Kbps.
As already stated, bandwidth is a diminishing problem, capacity is increasing as cable modem
and Digital Subscriber Line (DSL) technologies are beginning to enter the home. However, at
the current time the successful distribution of virtual environments is inexorably tied to the
ability to distribute within available bandwidth, thus file size is all-important.
Linking connectivity and file size is interaction. Interaction involves a sense of presence and
the ability to both access and modify content; it also defines the level of location and place in a
three-dimensional scene. The level of interaction is a key aspect to consider in the modelling
of the built environment. Despite the ability of networked three-dimensional graphics to
create imaginative and complex environments they are too often limited by low levels of
interactivity, either due to system limitations or interface design.
The level of interaction is tied in with the content and application (Figure 7.3). E-
commerce requires a lower level of interaction than, for instance, collaborative design. In e-
commerce applications the user normally only requires the ability to view the product from a
range of angles. Collaborative design requires a more sophisticated level of interaction with the
ability to walk and flythrough the environment in addition to the ability to move and add and
subtract objects. E-commerce is a recurring theme in the development of three-dimensional
graphics as it is the driving force behind the development of both web based 3d software and
the form of distribution. E-Commerce is therefore a limiting factor on the development of
collaborative design.
Personal impacts are linked with both the level of interaction and the content of the
environment. It can be seen as the sense of wonder or frustration when interacting and
navigating a virtual environment. Interlinking these components is interface design. How the
user navigates and explores the environment is crucial to the level of wonder or frustration
experienced.
Content can be seen as any information, dataset or stream that is transported via a
networking protocol, in our case models of the built environment. Content is an emotive issue
in the modelling and distribution of three-dimensional models as despite all the hype, there is
yet to be a ‘killer application’ for online three-dimensional graphics. Currently three main
content driven applications can be seen, these are multi-user gaming, shared virtual
environments and e-commerce. These 3 applications are shaping the delivery of virtual
environments.
Economics is all-important, if three-dimensional graphics are to become commonplace on
the Internet, entry requirements need to be at a consumer cost base. This has to include the
cost of the whole package, from cost of connection and hardware, to the cost of viewing and
development software. Economics is especially important in terms of use of three-dimensional
graphics by local authorities.
Three-dimensional communication of the built environment, in order to communicate a
sense of location and place online, is dependent of Brutzmans’ six factors and a seventh, file
size. For effective communication utilising networked three-dimensional graphics a rethink of
traditional modelling and distribution techniques is required.
The modelling of urban scenes as a portrayal of the existing environment is crucial to view
any proposed developments in context and portray a sense of location and place to the use. It
is the modelling of the existing environment that poses the most difficulty, as Kjems (1999) states.
It is much more difficult to build a 3d model of an existing environment than a new
development. With this in mind methods are examined based on four levels of detail and
abstraction, namely panoramic visualisation, prismatic primitive, prismatic with roof detail and
full architectural. Firstly we examine panoramic visualisation.
Panoramic visualisation is not three dimensional, in that it consists of a series of photographs
or computer rendered views stitched together to create a seamless image. Rigg (1998) defines
a panorama as an unusually wide picture that shows at least as much width-ways as the eye is
capable of seeing, if not a greater left-to-right view than we can see (i.e. it shows behind you as
well as in front). Figure 7.4 illustrates a sample panoramic image of Canary Wharf Square in
London Docklands.
Although panoramic images are two dimensional, as they are constructed from a series of
photographs, the effect is considerable realism (Cohen, 2000). Panoramic images are not a new
concept brought about by the rise of the digital age, indeed they have been made since the
1840’s with the introduction of the first dedicated panoramic cameras. However, it was not
until 1994 and the introduction of QuickTime Virtual Reality (QTVR) for the Apple
Figure 7.4 Panoramic image of Canary Wharf Square, London Docklands.
Macintosh that panoramic production, from the stitching of a number of photographs, became
available on consumer computers for the first time. QTVR works by taking a sequence of
overlapping images and automatically aligning and blending them together to create a seamless
panorama. The resulting picture is a photorealistic capture of a scene taken over the time
required to capture the images, typically between 30 seconds and two minutes. Panoramas are
viewed online via either a plugin or Java applet. The viewer renders a section of scene allowing
the user to pan and zoom the panorama using a combination of the mouse and keyboard. Each
single panorama is defined as a node, and hotlinking between a series of panoramas creates a
multi nodal tour.
In terms of Brutzmans’ components of networked three-dimensional graphics, panoramas
score highly on use of available bandwidth and file size. To view and navigate a panoramic
image all that is required is the plugin or Java applet and the image file. Based on a medium
level of compression, image file size for a typical panoramic scene is 150k, or 200k including
the Java viewing applet. Capture techniques are both rapid and low cost with stitching
software typically available for under £100. Panoramic images are thus well suited for the
communication of existing locations via the Internet, allowing photorealistic representations
with low files size. Interaction is however limited, users can pan and zoom or link to HTML
documents but as the image is two-dimensional no higher level of interaction is possible. The use
of panoramas becomes more problematic if new developments are to be visualised. This
involves integrating a three-dimensional object with an existing panorama or creating a
rendered photomontage, essentially augmenting reality. Augmented reality has been used for
the purpose of Online Planning in the example of the Bridge Lane Planning Enquiry, more
details can be found online at http://www.casa.ucl.ac.uk/olp/.
For the production of full three-dimensional models of the existing built environment there
are three critical factors—building footprints, roof morphology and height data. It is the
combination of these factors that allows the creation of realistic models. Building footprints are
Figure 7.5 Landline derived building outlines in ArcView.
Figure 7.6 LIDAR-based City Models (http://www.globalgeodata.com)

widely available in the UK, most commonly in the form of Landline data from the Ordnance
Survey. The data is however both prohibitively expensive and detailed for online usage.
Figure 7.5 illustrates building outlines derived from Landline data for a section of Central
London. At a cost landline data can be obtained, however the main problem is the acquisition
of suitable height data and roof morphology.
Average height data can be purchased ‘off the shelf from mapping companies such as Cities
Revealed. This data provides the average height according to building footprints, problems
have however been experienced with this data. Average height is used, calculated over
enclosed building footprints. This is not detailed enough to produce a convincing three-
dimensional model. In addition the fact that it is based on extruded footprints results in a high
level of detail in terms of final polygon count and thus in relation to Brutzman (1997) makes
the data unsuitable for web distribution. Comprehensive data can also be obtained from range
imaging methods, the most widely used being Light Detection and Ranging (LIDAR). LIDAR
provides a high-resolution three-dimensional surface, which can be imported into a GIS and
draped with an aerial photograph (Figure 7.6). LIDAR is at the high end of the data range scale
and as such is not suitable for the production of models aimed at online distribution, although
combined with building footprints average height can be extracted from the LIDAR dataset.
Figure 7.7 Extrusion by average height derived from LIDAR in ArcView.
Figure 7.8 PAVAN output from MapInfo, illustrating Roof Morphology.
Figure 7.7 illustrates a section of Central London extruded from building footprints up to
an average height derived from LIDAR data. The resulting model is a prismatic representation
of Central London. The model is both crude and unwieldy in terms of required processing
power and file size. Manageability of the model can be improved but considerable
generalisation of the base data is required. Generalisation of base data is problematic.
Prismatic models lack any significant architectural detail and thus do not convey any
compelling sense of the nature of the environment (Batty and Smith et al., 2000). Roof
morphology can be added either within a GIS or via a standard modelling package such as
Microstation or 3D Studio. There has been considerable research effort over recent years into
developing the capabilities of GIS to handle three-dimensional information of the built
environment (Faust 1995). This has often been achieved through the linkage of CAD
technologies to a GIS database (Ligget, Friedman and Jepson 1995). Linking GIS to CAD is
however a tentative process. Figure 7.8 illustrates the output of PAVAN a modelling package
for the MapInfo GIS package.
PAVAN enables roof morphology and texture maps to be added to height extrusion. While
this is adequate for basic modelling, the level of realism is low and it relies on knowledge of
the modelled areas roof structure, data which is not commonly available without a
Figure 7.9 Canoma modelling stage 1.

comprehensive area survey. Where roofing morphology is not known a new method for
modelling is required.
Methods to rapidly extract and texture map both building outlines and roof morphology
have become readily available in the last 18 months. A result of the increase in personal computing
power and the demand for realism, predominantly in gaming environments, has given rise to
packages such as Canoma from Metacreations, GeoMetra from AEA Technologies and Image
Modeller from RealViz. These packages, aimed at creating models optimised, in terms of file
size, while retaining a high degree of realism, are directly suited to the production of models
aimed for distribution on-line and fall within the requirements of Brutzman (1997).
The following example provides an illustrative walk through of creating a texture mapped
three-dimensional model, of a typical new build development in the UK, using Canoma from
Metacreations. Canoma is typical of the new range of low cost photogrammetric modelling
packages.
The model was constructed from two photographs, taken with a Nikon CoolPix 850 digital
camera, Figure 7.9. The photographs were framed to ensure that all four corners, and any
shared geometric features of the building, were in view. The first stage to constructing the
model is mentally dividing up the building into a series of primitives, these primitives will be
aligned, joined and stacked to create a wireframe version of the building. Once the building
has been divided into basic shapes the first primitive can be placed, in this case a box which
constitutes the main area of the building. The correct placement of the first primitive is all-
important. From the first primitive Canoma calculates the position of the ground plane and the
camera position. Pinning the corners to the corresponding points in each photograph places the
box, the pins are represented as triangles, Figure 7.9. Where a corner is not visible, as is the
case in the bottom right photograph, a bead, a round dot, is placed to guide the primitive to
roughly the correct position.

Stage 2 creates the central roof structure, this is created using a ‘stack’ command, placing
the selected roof shape primitive directly on top of the first box. A combination of pins and
beads are used to align the primitive with the actual photographs (Figure 7.10).
The third stage repeats the procedure of creating the first box primitive and stacking to
create the front section of the house. To ensure the new section of the model is correctly
aligned it is ‘glued’ to the first box primitive (Figure 7.11).
The wireframe is now taking shape and matching the two photographs, the front porch
section and chimney are added in the same manner as stages 2 and 3, using a combination of
pins, beads and glue, Figure 7.12. The model can now be automatically texture mapped and
exported in the desired distribution file format. The example provided is for a single building,
whereby a couple of images are sufficient to create the three-dimensional model. Two images
are sufficient for the wire frame model as there a number of linked geometric reference points
in each image, the model is also made up of basic standard primitives. For more complicated
large-scale urban areas the addition of oblique aerial photography is required to provide an
overview of the entire scene. Combined with street level views urban scenes can be
constructed, Plate 5 illustrates a model of Canary Wharf, London, modelled in Canoma. The
model was produced using a combination of aerial photography and streetlevel photographs
taken from the Canary Wharf Square panoramic example (Figure 7.4).
Once a scene is constructed, all-important to its successful placement online is the file
format it is saved in and the resulting format used for distribution. The format chosen is a
critical factor in the balance of Brutzman’s components for networked 3-dimensional graphics.
Our preference is for a combination of Virtual Reality Modelling Language 2.0 (VRML), an
open source language and Viewpoint from Metacreations. It is not intended to go into detail on
distribution formats here, a full comparison of web3d file formats can be found Online at
http://3dgraphics.about.com/compute/3dgraphics/.
7.4
DYNAMIC 3DGIS FOR ONLINE PLANNING: WOODBERRY
DOWN
We have established that to place three-dimensional models online the traditional ‘3D GIS’
route is not suitable, what is required is a low bandwidth compliant model produced from the
emerging range of photogrammetric software. This therefore rules out the common route of
developing the 3d model from within the GIS itself and integrating the three dimensional and
spatial data in one system.
It is more feasible to link a web enabled GIS or its Internet Map Server with web enabled 3d
modelling systems. This can be done by developing unique database key fields that can address
match a location in the GIS with an associated building or location in the 3d model. If the user
chooses to move within the 3d model then this movement is detected and the location in the map
interface updated, and vice versa. In addition a wealth of other data can be stored ‘beneath’ the
map interface. Users may only view a photorealistic 3d model of a building, but the map can
deliver a wealth of information about the building or the surrounding area. This could include
air pollution levels for the area, the name and address of the building, details of the history of
the building or even a link out of the system to the website of the architect’s practice who built
it.
We have developed a 3dGIS in the London borough of Hackney for collaborative planning
in the redevelopment of the Woodberry Down estate. The system is based on the ideal of
opening up the redevelopment system to the local residents, via the internet, through an
interface which gathers together communication and visualisation techniques aimed at
facilitating local democracy in the planning process. Funded by the Woodberry Down
Redevelopment team and supported by the Architecture Association and Hackney Building
Exploratory, the project demonstrates how in a short period of time, 3 months, it is possible
to implement a collaborative decision making system, running in people’s own homes. One of
the issues arising from previous work into this field, Kingston et al. (1999), is that people do
not have access to the tools that facilitate the communication, in our case the Internet. This has
been remedied by securing funding to ‘wire’ 30 resident representatives (one, from each block
of the flats in the redevelopment area) with free Internet access and a Pentium III 600Mhz
personal computer. This has ensured that the interface is tailored directly to its users, allowing
us to take into account Brutzman (1997) components with a set of known limitations. The
interface breaks down to the following components:
7.4.1
Arc Internet Map Server
ArcIMS, ESRI’s latest generation of Internet Map Server has been used to develop the GIS
interface to the Information System. On a basic level this needs to simply provide a navigation
tool around the area and a data structure to ensure a tight coupling between the 3d and 2d
elements of the system. However, on a more detailed level this can be used as a search engine
(for street names and postcodes), an information retrieval system (for example for discovering
air pollution levels in an area or information about a particular building or block of flats) and
finally as a way of allowing the user to interactively post location geo-referenced comments
about the area.
In order to avoid plug-in download times, the system has been restricted to an HTML site with
the standard ESRI Java applets to provide improved functionality (for example rubber band
selection of objects). The interface has then been customised by adapting the JavaScript files
provided by ESRI in order to adapt and extend some of the functionality of the system.
7.4.2
3D Models
Models of the area have been created in Canoma and 3Dstudio Max (version 4.0) to ensure a
photorealisitc representation of the area while ensuring compliance with the restrictions of a
low bandwidth connection. Plate 6 illustrates a model of Rowley Gardens, Woodberry Down.
The model is developed to be interactive, allowing ‘what if’ scenarios to be visualised in real-
time by the local residents. These scenarios will take place in association with the architects
and developers contracted to redevelop the site. The redevelopment of Woodberry Down is a
long term project, to ensure that the models remain compatible with emerging web based
technologies they are stored in 3Dstudio Max, an industry standard modelling package. Each
model is linked to the ArcIMS, allowing each section to be displayed according to any of the
parameters contained in the GIS.
7.4.3
Panoramic Visualisation
The redevelopment area has been captured in a series of interlinked (via the ArcIMS)
panoramas. Panoramas are an integral part of the site visualisation as they provide a 360×360
degree documentation of the area as is at the time of capture. As with the 3D models they can
be used to visualise future development in local context by augmenting reality.
7.4.4
Bulletin Board
Central to the system is the community-based bulletin board. The board allows local residents
to log in and express their views on each stage of the redevelopment. This enables the
redevelopment process to move away from the standard ‘village hall’ meeting scenarios in
which only a few residents are able to have their say. Using a bulletin, coupled with free
Internet access, all the residents are able to participate and have a fair say in the proceedings. The
bulletin board, is not however restricted to the issue of redevelopment. It also includes sections
on local/community news, a trading sections for good and general discussion. A range of
subjects underpins the project, it is not only about the development, it is about the community
as a whole, facilitating not only democracy but also community ties and relations over the
Internet.
All the separate elements are in place and it is intended that the work will go live by mid
2001, coinciding with the residents being ‘wired’. This will complete the picture of an Online
Planning system for the redevelopment of Woodberry Down.
This project is phase one of a larger scale project—a Virtual London. We have established
that large-scale models can be produced utilising a level of realism to ensure that a strong sense
of location and place can be achieved. These models can be linked to spatial data via an
Internet Map Server system which in turn can be linked to a bulletin board to facilitate
communication. Updates on the work may be found at http://www.onlineplanning.org and
http://www.casa.ucl.ac.uk.
7.5
FUTURE DEVELOPMENTS
Dynamic two-way interaction between the 3d models and ArcIMS is possible through the use
of XML. Both Viewpoint (the format of the models) and ArcIMS operate through XML,
enabling common ‘tags’ to be produced and thus enabling closer integration. It is also feasible
to run the Woodberry Down project on hand held devices, such as Palm Pilots, while maintaining
the required level of interaction. GIS, in the form of ArcPad, can be linked with both
panoramic images and 3d models optimised to run on either Windows CE or Palm operating
systems, leading the way to portable on-site visualisation and communication.
7.6
REFERENCES
Batty, M., and Smith, A., 2001, Virtuality and cities: Definitions, geographies, designs. In Virtual
Reality in Geography, edited by Fisher, P.F. and Unwin, D.B. (London: Taylor & Francis).
Brutzman D., 1997, Graphics internetworking: Bottlenecks and breakthroughs. Chapter 4, Digital
Illusions, Clark Dodsworth (ed.) (Reading, Ma: Addison-Wesley), pp. 61–97.
Cohen G., 2000, Communication and Design with the Internet (New York: W.W. Norton &
Company), pp. 128–129.
Coppock, J.T., Rhind, D.W., 1991, The History of GIS. In Geographical Information Systems:
Principles and Applications, edited by Maguire, D.J., Goodchild, M.F, Rhind, D.W., (Harlow:
Longmans), pp. 21–43.
Faust, N.L., 1995, The Virtual Reality of GIS, Environment and Planning B: Planning and Design, 22,
pp. 257–268.
Graphics, Visualisation and Usability Center, 1998, 10th WWW User Survey, Technology
Demographics, http://www.gvu.gatech.edu/user_surveys/survey-1998-10/graphs/
graphs.html#technology.
Harder, C., 1998, Serving Maps on the Internet: Geographic Information on the World Wide Web. ESRI
Press, pp. 16–24.
Kingston, R., Carver, S., Evans, A., Turton, I., 1999, A GIS for the public: Enhancing participation
in local decision making. Proceedings GISRUK 1999. http://www.geog.leeds.ac.uk/papers/99–
7/.
Kjems, E., 1999, Creating 3D-models for the purpose of planning. Computers in Urban Planning and
Urban Management, http://www.iuav.unive.it/stratema/cupum/pdf/C6.pdf.
Liggett, R., Friedman, S. and Jepson, W., 1995, Interactive design/decision making in a virtual
urban world: Visual simulation and GIS. Proceedings of the 1995 ESRI User Conference, pp. 308–335
Masser, I., Campbell, H. and Craglia, M., (editors), 1996, GIS Diffusion: The Adoption and Use of GIS
in Local Government in Europe. (London: Taylor & Francis), pp. 95–110.
Rigg, J., 1998, What is a Panorama? PanoGuide, http://www.panoguide.com/reference/
panorama html.
Rhyne, T.M., 2000, Internetworked 3D Computer Graphics: Overcoming Bottlenecks and Supporting
Collaboration II. Overview of 3D Interactive Graphics using the Internet, SIGGRAPH 2000
Course #28 Notes, http://www.siggraph.org/%7Erhyne/com97/netgraf3/overview-
netgraf3.html.
Telegeography, Inc., 2000, Telegeography 2000 Global Telecommuncations Traffic Statistics and
Commentary, Edited by G.C. Staple, (Washington: Telegeography).
8
Historic time horizons in GIS: East of
England historic landscape assessment
Lynn Dyson-Bruce
8.1
INTRODUCTION
Historic Landscape Assessment (HLA) is a new methodology being developed to assess the
historic dimension of the landscape which will enable the ‘time-depth’ and complexity of the
landscape to be assessed. The HLA process charts the surviving, visible historic components
within our landscape, to enable a better understanding of the processes of landscape genesis.
As a GIS project, HLA pays particular attention to metadata which documents the sources
and decisions made in interpreting each polygon. The methodology uses a succinct set of
attributes, within a simple database, to create a single layer. Within this layer are recorded all
components that share common and/or diverse histories to create a complete coverage of the
project area.
The metadata focuses on specific historic time horizons in conjunction with landscape types.
This may enable the development of spatio-temporal models of landscape change, which could
inform the decision making process to enable appropriate management of what is increasingly
becoming recognised as a finite resource. This paper reviews the techniques developed for a
project spanning six counties in the east of England.
8.2
BACKGROUND
Historic Landscape Assessment or Characterisation (HLA/C), is a broad-brush approach,
which has been developed to identify and map the historic ‘time depth’ of the landscape. This
approach is currently being applied across England on a county-by-county basis as part of a
national programme, by English Heritage (EH), the major funding body, in conjunction with
regional authorities. Work is in various stages of progress across the country.
HLA has developed from the traditional, ‘paper based’ methodology of Landscape
Character Assessment (LCA) (Countryside Commission 1993). LCA is a methodology devised
to assess the character of regions within the country, on regional and local scales, and has been
applied on a broad scale (usually 1:50,000), across the country on a county-by-county basis.
LCA has traditionally been a paper-based approach, but is becoming increasingly GIS based.
However for HLA, each county has devised its own methodology, either paper-based and/or
GIS-based, but without being couched within a national GIS or HLA methodological
HISTORIC TIME HORIZONS IN GIS 107
framework. It is therefore felt that results once collated on a national scale could reflect not only
historic landscape diversity, but also methodological differences.
This paper will discuss and illustrate various issues including the methodology and
metadata, which has been applied within the project, and its importance within GIS. Also a
proposal for a national series of historic landscape character types to be established based on
GIS national standards, to enable consistent comparative analysis. This combination of GIS and
historic analysis of the landscape may help to achieve a new understanding of our cultural
landscapes and their dynamics, and may assist in devising appropriate management strategies.
The methodology within this regional project has been re-defined to take into account the
basic principles of GIS, and to consistently reflect landscape complexity, across regions. The
result will be an apolitical seamless map, which will enable, once the project has been
completed, spatial and Digital Terrain Models (DTMs) to be developed and generated, as
required of historic landscape patterning regardless of these political boundaries. In
consequence results should reflect historic landscape diversity rather than methodology.
Traditionally archaeologists have used a ‘site-based’ approach in which archaeological sites
and features are viewed as discrete entities within the landscape, which has now been
recognised as having severe limitations for broad landscape analysis. These records may be held
within regional, or county based Sites and Monuments Records (SMR) or the National
Monuments Records (NMR) managed by the Royal Commission on the Ancient and Historical
Monuments of Scotland (RCAHMS), in Scotland, or English Heritage in England, or CADW
in Wales. This data is currently represented as point data, albeit increasingly within a GIS.
It is not only difficult but would be inappropriate to extrapolate this point data onto the
surrounding landscapes for wider landscape interpretation or assessment, as these may no
longer be historically contemporary with its surrounding landscape. Due to this traditional
‘site specific’ approach, the weighting is on the ‘site’ rather than its wider context the
‘landscape’, which until now has been difficult to assess. In addition, this implies importance to
the site, so that by default the surrounding areas or landscapes are of no, or of limited
historical significance, and are therefore less important. This lack of recognition and subsequently
appropriate protection, makes the landscape more vulnerable to future development.
However it has become increasingly clear that the landscape has an historic dimension, which
needs to be adequately assessed for future management purposes, especially in response to
current issues regarding sustainability. The landscape is an historic record in itself, also a finite
resource, which is being changed or lost in certain areas beyond all recognition due to modern
pressures, e.g. development, mineral extraction, farming.
Before we can devise appropriate management strategies, we must first assess the resource
in an appropriate manner. HLA seeks to redress this imbalance by applying a methodology for
assessing the landscape in terms of its historical development, within which this spot-site data
may then nest and be assessed as to its contemporaneity.
8.3
EAST OF ENGLAND PROJECT
8.3.1
Project Area and Remit
The East of England project was established to assess the whole of a broader administrative
region, comprising the six shire counties of Bedfordshire, Cambridgeshire, Essex,
Hertfordshire, Norfolk and Suffolk. This was considered to be sufficiently large and diverse in
terms of its geography and historical development for meaningful analysis. A key objective of
the project was to aid in the conservation of the historic environment by assessing the historic
development of the landscape and using this information to inform its most appropriate long-
term management.
8.3.2
Landscape History
In assessing landscape, in this case ‘landscape history’, it is necessary to be aware of certain
forces which determine how our landscape has been formed. These include natural and
anthropogenic forces, which are interdependent upon each other. In addition technological
events and processes, in conjunction with socio-political reforms and policies, may have far
reaching effects upon the landscape, and may be historically specific and therefore recordable.
Humans have manipulated the naturally occurring landscape, producing dramatic and long
lasting impacts on the landscape through time. Therefore it must be recognised that the
landscape is an historic record in itself, which requires to be appropriately understood.
We are aware of the variety of landscapes around us, but few realise the ‘time depth’ of
these landscapes, whether they have changed, and if so, how, the speed of change, or what
frequency and how they relate to each other and to other determinants. The landscape is
dynamic being in a constant state of flux.
HLA charts the surviving, visible historic components within our landscape. Results so far
indicate that the history of the landscape is dynamic, diverse, complex and unpredictable. So
far it has been shown that there are areas of continuity, discontinuity, contiguity, rarity, and
diversity. The only consistent result is this diversity and dynamism.
8.3.3
National, Regional, Local Landscape Type Series
It is proposed that by using a variety of scales of approach we may begin to identify National,
Regional and Local landscape types and thereby recognise areas of diversity and rarity, at
appropriate scales. Some events are strictly local and do not need to be preconceived in a
national taxonomy, but will fit into a broader grouping at the regional and national levels.
Therefore it is proposed that there should be a ‘national type series’ of historic landscape
types into which all areas of study could feed, at the appropriate scale (see Figure 8.1). This
need not deflect from other approaches and developments, nor each area’s individuality, they
only require prior thought and a linking field. It is proposed that this series should act as an
‘umbrella’ of generic types into which each region’s diversity and uniqueness may contribute
and support. This would result in ‘national’, ‘regional’ and ‘local’ types, a classification
reflecting at each appropriate scale landscape diversity and detail, in historic terms. This would
be taking the traditional ‘grandparent’, ‘parent’, ‘child’ model.
For example, at the national level one could have broad categories relating to period, e.g.
19th century, 20th century, with generic landscape type, e.g. field systems, woodlands, informal
parklands. At the regional level these may be more subdivided into a greater range of more
specific landscape types, e.g. irregular fields systems, co-axial fields systems, with more refined
time horizons. At the local level one may have very specific landscape types reflecting local
variations, with very specific historic sources, e.g. Tithe, Enclosure or Estate maps cited. The
Figure 8.1. Proposed concept as a schema for National HLA Types.
complex palimpsest may only be viable at the local/regional level. However, with a suite of
attributes one could select out those required, and group appropriately for analysis at
alternative scales, for example combining specific historic time horizons into broader categories,
or grouping basic landscape types together—hence taking local variants and combining them
into broader regional groups.
8.4
DEVELOPMENT OF HLA
8.4.1
Background
The HLA methodology has developed from the seminal work carried out in Cornwall
(Herring, 1998), which mapped the landscapes according to ‘Historic Character Types’, a
paper-based exercise, derived from the traditional LCA approach (Countryside Commission,
1993). It was the first attempt at assessing the landscape, rather than point data in historic terms.
Work in Scotland further developed the approach by mapping ‘Historic Landuse’ using a
Geographic Information System (GIS) (Dyson-Bruce, 1998: Dyson-Bruce et al., 1999), by
Historic Scotland and RCAHMS, now centrally managed and applied by RCAHMS. This
ensures consistency of application and methodology across Scotland. Wales has defined a
‘Register of Landscapes’ of specific or outstanding interest (CADW, 1998) which ranks and
values specific selected areas, but not the whole landscape.
EH has used a wide variety of paper or increasingly GIS based methodologies, to determine
‘Historic Landscape Character’ (HLC) in different counties (Fairclough, 1999) across England.
EH seeks to assess the entire landscape as an historical resource, rather than place import or
rank specific areas. This is a new, dynamic approach still under development, especially as a
GIS application.
8.4.2
Time Horizons within GIS
However it is the advent of GIS that has facilitated this approach, as the information may be
represented in four dimensions:
• 2nd Dimension horizontal space, i.e. N/S—E/W

• 3rd Dimension vertical depth, i.e. topography
• 4th Dimension time, i.e. time-depth or palimpsests
The horizontal, 2nd dimension definition of landscape components forms the backdrop of the
mapping project. The topographic or elevation aspect, is secondary and may be derived later,
using for example Ordnance Survey (OS) contour data.
The key element of the methodology is the treatment of historic events within the landscape
as a series of time-horizons to enable spatial-temporal modelling, the 4th dimension. This
together with the large quantity of complex spatial information is ideally suited to the GIS
platform, which has enabled HLA to become a powerful tool. As a consequence the original
paper-based methodology had to be adapted in response to the demands of various issues
including GIS, technology, landscape complexity and archaeology. Therefore the GIS provides
support to the HLA by including self-documentation regarding derivation of the data sources
within the metadata, thus providing a clear audit trail of the process of assessment and analysis.
Earlier approaches were based on established Landscape Character Assessment (LCA)
techniques and methodology (Fairclough, 1999). These were paper-based, which led, by their
inherent limitations, to the creation of a single map or series of simplistic maps, achieved by
the aggregation of data to create thematic historic landscape types. Paper-based HLA is
therefore difficult to be reliably and consistently replicated. In addition any detailed analysis or
changes in the representation or update of the data, are very difficult to achieve efficiently.
However, until recently GIS has only been used as an enhanced version of the original
paper-based approach to HLA, resulting in simple digital maps. This may solve some of the
problems of replication and representation of data, but these lack inherent intelligence, and
importantly metadata, which are key aspects of true GIS applications. These issues are now
being addressed within various county methodologies. However it is important to remember
that GIS is not an end-use in itself, but a tool that facilitates input, representation, access,
analysis and output of coherent data, to an established standard.
In response to these parameters, the methodology has been refined and developed in the
light of other HLC projects, to accommodate regional diversity and harness the capabilities of
GIS. This has been achieved, by creating a suite of ‘historic landscape types’ or ‘attributes’.
These may then be aggregated as required for different objectives, remits and end uses (Dyson-
Bruce et al. 1999: Fairclough 1999). The attributes are derived either directly from map
sources, e.g. woodlands, informal parklands, or indirectly by interpretation by informed

judgement, for example, field types forming identifiable ‘footprints’ or distinct morphology,
e.g. co-axial fields, parliamentary enclosure. However in reality, due to its complexity it may
not be possible to define the landscape into large morphologically consistent geographic units.
The methodology is flexible, so this suite of attributes may be expanded as required, in
response to landscape diversity. In addition the inclusion of dated sources, directly within the
database enables specific time horizons to be recognised.
HLA as a methodology has necessitated this fluid and dynamic approach to respond and
record the subtleties and dynamics within the landscape. This has had to be done in a suitably
sensitive manner, to ensure the development of a more rigorous analytical tool relating to
academia, data entry, analysis and metadata.
8.4.3
Application
The revised HLA methodology, used in the East of England Project is now GIS based, and may
be described by the following three interactive aspects:
• HLA—creates ‘historic landscape types’, attributes that may be aggregated to form

‘historic character areas’—the academic, archaeological and historical aspect. These
are recordable historical events within the landscape, derived from a variety of sources.
• GIS—which handles the data capture process and input—the practical aspect. Ensuring
consistent application and data management.
• Metadata—is the data informing the user of the process of, for example, data collation,
source, ownership and map creation, and is imbedded within the database. In addition it
creates an audit trail of data synthesis, analysis and interpretation.
8.4.4
Resources and Sources
Due to the constraints of staff, time and resources, the HLA methodology is by necessity a
quick, broad-brush and ‘desk-top’ based approach. It is meant to rapidly assess the current
landscape as to its historic origin and character. This is achieved by assessing primarily historical
information.
Key sources include maps (current and historic, paper and digital), aerial photographs and
documentary sources. The criteria used must be robust, definitive, replicable and meaningful
not only to the data creator, but to a variety of users and objectives. However the landscape
types must be sufficiently sensitive and definitive to reflect landscape composition, diversity,
variability, continuity and discontinuity, which enables the complex concept of ‘time-depth’
and palimpsests within the landscape to be assessed.
In moving from one county to another the availability of digital/paper datasets varies, as
does quality of that data, hardware, software and support and these can have a direct impact on
work progress and efficiency. For example, datasets may not have been created at an
appropriate scale, or may have used different criteria, or be without metadata, which at times
renders potentially useful data unusable. In addition other datasets, which may have been
created independently, usually at differing scales are difficult to combine due to among other
things edge-matching problems. It is hoped that with the migration to the new OS MasterMap
this will be a problem of the past.
8.4.5
Data Collation
The data is initially collated at 1:25,000 (data analysis), as this represents the smallest scale at
which field boundaries are represented on OS topographic maps (1:10,000 sheets were
deemed too detailed and cumbersome). The 1950 1:25,000 OS series (paper maps), have been
used to collate the first tranche of information, being compared to the current digital OS
landline and 19th century First Edition OS data, giving three distinct time horizons.
In certain parishes additional map sources were used for research purposes to try and
further understanding of the complexity of field morphology. Tithe, Enclosure, and early
estate maps were consulted, to inform the process of field shapes, types and land enclosure,
which has expanded the series of time horizons within these specific areas.
This information is then digitised, on screen, using a defined series of landscape types
(attributes) to create complete coverage. In the future with the forthcoming, polygonised OS
MasterMap, digitising will be much reduced, with data being seeded into the pre-existing
polygons.
At the generic level these landscape types distinguish the various landscape components, for
example field systems, woodlands, parklands, grazing, urban, industrial, extractive, military
(both current and relict of each type). These attributes must still be currently visible within the
landscape to be recorded. Recording the data in this way enables it to be used flexibly in
response to a variety of objectives. For instance, this data may then be aggregated, as desired,
to observe broad patterns in historic land-use at a county or regional level.
8.4.6
Metadata
The metadata has been incorporated directly within the database, with appropriate fields which
includes:
• historic landscape type—these fields allow short succinct data entry via pre-set ‘alpha-
numeric’ codes which are linked to another database providing full descriptors of landscape
type (e.g. code_c, relict_l or Hla_code). For example sf, represents sinuous fields;
ip=informal parklands; ba=built-up area; aw=ancient woodland; pf=prairie field. These
are based on direct or indirect evidence of landscape component type, either untampered
through time or as a palimpsest with various events recorded within a composite unit. It
follows that the more events recorded, the more fragmented the landscape. Conversely a
single relatively untouched event may cover a much larger area, e.g. a block of
parliamentary enclosure. So far up to four events have been recorded within a landscape unit,
however this does not reflect an older historic origin, purely that more events or changes
have been recorded within that unit of space.
• data source—the maps from which the data derived, whether current or relict, e.g. date_c,
date_rl. These fields again provide a single character link to that landscape’s component
source, e.g. L=OS Landline 2001, N=First Edition OS. This creates a clear audit trail of
source synthesis, as each polygon may reflect several sources in its creation. However it has
been noted that in some instances there have been changes or events within the landscape
which fall between these time horizons, and are therefore not able to be adequately
recorded, e.g. unrecorded mineral extraction, then recorded land restoration.
• creator—this records the persons responsible for assessing, assigning codes and deriving the
original polygon representing the landscape component.
• date of creation—date the data was digitised, which informs currency of update and data
creation.
• data owner—this reflects which county or regional authority has ownership and copyright of
the resultant material, e.g. Essex County Council (ECC), Hertfordshire County Council
(HCC).
• scale & consistency of data input—the scale of digitising is set at 1:10,000. Whereas the source
material is primarily at 1:25,000, it has been enhanced and checked at 1:10,000 with heads-
up, on-screen digitising at that scale, to ensure consistency.
• glossary of terms—e.g. code_c, code_r2. This is a linked database providing full textual
descriptions of landscape types, so that users can understand the landscape types without
having to use a lookup table to interpret the codes, whilst work is in progress.
8.4.7
Map Sources
In addition there are linked databases to correlate data, especially to the data sources, primarily
the maps used as a source to create the HLA dataset. It is important to recognise in assessing map
data, especially historical sources that these may reflect specific objectives and that paper maps
can only represent limited and selected information. The raison d’être of creation may be
different to its current use and interpretation, and thus an important caveat. For example some
of the early county maps are more a reflection of wealth than geographic reality, i.e. one paid
to have ones estate/manor represented on the map. Other maps are records of resources, for
example the Roy maps of Scotland, which records primarily resources of strategic use to the
military, e.g. woodlands, farmland, routes, water, populated areas.
In addition there is a major problem of data inconsistency not only between different
counties and regional authorities, but within, and this includes paper and digital sources. This
creates a practical problem in consistency and accuracy in data synthesis. With paper maps all
georeferencing is visual, whereas digital maps are georeferenced and therefore may be easily
overlaid for comparison and analysis. In reality, one is either assessing an avalanche of paper
maps, or a very cluttered GIS screen with many layers, or you have both—usually the latter.
There are benefits and disbenefits to having the data either digitally or as map sheets,
experience has proven that it is better to have both.
The datasets that are consistent between counties include the OS Landline 2001, 1:10,000K
raster maps and the 19th Century First Edition (either paper or digital), and 1950 1:25,000 OS
paper maps. The sources that are consistent within a county include, the 18th Century county
series, e.g. Dury 1760 for Hertfordshire or Andre & Chapman 1766 for Essex (all paper
sources). In addition sources that are inconsistent within counties have been referenced and
include parish based historical sources, e.g. Tithe and Enclosure maps, and individual Estate
maps. These vary in size, scale (if any is recorded), detail, accuracy and quality. Few have been
turned into digital sources, as many vary in size, are fragile, unwieldy and unique historical
documents, at times very large covering map tables requiring careful unrolling, one being able
to see only small sections at a time. Up to date aerial photography has only recently been
digitally available as a georeferenced source, otherwise they require visual referencing.
It is hoped there will be links within the database to other sources of information, which
will include other counties methodologies, e.g. Suffolk: hyper-links to photographs, scanned
images, maps, textual documents, videos. These databases in conjunction with the embedded
metadata, will therefore help to render the methodology as transparent as possible and be self-
documenting for future researchers and users. In addition the HLA data may be used as an archive,
the data may be ‘static’, providing uniform time horizons or ‘dynamic’ allowing constant
revision.
8.5
DISCUSSION
The East of England Project is unique, as other counties within England are being, or have
been assessed, independently using a variety of methodologies. The experience of the project
so far has shown that there are distinct advantages to the application of a single methodology to
a group of counties being assessed as a single geographic region. This ensures a seamless map with
consistency in methodology, application and analysis ensuring replicable results yielding
compatible analysis. Therefore analysis should reflect historic landscape diversity rather than
methodological differences.
Metadata is now being increasingly recognised as an important facet of GIS, to inform the
user of the very nature of the data. It is often a misunderstood or poorly understood concept
outside the realm of GIS practitioners. Its inclusion within the database of HLA ensures a clear
audit trail of synthesis, scale, ownership etc.
Broad historic landscape patterns are beginning to be identified which appear in some
instances to disregard county boundaries, an important consideration when one sees how an
individual county’s methodology and application respects each county’s boundary. Some
landscape types are scattered through the countryside, others have a geographically limited
spatial distribution, or are mutually exclusive. However, there may be localised pockets of
distinctive historic landscape character and development. This patterning reflects landscape
diversity in historic terms. It is proving to be a diverse, complex and dynamic landscape (see
Figure 8.2).
In collating the data the inconsistency and availability of datasets, GIS software, hardware,
resources, GIS support all have direct impacts on an application such as this. It is therefore
important to recognise this and take such realities into account when devising a methodology
which is not only meant to be a simple, direct, practical GIS application, but also a pragmatic
management tool, and is in addition an academic exercise.
When specific areas have been completed, with geospatial analysis and additional research it
is hoped to identify why some areas exhibit these differences in diversity and contiguity of
landscape character. Initial research indicates a range of factors including, for example, historic
tradition, social, economic, political, agri-environmental, geology, soils, slope, aspect, and
height. These all have a part to play in landscape development. It has yet to be established in
what proportion, why, and how these various factors influence the historic development of the
landscape, and could form the platform of future research. However any GIS research will be
limited to those datasets which are available digitally, at an appropriate scale, which again may
have disparate regional availability.
Figure 8.2 East of England Historic Landscape Assessment: Suffolk, Hertfordshire and Essex—
Illustrating work in progress.
HLA has already proven to be a valuable management tool, for example identifying and
informing:
• Local, regional, historic landscape patterns

• Chronological patterns, through time and space: landscape ‘time depth’ & palimpsests
• Landscape management and Landscape Character Assessment
• Management of the cultural resource
• Archaeological survey
• Research objectives
• Development control in planning
• Regional and Local Planning issues, policies and strategies
However the author also feels it is becoming increasingly important that a national GIS strategy
be implemented to ensure a consistency in standards and application of GIS within this field of
historic time-series analysis. In addition a consistent methodology and use of terms would
encourage use and respect of such a dataset, especially in controversial and contentious
circumstances such as Public Inquiries. Planners, landscape architects, developers and other
users could have faith in the data for improved land management, research and other
applications.
8.6
CONCLUSIONS
The HLA methodology has necessitated a flexible and dynamic approach to respond and record
the subtleties within the landscape in an appropriate manner. In addition the methodology has
been required to take into account GIS issues relating to metadata, including data source,
synthesis, entry, and analysis.
Experience is proving that appropriate hardware, software and a robust methodology are
essential for the success of any GIS project. It requires clarity of thought and application, for a
GIS map/theme/layer/coverage is only as intelligent as the database and metadata supporting
it. Realistic objectives and time-scales with appropriate resources need to be established and
made available, to enable the successful completion of any project.
The change of methodology from that of ‘paper-based’ to GIS approach, has not been
without difficulties, but is constantly under improvement and being refined with experience
and expertise. This is leading to a more robust, transparent, and flexible methodology. This
will facilitate a more reliable analysis not only within, but also between, different counties.
Thus results obtained will reflect real and meaningful historic patterns within the landscape,
rather than differences of methodology in the process of assessment, application, synthesis and
analysis.
Therefore the incorporation of metadata within the database is fundamentally important in
ensuring a transparent methodology to help maintain consistency of application and analysis
between regions, by a variety of practitioners. This particular application focuses on specific
time-horizons, incorporated within the database to widen the scope of current and future
synthesis and analysis.
Finally it is proposed that a national GIS set of standards and Historic Landscape Types be
formulated to enable appropriate and consistent GIS methodologies be established to ensure
compatibility of results. Thus local, national and regional landscape models may be developed,
which will consistently reflect true landscape diversity across England. There is no doubt that
in the future, the analysis and management of this form of historic spatial data lies within the
GIS platform.
8.7
ACKNOWLEDGEMENTS
The project has been funded by English Heritage, and supported by the relevant local
authorities.
I should like to thank Graham Fairclough (English Heritage), Stewart Bryant (Hertfordshire
County Council), Paul Gilman (Essex County Council) and Edward Martin (Suffolk County
Council) for their assistance and support in this project.
(The views expressed are those of the author and are no reflection of English Heritage (EH)
or the associated local authorities).
8.8
REFERENCES
CADW, 1998, Register of Landscapes of Outstanding Historic Interest in Wales, (Cardiff: CADW).
Countryside Commission, 1997, Landscape Assessment Guidance. Countryside Commission.
Dyson-Bruce, L., 1998, Historic Landuse Assessment, Unpublished Report for Historic Scotland.
Dyson-Bruce, L., with Dixon, P, Hingley, R., Stevenson J., 1999, Historic Assessing Historic Landuse
Patterns. Report of the pilot project 1996–98 , Research Report, (Edinburgh: Historic Scotland &
RCAHMS).
Fairclough, G., (editor), 1999, Historic Landscape Characterisation, (London: English Heritage).
Ford, M., 1999, Historic Landscape Characterisation in Suffolk 1998–1999. Draft Report.
Herring, P., 1998, Cornwall’s Historic Landscape: Presenting a Method of Historic Landscape Character
Assessment. (Truro: Cornwall Archaeology Unit).
Wrathmell, S., Roberts, B.K., 2001, Historic Settlement Atlas of England, English Heritage.
9
Using GIS to research low and changing
demand for housing
Peter Lee and Brendan Nevin
9.1
INTRODUCTION
This paper explores the sustainability and popularity of neighbourhoods using administrative
data at neighbourhood, local authority and regional level in the UK. The paper is based on a
number of research projects conducted by the authors on the issue of low and changing
demand for housing in the North West and West Midlands conurbation’s of the UK. The
research is principally dependent on the use of GIS in determining coterminous boundaries for
the analysis of social cohesion, social exclusion, access to community facilities and housing
popularity. The research identifies a number of problems associated with the identification of
unpopular or low demand housing areas based on the analysis of poverty at small area level and
illustrates the value of GIS in assisting with the targeting of resources and the formulation of
local urban and housing policy.
During the period 1992–2001 the UK experienced the longest duration of unbroken
economic growth since 1945. However, whilst regional disparities in unemployment rates fell,
highly significant differences in regional and sub-regional housing markets emerged. A shift in
housing demand away from the north to the south of England was partly explained by differential
migration patterns (Holmans and Simpson, 1999). This lead to social housing providers in
some localities attempting to manage significant increases in empty properties and turnover in
the social rented sector. But whilst, publicity surrounding this issue initially focused on the
difficulties being experienced by the social rented sector, it became apparent that parts of the
private sector owner-occupied market were also experiencing problems evidenced by low
sales prices and in extreme cases abandonment of properties. The spatial manifestation of these
housing market changes has increasingly been focused on disadvantaged neighbourhoods and in
some areas in the North and Midlands, this has resulted in the progressive abandonment of
neighbourhoods. The publication of a number of UK government and academic reports since
1999 (see DETR, 1999, 2000; Lowe et al., 1998) succeeded in raising the issue of low and
changing demand within the policy community. However for housing professionals the literature
at this stage did not greatly assist them with targeting resources at neighbourhoods or housing
markets in decline, particularly in districts where deprivation was widespread.
In this paper we want to demonstrate how the issue of low demand1 was researched in
different local and regional contexts using GIS and spatial analytical techniques. The paper
makes the point that it is important to differentiate between ‘deprived’ neighbourhoods that
are sustainable and those that are deprived and appear to be dysfunctional. GIS can help in
USING GIS TO RESEARCH LOW AND CHANGING DEMAND FOR HOUSING 119
differentiating between these spatial trends. We conclude that the issue of low demand is well
suited to spatial analysis as the dynamics of low and changing demand is influenced by a
number of processes taking place at different spatial scales, which cut across local
administrative and political boundaries. The paper by draws on three separate case studies
conducted by the authors at local and sub-regional level. The creation of new policy
boundaries cutting across pre-existing management/administrative/political boundaries is an
attractive feature of GIS tools, and one that policymakers find useful in clarifying issues or
highlighting new strategic priorities.
9.2
CAUSES OF UNPOPULAR HOUSING AND CHANGING
DEMAND
Explanations for low and changing demand differ from area to area. Not only are area specific
analyses important but the level of analysis is also important in understanding different process
occurring at regional, local and estate level. It is undoubtedly the case that at the micro-level
there are a number of factors that may explain popularity of neighbourhoods and housing
satisfaction and which in turn affects housing demand. The influence of ‘problem’ families on
an estate or the effect that management intervention/non-intervention has upon the popularity
of neighbourhoods are important in this respect. Socio-psychological explanations (Bruin and
Cook, 1997), preferences for different types of neighbourhoods according to resident
perceptions (Adams, 1992), the role of successful role models (Wilson, 1989) and ethnic
segregation (Galster, 1990) have also been explanatory factors used. Multivariate approaches
have been used to analyse household surveys in order to explain resident satisfaction (Canter
and Rees, 1982; Ginsberg and Churchman, 1984) but fail to incorporate an analysis of broader
housing market dynamics. In this respect, Lu (1998), shows that structural factors have a more
direct effect on housing mobility intentions independent of resident satisfaction.
In this paper we wish to focus on broader spatial patterns of neighbourhood failure and how
these relate to other contextual factors such as poverty. Murie et al. (1998) identify the most
significant macro level factors influencing whether low demand is present and these include:
• The quality and condition of the housing stock;

• Residualisation of social housing due to increased identification of the sector with lower
income groups;
• Reputation of the area;
• Aspirations of new entrants to the housing market;
• Changing markets resulting from a private sector building industry able to respond to the
changing aspirations of consumers in an improved economic environment where choice is
important;
• The age profile of parts of the housing market especially where it is heavily skewed towards
the younger and older age groups.
9.3
RESEARCH ON UNPOPULAR HOUSING AND LOW DEMAND
9.3.1
Measuring Sustainable Housing Areas in Liverpool
We begin with research in Liverpool conducted between 1999–2001, which has contributed
to the city’s housing strategy statement and guided decisions on unsustainable and unpopular
housing areas. Here we describe the way in which GIS was used to explore the manifestation
of problems at the neighbourhood level.
The housing market in Liverpool is highly differentiated by price, quality and residential
turnover. The city has been vulnerable to market change as it has a larger than average social
rented sector and a disproportionate number of low value terraced housing renovated during
the 1970s and 1980s, many of these properties are now nearing the end of their extended life
and the combination of these housing factors as well as long-term economic decline and
population loss meant that Liverpool has experienced both changing demand and low demand
for housing across all tenures. In 1999 the vacancy rates for all stock in the city was more than
double the national average. It was against this background that Liverpool City Council and the
principal Registered Social Landlords (RSLs) in the city commissioned the Centre for Urban
and Studies (CURS) to examine the issues surrounding the changing demand for housing and
the sustainability of neighbourhoods in the city.
We began by differentiating between contextual indicators of unpopularity and direct
outcome measures of housing and neighbourhood unpopularity. We developed 4 separate
domains of sustainability with which to measure neighbourhood sustainability and housing
unpopularity at the small area level:
9.3.1.1
Poverty and Social Exclusion
As poverty is widespread in Liverpool and as previous research in Birmingham had shown us
that not all deprived areas are unpopular (Lee, 1998), however, it tends to be a precondition
for circumstances leading to low and changing demand. Four poverty outcome measures were
used:
• Council Tax and Housing Benefit claimants: claimants in receipt of housing benefit identified
from the city’s Council Tax register and aggregated to enumeration district (ED) from
postcoded information;
• Children in receipt of free school meals: Local Education Authority supplied data on children in
receipt of meals aggregated to postcode and matched to ED;
• Unemployment: unemployment aggregated from postcoded information to ED;
• Standardised Mortality Ratios: aggregated to ED from health authority records on deaths.
1 We use the terms ‘low demand’ and ‘unpopular housing’ in an interchangeable way but
differentiate between low demand (housing in excess of those prepared to rent) and ‘changing
demand’ (a movement away from particular tenures or dwelling types).
9.3.1.2
Crime and Social Cohesion
Poor social networks at the neighbourhood level and its impact on social cohesion and crime
(Freudenburg, 1986; Perry 6, 1997; Gordon and Pantazis, 1997) as well as the known
relationship between crime, poverty and social cohesion (see: Hirschfield and Bowers, 1997)
lead us to develop a measure of social cohesion based on the following measures:
• Property Crime and Crimes Against the Person: the different geography and sociology of crimes
lead us to separate out property crimes and crimes against the person. Merseyside Police
supplied data at ED level. Since 1999 it has become easier to obtain data on crime at small
area level due to legislation promoting the community safety role of local authorities.
• Population Decline: Using 1991 census as a baseline, 1997 mid-year population estimates at
ED level were used to estimate change. Areas with a decline in population used as a proxy
for instability and unpopularity;
• Voting Turnout: ward level figures on voting turnout at elections in 1995–97 supplied by the
local authority and used as a proxy for social cohesion.
9.3.1.3
Environment and Infrastructure
Environmental infrastructure and amenities and their proximity or accessibility also determine
quality of life. Burrows and Rhodes (1998) found that the third most important factor in
determining neighbourhood satisfaction is the presence or absence of leisure facilities. This
domain therefore attempts to measure this dimension of popularity.
• Percentage of Derelict Land within an ED: the planning department supplied a polygon file of
known derelict land sites. This was over laid on to an ED polygon file and the proportion
of the surface area currently derelict calculated;
• Number of Amenities: the planing department also supplied detailed information on the
location and type of non-residential addresses. These were matched to ED object files in
MapInfo using SQL queries to select out different types of amenity.
9.3.1.4
Housing Unpopularity
The housing unpopularity domain was used as a direct measure of low demand and
unpopularity and treated as a dependent variable incorporating three measures: number of
voids and number of void days and house prices.
• Vacancy Quotient 1993/99: A vacancy quotient is a measure of both the length of time that a
property is void and the proportion of properties that become void in a given period (Smith
and Merrett, 1988). The Council Tax register for the period April 1993 to April 1999 was
used to identify periods over which properties had become vacant as well as vacant
properties as at April 1999. Properties were matched to ED using AddressPoint in
MapInfo.
• Average House Prices: House prices aggregated to Postcode Sector by type were obtained
from HM Land Registry for the period 1995–1997. EDs were assigned house price values
by matching ED centroids to postcode sector centroids in MapInfo.
All the data was standardised around a mean of 0 (Z Score) in SPSS before the values were
transferred to MapInfo. Using MapInfo we generated choropleth maps of the four separate
domains as well as observing the statistical overlap between them. Whilst poverty is
widespread across the city of Liverpool, Figure 9.1 shows that housing unpopularity is most
problematic in the Inner Core of the city as characterised by low house prices and high vacancy
rates and episodes. Figure 9.2 shows the coincidence of areas with high scores on the housing
unpopularity domain that also have low social cohesion and high rates of poverty are mostly
located in the Inner Core of Liverpool.
Using GIS to differentiate between elements of housing market weakness at small area level
and to highlight the spatial coincidence of area deprivation and unpopular housing areas
allowed several policy recommendations to be made (see Nevin et al., 1999). The City
Council have adopted the recommendations within the report and the Annual Housing
Strategy Statement for the years 1999–2001 have focused investment and resources as the
different housing issues prevalent within the housing zones highlighted in Figure 9.2.
9.3.2
The Relationship Between Poverty or Social Exclusion and Low
Demand for Housing
The policy debate on low and changing demand and unpopular housing within the UK has
largely been orchestrated from within the Cabinet Office’s Social Exclusion Unit (SEU). The
SEU’s Policy Action Team report on Unpopular Housing (DETR, 1999) puts a heavy emphasis
on the role that poverty plays in causing low demand (DETR, 1999, p.25). But, how far does
poverty, using the Liverpool case study, explain variations in housing unpopularity, or relate to
the other known drivers of unpopularity and low demand?
Table 9.1 shows the correlation coefficients between the housing unpopularity domain and
the other domains. Whilst all the coefficients are positive the overlap between the individual
domains is far from the maximum. This indicates that some poor areas are socially cohesive
with lower than average crime rates and population loss. The policy implications would
indicate a different set of responses in ‘low crime—high poverty’ areas than in areas where
crime and poverty are equally problematic. The standardised values of the individual domain
indicators were used in a multiple regression model to explain the variation in levels of housing
unpopularity in which this domain was used as the dependent variable. Across the city 56%
(R=0.7467) of the difference in levels of unpopularity was accounted for by the standardised
domain indicators. While this is a reasonably large amount of explained variation there remains
44% unaccounted variation in levels of popularity.
The unexplained variation will be accounted for by variables not measured or used in the
analysis (examples may include attractiveness of housing on offer, labour market role of the
area, number of students living in the area and so forth). Therefore, whilst we would not argue
against the idea that poverty is a precondition for low demand, as our analysis using GIS at
neighbourhood level in Liverpool shows, we would emphasise that not all poor areas are
unpopular or suffer from low demand. It is therefore important to differentiate between
observable measures of poverty and the dynamics of housing market change which can lead to
Figure 9.1 Unpopular housing areas in Liverpool.
Figure 9.2 Liverpool policy zones: multiple problems associated with housing market failure.
unpopularity and area blight. Additionally, how far these processes coalesce is important.
Areas with observable scores that are high but adjacent to relatively stable and popular
neighbourhoods will have a different trajectory and require a different set of responses
compared to areas that score highly and are surrounded by similar areas.
9.3.2.1
Adjacency Analysis
GIS is a useful tool in this regard as it allows us to develop a method by which we can compare
adjacent small areas (Enumeration Districts or EDs) and identify coalescence of problems of
unpopularity or areas at risk of low demand. In order to compare adjacent area scores we
created two identical object files with attached data fields and in MapInfo ran an SQL
Table 9.1 Correlation Coefficients for the Four Domains.
Figure 9.3 SQL Intersect command in MapInfo on two identical object files.
command intersecting all objects on the two files (Figure 9.3). The resulting query file
(subsequently saved as ed_y_h_pairs) is all the possible pairs of adjacent EDs with the original
data value (index_p) and its paired values (index_p_2) (Figure 9.4).
In the next stage of analysis we aggregated the values of all the adjacent EDs using an SQL
command in MapInfo (Figure 9.5). The resulting values are shown as the original index_p value
for each object as well as the average score based on all adjacent pairings of EDs (Figure 9.6).
9.3.3
Measuring Unpopularity in the North West of England
Using the same ‘adjacency’ methodology we showed that more than 95% of EDs within the
North West M62 corridor (8555 of 8987 EDs) are adjacent to or scored above expected on
the government’s area index of deprivation (the ILD—Index of Local Deprivation; DETR,
1998) (Figure 9.7). The implications confirm that deprivation appears to be widely spread
across the North West and that deprivation is a blunt instrument in targeting areas at risk of
low demand.
The M62 Corridor consists of a consortium of 18 local authorities (see Figure 9.7) that
experienced a 22% increase in local authority vacancies and a 72% rise in RSL voids over the
period 1995–1999. The authors were commissioned to identify the coalescence of the most
serious problems and the areas most at risk of low or changing demand. Drawing on previous
Figure 9.4 Resulting ‘paired’ objects with associated data.
Figure 9.5 SQL command in MapInfo aggregating adjacent object data values.
research (including the Liverpool study) we therefore identified a number of factors
influencing changing demand and helped us focus on the elements that we believe are
important in determining areas at risk of changing demand. These factors included:
• areas or parts of the city in which there is a predominance of rented housing or poor quality
stock in owner occupation;
• neighbourhoods in which there is a large-scale or monolithic provision of ‘obsolescent
housing’ of a certain type, i.e. high rise flats or terraces;
• areas with demographic characteristics likely to weaken demand such as a high
concentration of elderly residents;
• a concentration of households that are economically inactive or unemployed.
An index was produced which mapped the areas at risk of changing or low demand in the sub-
region. The index was constructed in two parts:
Figure 9.6 Resulting ‘aggregated’ data file and object label.

1. social housing areas (predominant housing is public sector) with higher than expected
combined standardised values on over 65s, economically inactive, unemployed and flatted
accommodation;
2. private housing areas (predominant housing is private) with higher than expected
combined standardised values on over 65s, economically inactive, unemployed and
terraced accommodation.
Indicators in both indexes were standardised using chi-square in SPSS and combined into one
single index by substituting the highest value in each case. The index values were adjusted in
MapInfo by taking into account adjacent ED values as in the methodology described above.
This method enabled us to identify coalescence of problems highlighting in particular clear
spatial concentrations of neighbourhoods at risk of changing demand in the core of the Greater
Manchester conurbation (Manchester, Salford and to a lesser extent Trafford) and in the inner
core of Merseyside centred on the City of Liverpool, Bootle, Tranmere and Birkenhead.
Smaller areas of potential decline are also highlighted in St Helens, Halton, Wigan,
Warrington, Blackburn with Darwen, Bolton, Oldham and Rochdale (see Figure 9.8). There are
280,000 households contained within the overall clusters of areas at risk of changing demand
(16.3% of the households in the study area). These areas contain a population of 690,000 people
(15.9% of the population of the M62 Corridor). Neighbourhoods at risk are predominantly
social housing areas, however there is clear evidence of multi-tenure problems with nearly
100,000 properties being privately owned. These multi-tenure issues are most pronounced in
the Merseyside Inner Core where 46% of households either rent privately or own their homes.
The research maps of areas at risk were produced to enable the identification of individual
enumeration districts and were distributed to each of the agencies in the Commissioning
Consortium. The responses received indicated that the Risk Index had identified
neighbourhoods where problems of low demand were entrenched and others where problems
were beginning to appear. The analysis is now being used to target housing investment
resources and has been given significant emphasis in the 2001 Regional Housing Statement for
the North West of England.
Figure 9.7 EDs not adjacent to deprived areas: average scores based on adjacency pairings (M62
Corridor).
Source: Nevin et al. (2001a).
9.3.4
Validating the Risk Index: A Case Study of Birmingham
The methodology was subsequently employed in a study of changing housing markets in the West
Midlands Region (Nevin et al., 2001b). The region has a population of around 5 million in
which Birmingham is the most densely populated city, housing more than 1 million residents.
As at June 2001 Birmingham City Council owned 88,000 properties and was pursuing a policy
of wholesale stock transfer to eleven new Registered Social Landlords. In October 2000 the
city commissioned the University of Birmingham to assess the popularity of housing in order to
make decisions concerning the transfer of the city’s housing stock.
To test the accuracy of the Risk Index for Birmingham (Figure 9.9) we brought together
two additional proxy indicators of ‘unpopularity’:
1. lettings of local authority stock between 1990–2000 (Figure 9.10);

2. transfer requests from local authority stock (as at October 2000) (Figure 9.11).
These were used to highlight local authority housing areas with market weakness measured on
all three dimensions of ‘unpopularity’ in which areas scored highly on all three measures. In
this study negative indicator scores were set to 0 in MapInfo before average indicator scores
were calculated. The scores were re-set to 0 in cases where the original score was zero. As the
study was primarily concerned with the popularity of local authority stock we also set to zero
scores in areas where there was no local authority stock as at 1 October 2000. Areas were
Figure 9.8 Typology of areas at risk of changing demand in the North West M62 Corridor.
Source: Nevin et al. (2001a).
assigned a score of 1 if they scored in the upper median on any one factor: ‘at risk’ of changing
demand, higher than expected lets or higher than expected registered transfers, the breakdown
for the city as whole is as follows:
• Of the 1,485 enumeration districts containing council housing, slight more than half
featured in the bottom median on all three measures;
• A further 32% (471 EDs) had a score above the median on at least one measure;
• The remaining 258 EDs (17.4% of the total containing council housing), were split as
follows: 187 EDs (12.6%) had scores above the median on two measures whilst slightly
under 5% of EDs (71) had scores above the median on all three measures.
The results of this exercise have shown a high correlation between those areas which we would
have expected to show signs of decline given their housing and socio-economic characteristics
in 1991 (using the ‘at risk’ index for the North West and West Midlands) and what has actually
happened over the period 1990–2000. For example, in most neighbourhoods highlighted in
Figure 9.12, neighbourhoods designated at risk of low demand have either high turnover or
transfers or both.
However, there is not a perfect fit and in a number of areas the problems of turnover and
transfer requests are more widespread than the Risk Index would predict. This latter point
illustrates the need to check data prepared using GIS against other contextual measures such as
crime, environment and property construction. These findings have been developed further to
Figure 9.9 Local Authority Areas ‘at risk’ of Low Demand.
Figure 9.10 Letting of council properties (fetandardised, weighted average), 1990–2000.

assess the drivers of change in different areas of the City and have been translated into policy
responses in the City’s housing strategy.
9.4
CONCLUSIONS
This paper has illustrated the use of GIS packages such as MapInfo in analysing administrative
data for the purpose of categorising areas with characteristics of unpopularity or experiencing
problems related to housing demand. It is clear that there is considerable difficulty in determining
the viability and sustainability of housing areas both empirically and methodologically. Whilst
Figure 9.11 Registered transfers (standardised, weighted average), 1990–2000.
Figure 9.12 Unpopular council housing areas.
the focus of this paper has been on the empirical and technical aspects related to this issue, the
task cannot simply be reduced to a technical exercise alone in which areas are ranked and
reduced to an index score on an agreed scale. The policy agenda and the management practices
of social and private landlords as well as broader regional and sub-regional trends have to be
incorporated into the analysis. It has not possible for us to explore these themes here.
However, we would argue that the use of these techniques and the differentiation between
poverty, social cohesion, housing popularity areas at risk of changing demand needs to be the
starting point for exploring policy solutions. In this way it enables policy makers to target areas
for investment or disinvestment on the basis of market and socio-economic characteristics.
This is particularly valuable in local authorities where deprivation is widespread. Additionally

the mapping of areas at risk using GIS has proved to be highly effective in conveying
complicated messages to a varied audience which has included politicians and tenants and
residents groups.
There are a number of implications for further research in this area and the use of other
administrative data and national data sets. What this paper sets out to show is that contextual
indicators such as poverty, although widely referred to in the literature, do not discriminate
sufficiently between popular and unpopular areas. For this reason, more direct measures of low
demand should be used. Moreover, where regeneration and housing strategies have failed in the
past is partly due to the focus on static geographies and narrowly drawn target areas. In this
paper we have illustrated ways in which a coalescence of problems associated with low demand
can be highlighted through the application of ‘adjacency’ analysis. It would not be possible to
conduct such analysis without the aid of a GIS or mapping package. Additionally, we have
attempted to avoid the problem of static geographies by conducting regional and sub-regional
analyses. This we feel is crucial to the study of low demand as it is the different geographical
levels of enquiry that help us understand the dynamics of low and changing demand and allows
new policy vehicles to be deployed which cut across existing administrative boundaries. Whilst
it is costly and time consuming to assemble data for a large number of authorities, modelling
census data has enabled us to construct a picture of risk across sub-regions at small area level.
By mapping direct indicators of low demand from administrative records at local authority
level in Birmingham and Liverpool we were able to validate this analysis. It will therefore be
possible to conduct a similar exercise using the 2001 census to identify areas at risk of changing
demand over the next decade.
9.5
REFERENCES
Adams, R.E., 1992, Is happiness a home in the suburbs—the influence of urban versus suburban
neighborhoods on psychological health. Journal Of Community Psychology, 20 (4), pp. 353–372.
Bruin, M.J. and Cook, C.C., 1997, Understanding constraints and residential satisfaction among
low-income single-parent families. Environment and Behavior, 29 (4), pp. 532–553.
Burrows, R. and Rhodes, D., 1998, Unpopular Places? Area Disadvantage and the Geography of Misery in
England. (Bristol: The Policy Press in association with the Joseph Rowntree Foundation).
Canter, D. and Rees, K.A., 1982, Multivariate model of housing satisfaction, International Review of
Applied Psychology, 31 (2), pp. 185–208.
DETR, 1998, The 1998 Index of Local Deprivation: A Summary of Results. Department of the
Environment, Transport and the Regions. (London: Stationery Office).
DETR, 1999, Report by the Unpopular Housing Action Team, Department of the Environment,
Transport and the Regions. (London: Stationery Office).
DETR, 2000, Low Demand Housing and Unpopular Neighbourhoods. Department of the Environment,
Transport and the Regions. (London: Stationery Office).
Freudenburg, W.R., 1986, The density of acquaintanceship—an overlooked variable in community
research. American Journal of Sociology, 92 (1), pp. 27–63.
Galster, G.C., 1990, White flight from racially integrated neighborhoods in the 1970s—the
Cleveland experience. Urban Studies, 27 (3), pp. 385–399.
Ginsberg, Y. and Churchman, A., 1984, Housing satisfaction and intention to move—their
explanatory variables. Socio-Economic Planning Sciences, 18 (6), pp. 425–431.
Gordon, D. and Pantazis, C. (editors), 1997, Breadline Britain in the 1990s. (Aldershot: Ashgate).
Hirschfield, A. and Bowers, K.J., 1997, The effect of social cohesion on levels of recorded crime in
disadvantaged areas. Urban Studies, 34 (8), pp. 1275–1295.
Holmans, A. and Simpson, M., 1999, Low Demand: Separating Fact from Fiction. (Coventry: Chartered
Institute of Housing for the Joseph Rowntree Foundation).
Lee, P., 1998, Popular and Unpopular Housing Areas: A Report to Urban Renewal. Birmingham City
Council.
Lee, P. and Murie, A., 1997, Poverty, Housing Tenure and Social Exclusion. (Bristol: The University of
Bristol (The Policy Press) in association with the Joseph Rowntree Foundation, York).
Lowe, S., Spencer, S. and Keenan, P. (editors), 1998, Housing Abandonment in Britain. (York:
University of York Centre for Housing Policy).
Lu, M., 1998, Analyzing migration decision making: relationships between residential satisfaction,
mobility intentions, and moving behavior. Environment and Planning A, 30 (8), pp. 1473–1495.
Murie, A., Nevin, B. and Leather, P., 1998, Changing Demand and Unpopular Housing. Working
Paper 4, (London: Housing Corporation).
Nevin, B., Lee, P., Phillimore, J., Burfitt, A. and Goodson, L., 1999, Measuring the Sustainability of
Neighbourhoods in Liverpool. Research report by the Centre for Urban and Regional Studies, The
University of Birmingham, for Liverpool First Partnership Group/Liverpool City Council.
Nevin, B., Lee, P., Goodson, L., Murie, A. and Phillimore, P., 2001a, Changing Housing Markets and
Urban Regeneration in the M62 Corridor. (Birmingham: CURS, The University of Birmingham).
Nevin, B., Lee, P., Murie, A., Goodson, L. and Phillimore, P., 2001b, The West Midlands Housing
Markets: Changing Demand, Decentralisation and Urban Regeneration. (Birmingham: CURS, The
University of Birmingham).
Perry 6, 1997, Escaping Poverty: From Safety Nets to Networks of Opportunity. (London: Demos).
Smith, R. and Merrett, S., 1988, The Challenge of Empty Housing: Void Control in the Public Sector.
Occasional Paper No 31. (Bristol: SAUS, The University of Bristol).
Wilson, W.J., 1987, The Truly Disadvantaged: The Inner City, the Underclass and Public Policy.
(Chicago: University of Chicago Press).
PART III
GIS and Urban Applications

10
Geographical visual information systems
(GVIS) to support urban regeneration:
design issues
Xiaonan Zhang, Nigel M.Trodd and Andy Hamilton
10.1
INTRODUCTION
Urban planning is complex. It combines data and opinion on social, political, economic and
environmental issues. In western societies, planners are responding to new challenges to make
planning more transparent, e.g. e-democracy, open government, and it has been argued that
any solution should increase participation in the process. New technologies could be employed
to promote dialogue and make the process less confrontational and easier to understand. The
research at the University of Salford is designing and building a prototype which links emerging
virtual reality (VR) and Internet technologies with (more mature) geographical information
technologies. These geographical visual information systems (GVIS) are being designed to
facilitate participation by multiple stakeholders in urban regeneration projects. In this chapter
we consider how such a prototype can adopt a framework based on theories of learning and
how that framework can be used to evaluate impacts of GVIS in urban planning.
The chapter introduces public participation in urban planning and reviews developments in
GIS, VR and Internet technologies. A learning theory is outlined based on Kelly’s theory of
personal constructions (1955) and is mapped to functions available in GIS, VR and Internet
technologies. Research is based on a case study of the Chapel Street Regeneration Project,
Salford. The study identifies two stages of planning process—developing and selecting
planning options—that offer opportunities for these technologies to engage stakeholders, and
in particular the public, with the planning process. Design options for a GVIS are considered
and methods of evaluating a learning system approach to GVIS are presented.
10.2
URBAN PLANNING AND PUBLIC PARTICIPATION
If a “city is a drama in time” (Geddes, 1905, in Cowan, 1998), then it is important to involve
every citizen in writing the scenario and playing active roles in the drama. The sense of
involvement not only gives citizens a greater meaning to their lives but also fosters a sense of
responsibility which is often lacking in modem society (Ingram, 1998). These opinions are
reflected in a movement from ‘planning for the public’ to ‘planning with the public’ (Klosterman,
1999; Roberts and Lloyd, 1999).
GEOGRAPHICAL VISUAL INFORMATION SYSTEMS (GVIS) 135
Whittick, (1974) defined public participation as ‘the means by which members of the
community are able to take part in the shaping of policies and plans that will affect the
environment in which they live’. Participation should involve the public putting forward ideas
and comments at the early stages of planning process (Sarjakoski, 1998). It should be a
continuous dialogue between the public and other stakeholders (McConnell, 1981). Rydin
(1999) argues that in most planning activities participation is still narrow and low-level. It is
usually undertaken in the late stages in the planning process and mostly based on consultation
documents and public meetings (DETR, 1998; Bickerstaff and Walker, 2001). There is an
argument for the public to be involved at earlier stages (Alterman et al., 1984).
Two key factors have been identified for effective public participation:
• Availability of, and Access to, Information

To participate in planning process, public require access to the necessary information
about the planning work and have opportunities to express their opinions. Lack of
information about planning activities and the limited opportunities for participation in
planning policy decisions have been highlighted as key problems (Barlow, 1995). Many
authors propose that GIS technology could be used to increase public access to information
and optimise their participation in the planning and policy-making process (e.g. Weiner et
al., 1995, Myers et al., 1995 and Nedovic-Budic, 2000). They envisage that people should
have access to information presented at a level they understand and through media with
which they are familiar.
• Communication and Interaction
Planning requires a dialogue between and among stakeholders. Dialogues between
public and professionals can be inhibited by the lack of a common terminology and the alien
jargon and alienating media used to convey ideas. Three-D representations of cityscapes
offer more natural fora to exchange ideas (Sarjakoski, 1998) and Al-Kodmany (1999) has
shown that visualisation tools allow residents to directly participate in the design of their
neighbourhood.
Nowadays, tools such as GIS, VR and Internet could be employed to achieve the goals of effective
participation.
10.3
DESIGN ISSUES 1: TECHNOLOGY
The last two decades have seen the dramatic development of information technologies and
their ubiquitous adoption. In this section, the current use of three information technologies,
namely GIS, VR and Internet, in urban planning are outlined and the integration of these
technologies is considered.
10.3.1
GIS
GIS is a special information system, as the data it handles are all referenced with geographical
location. It focuses on spatial entities and relationships and pays specific attention to spatial
analytical and modelling operations (Maguire, 1991). GIS are a powerful tool for storing and
handling geo-spatial data and have been adopted in many market sectors, such as
telecommunications and natural resource management.
GIS research has, since the early years, “moved from primitive algorithms and data
structures to the much more complex problems of database design, and the issues surrounding
the use of GIS technology in real applications” (Goodchild, 1992). Although GIS are adopted
widely, their potential remains unfulfilled (Batty, 1993; Douven et al., 1993).
Until recently GIS users have been limited to specialists and professional users. This is
attributed to two reasons:
• Low Accessibility
Expensive software and data, poorly catalogued and protected databases are barriers to
non-profit organisations and the general public (Nedovic-Budic, 1998). As a user group the
public requires tailored small and beautiful GIS by which they can solve some simple spatial
problems like where is…?, what is at location…? and what if…? by themselves.
• Weak Visualization
The user interface is crucially important as it is the only part directly seen and ‘is’ the
system for the user (Frank & Mark, 1991). Most commercial GIS-user interfaces are based
on the use of windows, icons, menus, and pointing devices (Egenhofer & Kuhn, 1999).
These interfaces are too often an impediment to effective problem solving or decision-
making (Medyckyj-Scott & Hearnshaw, 1993).
The emergence of web-based GIS increases access. Through the Internet, people can transmit
data and access tools to conduct analysis and create GIS presentations (Peng, 1997). Although
the Internet-based GIS creates many benefits for the public, such as the convenience of access
and the low cost, problems of tedious interface and difficulty of use are still not solved. Virtual
reality (VR) offer potential to facilitate public use of GIS tools as it increases the engagement
of the user by coming closer to natural ways of interacting with the world than would happen
with maps or other static models (Jacobson, 1992; Neves and Camara, 1999).
10.3.2
Virtual Reality
VR is a human-computer interface in which the computer creates a three-dimensional, sensory
immersing environment that interactively responds to and is controlled by the behaviour of the
user (Pimmentel and Teixeira, 1995). Its characteristics are response to user actions, real-time
3D graphics and a sense of immersion. By using multimedia, users can gain real time response
to their actions by graphic and sound in a virtual environment. An example is that as a ‘person’
walks along a street, the sound heard by that person would change continuously based on their
location relative to the sound source. The richness of this experience facilitates a users’
learning and understanding (Pont, 1993).
As with GIS, links between VR and the Internet have been developing. Virtual Reality
Modelling Language (VRML) is a standard format for the web (Rohrer and Swing, 1997) that
lets you quickly build virtual worlds incorporating 3D shapes, animation and sound effects. Since
1994, the most important developments in VR have not been in technologies, but in the
adoption of VR technologies and techniques to increase productivity, improve team
communication, and reduce costs (Brooks, 1999).
10.3.3
Internet
In the last decade of the twentieth century, the Internet emerged as a new information and
communication technology. At the end of 1999, 1 in 5 households in the UK had Internet
access compared with 1 in 20 only 2 years earlier (Corrigan and Joyce, 2000). It provides a
more efficient way for people to access information and disseminate their opinions as it not
restricted by time or physical distance. Corrigan and Joyce (2000) and Craig (1998)
demonstrate the usefulness of the Internet to improve the productivity of public services and
contact with government representatives. Local planning authorities are beginning to realise
the potential of the Web as a communication device. A number of examples exist where local
authorities have placed important planning documentation such as Structure Plans and
Development Plans on the web for public consultation (Carver and Peckham, 1999).
10.3.4
GVIS
It is a widely held view that integrating GIS, VR and Internet technologies can facilitate greater
and more effective participation in planning activity and thereby strengthen and democratise
the process. Research in this field is attracting considerable attention, e.g. the Centre of
Advanced Spatial Analysis (CASA) of UCL and the urban simulation team in Los Angles (Batty
et al., 1998; Dodge et al., 1998; Jepson et al., 1996). Although these demonstration systems
suggest that the technologies exist to provide functions for public participation the published
literature is notable for the absence of formal theory in the design of this type of systems. A
lack of theory may undermine the development of GVIS and inhibit longer-term progress
through rigorous evaluation. Formal theory helps to explain the success (or failure) of these
systems and better understand the likely impediments to future system. An objective of the
University of Salford research is to develop a robust framework for the design of GVIS.
10.4
DESIGN ISSUES 2: LEARNING SYSTEMS AND GVIS
10.4.1
Learning System Theory
Learning can be defined as the synthesis and analysis of information obtained through
communication and interaction. It can be argued that urban planning is a learning process as it
is information-rich, complex and benefits from stakeholders sharing a greater understanding.
The exchange of information and ideas between stakeholders creates an informal learning
environment. As such, the planning process can be considered within the framework of a
theory of learning. Of the many theories of learning, Kelly’s ‘personal construct theory’
(Kelly, 1955) is an appropriate approach for urban planning (Hamilton et al., 2001). Kelly
states that people look at their world through patterns that they construct and try to fit to the
real world. Without these patterns the world would make little sense to people. Patterns are
constructed based on an individual’s experiences, i.e. personal constructs. People change and
revise their patterns in order to explain better their view of the world. It is noted that personal
Figure 10.1 Effect of a learning system to enhance participation in planning.
experiences include interaction with both tangible and intangible features of the world (Kelly,
1955).
To relate the learning theory to urban issues, personal constructs are formed by making
sense of our direct experiences of life in the city, or indirect experiences through newspapers,
books, TV and other informing media. The social interaction also lead to the building of
personal constructs. Furthermore, when social interactions take place with a group, it is possible
for us to envision new and more creative ways of dealing with a problematic situation by actively
considering alternative constructs (Figure 10.1).
Three aspects of the learning system are needed to enhance participation in the planning
process, namely access, analysis, and comprehension (Zhang et al., 2001). The current system
to gain aaccess to information for a particular proposal can be a burden to the public because
the time involved may be incompatible with their lifestyle or incur a financial penalty due to
loss of earnings. Furthermore, gaining access does not improve matters unless the information
is presented in a way that they can comprehend. Experimenting with alternative scenarios is
another essential part of the learning system. To fully understand the planning issues, people
need to analyse the information they get. The alternatives also need to be analysed and
evaluated. In order to do that, public need tools to interact and refine the information. The
tools may not be as complicated as the ones for the professionals but at least they can achieve
some analysis functions. It has been postulated that allowing people to analyse planning
proposals followed by debate between public and other stakeholders can lead to greater
consensus in the final plan.
It has been observed that many people find it difficult to participate effectively in planning
systems because they lack the necessary skills and knowledge (Hamilton et al., 2001). A
learning system could be built to bridge the skill and knowledge gap identified (Figure 10.1).
On the one side, the planning process and participation issues are ‘the content’ of the system.
On the other side, technologies like GIS, VR and Internet provide functions that are needed to
build a learning system.
Figure 10.2 GIS, VR and Internet in Information Learning System (Source: modified from
Hamilton et al., 2001).
10.4.2
The Strengths and Limitations of Technologies to Enhance
Learning
In the light of the learning system theory, the strengths and limitations of each technology in
each of the aspects of the learning system have been evaluated (Figure 10.2). GIS allows people
to process information and detect spatial patterns and relationships. GIS is classified as high in
terms of analysis. It is not rated as highly for access or comprehension. VR is classified as high
for comprehension and the Internet provides an effective way for people to access information.
10.5
PROTOTYPE DESIGN AND EVALUATION
This chapter has identified design issues associated with the technologies and learning systems
methodology. It follows that a prototype is designed to combine the strengths of the
technologies and implement features of a learning system. The research prototype is being
designed around a common geospatial database for the Chapel Street Regeneration Project,
Salford.
Chapel Street is the main thoroughfare through the city of Salford (Figure 10.3). It is also
one of the main approaches to the centre of Manchester. Over the past thirty years Chapel
Street has declined as a commercial and retail centre and in 1998 the City Council launched a
major regeneration project. The project is funded by decline public and private sector
organisations and involves residential and business communities.
The Chapel Street case study provides a complex platform for the design and evaluation of a
learning system-based GVIS. The prototype will be developed in parallel with regeneration
plans and is intended to complement existing activity rather than replace it. The main benefit
of using an established urban regeneration project is the ability to compare old and new
approaches to participation. Before the prototype can be developed, however, it is necessary to
identify those aspects of the planning process that it will address.
Figure 10.3 Chapel Street Regeneration Corridor, Salford, UK.

10.5.1
Urban Planning System: User Needs
In keeping with the aim of this research the planning process was studied for key stages which
represent the best opportunities for the use of GVIS to support public participation. Kammeier
(1999) suggested that planning support systems should support clarifying the planning options,
simulating alternative proposals, assessing shortlisted projects and implementing the decision.
These coincide fairly closely with Skeffington’s observations (1969) that the main
opportunities for public participation in a local plan are at the stages of surveys of facts,
developing planning options and discussing favoured proposals. If these proposals are mapped
to Yeh’s model of the planning process then 4 stages can be identified which are most
conducive to public participation (Figure 10.4). By comparison GIS is most useful whilst
analysing the existing situation, and developing and selecting planning options because of the
need for spatial analysis in these stages. The stages of modelling the existing situation and
developing and selecting options offer greatest potential for using VR because it facilitates
presentation and interaction.
Combining these assessments allows us to identify 2 stages in the planning process that are best
suited to developing and testing a GVIS prototype, namely developing planning options and
selecting planning options (Figure 10.5).
Figure 10.4 GIS, VR and public participation in planning process (modified from Yeh, 1999).
Figure 10.5 Planning stages for developing and evaluating a GVIS.
10.5.2
Learning System Components
The prototype has been designed as 3 modules that each focus on an element of the learning
system.
10.5.2.1
Analysis Module
The analysis module uses spatial analysis functions to process data and support queries such as
where is…?, what is at location…? and what if…?. In particular the module will develop decision
support functions such as multi-criteria analysis.
10.5.2.2
Comprehension Module
A 3D model of Chapel Street Corridor has been built based on the same geo-spatial database of
the analysis module. In this module, user can navigate and query the planning area. Main
features are the provision of functions to switch between before and after views of the area
based on possible scenarios and the geo-referencing of multimedia information such as video
clips and panoramic photographs. The module is also designed to improve communication
between stakeholders by allowing them to comment on various aspects of the regeneration
plan, attach those comments to a visual object and to retrieve the comments of other
participants.
10.5.2.3
Access Module
The module is mainly about the access of the planning related information through the Internet
and/or Intranet. Users are allowed to access the information via a popular Internet browser. All
the functions of the former two modules would be transplanted/linked in the user interface
that means uses can access them by Internet or Intranet.
10.5.3
Evaluating GVIS as a Learning System
The question of whether or not new technologies improve complex tasks such as urban
planning is often very difficult to answer. The evidence is usually qualitative, may be
contradictory and rarely allows the system developer to make an objective evaluation. In
particular it is difficult to assess the impact of combined technologies. Implementing learning
system theory in the design of a GVIS prototype enables the research to reduce the system into
three objectives. These objectives transcend the technologies and therefore provide a useful
frame of reference for evaluating combined GVIS. Failure to achieve one of objectives can be used
to focus further research and development of the prototype system.
10.6
SUMMARY
Stakeholders, including the public, are being encouraged to participate in urban regeneration.
To do so they need to be able to contribute to and benefit from the planning process. Recent
studies have created virtual environments as a mechanism to facilitate the communication of
planning information and postulate on the near-future appearance of cityscapes. At the same
time community-based GIS researchers have argued that GI technologies can help to secure
active participation of leading individuals and groups at early stages in the planning process and
numerous authors promote the Internet as a forum for interaction between participants. In this
chapter the authors have presented the case that integrating GIS, VR and Internet technologies
can increase participation in planning process. More importantly, applying learning systems
theory to the design of GVIS creates a framework that can be used to optimise participation.
10.7
REFERENCES
Al-Kodmany, K., 1999, Using visualization techniques for enhancing public participation in
planning and design: process, implementation, and evaluation. Landscape and Urban Planning, 45,
pp. 381–392.
Alterman, R., Harris, D. and Hill, M., 1984, The impact of public participation on planning. Town
Planning Review, 55 (2), pp. 177–196.
Barlow, J., 1995, Public Participation in Urban Development: the European Experience. (London: PSI).
Batty, M., 1993, Using geographic information systems in urban planning and policy-making. In
Geographic Information Systems, Spatial Modelling and Policy Evaluation, edited by Fischer, M. and
Nijkamp, P., (New York: Springer-Verlag), pp. 51–69.
Batty, M., Dodge, M., Doyle, S. and Smith, A., 1998, Modelling virtual environments. In
Geocomputation: A Primer, edited by Longley, P., Brooks, S.M., McDonnell, R., and MacMillan,
B., (Chichester: John Wiley & Sons), pp. 139–161.
Bickerstaff, K. and Walker, G., 2001, Participatory local governance and transport planning.
Environment and Planning A, 33, 431–451.
Brooks, P.P., 1999, What’s real about virtual reality?. IEEE Computer Graphics and Applications, 19
(6), 16–27.
Carver, S. and Peckham, R., 1999, Internet-based applications of GIS in planning. In Geographical
Information Systems in Planning: European Perspectives, edited by Stillwell, J., Geertman, S., and
Openshaw, S., (New York: Springer-Verlag), pp. 371–390.
Corrigan, P. and Joyce, P., 2000, Reconnecting to the public. Urban Studies, 37 (10),
pp. 1771–1779.
Cowan, R., 1998, The people and the process. In Introducing Urban Design: Interventions and
Responses, edited by Greed, C., Roberts, M., Longman.
Craig, W.J., 1998, The Internet aids community participation in the planning process. Computers,
DETR, 1998, Guidance on Enhancing Public Participation in Local Government, (London: The Stationery
Office).
Dodge M., Doyle S., Smith A., and Fleetwood S., 1998, Towards the virtual city: VR & Internet
GIS for urban planning, paper presented at the Virtual Reality and Geographical Information Systems
Workshop, Birkbeck College, London.
Dollner, J. and Hinrichs, K., 1998, An object-oriented approach for integrating 3D visualization
systems and GIS. Computers and Geosciences, 26 (1), pp. 67–76.
Douven, W., Grothe, M, Nijkamp, P., and Scholten, H., 1993, Urban and regional planning
models and GIS. In Diffusion and Use of Geographic Information Technologies , edited by Masser, I.
and Onsrud, H., NATO ASI Series, (Dordrecht: Kluwer Academic).
Egenhofer, M. and Kuhn, W., 1999, Interacting with GIS. In Geographical Information Systems, 2nd.,
edited by Longly, P., Goodchild, M.F., Maguire, D.J., Rhind, D.W., (New York: John Wiley &
Sons), pp. 401–412.
Frank, A., and Mark, D., 1991, Language issues for geographical information systems. In
Geographical Information Systems: Principles and Applications, 1st., edited by Maguire, D.J.,
Goodchild, M.F., Rhind, D.W., (London: Longman), pp. 227–237.
Geddes, P., 1905, Civics: as applied sociology. Sociological papers 1904, (London: Macmillan).
GeoWorld, 1999, Industry trends: GIS industry outlook 2000—the birth of a new millennium.
GeoWorld, December, http://www.geoplace.com/gw/1999/1299/1299ind.asp.
Goodchild, M.F., 1992, Geographical information science. International Journal of Geographical
Information Systems, 6 (1), pp. 31–45.
Hamilton, A., Trodd, N., Zhang, X., Fernando, T., Watson, K., 2001, Learning through visual
systems to enhance the urban planning process. Environment and Planning B: Planning and Design,
Oct., 2001.
Ingram, R., 1998, Building virtual worlds: a city planning perspective. Online Planning Journal,
http://www.casa.ucl.ac.uk/planning/articles3/vcity.htm.
Jacobson, B., 1992, The ultimate user interface. Byte, 17, pp. 175–182.
Jepson, W., Liggett, R. and Friedman, S., 1996, Virtual modelling of urban environments. Presence,
5 (1), pp. 72–86.
Kammeier, H., 1999, New tools for spatial analysis and planning as components of an incremental
planning-support system. Environment and Planning B: Planning and Design, 26 (3), pp. 365–380.
Kelly, G.A., 1955, The Psychology of Personal Constructs, 2 Vols., (New York: Norton).
Klosterman, R., 1999, The what if? Collaborative planning support system. Environment and Planning
B: Planning and Design, 26 (3), pp. 393–408.
Maguire, D.J., 1991, An overview and definition of GIS. In Geographical Information Systems: Principles
and Applications, 1st., edited by Maguire, D.J., Goodchild, M. F., Rhind, D.W., (London:
Longman), pp. 9–20.
McConnell, S., 1981, Theories for Planning, (London: Heinemann).
Medyckyj-Scott, D. and Hearnshaw, H., (editors), 1993. Human Factors in Geographical Information
Systems, (London: Belhaven Press).
Myers, J., Martin, M. and Ghose, R., 1995, GIS and neighbourhood planning: a model for
revitalizing communities. Journal of Urban and Regional Information System Associations, 7 (2),
pp. 63–67.
Nedovic-Budic, Z., 1994, Effectiveness of geographic information systems in local planning. Journal
of the American Planning Association, 60 (2), pp. 244–263.
Nedovic-Budic, Z., 1998, The impact of GIS technology. Environment and Planning B: Planning and
Design, 25 (5), pp. 681–692.
Nedovic-Budic, Z., 2000, Geographic information science implications for urban and regional
planning. URISA Journal, 12 (2), pp. 81–93.
Neves, J.N. and Camara, A., 1999, Virtual environments and GIS. In Geographical Information Systems,
2nd., edited by Longley, P.A., Goodchild, M.F., Maguire, D.J., and Rhind, D.W., (New York:
John Wiley & Sons), pp. 557–65.
Peng, Z., 1997, An assessment of the development of Internet-GIS, presented at 1997 ESRI User
Conference.
Pimentel, K. and Teixeira, K., 1995, Virtual Reality: Through the New Looking Glass, 2nd., (New
York: MacGraw-Hill).
Pont, P., 1993, Applied virtual reality. In Virtual Reality in Engineering, edited by Warwick, K.,
Gray, J. and Roberts, D., (London: Institute of Electrical Engineers).
Roberts, P. and Lloyd, G., 1999, Institutional aspects of regional planning, management, and
development: models and lessons form the English experience. Environment and Planning B:
Planning and Design, 26 (4), pp. 517–531.
Rohrer, R. and Swing, E., 1997, Web-based information visualization. IEEE Computer Graphic and
Application, pp. 52–59.
Rydin, Y., 1999, Public participation in planning. In British Planning-50 Years of Urban and Regional
Policy, edited by Cullingworth, B., (London: Athlone), pp. 184–197.
Sarjakoski, T., 1998, Networked GIS for public participation emphasis on utilizing image data.
Computers, Environment and Urban Systems, 22 (4), pp. 381–392.
Skeffington Report, 1969, People and Planning. (Report of the committee on public participation in
planning), (London: HMSO).
Weiner, D., Warner, T., Harris, T., Levin, R., 1995, Apartheid representations in a digital
landscape: GIS, remote sensing and local knowledge in Kiepersol, South Africa. Cartography and
Geographic Information Systems, 22 (1), pp. 30–44.
Whittick, A., 1974, Encyclopaedia of Urban Planning. (New York: McGraw-Hill).
Yeh, A. 1999, GIS and Urban Planning. In Geographical Information Systems, 2nd., edited by Longley,
P., Goodchild, M.F., Maguire, D.J., Rhind, D.W., (New York : John Wiley & Sons),
pp. 877–888.
Zhang, X., Trodd, N. and Hamilton, A., 2001, Geographic & visual information system (GVIS) to
support advanced urban planning: a research framework. In Proceedings of the First International
Conference of Postgraduate in the Build and Human Environment, pp. 622–633.
Plate 1 The map applies incremental standard deviations above the mean to define crime hotspot
legend thresholds. The crime data mapped is robbery (June–August 1999). The map was produced
using a bandwidth of 207m (K order 5).
Plate 2 An incremental mean approach for defining legend thresholds provides an appealing method
for describing concentrations of crime. The crime data mapped is robbery (June–August 1999). The
map was produced using a bandwidth of 207m (K5).
Plate 3 Gi* LISA statistic output (grey areas) derived from the robbery (117m (K2) bandwidth)
quartic kernel density estimation surface. The LISA output matches very closely to the ranges above
the 3 mean threshold, and is useful in adding definition to our original continuous surface crime hotspot
map, indicating the level at which hotspots can be more clearly distinguished against other levels of
crime concentration.
Plate 4 Crime hotspot for all crime, robbery, residential burglary and vehicle crime. The maps show the application of
the incremental mean approach for defining crime hotspot legend thresholds. The method is also sensitive to map design
flexibility. where users require the need to view hotspot output at different scales.
Plate 5 Canary Wharf modelled in Canoma.
Plate 6 Canoma model of Rosemary Gardens, displayed in Viewpoint.
Plate 7 Correlation of Peripherality with respect to population against Peripherality with respect to GDP
(NuTS 2).
Plate 8 Peripherality index with respect to GDP by lorry (NUTS 3).
11
Using measures of spatial autocorrelation to
describe socio-economic and racial
residential patterns in US urban areas
Andrea I.Frank
11.1
INTRODUCTION
Residential patterns are investigated using an exploratory approach with GIS-based thematic
maps and spatial statistics. The research aims to contribute to the debate whether geopolitical
fragmentation enhances spatial segregation of households based on racial and socio-economic
characteristics. Using 1990 US census data for three key variables (income per capita, percent
black and percent hispanics), measures of spatial association serve as quantitative index to
describe the level of homogeneity or heterogeneity of residential neighborhoods. Statistical
results are mapped and the correlation of significant homogeneous clusters with political
entities is assessed. The results show considerable variation of household cluster sizes
depending on the geopolitical structure as well as the population composition of the urban
area. Also, cluster sizes and locations are correlated with municipal boundaries in some of the
areas; this supports the proposition that the location of segregated communities is influenced
by political and policy frameworks.
11.2
BACKGROUND
Within the fabric of the city, households of similar socio-economic, racial or ethnic
background tend to cluster spatially forming distinct neighbourhoods and communities.
Human geography and urban planning have a long tradition in observing and analysing the
spatial pattern of residential arrangements and the change of such arrangements in time and
space as a means to develop and test explanatory models of social interaction and urbanisation
processes. It is from such observations that theories of human ecology and the concentric zonal
model (Burgess, 1925) of urban structure were developed in the 1920’s by scholars from the
Chicago School (e.g. Park et al., 1925; Park, 1936; Hawley, 1950). In the context of movements
such as sustainability and social justice, knowledge of patterns of residential differentiation is
vital to raise awareness of the spatial isolation of disadvantaged households lacking access to
health care, jobs and transportation.
Recent research (e.g. Foster, 1997; Pendall, 2000) suggests that the spatial sorting of social
and racial groups in the US is not necessarily based on free choice, but is induced and facilitated
in part by local government policies on land use, zoning, taxation and restrictive policies like
minimum floor space requirements for new homes. As a result of the influence of local
USING MEASURES OF SPATIAL AUTOCORRELATION TO DESCRIBE 147
governments on the residential mosaic we might expect in turn that the socio-economic
structure and patterns mirror the underlying spatial political structure. In other words,
household characteristics might be changing significantly when crossing jurisdictional
boundaries. This study continues the tradition of research in urban residential patterns looking
specifically at linkage of segregation and geopolitical structures. The objective is to a) evaluate
and compare different patterns and levels of residential segregation in US urban areas and b)
test the hypothesis of the association of political structures with patterns of residential
segregation.
The paper consists of three parts. The first part briefly reviews theories of urban residential
segregation and the potential influences of geopolitical structures on these patterns. The
second part discusses the analysis approach, which is based on spatial statistics and pattern
analysis techniques. Spatial statistics is a helpful tool to assess the distribution of household
types quantitatively using spatial association measures. The level of integration (co-location)
and segregation of socio-economic and racial groups is analysed using multivariate spatial
correlation, whereas the correspondence of jurisdiction boundaries with the extent of statistically
significant patches of similar households is evaluated through visual assessment.In the third
part, the analysis is applied to eight US urban areas with different political structures. Spatial
clusters of socio-economic indicators are evaluated and compared with the areas of local
political entities.
11.3
URBAN RESIDENTIAL MOSAIC AND POLITICAL STRUCTURE
Residential segregation or congregation (depending one’s point of view) is primarily based on
socio-economic status, ethnicity, race, and to a lesser degree life-cycle status, lifestyle, and age
(Knox, 1994). A range of theories on residential segregation exists. On one hand, the
phenomenon is seen as a result of the complex dynamics of social interaction. Lasting social
contacts are based on commonalties, such as hobbies, political and cultural values.
Commonalties often derive from similar educational backgrounds, occupations and socio-
economic status. The shared experience of university life or facing the same issues at work
often forms the basis of a friendship. On the other hand, social interaction is greatly facilitated
by physical closeness (Johnston, 1982); this holds for significant proportions of the population
despite technology that renders distances increasingly obsolete such as telecommunications,
improved personal mobility and the Internet (Dodge and Kitchin, 2000). Thus, it seems
reasonable that people locate in proximity of others with whom they share common interests
and values. Benefits of residential segregation go beyond the convenience of proximity. Suttles
(1970), for example, argued that residential segregation leads to minimisation of inter-group
conflict, maximisation of political voice, and an increased degree of social control. Ethnic
segregation helps to preserve cultural values, provides a haven of moral, spiritual and practical
support and reduces the impact of discrimination (e.g. Boal, 1976).
Residential segregation is a matter of fact in US metropolitan areas; more interesting is
therefore where certain groups locate within the metropolitan fabric and how different
neighbourhoods relate to each other. In general, the location of different social and racial
groups within a metropolitan area is determined through an approximate match of land value
and resources available to a group. This has led to the development of a generic model of urban
structure with poor, black or minority urban centres and wealthy (white) urban fringes. Yet,
the urban residential mosaic is not static and local conditions vary. Downtown neighbourhoods
are being gentrified by young urban professionals displacing less affluent former residents, for
example, while inner city industrial land is converted to housing and so forth.
Residential self-sorting may be facilitated by municipal and school district boundaries
(Weiher, 1991). The statement resonates the view offered by public choice theory (e.g.
Tiebout, 1956). According to Tiebout (1956), households will choose and locate in the
municipality, which offers the closest match to the households preferred level of services, costs
and environmental conditions. Thus, municipalities will over time attain a population that is
relative homogeneous by taste. A factor analysis of demographic variables led Heikkila (1996)
to conclude that the various municipalities of Los Angeles indeed resemble Tieboutian clubs of
relative homogeneous population groups. As households with similar characteristics often
share taste and preference, we may assume relative high levels of homogeneity by household
type, economic class, and race as well.
Municipalities are not passive players in this game but compete like ‘municipal firms’ for
residents. To the extent that municipalities influence land values and access to their territory
by means of policy and zoning, they have a direct impact on the socio-spatial pattern of the
metropolitan area. It may be fiscally advantageous for municipalities to specialise. Newly
incorporated cities and municipalities, that cater to specific population groups (e.g. the
retirement communities of the Sun Belt), or land uses (e.g. City of Industry, CA) are examples
for this specialisation. In effect municipalities and local governments act as territorial containers.
Through specific local government legislation certain land uses or population groups can be
offered preferential treatment or denied access. The size of jurisdictions may be of importance
for the facilitation of the segregation process. Access control becomes more difficult, the
larger the area. Hence, larger jurisdictions are less likely to be highly homogeneous by class,
race or along any other dimension (Madison cited in Sack, 1986:147). From this we might
conclude that in urban areas with lots of small jurisdictions, segregation might be more easily
attained than in urban areas with large jurisdictions.
To summarise, the reasons for residential segregation are diverse ranging from different
tastes and preferences, racial prejudice and discrimination to economic barriers. There is some
indication that the self-sorting process of households is facilitated by the differential conditions
in local governments. As there is only a certain proportion of an urban population that fits into
each grouping (e.g. the Asian community, the wealthy, the black entrepreneurs, etc.), spatial
containers and jurisdictions ideally should be small for better fit. It is expected that in urban
areas with many local government entities: (i) segregation patterns reflect the small-scale
political structure; and (ii) boundaries of clusters of similar households are often matching
local government boundaries. In contrast, in urban areas with few government entities (fewer
choices), clusters of households with similar race, ethnicity or social status are likely to form
larger clusters that not necessarily match up with the boundaries of jurisdictions.
11.4
SPATIAL PATTERN ANALYSIS
Describing and comparing the residential mosaic of different cities is not a simple task. In a
research context, social area analysis (a technique based on factor analysis) has successfully
been used to generate generic descriptions of the socio-spatial structure of a wide range of
cities. The results showed that residential differentiation is primarily dependent on three
dimensions: socio-economic status, life cycle and ethnicity/race1 of households. The technique
is difficult to interpret, however, and critics voiced concerns over the dependency of analysis
results on the underlying spatial structure. Moreover, phenomena in space tend to be

interdependent (e.g. Tobler, 1970; Gould, 1970). Households with similar characteristics
(socio-economic status, age, ethnicity etc.) are likely to be located near each other, and are
thus in a statistical sense dependent or autocorrelated. This spatial association of household
locations casts further doubt on the validity of social area analysis results as variable
autocorrelation may bias the results of the underlying factor analysis.
Scholars have been aware of the tendency of spatial autocorrelation in geographic data
(Tobler, 1970; Gould, 1970) and have developed a variety of approaches to deal with the
phenomenon. One response is to determine the degree of interdependence via measures of
autocorrelation and allow for the necessary corrections to avoid inferential errors (Cliff and
Ord, 1981; Anselin, 1996). A second approach is to accept autocorrelation as a general
property found in variables of any system, man-made or natural, observed over time
(temporal autocorrelation) and/or space (spatial autocorrelation) (Legendre, 1993). The
spatial associations of variables are in fact valuable indicators of life supporting processes and
organisation and spatial association measures can be viewed as a quantitative index for the
description of spatial patterns and processes (Cliff and Ord, 1981; Goodchild, 1986;
Legendre, 1993; Shen, 1994).
Patterns in space can be defined based on the heterogeneity or homogeneity of adjacent
observations and the size and arrangement of patches of similar values in relation to each other.
Spatial association measures, which determine the degree of heterogeneity or homogeneity of
variables in space, can therefore be used to determine the degree of segregation of households
in an urban area.
Overall two aspects of urban residential patterns are investigated. On one hand,
characteristics of today’s socio-economic pattern in urban areas are explored. On the other hand,
the correlation between boundaries of local government entities and the spatial arrangements
and size of segregated residential clusters is evaluated. In a step-by-step analysis, socio-economic
patterns are first visualised on thematic maps and second described both qualitatively as well as
quantitatively via the level of spatial association of three key indicators of residential
differentiation: income, race and ethnicity. The measure of spatial autocorrelation indicates the
level of segregation for each variable. Positive spatial association indicates that similar values of
a variable under study are clustered in space; this would be an indicator of segregation.
Negative spatial autocorrelation indicates that similar values of a variable under study are
dispersed to a degree that is unlikely under randomisation. No spatial autocorrelation means
the pattern observed is likely to be random. Third, a visual inspection of map overlays of
significant clusters of segregated households with municipal boundaries addresses the question
of the correlation of the political structure and identified residential segregated clusters. The
analysis compares residential patterns of eight different urban areas.
The analysis is conducted using two software applications working in tandem via a loose
coupling approach (Anselin, 2000): SpaceStatTM (Version 1.90), a spatial statistics software
and ArcView™ (Version 3.1), a vector-based GIS from the Environmental Systems Research
Institute (ESRI). Geographic co-ordinate data of enumeration district centroids and topology
are exported for use in a spatial weights matrix and statistical analysis results are fed back into
the GIS for visualisation.
1 In the US context White, Black or Asian, represent racial population categories, whereas Hispanic
indicates ethnic heritage; Hispanics may be of any race.
Table 11.1 Urbanized Area complexity/fragmentation and racial/ethnic composition
11.4.1
Analysis Unit
The census designated Urbanized Area (UA) rather than the Metropolitan Statistical Area
(MSA) is used as primary analysis unit. MSAs, which are based on county boundaries,
frequently include large areas of undeveloped land. This could distort the residential pattern
analysis and was thus found unsuitable for the analysis. The census designated UA was
introduced to help distinguish between urban and rural land use and is a standard enumeration
unit since 1960. A UA comprises of a minimum of 50,000 persons, one or more central places
and surrounding urban fringe territory. Fringe territory must have a population density of at least
1000 persons/square mile to be included (GARM, 1994). As a consequence, the UA describes
best the extent of a densely populated area with a territorial definition that is largely
independent of administrative boundaries such as county or municipal boundaries.
Case study UAs were selected for comparison along two aspects: political structure and
population mixture. The political structures of the selected UAs’ range from very simple with
a few governments to ones that stretch over several states and consist of 30+ local
governmental entities. The selected UAs are located in different regions of the US displaying
different racial and ethnic composition. Southwestern UAs (e.g. Albuquerque) have a high
level of Hispanics in the overall population, whereas Eastern UAs (e.g. Richmond) house a
higher percentage of black population (Table 11.1). The analysis was conducted on midsize
urban areas with a population between 250,000 and 1,000,000.
11.4.2
Data
For the analysis, 1990 US census data was extracted from ArcData Online2. The database
provides geographic and attribute data. At the time of download data were based on the
Tiger® 1995 products of the US census. Block groups serve as the spatial units for the
2 ArcData Online from ESRI (Environmental Systems Research Institute) can be accessed via the
Internet from URL: http:www.esri.com/data/online/tiger.
Figure 11.1 Distance-based Neighbourhood Relationship.
representation of the residential pattern. Block groups are the smallest enumeration unit for
which the US Census provides socio-economic and demographic data on a general basis.
Boundaries of independent local government units (towns, townships, parishes, villages,
cities), counties and states are used to describe the geopolitical structure of the areas. Three
variables were used to portray the socio-economic pattern:
• Income per Capita (INPRCAP),

• Percent Blacks (PERBLACK),
• Percent Hispanics (PERHISP).
11.4.3
Spatial Statistics
The evaluation of the level of spatial association of variables values from different enumeration
districts is influenced by the definition of nearness or proximity. For object-oriented data
models like the TIGER® census files, data are aggregated for each enumeration district and
conceptualised at discrete locations. Thus, in order to assess the spatial autocorrelation of
income per capita, for example, the aggregate value of one census block group is compared to
the equivalent income per capita values of neighbouring block groups.
For this study, neighbourhood is defined as a) dichotomous, assuming that boundaries
between ethnic, racial or socio-economic enclaves are relative sharply defined and b) distance-
dependent, assuming that segregated communities are not infinitely large. In other words, every
spatial unit whose centroid is within a specified distance of another spatial unit’s centroid is
considered a neighbour and every spatial unit with a centroid outside the specified distance is
not. The concept of this neighbourhood relationship is depicted in Figure 11.1. For block
group A (centre) and the distance r, only one block group is defined as neighbour (dark grey
polygon); for distance 2r, 21 block groups fall into the neighbourhood range.
The neighbourhood definition results in small block groups having more neighbours and
large rural block groups at the fringe of the analysis regions being eliminated, which is
desirable. Perry (1929) has defined the spatial extent of a neighbourhood unit as an area
enclosed by arterial, high-volume roads and served by a school. Considering the relative low
density of US urban areas and the block group size, an idealised neighbourhood is assumed to
be not larger than a circle with r=2000m (approx. 4.8 square miles). The distance r=2000m
is also greater than the average distance between centroids for all UAs, which is important
methodologically as distance cut-offs smaller than the average will lead to a non-continuous
analysis surface and missing values in the spatial weights matrix.
The level of spatial autocorrelation is expressed by the global Moran’s I coefficient, which is
calculated in SpaceStat™ as outlined in Equation 11.1, whereby N depicts the number of
observations, wij represents an element in the spatial weights matrix corresponding to an
observation pair i, j, while xi, xj are observations for locations i and j (with mean µ) and S0 is a
scaling constant of the sum of all observation pairs (Eq. 11.2).
(11.1)
(11.2)
For any number of polygons N, a Moran’s I coefficient approaching +1 indicates strong
positive autocorrelation and • 1 indicates negative autocorrelation. A coefficient of • l/(N-l)
indicates zero autocorrelation or a random pattern. The Moran’s I is a global measure of
autocorrelation, which is strictly speaking only valid under the assumption of spatial
stationarity. As the assumption of spatial stationarity is unlikely to hold for regions with several
dozen and more spatial observations, a local measure of spatial association—the local Moran’s
Ii (Equation 11.3)—was introduced to detect local instability and nonstationarity. Values zi, zj
represent deviations from the mean, while the summation over j includes only neighbouring
values j ε Ji. The average of all li will equal the global Moran’s I. Extreme local Moran values
represent outliers with strong deviation from the mean of the overall sample.
(11.3)
Moran scatterplot maps are used to depict the analysis results; they are deemed instrumental in
the exploration of spatial patterns (Anselin, 1996; Anselin and Bao, 1997). A scatterplot map
is based on a statistic that can be visualised as the slope of a straight line that indicates the level
of global association. When the statistic is decomposed into four types of association, the
lower left and the upper right quadrant indicate positive spatial autocorrelation, i.e. the
presence of similar (low or high) values in neighbouring locations. The upper left and lower
right quadrants indicate negative spatial association or the presence of dissimilar values in
neighbourly locations. The results are mapped by assigning different colour schemes to values
from the four quadrants. Outliers, i.e., values significantly above or below the mean can be
identified and shown in additional maps. Scatterplot maps together with maps of significant
local indicators of spatial association are key components of the analysis.
11.5
CASE STUDY RESULTS
The residential pattern of eight different UAs were analysed and compared. As expected,
significant clustering of households was discernible from mapping the racial and ethnic
variables. Spatial association for the income variable is not as significant. Furthermore,
patterns vary depending on the political structure and population composition.
These differences are illustrated below contrasting two UAs: Tucson, with a simple political
structure and a large Hispanic population and Harrisburg, with a complex political structure
and a high percentage of blacks in the population.
11.5.1
Tucson
Figure 11.2 shows the UA boundary with a thick black line. Jurisdiction over the area is shared
by only 4 different governments: the City of Tucson, South Tucson an enclave within the City
of Tucson, Oro Valley town in the Northwest and Pima County.
Figure 11.3 shows the location of block groups with higher or lower than average levels of
Hispanic households in the Tucson UA. The scatterplot map shows that the UA is divided in
two parts: one which is dominantly Hispanic (high-next-to-high values of PERHISP), and one
which is dominantly non-Hispanic (low-next-to-low values of PERHISP). The large grouping
of dark grey block groups in the south of the UA depicts the part that is dominated by Hispanic
households. While circa 20% of the residents of the entire UA claim Hispanic origin, the
proportion of Hispanics in this contiguous patch of block groups is 40 to 95% indicating
significant homogeneity for ethnic origin. This is clearly not random. Block groups that are
highly Hispanic next to block groups with little or no Hispanic population are shown in middle
grey shades, whereas the light grey block groups depict areas for which the proportion of
Hispanics is consistently below 19%.
Figure 11.4 shows again two large clusters. The cluster of wealthy households (high next to
high INCPRCAP values) is located in the north of the UA—outside the city in Pima County.
The per capita income in the Tucson UA in 1990 was US$ 13,105. The average per capita
income for the block groups in the cluster located in Pima county is over US$ 28,000. The
cluster of poorer households (low next to low INCPRCAP) is located to the south in the city
of Tucson and South Tucson. The boundary between high and low-income areas is not as clear
as for the ethnic variable.
Only 3.4 % of the UA population in Tucson are black; most of them live north of the
Hispanic neighbourhood but within the city limits. It seems that the black community is part of
mixed income and racial area that forms a buffer between the rich white and poorer Hispanic
population.
The clustering, i.e. spatial autocorrelation, of households of similar ethnicity and income is
confirmed by the high global Moran’s I coefficients for both the PERHISP (=0.859) and
INCPRCAP (0.551) variables (see Table 11.2 below). Negative multivariate spatial correlation
of the PERHISP and INCPRCAP variables points also to the separation of high-income from
Hispanic households. The Hispanic households in the southern part of Tucson are not only
Hispanic but also much poorer on average.
Maps and statistics show a stark pattern of economic and ethnic segregation. The clarity of
the political structure reflects the large-scale political structure of the UA. Residents do not
have many choices except for Oro Valley town and South Tucson. There is only the choice to
Figure 11.2 Spatial political structure of Tucson UA.

live inside or outside the city limits of Tucson. The choice of wealthier households to locate
outside the city proper is common in US metropolitan areas and consistent with theories of white
flight from redistributive urban taxes.
Figure 11.3 Moran scatterplot for Hispanics in Tucson (r=2000m).

11.5.2
Harrisburg
The political structure of the Harrisburg UA is complex, consisting of 38 local government
entities including townships, counties, towns and villages (Figure 11.5). Towns and villages
denoted by letters a—r on the map are fairly compact and a good proportion of the populations
lives in townships.
Overall clusters of adjacent block groups with significant homogeneous population groups
(race, economic status or ethnicity) are much smaller in comparison. Figure 11.6 depicts four
clusters for income per capita (INCPRCAP). There are two larger contiguous areas to the east
of the central city of Harrisburg and two to the west. Clusters are not as clearly delineated as
in the Tucson UA. As a result the pattern for income is much less spatially autocorrelated than
in Tucson, which is reflected in the lower Moran’s I coefficient (0.206) for INCPRCAP. The
pattern for PERBLACK, however, is spatially correlated as indicated by a Moran’s I coefficient
of 0.668 (Figure 11.7 and Table 11.2). Blacks are mostly confined within the boundaries of the
central city of Harrisburg. This is also the poorer area. The pattern is almost a perfect reversal
to that of the income pattern in Figure 11.6. Multivariate spatial correlation of the INCPRCAP
and PERBLACK variables is negative with • 0.50 pointing also to the spatial segregation of
Blacks and high income households. The correlation of boundaries of racially segregated areas
with the boundaries of jurisdictions is considerable.
11.5.3
Other Results
When comparing spatial autocorrelation of the variables across the eight different case studies
autocorrelation levels for the ethnic/racial variables show a significant geographical bias
Figure 11.4 Moran scatterplot for Income per capita in Tucson UA (r=2000m).
(Table 11.2). All five Eastern UAs show significant and in the case of Oklahoma City and
Richmond very high positive autocorrelation for the PERBLACK variable. In five of the eight
urban areas the autocorrelation of Hispanics is a dominant force in shaping the residential
pattern. Three of these urban areas are located in southwestern states, which have a high
proportion of Hispanics in the urban population. The other two UAs with a significant
Moran’s I coefficient for Hispanics are Oklahoma City (0.658) and Providence-Pawtucket (0.
635). These two UAs display significant autocorrelation for PERBLACK and PERHISP.
Interestingly, the level of global spatial association for INCPRCAP is less significant and
only for four of the eight test cases is the Moran’s I coefficient greater than 0.4. One of them is
in the Tucson UA, where a large cluster of high-income households is located in the north of
the UA (Figure 11.4).
In two of the UAs, Richmond and Akron, the PERHISP variable approaches a zero level of
autocorrelation or random distribution of households. This result may be an artefact of the overall
low level of Hispanics in the population (Table 11.1). The differences in spatial structure of
ethnic and racial groups is likely to be related to the overall population mixture, i.e. the
proportion of blacks or Hispanics as part of the overall urban population.
The resolution of the pattern, i.e. the size of relative homogenous groupings of households
seems to vary as well. In older, northeastern urban areas groupings are smaller and more
numerous, whereas in the newer southwestern areas the socio-spatial structure tends to
consist of few large districts. This may be a function of the different political structures, which
may influence residential choice and foster population segregation. Political structures on the
whole tend to be more small-scale and fragmented in the North and Northeast of the US and
large-scale in the Southwest. In any case, distinctly high-income block groups are repeatedly
found outside but abutting central city limits.
Figure 11.5 Spatial political structure of Harrisburg UA

Based on the small sample, no clear relationship can be determined between the level of
spatial association (as in the global Moran’s coefficient) and the complexity of the political
structure. It was expected that groupings of block groups of similar households would be
smaller in the UAs with a more complex urban political structure and therefore that the
Moran’s coefficients would be less significant as well. In the case of INCPRCAP, Harrisburg,
Oxnard-Ventura and Providence Pawtucket (all with complex small scale political structures)
show little spatial association as expected but Albuquerque has a similarly low Moran’s I
despite a much simpler political structure. Reversibly, Oklahoma City has an equally high
positive spatial autocorrelation for INCPRCAP than Tucson, despite a much more complex
political structure. The relationship for PERHISP and PERBLACK are also inconsistent,
however when separated by geographic region a negative relationship between the number of
political entities and the level of spatial association can be shown for the Northeastern UAs and
Figure 11.6 Moran scatterplot for INCPRCAP in Harrisburg (r=2000m).
Figure 11.7 Moran scatterplot for PERBLACK in Harrisburg (r=2000m).

PERBLACK and the Southwestern UAs and PERHISP. This is in support of the reasoning that
more complex structures facilitate sorting into smaller entities that might not be adjacent and
therefore result in a low global spatial association.
11.6
PLANNING IMPLICATIONS AND FUTURE RESEARCH
The study revisits residential pattern research using widely available census data and a spatial-
statistical analysis approach. The goal was twofold: a) to investigate contemporary patterns of
Table 11.2 Level of Spatial Association for sample Uas.
residential segregation in US urban areas, and b) examine whether the location and size of
segregated communities coincide with local political entities. This research and its results are of
interest to urban planners and planning agencies for it may provide relevant insight into the
location of minority households and patterns of segregation.
In terms of segregation and residential patterns, the research findings indicate that global
spatial association for race and ethnicity tend to be higher than spatial association of income.
The degree of spatial association is somewhat dependent on the proportion of racial or ethnic
minorities in the area and geographically biased. In addition, multivariate analysis shows that
locational overlap of high-income with PERBLACK or PERHISP tends to be negative, i.e.
Hispanics and Blacks tend to reside in more deprived areas. Segregation by economic status, it
seems, is not as significant a factor than race or ethnicity or as Harvey stated segregation by
ethnicity and race tends to weaken with higher income.
In many instances, cluster dimensions and location indicating similar ethnic, racial or high-
income households correlate significantly with governmental entities. For example in the
Tucson UA, where a large group of high-income block groups is found outside the city limits or
Harrisburg, where the central city holds most of the black population. This correlation
supports the hypothesis that local government policy has some influence on residential choice
and land use. Further research at the local level is needed to pursue the reasons for these socio-
economic boundaries and divisions. This research could be in form of additional case studies to
explore geographically determined trends further, or it could be in form of detailed local,
ethnographic studies. As clusters of high-income households that coincide with jurisdictional
boundaries may be indicators of discriminatory development practices, an examination of the
local policies for discriminatory practices and regulations may be helpful.
There is value not only in the content or results, however. In contrast to standard thematic
mapping, the approach takes into account spatial interdependencies and evaluates the level of
chance at which a certain pattern may occur. Although, unfortunately, the issue of a
predetermined spatial framework persists by using census block groups3, credibility and
interpretation of patterns is nonetheless improved over mere visual observation as spatial
autocorrelation measures such as global and local Moran’s I coefficients are used to verify the
visual impression of clustering. The spatial autocorrelation measures also can give indications of
the level of segregation and integration of certain population groups via the multivariate spatial
autocorrelation analysis.
The approach uses widely available census data and is easy to replicate for other places.
Despite the methodological sophistication, results remain accessible by planners, policy
makers and lay people as the statistics are translated into maps. Thus, the approach provides a
feasible and easy to use tool for quick assessments of the segregation and residential patterns in
urban regions. Planning agencies may use the approach and maps to outline potential zones of
conflict and discrimination or evaluate whether certain neighbourhoods face problems in terms
of access to public transport, health care or exposure to environmental risks.
11.7
REFERENCES
Anselin, L., 1996, The Moran scatterplot as an ESDA tool to assess local instability in spatial
association. In Spatial Analytical Perspectives on GIS, edited by Fischer, M. et al. (London: Taylor &
Francis), pp. 111–125.
Anselin, L., and Bao, S., 1997, Exploratory Spatial Data Analysis: Linking SpaceStat and ArcView.
In Recent Developments in Spatial Analysis, Chapter 3, edited by Fischer, M. and Getis, A. (Berlin
and New York: Springer), pp. 35–59.
Anselin, L., February 25, 2000, SpaceStat™ software for spatial data analysis and related services,
Distributed by Biomedware Inc. Copyright Luc Anselin 1999–2000, WEBSITE, Available:
http://www.spacestat.com [Accessed May 18, 2000].
Burgess, E.W., 1925, The growth of the City: An Introduction to a Research Project. In The City,
edited by Park, R. et al., (Chicago: Chicago University Press).
Cliff, A.D., and Ord, J.K., 1981, Spatial Processes: Models and Applications, (London: Pion).
Dodge, M. and Kitchin, R., 2000, Mapping Cyberspace, (London: Routledge).
Foster, K.A., 1997, The Political Economy of Special-Purpose Government, (Washington:Georgetown
University Press).
GARM—Geographic Areas Reference Manual, 1994, U.S. Department of Commerce. Economics and
Statistics Administration. Bureau of the Census. ONLINE. Available: http://www.census.gov:
80/geo/www/garm html [Accessed May 1999].
Goodchild, M.F., 1986, Spatial Autocorrelation. In Concepts and Techniques in Modern Geography
(CATMOG), 47, (Norwich:Geobooks), pp. 1–56.
Gould, P.R., 1970, Is Statistix inferens the geographical name for a wild goose? Economic Geography,
46, pp. 439–448.
Hawley, A., 1950, Human Ecology: A theory of Community Structure, (New York: Ronald Press).
Heikkila, E., 1996, Are Municipalities Tieboutian clubs? Regional Science and Urban Economics, 26,
pp. 203–226.
Johnston, R.J., 1982, The American Urban Systems: A Geographical Perspective, (New York: St Martin’s
Press).
Knox, P., 1994, Urbanization: An Introduction into Human Geography, (Englewood Cliffs: Prentice
Hall).
Legendre, P., 1993, Spatial Autocorrelation: Trouble or New Paradigm?, Ecology, 74 (6),
pp. 1659–1673.
3 Dependence on the spatial framework could be weakened by a resampling of values into a grid
structure.
Park, R., 1936, Human Ecology, American Journal of Sociology 42 (2), pp. 1–15.
Park, R., et al. (eds.) 1925, The City, (Chicago: University of Chicago Press).
Pendall, R., 2000, Local Land Use Regulation and the Chain of Exclusion. Journal of the American
Planning Association, 66 (2), pp. 125–142.
Perry, C.A., 1929, The Neighborhood Unit, In Regional Study of New York and its Environs, VII,
Neighborhood and Community Planning, Monograph One (New York: Regional Plan of New York),
p. 88.
Sack, R D., 1986, Human Territoriality: Its Theory and History, (New York: Cambridge University
Press).
Shen, Q., 1994, An Application of GIS to the Measurement of Spatial Autocorrelation. Computers,
Suttles, G., 1970, The Social Construction of Communities, (Chicago: University of Chicago Press).
Tiebout, C.M., 1956, A pure theory of local expenditures. Journal of Political Economy, 64 (5),
pp. 416–424.
Tobler, W., 1970, A computer movie simulating urban growth in the Detroit region, Economic
Geography, 46 (Supplement), pp. 234–240.
US Bureau of the Census, 1990 Census of Population, Social and Economic Characteristics, Urbanized
Areas, 1–6, (Washington: Government Printing Office).
US Bureau of the Census, 1990 Census of Population, Urbanized Areas, U.S. Supplementary Report,
1–2, (Washington: Government Printing Office).
Weiher, G.R., 1991, The Fractured Metropolis: Political Fragmentation and Metropolitan Segregation,
(Albany: SUNY Press).
12
Georeferencing social spatial data and intra-
urban property price modelling in a data-
poor context: a case study for Shanghai
Fulong Wu
12.1
INTRODUCTION
Despite enormous progress in GIS techniques, the lack of geo-referenced spatial data is still a
major constraint to the development of more accurate and in-depth/sophisticated spatial
analysis in developing countries. Even when some spatial data are available in theory, the use
of geo-referenced data is constrained due to two reasons. First, the digital data may not be
available to researchers (especially for those are not affiliated to a local institution) for the sake
of confidentiality. For example, in the case of China, population census is not released to the
public at the sub-district, i.e. Street Office level or lower levels. Even at the sub-district level,
the data are not publicly available except for the total population. However, the sub-district is
probably equivalent to wards rather than enumeration districts (EDs) in UK, which is still too
coarse for the purpose of examining subtle intra-urban spatial variation. Ideally, disaggregated
data should be obtained for urban and rural settlements because residential quality varies at the
scale of neighbourhoods. In case of China, these units are residents’ and villagers’ committees.
Second, even when some spatial data are available, it is difficult to directly apply them in
spatial analysis because they are collected from different spatial units. For example, population
data are usually collected on the basis of subdistricts, while the data of firms are associated with
postal codes and land use are interpreted from parcel-based images or maps. The need for
cross-referencing different spatial data layers then becomes an issue. In general, the availability
of spatial data in developing countries is undesirable. A more spatially sensitive methodology is
therefore critical for empirical research in this data-poor context. This project uses
disaggregate property data from Shanghai, PRC to study intra-urban spatial variation of
property prices. This is achieved through geo-referencing the address information and
interpolation of residual surfaces, which highlights the way forward in developing high-
resolution spatial data for decision-support in developing countries.
12.2
GEO-REFERENCING PROPERTY DATA
From the social scientist’s perspective, the key issue of spatial representation is geo-referencing
socio-economic phenomena (Martin, 1999), through for example, associating the property
address with the national grid reference (Martin and Higgs, 1996). Over time there has been a
massive improvement in the resolution and quality of geo-referencing socio-economic data
GEOREFERENCING SOCIAL SPATIAL DATA AND PROPERTY PRICE MODELLING 163
(Martin, 1999). The resolution for population data increased from 1-km grid squares in 1970s,
to digitised ED boundaries including about 400 persons in the early 1990s, and now to 0.1 m
Address-Point information. Thanks to the spatial database functions, GIS is naturally the ideal
tool for real estate valuation. Higgs et al. (1992) map the distribution of properties in council
tax bands, based on a survey of property prices. Using GIS analysis techniques, Longley et al.
(1994) assess the accuracy of council tax banding in Cardiff. GIS-based property information
system has been developed by integrating various sources of spatial data (Wyatt, 1997). The
digital geo-referencing and map product such as ADDRESS-POINT and Land-Line.Plus
produced by the Ordnance Survey (OS) together with population census at the Enumeration
District (ED) level have made it possible to develop property price models (Lake, et al., 1999,
2000; Orford, 1999; 2000). Through GIS operations and processing functionality, a wide range
of neighbourhood attributes describing the character of a property’s surroundings can be
inferred. For example, based on DEM, viewsheds can be calculated to assess the visual impact
of amenity and nuisance land uses (Lake, et al., 1998).
In contrast, the spatial analysis of property price variation is difficult in developing countries
because of the lack of high resolution spatial data. Efforts have been made by using indirect
measurements from aerial photographs and remote sensing imagery to study the settlement
structures (e.g. de Bruijin, 1992) but these are only within the remit of photogrammetric
engineering and land cover studies. In the case of Chinese cities, there have been a few
attempts to overcome the data constraint by deriving information directly from aerial
photographs and satellite imagery (such as Landsat TM images) (Wu and Yeh, 1997; 1999; Yeh
and Li, 1997; Li and Yeh, 2000). Ortho-adjusted aerial photographs so far provide the most
accurate information on land uses. On a 1:10,000 scale, roads and large buildings are
identifiable. Landsat TM at a resolution of 30×30 metres can be used to identify the
conversions from the rural to urban land. This is particularly useful in finding construction
sites because of significant changes in the spectral characteristics of land surface. Through
making reference of factory locations and centroid of urban sub-districts, the population and
firm potentiality surfaces were derived through spatial interaction formula. Based on these
variables, intra-urban land use probability models of Chinese cities have been developed (Wu,
1999). However, the potential of the rich socio-economic information source has not been
fully explored. Recently, intrametropolitan location of foreign investment firms has been
modelled by geo-referencing their post-codes (Wu, 2000). Registered on the same co-ordinate
system, this allows the cross-reference of foreign investment with the land use information
derived from aerial photographs and Landsat TM imagery. Previous studies suggest the geo-
referencing of socio-economic data could provide high potential for developing intra-urban
models.
12.3
INTRA-URBAN PROPERTY PRICE MODELLING
Traditionally, urban property price has been modelled under the hedonic price framework
developed by Rosen (1974), in which the property is regarded as a bundled commodity
consumed in a competitive market. The complication of housing price modelling is that the
attributes are not limited to the product itself because of the existence of externalities. This
has led to the inclusion of attributes at different levels of resolution which can be conceptually
divided into structural characteristics, accessibility characteristics, and neighbourhood
characteristics (Olmo, 1995). The latter is often a determinant of the dynamics of local
housing market. While empirical studies in the 1970s used very simplistic measures of locational
features (such as the distance to CBD), the advance of spatial database technology is driving
recent research towards more sophisticated way of capturing the characteristics of the
environment in which the property is located. For example, Lake et al. (1998) used GIS
viewshed analysis to include the visual impact of various land uses on property prices. Orford
(1999) developed explicit measures of the attributes at the levels of property, street, and
community. GIS is naturally an ideal useful tool to derive various ‘spatial’ attributes through
spatial operations (such as distance proximity, buffer, overlay, network analysis, visibility
analysis). By cross-referencing different layers, new attributes can be generated, for example
from the Census, unit postcode areas, and other utility databases. In a sense, the utilisation of
GIS can help develop more context-sensitive property price modelling (Orford, 2000). The
list of attributes that can be generated from GIS operations seems endless, although this can be
a problem when determining which ones are significant.
The rationale behind using various locational variables is that, in order to estimate a proper
contribution of the attributes to property prices, one has to consider local housing market
dynamics. Orford (2000, p. 1644) argues that, ‘if the hedonic house price function is to
generate estimates that properly reflect the implicit price of attributes, the model specification
must capture sufficiently the dynamics of the local housing market. In particular, the
specification must reflect the variations in supply and demand of housing attributes and also the
ad hoc nature of the valuation process’. The problem underlying the use of ordinary least-
squares (OLS) regression in the estimation of the hedonic property price model is the
existence of heteroscedasticity and spatial autocorrelation in the data, which violates the
assumptions of independent, identically distributed errors. In other words, the contribution of
a factor (such as the size of the property) to the price may vary spatially—in accordance with
different dynamics of a local housing market. Based on this understanding, Orford (2000)
proposed a multilevel modelling approach. The specification of multilevel models includes the
attributes at the property, community (wards) and EDs levels. The model therefore allows more
explicit considerations of spatial variation and interaction among different levels of attributes.
In particular, it is important to justify that the existing boundaries of wards and of EDs
delineate the sub-markets of housing. In other words, the development of a property price
model requires a much deeper understanding of a particular housing market which is different
from place to place.
12.4
METHODOLOGY
The classic specification of the hedonic price model includes the attributes of property
structure and locational characteristics (Can, 1990; Olmo, 1995):
(12.1)
where i=1 to N, N is number of properties, pi=the price of ith property; sk=kth attribute of
property structure; Lj=jth locational characteristics of the ith property; α, βk, and
δj=regression coefficients, εi=disturbance term that is only related to the ith property.
This hedonic approach, however, does not consider the interaction between property
attributes and neighbourhood characteristics. Oxford (2000) used a multilevel modelling
approach to capture the influence of local community on housing prices. However, it is

difficult to apply this approach to Shanghai because the definition of ‘community’ is unclear.
The urban districts designated by Shanghai municipal government for the administrative
purpose are larger than the wards defined in Britain. In Cardiff, Orford (2000, p. 1651–1652)
finds that the ‘communities’ used by the city council are defined by major thoroughfares,
rivers and railway line, and that these communities are often used by estate agents as a basis for
defining residential neighbourhoods. In contrast, urban districts in Shanghai range from 0.25
million (Huangpu District) to 1 million (Yangpu District) (SSB, 1998). It is unlikely that the
districts of this size could form a unitary housing market. Below the urban districts are sub-
district offices. In 1997, there are totally 99 sub-district offices in the main urban area.
However, the sub-district offices are traditionally not a substantial unit for urban management
because of centralised administration. Data at the sub-urban district offices are sparse and
difficult to collect. The central point is that it would not be very useful to include ad hoc
defined administrative units into property price modelling. What is needed, however, is to
show the city-wide distribution of the deviation of property price at the resolution of individual
properties.
In this study, we define the structure attributes as those that are only related to the property
itself. All locational features are not included, because these attributes are difficult to measure
and the spatial structure of these attributes are unclear. The regression thus gives an error term
that might be spatially autocorrelated. As a result, the regression cannot be interpreted in a
way it is normally used. But the purpose is to examine the residuals once the structure variables
are controlled.
(12.2)
where, x, y=location co-ordinates of the property i.

The method is therefore to separate the environmental influence from that of property
structure, while the interaction of environmental and structural attributes is ignored. The
underlying assumption is that the adjustment of the property price would not be dependent on
the property but rather on the environmental amenity. However, if needed, it would be
possible to classify residuals according to the type of property that generates the residual. As a
result, there will be a series of residual surface according to property types. However, certain
property types may be located in part of the city and therefore it is likely that the surface will
only cover part of the city.
The influence of spatial factors can be reflected through the deviation of property price from
the metropolitan-wide ‘standard’ price that based on the structure of the house (Figure 12.1).
The deviation of the observed price from the price predicted through the attributes is the
‘residual’ of regression, which is composed of the ‘unknown’ disturbance in property
transaction and a locational premium (positive or negative). There is no reason to believe that
the disturbance in property transaction should present a systematic spatial variation.
In order to examine the spatial distribution of ‘residuals’, the residuals are interpolated from
the sample points into a surface. This interpolation of residual surface is similar to the
development of Digital Terrain Model (DTM). The residuals can be interpolated because they
reflect the contribution from the environment, which is seen as continuous over space. Certain
spatial barriers such as the river and major railway lines may separate communities into distinct
sub-markets. However, this is not modelled in the current study.
Figure 12.1 The regression line of property price and positive and negative residuals.
There are a variety of interpolation methods (Burrough and McDonnell, 1998). The most
common one is the inverse distance weighted (IDW) interpolation method, which is used to
generate the residual surface from the point coverage of the regression model. The formula for
calculating the output cell value is:
(12.3)
where x1, y1=co-ordinates of the centre of the cell; c=the value of interpolated residual; d
[(x,y), (x1,y1)]=the distance between the centre of the cell to the point location; s=distance
decay parameter.
The EDW interpolation can also specify a neighbourhood range. This assumes that the
contribution of the samples outside this range should be neglected. We use different ranges to
examine the contribution of environmental factors at the local (400 m) and community (1,000
m) resolution. When the distance decay parameter is specified as zero, the distance decay
within the neighbourhood is not considered. In essence, the cell value produced by the
interpolation becomes the average of the residuals within the moving kernel. This is
conceptually plausible, as this will in fact ‘smooth’ out the random effect of property
transaction and leave the ‘regional’ variation (Figure 12.2).
The fixed neighbourhood range only generates the grid cells where at least one sample point
exists within the kernel. From the property price modelling point of view, it means we cannot
predict the property value where no sample is available. The missing sampling locations are
mainly located in the suburbs of the city. However, it is known that because of the large size of
land plot and relatively homogenous price variation, it is possible to use the sample from a
longer distance than in the inner cities. Alternatively, the minimum number of points can be
Figure 12.2 The interpolation of residual surface from the property sample by specifying a moving
kernel of 400 m or 1,000 m influence.
specified to interpolate the residual surface. When the specified minimum points cannot be
found in the kernel, the search radius (equivalent to the neighbourhood range) is enlarged to
ensure that the minimum number of points is used. If the interpolation of fixed neighbourhood
range is applied to the observed property price, this gives the average property price within the
moving kernel.
12.5
CASE STUDY OF SHANGHAI
12.5.1
Data Source
This project aims to find the spatial differentiation of Chinese cities in the transition towards a
more-market oriented economy. Because of the lack of high-resolution population data in
China, it is extremely difficult to study intra-urban spatial differentiation. However, the
information of property price can be used to examine the urban spatial changes which are
becoming more differentiated in the newly established real estate markets.
The main information comes from the Shanghai Housing Market released by the Shanghai
Real Estate Exchange Centre (SREEC). The data consist of a total of 3,207 records of the
residential properties for sale in August of 2000. The size of the sample was then reduced to 1,
604, using stratified sampling according the total number of properties available for sale in
each district. This was due to the heavy workload of geo-referencing their locations. The
distribution of properties in the urban districts of Shanghai is shown in Table 12.1. The
location of each property was geo-referenced by hand through its address because no digital
product is available. This is a very tedious and time-consuming process.
Table 12.1 The distribution of the sample of properties in Shanghai, 2000.
Figure 12.3 Identification of the location from the street address.
12.5.2
Map Registration and Transformation
The basic location information is based on Shanghai Atlas (1997), which is in turn mainly
derived from the 1:10,000 topographic map produced by Shanghai Survey Academy in 1996.
The maps from the Atlas were scanned, registered with control points and then rectified. The
Atlas also provides an address index, which allows a quick finding of a street through its name,
and some key road numbers along each street. Therefore, the precise address location is
unknown and has to be inferred from nearby key addresses (Figure 12.3).
This process is subject to error and distortion. In order to assess the degree of error, some
tests are carried out to infer known addresses through this method. First one researcher was
given an address and asked to infer ‘blindly’ its possible location using the nearby key
addresses. Then, the actual location is revealed and the distance of deviance was recorded. The
accuracy of inferring a location by the post-code area and sub-districts is also assessed. The size
of post-code areas increases from the central area to rural areas and therefore the error
produced will depends on the distribution of the location. But from the test the manual
reference of street address is within the accuracy of 50 to 100 metres. In some extreme cases,
the variation can be as large as 500 metres. Compared with the geo-referencing product such as
ADDRESS-POINT, the result is not accurate but much better than other possible solutions in a
data-poor situation.
The study area consists of 14 districts. Each map sheet in the Atlas covers one urban district.
The co-ordinates derived from the referencing address on each map sheet thus are measured in
map projection. These co-ordinates need to be transformed into a single common co-ordinate
system in order to view the metropolitan wide distribution of property price. The common co-
ordinate system is referred through the large-scale map provided by the Shanghai Survey
Academy. Because the map sheets are derived from ortho aerial photographs, ‘rubber
sheeting’ is not required. The registration of map sheets only requires the transformation of
map units into the units of the common co-ordinate system. For each map sheet, the following
steps are applied:
1. establish the tics (control points) on the basis of obvious land marks such as road
intersection and bridges;
2. measure the co-ordinates of these tics in map unit as well as real-world unit;
3. create a new coverage using the tics with their real-world co-ordination;
4. transform the co-ordinates of map projection into the newly created coverage that is
measured in the co-ordinates of all districts.
Finally, all transformed coverages are merged into a single coverage of property location. On
the basis of property-ids, the property attributes are then joined with locations through point-
attribute table (PAT) to form the database of properties, which are then used for property
price modelling. Figure 12.4 shows the location of the properties. In Pudong, a new district
created in 1990, some properties are developed near the newly built roads, and so it is not
possible to refer the location through their addresses. Consequently, these properties are
dropped out from the analysis and the number of properties in the database is reduced to 1,
431.
12.5.3
Price Modelling
In this study, the use of the hedonic price approach is not to provide the relationship between
the property price and the property attributes. Rather, the regression here is to control the
effect that only depends on the property itself, thus separating the spatial effect through the
examination of the residuals. The deviation of the observed price from the predicted price is
referred to as a residual. The residual can be due to ‘unknown’ disturbance purely related to
the property itself and location premium. The former should not present a systematic spatial
variation and the latter can be visualised to show the intra-urban spatial changes because the
properties are ‘geo-referenced’. The spatial variation can be imagined as a rough surface. This
is because the unknown and random noise also contributes to the shape of the surface. The
point-based observation is then interpolated into a surface. The model, to our current
knowledge, is the first of its kind to be developed in a data-poor context. While the hedonic
Figure 12.4 The distribution of sampled properties in Shanghai, 2000.
price approach has been widely applied, only through geo-referencing the observed data is it
possible to visualise the spatial variation of location premium.
The regression uses the physical and property right attributes of housing. The first category
includes the number of rooms (ROOM), the number of receptions (REPT) and the second
category includes two dummy variables SRT and PRT, which represents respectively the full
property right and partial property right in a ‘commodity housing’ market. The full property
right means the property can be sold openly in the housing market without any restriction,
while the partial right housing means the use right can be sold but the owner does not have the
full right of the land premium associated with the house. If both dummy variables are zero, the
property belongs to the ‘public housing’, which cannot be sold in housing market. These
properties can be exchanged if certain compensation is paid to their occupants.
Table 12.2 suggests that the physical attributes of housing are good predicator of housing
price with R2 of 0.619, which the property right alone only partly explains the housing price,
with a low R2 of 0.052. The full attribute model gives the R2 of 0.619 and is a controlled
prediction. The residual of the model is calculated accordingly.
12.5.4
Residual and Property Price Surface
The residual interpolation was undertaken using ESRI Arc/Info GRID module. The output
table of statistical analysis in SPSS is joined with the PAT of the processed property sample
coverage. The residual value is used as the ‘spot’ item for the interpolation command. Two
Table 12.2 Multiple regression of housing price (dependent variable: housing price in 10,000
Yuan).
Notes: B=coefficient of regression; t=t-values; ** significant at the 0.01 level and above; *
significant at the 0.05 level.
neighbourhood ranges are used to represent price variation at different scales. The command
used for generating the surface is:
Residual_surface=pointinterp (housing_sampling_cover, residual_value,

cell_size, idw, decay_parameter, plateau_parameter, the
neighbourhood_shape, size of neighbourhood)
The cell size of the surface is 100 m, the decay parameter and plateau parameter (within the
distance the weighting of points is constant) are set to zero. The size of neighbourhood is set
respectively as 400 m and 1,000 m. Figure 12.5 and 12.6 show the generated residual
surfaces.
The surfaces clearly show that the areas in the Xuhui District and part of Jing’an district,
where the previous French International Settlement was located, have above average positive
residuals. The residential quality of these areas is higher than in such places as Yangpu and
Zhabei District where there is a high proportion of industrial land. A second clustering of positive
residual is near North Shichuan Road in the Hongkou District. For the new urban district,
Pudong District, because there are not enough sample points, the residual surface covers only
some locations and does not show the trend of variation.
To measure the spatial correlation, the Moran I is used as a spatial statistical indicator
reflecting the degree of spatial autocorrelation (Goodchild, 1986). The indicator is used to
reveal the pattern of clustering of the same value at adjacent cells. The measure provided by
the ARC/INFO Moran I function does not describe the clustering at the level of spatial object.
The absolute concentration of value within a kernel generates a Moran I close to unity, while a
more even distribution than can be expected by chance gives a value below zero.
Figure 12.5 The residual surface at the resolution of 400 m neighbourhood range (unit: 10,000
Yuan).
Figure 12.6 The residual surface at the resolution of 1,000 m neighbourhood range (Unit: 10,000
Yuan).
Although the absolute value of Moran I may not correspond to a fixed scale of spatial clustering,
the indicator can be used to assess the degree of spatial autocorrelation. Table 12.3 shows the
Moran I for the four surfaces.
For the interpolation of the residual surface using the minimum point within a radius, the
command used is:
Table 12.3 The Moran I measurement of the interpolated residual and price grids.
Residual_surface=idw (housing_sampling_cover, residual_value, barrier_cover,

decay_parameter, radius, radius_parameter, minimum_points, cell_size)
The decay parameter is set to 2; the radius parameter sets the initial size of the search radius
(the default is 5 times the cell_size); minimum points are set to 8. This command generates
the interpolated surface, which covers the area where the sample points are not available.
The surface, presented in Figure 12.7, clearly reveals the clustering of the positive
contribution of environment factors to housing prices. The darker areas have positive
residuals, which means the environment enhances the property price; the lighter areas
(including Huangpu) have the negative residuals, which means the areas are less attractive in
terms of property value. The positive area are generally distributed along with the circular
road, clustering at some particular locations with high residential quality; these areas to a
certain extent conform to the legacy of high-class residential areas before the establishment of
socialism in 1949. The negative areas are distributed in the inner areas (such as the Huangpu)
and industrial areas (such as Putuo, Zhabei, Yangpu), and peripheral districts (Baoshan and
Minhang) where infrastructure is less developed. The situation in the Pudong new district is
less clear because the value is interpolated over a longer distance (than 400 m).
The same approach is applied to the property price to derive the average price within a
moving kernel. The size of the kernel is set to 400 m and 1,000 m respectively for
neighbourhood and community levels. Figure 12.8 shows the average price in terms of
standard deviation from the mean. The darker values indicates the average price above the
mean, while the lighter shades reflects the opposite. It can be seen that property prices vary
within each district. In general, the property price in peripheral locations and industrial areas
such as the Baoshan, Putuo, and Yangpu District is lower than in Jing’an, Luwan, and Xuhui
District. In the inner city (the Huangpu District) property price is low because of the high
proportion of old ‘alley housing’ (terrace housing). For the range of 1,000 m kernel, the
concentration of high property price becomes even more obvious (Figure 12.9). Because of
internal price variation, the average price in Hongkou District is lower than some places in the
Jing’an, Xuhui, Luwan and Changning District. If we compare the prices of Zhabei and
Hongkou District, it is obvious that the mean housing price is different. This is consistent with
the understanding of the different images of these two districts.
12.6
CONCLUSION
This work is the first of its kind to explore social-spatial differentiation in the context of a
developing country where socio-economic data at a high spatial resolution are usually difficult
to obtain. Through geo-referencing the address information, it is possible to construct a database
of the distribution of activities at the highest possible spatial resolution. Under the data-poor
context, small-area data (such as these based on census tract and enumeration district) will likely
take many years to develop, while rapid social and economic changes require that up-to-date
information be provided to support decision-making. However, to speed up data capturing, it
Figure 12.7 The residual surface interpolated using the varying radius with a minimum number of
eight sample property locations (unit: 10,000 Yuan).
Figure 12.8 The average property price surface with the 400 m neighbourhood range. (Unit: 10,
000 Yuan).
is critical to automate the geo-referencing process. Manual processing is still too time
consuming to provide a feasible solution. A possible way forward is to register all the addresses
and use ‘address matching’ to automate the process. This may prove to be very costly in a data-
poor context because this requires a large amount of financial support. Alternatively, it is
possible to use ‘key addresses’ as shown in this study to infer the location of address between
them. Although this is subject to error and distortion, this approach is more practical and low-
Figure 12.9 The average property price surface with the 1,000 m neighbourhood range (unit: 10,
000 Yuan).
cost in the development of disaggregated spatial data. Once the key addresses are digitized, the
new address can be created using ‘interpolation’. In fact, the new address between the key
addresses can be created during the data automation and thus do not need to be stored.
Regarding the property price modelling, in the data poor context, the first step should be
the visualisation of the distribution of positive and negative environmental contribution to
property prices. The study shows that by geo-referencing the location of properties and the
development of a spatial property database, it is possible to quickly use the interpolation
method to display the spatial variation of property price distribution. The results of the study
reveal that property market in Shanghai shows the feature of spatial differentiation. Future
research should use various additional sources of data about the distribution of facilities,
services, population, and infrastructure to study the reasons for spatial differentiation of the
housing market. Methodologically, the automation of geo-referencing socio-economic data
will be important, as this will open up a rich source of information in addition to government
statistics.
12.7
ACKNOWLEDGEMENT
The study is supported by the Nuffield Foundation project (SGS/LB/0488).
12.8
REFERENCES
Can, A., 1990, The measurement of neighbourhood dynamics in urban house prices. Economic
Geography, 66, pp. 254–272.
de Bruijin, C.A., 1991, Spatial factors in urban growth: towards GIS models. ITC Journal, 4,
pp. 221–231.
Goodchild, M.F., 1986, Spatial Autocorrelation: Concepts and Techniques in Modern Geography 47,
(Norwich: Geo Books).
Higgs, G., Longley, P. and Martin, D., 1992, Analysing the spatial implications of the Council Tax:
a GIS approach. Mapping awareness and GIS in Europe, 6, pp. 42–46.
Lake, I.R., Lovett, A.A., Bateman, I.J. and Day, B., 2000, Using GIS and large-scale digital data to
implement hedonic pricing studies. International Journal of Geographical Information Sciences, 14,
pp. 521–541.
Lake, I.R., Lovett, A.A., Bateman, I.J. and Langford, I.H., 1998, Modelling environmental
influences on property prices in an urban environment. Computers, Environment and Urban Systems,
22, pp. 121–136.
Li, X. and Yeh, A.G.O., 2000, Modelling sustainable urban development by the integration of
constrained cellular automata and GIS. International Journal of Geographical Information Sciences, 14,
pp. 131–152.
Longley, P., Higgs, G. and Martin, D., 1994, The predictive use of GIS to model property
valuations. International Journal of Geographical Information Systems, 8, pp. 217–235.
Martin, D., 1999, Spatial representation: the social scientist’s perspective. In Geographical
Information Systems: Principles, Techniques, Management and Applications, edited by Longley, P.A.,
Goodchild, M.F., Maguire, D.J. and Rhind, D.W. (Chichester: Wiley), pp. 71–80.
Martin, D. and Higgs, G., 1996, Georeferencing people and places: a comparison of detailed
datasets. In Innovations in GIS 3, edited by Parker, D. (London: Taylor and Francis), pp. 37–47.
Olmo, J.C., 1995, Spatial estimation of housing prices and locational rents. Urban Studies, 32,
pp. 1331–1344.
Orford, S., 1999, Valuing the Built Environment: GIS and House Price Analysis, (Aldershot: Ashgate).
Orford, S., 2000, Modelling spatial structures in local housing market dynamics: a multilevel
perspective. Urban Studies, 37, pp. 1643–1671.
Rosen, S., 1974, Hedonic prices and implicit markets: product differentiation in pure competitions.
Journal of Political Economy, 72, pp. 34–55.
State Statistical Bureau (SSB), 1998, Shanghai Statistical Yearbook 1998, (Beijing: China Statistical
Press).
Wu, F., 1999, Polycentric urban development and land use change in a transitional economy: the
case of Guangzhou. Environment and Planning A, 30, pp. 1077–1100.
Wu, F., 2000, Modelling intrametropolitan location of foreign investment firms in a Chinese city.
Urban Studies, 37, pp. 2441–2464.
Wu, F. and Yeh, A.G.O., 1997, Changing spatial distribution and determinants of land
development in China’s transition to a market economy: the case of Guangzhou. Urban Studies,
34, pp. 1851–1879.
Wu, F. and Yeh, A.G.O., 1999, Urban spatial structure in a transitional economy: the case of
Guangzhou, China. Journal of the American Planning Association, 65, pp. 377–394.
Wyatt, P.J., 1997, The development of a GIS-based property information system for real estate
valuation. International Journal of Geographical Information Sciences, 11, pp. 435–450.
Yeh, A.G.O. and Li, X., 1997, An integrated remote sensing and GIS approach in the monitoring
and evaluation of rapid urban growth for sustainable development in the Pearl River Delta.
International Planning Studies, 2, pp. 193–210.
PART IV
GIS and Rural Applications

13
Accessibility to GP surgeries in South
Norfolk: a GIS-based assessment of the
changing situation 1997–2000
Andrew A.Lovett, Gisela Sünnenberg and Robin M.Haynes
13.1
INTRODUCTION
A guiding principle of the National Health Service in the UK is the ideal of equal access to
health services for those in equal need. Health services, however, are inevitably concentrated
in certain places, and are consequently more accessible to nearby residents than people living
further away. Inequality in accessibility due to distance is, of course, only one element of the
problem of equal access to health services. Access to services can be represented as a
continuum, across which many social, economic and cultural factors may contribute (Ricketts
and Savitz, 1994), but the difficulty of overcoming distance tends to be a dominant factor in
rural regions (Joseph and Bantock, 1982). Poor physical accessibility is known to reduce the
use of services, and may lead to poorer health outcomes (Joseph and Phillips, 1984; Haynes,
1986; Jones and Bentham, 1997). Low utilization of primary care services is of particular
concern because of the gateway role of general practitioners (GPs) in terms of referral to
hospitals. There is a considerable body of research on variations in physical accessibility within
rural areas of the UK (e.g. Moseley, 1977; Higgs and White, 2000), and the problem is still an
important concern for several government agencies (Department of Environment, Transport
and the Regions, 2000; Countryside Agency, 2001).
Developments in Geographical Information Systems (GIS) and digital map databases have
made it possible to calculate measures of physical accessibility such as travel time in a more
automated and sophisticated manner than was previously practical (e.g. Higgs and White, 1997;
Naude et al., 1999). Another innovation has been the use of GIS to assess accessibility by public
transport, taking into account the spatial distribution of bus routes and frequency of services
(O’Sullivan et al., 2000; Higgs and White, 2000). Both these types of GIS techniques were
used in a study of access to primary health care services in East Anglia during Autumn 1997
(Lovett et al., 2000). The results of this analysis included evidence of an inverse care law; in
the sense that the rural populations with the poorest accessibility were also more deprived and
with evidence of higher health needs than other populations.
Since 1997 there have been considerable changes to public transport services in rural areas of
the UK. In the Spring 1998 Budget the government announced substantial additional funds to
support rural bus services and later that year the concept of Rural Transport Partnerships was
introduced to encourage local community transport initiatives. Both these developments have
resulted in new services for rural populations in a region such as East Anglia, but there have
Figure 13.1 The study area.
also been instances where services have been withdrawn or altered only a few months after
being introduced. It was therefore thought appropriate to assess what had been achieved and
examine the extent to which variations in accessibility still remained. To achieve this objective,
an investigation of changes in accessibility to GP surgeries between Autumn 1997 and Autumn
2000 was undertaken for the district of South Norfolk, working in co-operation with the local
authority and South Norfolk Primary Care Group (PCG).
South Norfolk covers an area stretching from several suburbs of Norwich to the county
border with Suffolk (see Figure 13.1). It includes several market towns and had a population
of approximately 104,000 in the 1991 Census. South Norfolk was placed in the ‘Prospering
Areas’ cluster in the Office for National Statistics (ONS) classification of local authorities
(Wallace and Denham, 1996), but in the previous East Anglian study (Lovett et al., 2000) it
ACCESSIBILITY TO GP SURGERIES: A GIS-BASED ASSESSMENT 181
was found to include several groups of parishes with the poorest accessibility to primary health
care services. The district was therefore considered to encompass a good range of accessibility
conditions, although it is fair to note that even the most rural resident would not be more than
20 km from a market town and so would not be classed as remote in comparison to some
regions of the UK.
13.2
DATA
13.2.1
Primary Care Facility Locations
Details of GP surgeries were supplied by the health authorities and PCGs covering South
Norfolk and the surrounding area. Information was obtained on main and branch surgeries
existing in Autumn 1997 and Autumn 2000. Outlying consultation facilities that were typically
open for only a few hours each week were not included in the analysis. All the surgery
addresses were converted to 100 m resolution grid references via the Postcode to Enumeration
District Directory (Martin, 1992). Between the two dates there were only three short distance
moves by practices to new premises so the overall pattern of facility provision did not alter
appreciably.
13.2.2
GP Patient Registers
Information on the resident population of South Norfolk in Autumn 1997 and Autumn 2000 was
derived from GP patient registers. This source was used because it supplied population
estimates that were as contemporaneous as possible with the other data and provided a level of
geographical detail that was appreciable better than alternatives such as ONS mid-year
population estimates or 1991 Census products. The geographical resolution of population data
is an important consideration in accessibility research, because analyses based on centroids of
areas such as enumeration districts or parishes inevitably tend to over-concentrate residents in
locations where services are more likely to be present (Lovett et al., 2000).
Downloads of selected fields in the patient registers for South Norfolk in Autumn 1997 and
2000 were obtained from the health authorities and PCGs. All the records were anonymous
and contained details of number of residents for each unit postcode and GP practice code
combination. The unit postcodes were converted to 100 m resolution grid references using the
Postcode to Enumeration District Directory and the positions were checked using point-in-
polygon techniques with enumeration district boundaries (Gatrell et al., 1991) and distances
from ED centroids. Corrections were made using Internet resources such as http://
www.streetmap.co.uk where necessary.
Some doubts have been expressed about the reliability of GP patient registers as a source for
local population estimates, but previous research (Haynes et al., 1995; Lovett et al., 1998) has
concluded that those in East Anglia are of increasingly high quality. Table 13.1 compares the
two patient register totals for South Norfolk with recent ONS mid-year estimates. The patient
register figures are clearly slightly higher than those from ONS, and this may be due to list
inflation arising from administrative delays in removing outdated records (Simpson, 1998). It
is worth noting, however, that when the Autumn 2000 patient register figures were compared
Table 13.1 Population estimates for South Norfolk.
with parish mid-1999 estimates produced by Norfolk County Council (2000), the largest
differences were all concentrated in locations where there is currently considerable house
building activity (e.g. Wymondham and Diss) and the variation for some other urban areas (e.g.
Costessey on the edge of Norwich) was much less pronounced. This suggests that the patient
register data may be reflecting recent movements into new residential developments in a
quicker manner than is possible in the mid-year estimation methodology. Overall, therefore, it
was concluded that the register data were sufficiently reliable to form the basis of the
accessibility assessment.
13.2.3
Area Boundaries
For some elements of the research it was necessary to use classifications of geographical areas.
Discussions with local authority staff revealed that many aspects of public transport provision
used parishes, rather than census wards, as a basic spatial unit for planning and management
purposes. Parishes also had the advantage of providing greater geographical disaggregation than
wards and so were selected as the unit of analysis. Digitised boundaries for the parishes in
South Norfolk were constructed from map databases available via the MIMAS service at the
University of Manchester. There are officially 118 parishes in South Norfolk, but that around
Wymondham (NG117) is far larger (in both areal extent and population) than any of the
others. For the purposes of investigating accessibility issues it was thought best to subdivide
this parish and treat the southern part as a separate zone. In this chapter, therefore, the term
parishes is used to refer to a set of 119 areas. All the GP patient register postcodes were
matched to parishes via a point-in-polygon procedure and these assignments were checked
using the enumeration district-to-parish lookup table in the Area Master File available from
MIMAS.
13.2.4
Road Network and Travel Speeds
The road network in the study area is characterised by main roads radiating out from Norwich
(see Figure 13.1). Details of the road network were taken from the 1:250,000 scale Bartholomew
digital map database for Great Britain. The data included codes for different road categories
(e.g. A-road dual carriageway, B-road single carriageway) and average car speeds for each road
class were used to calculate how long it would take to drive along any particular segment. The
road speeds were based upon national off-peak estimates published by the Department of
Environment, Transport and the Regions (DoT, 1993), but reduced slightly to take account of
local conditions. These modified speeds had previously proved realistic in other research
(Bateman et al., 1999).
13.2.5
Bus Services
Information on bus services across South Norfolk was obtained by examining published
timetables and route maps supplied by the Norfolk Bus Information Centre. Where necessary,
checks were made with local authority staff to produce lists of services as of Autumn 1997 and
2000. For the purposes of assessing accessibility to GP surgeries, particular attention was
focused on routes where there was at least one daytime return journey every weekday (i.e. a
viable level of provision for many activities), but overall the bus services were divided into the
following five categories:
1. Routes where there were four or more buses in each direction between approximately
Sam and 6pm every weekday and a return journey was possible within three hours.
2. Routes meeting similar criteria to those above, but with only one to three buses in each
direction. Services restricted to school terms were not included in this category.
3. Routes providing a journey-to-work service every weekday. These were defined as
situations where the outward journey was early in the morning and the return trip did
not occur until the evening (e.g. someone reliant on such a service would have to spend
the whole day at the destination).
4. Routes with a daytime return service (as defined above) on two to four weekdays.
5. Routes where there was a daytime return service on one weekday.
Bus routes were defined on the digital road network using editing facilities in the Arc/Info GIS
software. Figure 13.2 depicts the pattern of services in Autumn 2000, individual sections of
road being symbolised according to the combinations of provision along them. All of the South
Norfolk parishes were also classified into six categories of bus service provision for both
Autumn 1997 and 2000. The classifications were based on the minimum level of service
available to the majority of the population in each parish, five of the categories corresponding
to those for the routes listed above and the sixth representing no bus service. Such an approach
is helpful for assessing change, but it needs to be recognised that there were a number of cases
where one part of a parish was rather better served than the rest.
13.2.6
Community Transport Provision
Details of community car and dial-a-ride schemes operating within South Norfolk were
obtained from service providers and local authority staff. Community car schemes typically
involve volunteer drivers using their own cars to provide door-to-door journeys for people
without transport, while dial-a-ride services often use minibuses or taxi-style vehicles. The
extent of community transport in the district was known to have increased considerably since
1997 and the research focused only on the provision in Autumn 2000. Most of the schemes had
defined catchment areas and information on these was used to classify parishes according to the
services available. One scheme (CarLink) run by Norwich and Norfolk Voluntary Services
Figure 13.2 Bus routes in South Norfolk, Autumn 2000.
operated over the whole of the district and supplied data on the locations of their volunteer
drivers as of October 2000. This information was subsequently used to identify gaps in driver
availability.
13.3
METHODS
13.3.1
Calculating Journey Times
To calculate travel times along roads for each year it was necessary to first assign the GP
surgeries to their nearest nodes on the network. An allocate operation was then implemented
in Arc/Info to identify the shortest travel time to a GP surgery from each node on the road
network. Travel times from patient register postcodes were calculated using the following
procedure. The first step was to identify the closest node on the road network for each
postcode and then the travel time from that node across the road network to the nearest GP
surgery. Next, a straight-line distance from the node to the postcode was determined and
subsequently converted to a time assuming a vehicle speed of 10 miles per hour. This stage
was considered necessary because many postcodes were several hundred metres away from the
nearest road on the network (the database excluded most small residential streets). Adding the
two times produced an overall value for the postcode and, though the use of straight lines to
nodes inevitably introduces a small element of approximation into the calculations, this
approach was found to be the most robust option available given the restrictions of the
available road network.
Visual representations of travel time by car to GP surgeries in 1997 and 2000 were
generated by defining contours around the estimated time values for the road network nodes.
This was done within Arc/Info by first undertaking a Delaunay triangulation of the nodes and
then creating a continuous surface known as a Triangulated Irregular Network (TIN) (see
Worboys, 1995). A regular lattice of estimated values spaced 100 m apart was then extracted
from the TIN and subsequently processed to identify lines of equal travel time (known
technically as isochrones, see Brainard et al., 1997).
13.3.2
Measuring Accessibility by Bus
Estimating bus journey times was not regarded as feasible or appropriate in this study and
instead an approach was developed which sought to determine whether residents could walk to
a bus route that would take them close to the nearest facility of a particular type. The
procedure implemented within the Arc/Info GIS involved taking each surgery location in turn
and first defining an 800 m radius buffer zone around it. A radius of 800 m (approximately half
a mile) was used to represent a distance that the great majority of the population would find
acceptable to walk. The GIS then selected all bus routes with a particular service frequency
(e.g. at least four return journeys per day) that passed through the buffer zone. Next, a second
buffer zone extending 800 m each side of the relevant routes was determined and the GP
patient register postcodes within this corridor were identified. Residents with these particular
postcodes were subsequently classified as populations, which had the relevant level of bus
access to the surgery being examined. The procedure was repeated for all surgeries in both
years using the two categories of route where there was at least one daytime return service
every weekday.
There are clearly some limitations in the method described above. The selection of a maximum
walking distance of 800 m was inevitably arbitrary, but reflected advice from local bus
operators and transport planners. Location of bus stops were essentially ignored, but most
services would stop in towns and villages and it is common practice to flag down a bus from
the roadside in many rural areas. It would also have been possible to extend the analysis to
include less frequent services, but it was decided that it would be more appropriate to use the
parish classifications to represent a lower level of bus availability (e.g. this allows for longer
walking distances). All these considerations mean that the results obtained need to be
interpreted with care, but several of the simplifications counterbalance each other and the
procedure is certainly sufficiently robust to identify general contrasts in accessibility by bus.
13.4
RESULTS
13.4.1
Travel Time to Nearest Surgery
In both Autumn 1997 and Autumn 2000 the average population-weighted car travel time from
the patient register postcodes to the nearest GP surgery was approximately five and a half
minutes. Maximum values in both years were just over 21 minutes. Figure 13.3 shows the
pattern of isochrones for Autumn 2000, with shorter travel times occurring around the towns
or along main roads, and longer journeys in a number of more rural areas. Table 13.2 confirms
the skewed distribution of physical accessibility, around 58% of residents having a travel time
under five minutes and 17% a journey longer than 10 minutes.
13.4.2
Accessibility by Bus
Figure 13.4 shows the classifications of parishes by majority bus service in 1997 and 2000. As
already acknowledged, the classifications conceal variations in provision within some parishes,
but the maps certainly imply an improved situation by Autumn 2000. This is particularly
apparent in terms of the reduction in parishes with the poorest levels of bus service provision.
Results from the more detailed assessment of accessibility to GP surgeries using the two
most frequent categories of service are listed in Table 13.3. These figures indicate an increase
in the number of South Norfolk residents with the most frequent category of bus service
between 1997 and 2000, although the overall proportion of the district population covered by
daytime return services every weekday remained stable at around 86%. This implies an
improved frequency of service, though it actually reflects a situation where improvements in
many parishes were partly counterbalanced by declines in provision for a few larger villages.
This point is illustrated in Figure 13.5 which shows the percentage of residents in each parish
within 800 m walking distance of a weekday daytime return bus service for both 1997 and
2000. The trend is not especially pronounced, but a wider coverage of parishes by Autumn
2000 is nevertheless apparent.
Figure 13.3 Estimated travel time by car to nearest GP surgery, Autumn 2000.
Table 13.2 Population by car travel time to nearest surgery.
13.4.3
Availability of Community Transport
The possible impact of the CarLink driver distribution was examined by calculating the
percentage of population in each parish who lived within three miles of the nearest volunteer
driver. A three mile limit was used because the CarLink pricing structure involves a fixed
minimum fee when the driver travels up to six miles, and a per mile rate beyond that distance
Table 13.3 Populations with daytime return bus services to GP surgeries, 1997 and 2000.
threshold. It would therefore be more expensive for anyone living more than three miles from
a CarLink driver to use the service.
Figure 13.6 displays the resulting percentage values. Most parishes have a high level of
coverage (i.e. all residents are within three miles of a driver), but there are obvious gaps in
some southern and eastern parts of South Norfolk. A 75% threshold was used to classify
parishes as having reasonable proximity to a CarLink driver and these data were then combined
with details of the catchments for other services to produce the overall categorisation of
community transport provision shown in Figure 13.7. The results of this assessment are
generally reassuring since they indicate that in Autumn 2000 there were no parishes without
some form of community transport provision and many areas had at least two service
providers. Table 13.4 lists the numbers of parishes and the size of the resident population for
the different categories in Figure 13.7.
Although it is difficult to precisely quantify the change in community transport services
since Autumn 1997, there seems little doubt that the situation has improved. New schemes
have been established in a number of small parish groupings and the introduction of the
Wymondham Flexibus has provided additional services in a significant number of rural
parishes. It is worth noting, however, that organisations such as CarLink and Bungay
Community Transport are very dependent on the availability of volunteer drivers. Only one
parish was classed as having a ‘Limited Carlink Service’ (because fewer than 75% of residents
were within three miles of a driver), but a number of others could easily drop to this level if a
few key volunteers were no longer available.
13.4.4
Overall Accessibility by Private and Public Transport
In order to examine the overall improvement in accessibility, the postcoded data for each year
were crosstabulated in terms of car travel time and the availability of bus services. Table 13.5
presents the result for Autumn 1997, the district population being subdivided into three car
travel time bands and five bus service categories. The first two of the latter represent
populations living close to the most frequent buses, while the third consists of residents more
than 800 m from a route, but within a parish classed as having a daytime return service each
weekday. Category four contains populations over 800 m from a main route but in a parish
with a less frequent bus service (e.g. on only one or two days a week), while the final group are
residents of parishes classed as having no buses at all. The progression down the rows of the
table therefore reflects a decreasing access to bus services.
One interesting feature of Table 13.5 is that all 15 cells of the crosstabulation are occupied.
This emphasises the point that car travel time and availability of bus services can be rather
different dimensions of physical accessibility and also highlights the variations in accessibility
Figure 13.4 Classification of parishes by bus provision, Autumn 1997 and 2000.
Figure 13.5 Percentage of residents within walking distance of a daytime return bus service.
Figure 13.6 Percentage of residents within three miles of a CarLink driver, October 2000.
within South Norfolk. Table 13.6 presents a similar crosstabulation based on the Autumn 2000
results and comparison with the 1997 data indicates a particular shift in population away from
the poorer accessibility categories. One illustration is provided by examining the totals within
the over 10 minutes car travel time columns. Another approach is to regard the four cells in
the bottom-right corner of the tabulation (i.e. those with car travel times of at least five
minutes, no nearby main bus routes, and in parishes without a daytime return service each
weekday) as populations for whom physical accessibility could begin to become more difficult.
In Autumn 1997 there were 8,070 South Norfolk residents in these categories and by Autumn
2000 the total had declined to 6,945. Of the latter, 54% were in parishes where at least two
community transport schemes were operating and there were only 93 people in areas more
than 10 minutes car travel time from a GP surgery, without a bus service, and with only one
community transport provider.
Table 13.4 Classification of parishes by community transport provision in Autumn 2000.
Figure 13.7 Classification of parishes by community transport provision, Autumn 2000.

The structure of the patient register data also makes it possible to examine the distribution
of population with potential accessibility difficulties by GP practice. Table 13.7 lists the
percentage of registered patients in the four poorer accessibility categories discussed above for
a sample of South Norfolk practices in Autumn 1997 and 2000. The contrasts between
practices and over time are evident, demonstrating the usefulness of such information for
primary health care planning and management.
Table 13.5 Population by car and bus accessibility to nearest GP surgery, Autumn 1997.
13.5
CONCLUSIONS
The results of the research described in this chapter indicate a broad improvement in levels of
accessibility to GP surgeries between Autumn 1997 and Autumn 2000. There was relatively
little change in travel times by car, but more residents had higher frequency bus routes within
walking distance and there was enhanced bus provision in a number of the remoter rural
parishes. By Autumn 2000 there were community transport schemes covering all South
Norfolk parishes and over half the district population had at least two service providers
available.
These findings suggest that the public sector investments in rural transport since early 1998
have had a definite beneficial impact on accessibility. There are still some substantial contrasts
in accessibility to GP surgeries within the district, but it appears that the populations for whom
travel to facilities remains difficult are now smaller and dispersed across many parts of the
district. The best means of helping such residents is likely to be through further expansion of
community transport, perhaps through the creation of new schemes or the recruitment of
additional volunteer drivers in areas where availability is currently limited. There may also be a
case for some investment in information dissemination and co-ordination, particularly so that
residents know how to request such transport services.
Although the period since early 1998 has seen the introduction of many new public
transport services, it is fair to note that actual usage has sometimes been disappointing. This, in
turn, has lead to adjustments in timetables and routes where, if anything, it is probably stability
in service provision that would help to encourage greater use. Recent increases in fuel costs
have also created problems, with several bus operators re-organising routes to provide a
greater frequency of service between towns, but fewer detours to villages away from main
Table 13.6 Population by car and bus accessibility to nearest GP surgery, Autumn 2000.
Table 13.7 Percentage of population with potential accessibility difficulties for selected GP
practices.
roads (Clement, 2000; Hill, 2000). This means that some of the results presented in this
chapter are already out-of-date. More positively, the Rural White Paper published in
November 2000 proposed further funding for rural transport (see http://www.wildlife-
countryside.detr.gov.uk/ruralwp/index.htm). The new Parish Transport Grants scheme may
offer particular opportunities for districts such as South Norfolk as it explicitly mentions support
for community transport or for existing bus services to divert through a village.
The results presented in this chapter have focused on one local authority, but the recent
experience of South Norfolk is likely to be reflected in many other lowland rural areas of the
UK. The research certainly demonstrates the analytical power provided by combining GIS
techniques with data sources such as patient registers and digital road networks, and the
methodology presented in this study could certainly be applied in other locations. It must be
acknowledged that the work required to produce digital bus route information was time
consuming, and it is also important to recognise that there are confidentiality issues associated
with the use of patient register data. Studies such as this nevertheless demonstrate the
considerable research resource that patient registers could provide to the NHS, and it is to be
hoped that developments such as the establishment of Regional Public Health Observatories
(Department of Health, 1999) will help to make anonymised extracts or summaries more
widely available for planning and management purposes.
13.6
ACKNOWLEDGEMENTS
The research described in this chapter was undertaken during parts of projects funded by South
Norfolk Council, Norfolk Rural Community Council and the NHS Executive Eastern Region.
We would like to thank Alexandra Bone for her assistance in compiling the information on bus
services, and the following for their help with data or guidance: Jen Wingate and Jeananne
Waugh, South Norfolk Council; Lyn Reynolds, South Norfolk Primary Care Group; Julian
Bester, East Norfolk Health Authority; and Andrew Coleman, Liz Joyce and Wendy Pontin,
Norfolk County Council. For census products and digitised boundaries, we acknowledge the
source: The 1991 Census, Crown Copyright, ESRC Purchase.
13.7
REFERENCES
Bateman, I.J., Lovett, A.A. and Brainard, J.S., 1999, Developing a methodology for benefit
transfers using geographical information systems: modelling demand for woodland recreation.
Regional Studies, 33, pp. 191–205.
Brainard, J.S., Lovett, A.A. and Bateman, I.J., 1997, Using isochrone surfaces in travelcost models.
Journal of Transport Geography, 5, pp. 117–126.
Clement, B., 2000, Bus firms blame high cost of fuel for scrapping of routes. The Independent, 28th
November 2000.
Countryside Agency, 2001, State of the Countryside 2001, (Cheltenham: Countryside Agency ).
Department of Environment, Transport and the Regions, 2000, Our Countryside: The Future—A Fair
Deal for Rural England, (London: The Stationary Office).
Department of Health, 1999, Saving Lives: Our Healthier Nation, (London: The Stationary Office).
Department of Transport, 1993, Vehicle speeds in Great Britain 1992, Statistical Bulletin 93 (30),
(London: Department of Transport).
Gatrell, A.C., Dunn, C.E. and Boyle, P.J., 1991, The relative utility of the Central Postcode
Directory and Pinpoint Address Code in applications of geographical information systems.
Environment and Planning A, 23, pp. 1447–1458.
Haynes, R.M., 1986, The Geography of Health Services in Britain, (London: Croom Helm).
Haynes, R.M., Lovett, A.A., Bentham, C.G., Brainard, J.S. and Gale, S.H., 1995, Comparison of
ward population estimates from FHSA patient registers with the 1991 Census. Environment and
Planning A, 27, pp. 1849–1858.
Higgs, G. and White, S.D., 1997, Changes in service provision in rural areas. Part 1: The use of
GIS in analysing accessibility to services in rural deprivation research. Journal of Rural Studies, 13,
pp. 441–450.
Higgs, G. and White, S.D., 2000, Alternatives to census-based indicators of social disadvantage in
rural areas. Progress in Planning, 53, pp. 1–81.
Hill, P., 2000. Give and take on the buses. Eastern Daily Press, 20th November 2000.
Jones, A.P. and Bentham, G., 1997, Health service accessibility and deaths from asthma in 401
local authority districts in England and Wales, 1988–92. Thorax, 52, pp. 218–222.
Joseph, A.E. and Bantock, P.R., 1982, Measuring potential physical accessibility to general
practitioners in rural areas: a method and case study. Social Science and Medicine, 16, pp. 85–90.
Joseph, A.E. and Phillips, D.R., 1984, Accessibility & Utilization: Geographical Perspectives on Health
Care Delivery, (New York: Harper & Row).
Lovett, A.A., Haynes, R.M., Bentham, C.G., Gale, S., Brainard, J.S. and Sünnenberg, G., 1998,
Improving health needs assessment using patient register information in a GIS. In GIS and Health,
edited by Gatrell, A.C. and Loytonen, M. (London: Taylor & Francis), pp. 191–203.
Lovett, A.A., Haynes, R.M., Sünnenberg, G. and Gale, S., 2000 Accessibility of Primary Health Care
Services in East Anglia, Research Report No. 9, (Norwich: School of Health Policy and Practice,
University of East Anglia)
Martin, D., 1992, Postcodes and the 1991 Census of Population: issues, problems and prospects.
Transations Institute of British Geographers, 14, pp. 90–97.
Moseley, M.J., 1977, Accessibility: The Rural Challenge, (London: Methuen).
Naude, A., de long, T. and van Teeffelen, P., 1999, Measuring accessibility with GIS tools: A case
study of the wild coast of South Africa. Transactions in GIS, 3, pp. 381–395.
Norfolk County Council, 2000, Mid-1999 Small Area Estimates for Norfolk Parishes, Urban Wards,
County Electoral Divisions and Built-Up Areas, Demographic Information Note 5/00, (Norwich:
Planning and Transportation Department, Norfolk County Council).
O’Sullivan, D., Morrison, D. and Shearer, J., 2000, Using desktop GIS for the investigation of
accessibility by public transport: an isochrone approach. International Journal of Geographical
Information Science, 14, pp. 85–104.
Ricketts, T.C. and Savitz, L.A., 1994, Access to health services. In Geographic Methods for Health
Services Research, edited by Ricketts, T.C. et al., (Langham, Maryland: University Press of
America, Inc.), pp. 91–112
Simpson, S., editor, 1998, Making Local Population Statistics: A Guide for Practitioners, (Wokingham:
Local Authorities Research and Intelligence Association).
Wallace, M. and Denham, C, 1996, The ONS Classification of Local and Health Authorities of Great
Britain, Studies on Medical and Population Subjects No. 59, (London: HMSO).
Worboys, M.F., 1995, GIS: A Computing Perspective, (London: Taylor & Francis).
14
Measuring accessibility for remote rural
populations
Mandy Kelly, Robin Flowerdew, Brian Francis and Juliet
Harman
14.1
INTRODUCTION
In the UK services such as Police, Fire, Education, Social Services and Health are supplied by
government to the population. Funds are generated locally (through means such as Council
Tax) and are topped up by Central Government to provide a standard level of service. The
way in which funds are allocated varies across the UK but in this paper we are focusing on
England, where funds for services (except for health) are allocated by the Standard Spending
Assessment (SSA) formula. The Standard Spending Assessment system was devised to
determine how allocations should be modified from a strictly per capita basis. The system is
based mainly on a regression analysis of costs incurred by authorities on a set of explanatory
variables that might seem relevant to costs.
This paper demonstrates the calculation of a set of new measures of population remoteness,
using GIS techniques to measure travel distance and time on the road network, and examines
the results of the incorporation of these measures into the existing SSA formula.
14.2
BACKGROUND
The premise of this work is the argument that the cost of service provision is strongly affected
by population distribution (Shropshire County Council (1996), Northumberland County
Council (1997), Hale and Capaldi, (1997)) and it is the development of earlier work by
Flowerdew and Gill (1998), which used straight line distance to create similar measures. The
assumption is that if the service involves travel, either by clients to service delivery points or
by service workers to the people, the distances travelled are likely to be longer in a rural area
than an urban one, as are travel times, and the travel costs will be higher.
The most common approach to measuring population remoteness is through calculating a
sparsity measure. At present sparsity measures are included in SSA calculations for some
services but not for others, and are included in different ways for different services (see DoE,
1997). The measure of sparsity of population used in SSA calculations seeks to take account of
both the overall sparsity in an authority and areas of extreme sparsity within authorities that
are not generally so sparsely populated. In the Education SSA, it is based on a weighted sum of
‘ordinary sparsity’ and ‘super-sparsity’. Ordinary sparsity is defined as the proportion of the
authority’s population that lives in wards where the density of population is between 0.5 and 4
people per hectare. Super-sparsity is defined as the proportion of the authority’s population
that lives in wards where the density of population is 0.5 or fewer people per hectare. Super-
sparsity is given double the weight of ordinary sparsity. For EPCS services, the definition is the
same except that calculations are based on enumeration district (ED) rather than ward data
(DoE 1997).
It is argued that sparsity, as defined above, is not a good measure of the additional costs
incurred in delivering services to a rural area (Flowerdew and Gill, 1998). Sparsity does not
take into account the geographical distribution of the population, in particular the relative
location of the population being served and the centres from which the service is provided.
Thus a sparsely populated ward or ED close to a service delivery centre would not be nearly as
expensive to serve as one with a similar degree of sparsity located far away from the centre. A
better measure would be based on the distance or travel time between the service delivery
centre and the population.
Until recently, such a measure would have been time-consuming and difficult to construct.
However, with the rapid advance of geographical information systems (GIS) technology and
suitable data sets, we hoped to supply a range of measures of this type and to try them out in
the SSA regressions. Similar approaches have been taken by other recent studies attempting to
assess accessibility in different contexts, such as work by Lovett et al. (2000) on accessibility of
primary health care services in East Anglia, and Higgs and White (1997) on accessibility to
services in Wales.
This paper will provide an overview of the GIS techniques used to derive a series of
alternative measures of population remoteness for local government areas using weighted
travel times and distances from each ED to assumed service delivery points, based on different
assumptions about where the services are located and on the local authorities for which the
measure was to be computed. Once these measures had been constructed, the second
objective was to examine the impact of using them (rather than the conventional sparsity
measures) in the calculation of Standard Spending Assessment (SSA). The methodology of this
objective is not outlined in detail here, although some of the final results are discussed.
14.3
METHODOLOGY
The construction of measures of remoteness required the use of a GIS to calculate the lengths
of the shortest paths along the road network between each ED centroid and the nearest service
delivery point. This was done using Bartholomew’s 1991 1:250,000 scale digital map database
of the road network obtained under the CHEST agreement. ED populations and grid
references for ED centroids were obtained from the MIMAS system at Manchester University,
using data purchased by ESRC and JISC on behalf of the academic community. Accessibility
measures were computed as if all trips were made to or from the population weighted centroid
of the ED, or the nearest point on the road network.
A considerable amount of time was necessary for constructing coverages of the road system.
One unanticipated problem was the necessity to specify locations at which limited-access roads
like motorways could be joined—otherwise the algorithm would allow traffic to join or leave
wherever the limited-access road was bridged by another road. Routes were not constrained to
be completely within the local authority for which the measures were being calculated. Thus it
was possible for a route from an ED to a service delivery point within the same authority to
pass through a different authority if distances or times were reduced by so doing.
MEASURING ACCESSIBILITY FOR REMOTE RURAL POPULATIONS 199
Calculation of travel times required estimates to be made of the speeds of travel on roads of
particular quality. Standard estimates of travel times, such as the national off-peak estimates
produced by the Department of Transport (1993), seemed inappropriate in the context of
travel much of which would be done at or near peak hours. Instead, we used average speeds
computed for four types of roads based on field testing in North West England, conducted in
order to determine accessibility for private sector service provision. These are 46.55 mph on
motorways, 28.60 mph on A roads, 22.20 mph on B roads, and 15.13 mph on other roads.
These speeds are intended to be averages reflecting different degrees of traffic density, different
driving styles, and variable effects of weather, congestion, joining and leaving delays, speed
limitation and traffic lights, the speeds themselves are less important than the ratios between
the speeds obtained for different types of road.
Islands were excluded from the analysis. In many cases they are not dealt with appropriately
in GIS coverages of the census data. A small offshore island may be part of a mainland ED with
the same ED code assigned to it, and may incorrectly be assigned the total population for that
ED. The census data do not provide information on how the population is split geographically
within the ED, and it is not possible to tell from the data whether an island is inhabited or not.
In addition, if an island is not connected to the mainland as part of the road network, it would
be necessary to assign some arbitrary measures of travel distance and time.
Initial experiments had investigated the use of Arclnfo, as functions exist to calculate the
distance between nodes on the network (Arclnfo command NODEDISTANCE). Each ED and
service centre could be allocated to a node (using Arclnfo command NEAR). However, it was
found that the robustness of the road network was seriously limiting. Any change to the road
network would result in a renumbering of the nodes and all the preparation stages had to be
repeated. There were also problems with nodes that had an ID of zero, due to truncation or
clipping of the network, and these were indistinguishable from each other. The use of
Bartholomew’s data may have been an influencing factor but even with the more accurate
Ordnance Survey road data (such as OSCAR) the work may not have been problem-free.
Due to the problems encountered in Arclnfo, an extension available for ArcView (Shortest
Network Paths vl.2 developed by Neudecker (1999)) was used to calculate the travel distance
and time. This associated points with nodes on the network, identified all the possible paths
along the network and then found the minimum cost (either distance or time). Since this software
existed and was generally suitable and in Arclnfo further programming would be necessary to
facilitate automation, the analysis was continued in ArcView. The extension was further
adapted using Avenue programming language to aggregate the measures, weighted by ED
population, to produce a single measure for each local authority. This measure could be
interpreted as the total travel distance (or time) needed for each resident to reach the nearest
service delivery point. If desired, this could be divided by population to get a per capita
measure of remoteness that would be comparable between local authorities of different size.
Similar computations were done using time rather than distance as the basis of calculation.
Because different levels of local government deliver different services, it was appropriate to
reproduce the measures for different types of authority. All services included in SSA are
delivered by Metropolitan Districts (including London Boroughs) and Unitary Authorities.
Some services, including education and Personal Social Services (PSS), are delivered by the
shire counties while others, including Environmental, Protective and Cultural Services
(EPCS), are delivered by districts within the shire counties. Different databases were therefore
necessary for consideration of different services. To examine services at an upper tier level
(Metropolitan Districts, Unitary Authorities and shire counties, see Figure 14.1) we looked at
the Other Personal Social Services component of the PSS Standard Spending Assessment. To
Figure 14.1 New upper tier local government geography of England 1998.
investigate the impact of new measures of remoteness for authorities at the ‘lower tier’ level
(Metropolitan Districts, Unitary Authorities and districts within shire counties, see
Figure 14.2), EPCS were also examined.
In addition, the regressions used in the development of SSA were based on calculations
performed using the local authorities in existence before the introduction of unitary authorities
in the late 1990s. Accordingly the measures were also calculated for the ‘old’ local authorities,
in which unitary authorities were included within the county in which they were located. In
practice these calculations were actually easier than those for the ‘new authorities’ as some
boundary changes could not be constructed easily directly from the previous system.
Population data for these new boundaries were supplied by look-up tables by Wilson and Rees
(1999).
Figure 14.2 New lower tier local government geography of England 1998.
The pattern of service provision is likely to differ substantially between different service
sectors. In particular, the number of service delivery points may differ, and be either clustered
or dispersed and single or multiple.
A series of meetings with local authorities was scheduled, intended to get their views about
the role of geographical distance in the organisation of services and their costs. Respondents
were chosen to give a representative selection of different types of local authority. The
selection was also influenced by the responsiveness of those authorities invited to take part.
Interviews were held with representatives of Lancashire County Council, Barrow District
Council, Norfolk County Council and North Lincolnshire Unitary Authority. They were asked
a series of open-ended questions about the organisation of services, including issues concerned
with the size of settlement for which a particular service might be available, and with the role
of distance in service provision.
It is clear from these examples that remoteness and accessibility are important to these
authorities, not just in terms of the distances that must be travelled for clients to meet service
suppliers, but in a range of more subtle ways. The interviews also strengthened our view that
distance and travel time measures are more relevant than sparsity. Unfortunately the situation
is complex and the size of the effects is difficult to measure.
To model some of these possibilities we used the following:
1. the (population-weighted) centroid of the local authority only. The authority centroid was
used as it is the simplest model (although somewhat unrealistic) and enabled the
development and assessment of the efficacy of the methodology.
2. the nearest district centroid. These measures were equivalent for Metropolitan Districts
and Unitary Authorities (which could be regarded as single districts) but differed
substantially for counties, where service delivery from a single point seems most likely to
lead to very long travel distances.
3. distance to the nearest settlement of a given size (using ONS definition of urban/rural
areas).
14.4
RESULTS
There are three main areas of results. Each analysis creates a set of aggregate travel distances
and times for each local authority (see Figures 14.4 to 14.6). These can be compared with the
sparsity measures currently used in SSA calculations (Figure 14.3). The redistribution of
resources that would be implied by substituting the remoteness measure for sparsity can also
be computed, to identify the ‘winners’ and ‘losers’.
The number of sets of measures produced was not as extensive as initially hoped. There
were several contributory factors during the data preparation and analysis. The machine ran to
capacity whilst calculating the measures. Individual runs were very time-consuming to the
extent that there were major delays in completing this part of the analysis, finally forcing us to
reduce the number of measures for which remoteness indices could be calculated. In
particular, despite our intention to produce measures for distances and times to settlements of
a range of threshold sizes, time only permitted these calculations for a threshold size of 100,
000.
Complications are introduced by the availability of data for some set of boundary units but
not others. It is more useful to display the comparisons between the traditional sparsity
measure and our measures of travel distance and time for existing local government units.
However recalculation of the regression equations used for deriving the SSA formulae should
be done using the old units (both for comparability, and because some of the data are only
available for these units). Accordingly, we computed distance and time measures for
accessibility to local authority centroids, district centroids, and settlements of 100,000
population separately for the four sets of local authorities—the old upper tier, the new upper
tier, and old lower tier and new lower tier. This was a considerable increase in the workload
from what had originally been envisaged.
One problem that arose in incorporating our new measures into SSA methodology is that
detailed information on recent changes in the methodology was not available until late in the
Figure 14.3 Existing sparsity measure.
project. Although we are indebted to DETR staff for supplying the information necessary to
replicate their calculations and hence to examine the effects of our measures, problems were
encountered in understanding exactly what had been done and in reproducing the DETR’s
results to sufficient accuracy.
It should be noted that most of the data in the SSA calculations are taken from the 1991
census, and hence are nearly ten years old. There is also a minor difference in that the
population data for the ‘old’ areas are based on the 1991 census, while the data for the ‘new’
areas are based on a 1991 mid-year estimate. This leads to small differences in population
totals even for areas whose boundaries have been unchanged in the reorganisation, and also to
small changes in the locations of their population-weighted centroids.
Figure 14.4 Measures from EDs to the local authority centroid.

14.4.1
Measures Computed
The maps (Figures 14.3 to 14.6) show the values of the various measures of remoteness used in
the analysis. Figure 14.3 shows the existing sparsity measure based on calculations at the ED
level, mapped for the upper tier of authorities as currently constituted. Generally the pattern
is a reasonable reflection of remoteness, though it may be surprising that areas like Rutland and
Shropshire appear to be more sparsely populated than Northumberland, Cumbria, East Anglia
and the South West.
Following is a series of maps showing the measures we have computed for the upper tier.
There are three pairs of maps, representing an average distance measure and an average travel
time measure under three sets of assumptions. First, distance and travel time to the authority
centroid are calculated (Figure 14.3). Generally there is a fairly close relationship between the
distance and travel time results (with some significant differences). In comparison to the
sparsity map, these maps show a strong relationship with the area of the county. Kent, for
example, has a fairly high value because it is assumed that a fairly dense population has to
travel considerable distances to a central point in a large county. It may be noted however that
Kent is less prominent in the travel time map, presumably because the motorway network
makes the centroid more accessible than is the case for other large counties with less motorway
mileage.
The second pair of maps (Figure 14.5) shows distances and travel times to the nearest
district centroid. These will be identical for metropolitan districts, London boroughs and
Figure 14.5 Measures from EDs to the district centroid.
Figure 14.6 Measures from EDs to urban areas over 100,000 population.
unitary authorities but will be shorter for the shire counties. Relatively sparsely populated
unitary authorities, like Herefordshire and Rutland, are more prominent in the district maps
while less sparsely populated but larger counties like Dorset, Kent and Suffolk appear less
prominent.
The third pair (Figure 14.6) shows distances and travel times to the nearest urban area with
a population of over 100,000, on the grounds that such places are likely to have service
facilities not present in smaller cities. The maps again show large and peripheral counties as
Table 14.1 Remoteness measures.
having high values, with the most prominent being counties which do not contain any large
cities, such as Cornwall, Cumbria and Lincolnshire.
14.4.2
Comparision With Existing Sparsity
As one example of how our remoteness measures might impact calculations of SSA, we looked
at the Other Personal Social Services component of the Personal Social Services (PSS) Standard
Spending Assessment (‘Other’ refers to aspects of PSS besides those provided for children and
older people). ‘Upper tier’ authorities (using the nomenclature introduced above) provide
these services. The client group is the population aged between 18 and 64. The formula for the
cost per client includes two indicators, estimated by regression, that reflect social and health
conditions in each authority by taking into account the difference in the cost of service
provision.
The regression formula was successfully replicated to 3 significant figures. Several
remoteness measures were added, one at a time, as an extra explanatory variable into the
regression formula. The effect of remoteness was negative in all regressions.
As it was not possible to calculate road distances to an urban centre of 100,000 or more for
the Isle of Wight, this authority is omitted from the last two regressions, which therefore use
106 observations rather than 107. Note that the relative size of the coefficients is dependent on
the scale at which the measurements are conducted, and does not reflect the magnitude of the
effects, which is stronger for the new measures than for the traditional sparsity measure. The
distance measures seem to perform slightly better than the travel time measures (p=0.04), and
there is little difference between the measure based on distance to the centroid and distance to
the nearest large urban area. Only two of these five measures are statistically significant (distance
to local authority centroid and distance of urban area of 100,000 plus p=0.04).
14.4.3
Applying the New Measures to the SSA Formula
Because the effect of the remoteness measures in the regression analysis was negative when
these figures were put in the SSA formula, the redistributive effects work against the interests
of the remote rural areas. The largest gainers in percentage terms from introducing the
distance to local authority centroid variable are Wokingham (13%), Rutland (10%) and
Windsor and Maidenhead (8%); the biggest losers in percentage terms are Cumbria and
Northumberland (• 9%) and North Yorkshire (• 8%). In absolute terms, Leicestershire and
Bromley are the biggest gainers (over £500,000 per year) and Kent, Cumbria, North
Yorkshire, Essex and Lancashire would all lose over £1 million per year. Using distance to the
nearest urban area of over 100,000 gives the highest percentage increases to Wokingham
(8%), Poole and South Gloucestershire (6%), with the largest percentage losses experienced
by Cumbria (• 21%), Cornwall (• 13%) and Devon (• 11%). The greatest absolute gains
would accrue to Hampshire, Staffordshire, Surrey and Leicestershire, while the greatest absolute
losses would occur in Kent, Cumbria, Devon, Cornwall, Lincolnshire and Somerset.
An overall conclusion from this analysis is that inclusion of a variable representing remoteness
acts against the common-sense view that service delivery in the case of Other Personal Social
Services would be more expensive in rural areas. The most likely explanation of this may be
that the existing regression formula does not fully represent the costs relating to social
conditions, which tend to be higher in the urban areas. The negative coefficient of the
remoteness variable suggests that it may be acting as a surrogate for this urban effect. Only if
the urban effect was adequately represented by variables more obviously reflecting it would
the remoteness variable become positive and reflect the increased costs of providing services to
a dispersed population in the way that was intended. It is interesting that our new measures
reflect this effect better than the conventional sparsity calculation.
14.5
CONCLUSIONS
We feel that the methods used have been successful and that the measures produced represent
a more sensitive approach to the problems of measuring remoteness than the sparsity index
generally in use. Although we have not been able to complete all the analyses that we had
intended to do in the project we have demonstrated that it is possible, though still time-
consuming, to use GIS methods to construct a set of measures of accessibility for local
authorities in England. However, we are not happy that the measures used are fully successful
in capturing remoteness. In particular, they appear to give values that seem too high for large
urban areas, where the assumption that services are provided from one centre may lead to
excessive aggregate travel times and distances. It may be appropriate to develop a method
where the number of service centres is assumed to be greater for urban areas with a larger
population. Another alternative to modelling the pattern of service provision is to use the
relative size of an urban area within an authority, or to construct a database of actual service
delivery points using gathering data from Local Authorities. An additional alternative is to
apply a non-linear function to the travel times to allow for the difference in roads passing
through urban and rural areas.
The new measures have also failed to account for the discrepancy between intuitive
expectations that service delivery costs would be greater for rural areas and analytical results
from the SSA regression analyses, which are inconsistent at best and if anything tend to show
the reverse. Our analysis of Environmental, Protective and Cultural Services SSA indicates that
the judgmental allocation of SSA based on sparsity is too high when compared to the results of
including sparsity in the regression analysis. It is also the case that the sparsity measure is more
significant in the SSA regression than our travel time and distance measures. In our analysis of
Other Personal Social Services SSA, travel time and distance are more significant than is
sparsity, but the relationship is negative; i.e. service delivery costs appear to be lower for remote
areas! This may be because our measures do not deal adequately with large urban areas.
However it seems more likely that this counter-intuitive relationship reflects a failure of other
variables in the model to account adequately for socio-economic variables leading to costs
being higher in urban areas. A further possibility is that the relationship between costs and
remoteness may be non-linear. There is some evidence to suggest that costs are higher for the
most remote and the least remote areas, and lower for intermediate areas.
14.6
FUTURE RESEARCH PRIORITIES
We would like to do more work on the measures, in particular on extending the analysis for a
greater range of settlement sizes and investigating different alternatives to represent the
pattern of service delivery. We had hoped to do this during the project, but did not have
sufficient time. There is also a problem with the assumptions made in our methodology about
service provision in large cities. This was recognised during the project and meetings with
local authorities proved that to quantify the pattern of service provision for the whole country
would be a difficult task. Accessibility appears to be more of a problem according to our
measures than seems likely in practice, probably because we assume that accessibility is
calculated from the district centroid, even when the district concerned has a large enough
population to support several service delivery centres. We would like to develop methods for
dealing with this problem, and with other more minor technical problems, before presenting
our results as ready for use in the policy arena.
Some of our results suggest that the relationship between remoteness and service delivery
costs may be non-linear. It appears that costs may be higher in remote rural areas but also in
inner cities, perhaps as a result of congestion. We would like to explore this possibility in
more detail, both through further statistical analysis and through further discussions with local
government service providers.
The nature of the system for central government support of local government activities is
currently under review. It is generally accepted that additional costs related to issues of
rurality and remoteness are problematic in the current system, so we are keen to contribute
towards a more satisfactory way of taking account of them. We would also be interested in
reviewing how similar issues are treated in other countries (including the other component
countries of the United Kingdom).
Similar considerations are also relevant to the provision of health services, where it is clearly
important for everybody to be within reach of medical facilities. Rurality and remoteness are
therefore again likely to be crucial in the funding formula for devolution of central funds to
local areas. Current National Health Service funding takes this account in the funding of
ambulance services but there is a strong case for it to be more prominent in other aspects of health
service funding.
14.7
REFERENCES
DOE, 1997, Standard Spending Assessments: Guide to Methodology 1997/98. Local Government Finance
Policy Directorate, (London: Department of the Environment).
Flowerdew, R. and Gill, M., 1998, Geographical measures of accessibility for modelling costs of
service delivery. Paper presented at the Geocomputation 98, Bristol.
Hale, R. and Capaldi, A., 1997, Local Authority Services in Rural England. Rural Development
Commission.
Higgs, G. and White, S.D., 1997, Changes in service provision in rural areas. Part 1: The use of
GIS in analysing accessibility to services in rural deprivation research. Journal of Rural Studies, 13
(4), pp. 441–450.
Lovett, A., Haynes, R., Sunnenburg, G. and Gale, S., 2000, Accessibility of Primary Health Care
Services in East Anglia. Research Report 9, School of Health Policy and Practice, (Norwich:
University of East Anglia).
Neudecker, 1999, Shortest Network Paths V1.2. Developed from vl.l by Kevin Remington, available
from ArcScripts on http://www.esri.com.
Northumberland County Council, 1997, Sparsity and the SSA. Finance Department, Northumberland
County Council.
Senior, M.L., 1994, The English Standard Spending Assessment system: an assessment of the
methodology. Environment and Planning C: Government and Policy, 12 (1), pp. 23–52.
Shropshire County Council, 1996, SSA and Sparsity: Executive Summary.
Wilson, T. and Rees, P., 1999, Linking 1991 population statistics to the 1998 local government
geography of Great Britain. Population Trends, 97, pp. 37–45
15
Assessing the transport implications of
housing and facility provision in
Gloucestershire
Helena Titheridge
15.1
INTRODUCTION
With the rapid housing development expected to take place over the next decade, there is a
real opportunity to place new housing in such a way that it supports existing or new services
and facilities.
The hypothesis that allocating housing and other development based on the level of facility
provision could result in reduced travel and fuel consumption was tested by applying a
transport model to Gloucestershire. A number of development strategies based on balancing
population and facility provision were modelled and the results compared with current travel
patterns and those likely to result from housing allocations in the structure plan.
The greatest reductions were achieved through the strategy to expand as many towns as
possible to above a threshold population size of 25,000. At this point a town was considered to
be of sufficient size to support a number of higher order facilities as well as a wide variety of
low order services and facilities. It is concluded that allocating housing to support service and
facility provision could not only reduce travel but also significantly increase the level of access
to key services and facilities for rural populations.
15.2
BACKGROUND
Planners in the United Kingdom are currently facing difficult decisions concerning the siting of
new housing developments. UK Government projections predict that 3.8 million new
dwellings will be needed between 1996 and 2021 (DETR, 1999). Pressure has been placed on
local authorities to find suitable sites for this housing. Reducing traffic being just one of the
considerations. Increasing access to employment, services and facilities reducing car
dependency and social isolation being others. The later is particularly important in rural areas,
where the level of service and facility provision if often poor. A recent survey by the Rural
Development Council for the Countryside found that nationally 42% of parishes were without
a permanent shop, 83% were without a general medical practitioner and 75% were without a
daily bus service (CPRE: Council for the Protection of Rural England, 1998). This has a
number of implications. Rural residents tend to travel longer distances than their urban
counterparts, leading to a higher expenditure on fuel. CPRE (op. cit.) suggest that rural
households spend an average of 10% more a week on fuel than their urban counter parts, while
ASSESSING THE TRANSPORT IMPLICATIONS 211
ACRE (Action for Communities in Rural England, 1998) found that those living in rural
communities travel 50% further than those living in urban areas. For those with restricted or
no access to a car the accessibility of services and employment opportunities can be extremely
limited.
Government Planning Policy Guidance for Housing, PPG3 (DETR, 2000a) lists a number
of criteria for selecting suitable development sites:
• The availability of previously developed sites and empty or under-used buildings;

• The accessibility of local services;
• The capacity of existing infrastructure to absorb further development;
• The ability to build communities, to support new physical and social infrastructure and to
provide sufficient demand to sustain appropriate local services; and
• The physical constraints on development land.
Current emphasis in Government policy is to place new housing development in or on the

edge of existing towns wherever possible. However, under certain circumstances substantial
sized development in villages are to be permitted along with new settlements. One of the criteria
for village development is that “the additional housing will support local services, such as
schools or shops, which could become unviable without some modest growth” (op. cit.,
paragraph 70). Similar policies govern the development of new settlements, where new
settlements must be of a size that can support a number of local services.
With the rapid housing development expected to take place over the next decade, there is a
real opportunity to place new housing in such a way that it supports existing or new services
and facilities in line with PPG3. In order to achieve this Local Authorities need more
information on the size of population and catchment areas needed to support these services.
Techniques are also needed to identify sites with development potential based on this and the
other criteria set out in PPG3.
This paper describes work undertaken as part of the URBASSS (URBAn Sustainability and
Settlement Size) project, funded by the EPSRC Sustainable Cities programme. The aim of the
research discussed in this paper was to test whether allocating housing and other development
based on the level of facility provision could result in reduced travel and fuel consumption.
This was done by applying the ESTEEM (Estimation of Travel, Energy and Emissions Model)
transport model to Gloucestershire and comparing the travel patterns resulting from a number
of development strategies for Gloucestershire based on balancing population and facility
provision with current travel patterns in the region and with travel patterns likely to result
from the housing allocations given in the Gloucestershire Structure Plan (Gloucestershire
County Council, 1999).
15.3
ESTEEM
ESTEEM was developed as part of EPSRC funded research carried out at UCL between 1997
and 2000. The model assesses the sustainability of new developments in terms of personal
travel demand, and associated energy consumption and emissions. ESTEEM has been tested
through application to Leicestershire and Kent (see Titheridge et al., 2000).
ESTEEM operates as an extension to ESRI’s ArcView GIS package (Figure 15.1). ArcView
was chosen because of its relatively low cost to Local Authorities, its network analysis capabilities
Figure 15.1 ESTEEM Structure.
Table 15.1 Population bases for calculating trip production rates.
and easy of customisation. An RTPI survey (RTPI, 1998) of Planning Departments, carried
out in 1995, found that 7% of County/Region planning departments had access to ArcView.
Over 30% had access to its sister product Arclnfo. A small survey conducted by the Bartlett
School of Planning in 1997 (Titheridge et al., 1998a) of County planning departments found a
migration towards desktop pc and windows-based packages since the RTPI survey had been
carried out. A more recent survey by the RTPI (RTPI, 2000) found that the number of Local
Authorities using networked PC systems had increased from 50% in 1995 to 79% by 2000.
An origin-constrained gravity model is used to simulate travel patterns by car, bus and rail
for commuting, education, shopping, personal business and leisure trips. ESTEEM models
approximately 80% of all motorised trips1. The modal split and the number of trips generated
per person is determined by car ownership levels, age characteristics, employment levels, and
the proximity to a bus route and to a rail station of each origin. The fuel consumption and
emissions calculations take into account fleet characteristics, as well as estimates of cold start
distances. See Titheridge and Rana (2000) for a fuller description of ESTEEM.
15.4
APPLICATION OF ESTEEM TO GLOUCESTERSHIRE
15.4.1
Data Input
Different population bases were used for calculating the trip rates for each type of journey
purpose modelled, based on the demographic group most likely to make a trip for that
purpose, see Table 15.1.
1Based on data from the National Travel Survey 1997/99 on journeys per person per year by main
mode (DETR, 2000b).
Table 15.2 A summary of the data used in ESTEEM to represent trip destinations within
Gloucestershire.
Populations from the 1991 Census were used at enumeration district level and attributed to
population centroids to give the data in the required point format, with each centroid acting as
a ‘trip origin’. Census data was used despite the fact that it is now 10 years old because of its
availability at such a fine scale and the completeness of the sampling. Data on the fraction of
the population living in households with access to a motor vehicle, was used to vary trip
production rates for each mode, according to car availability. Trip production rates were also
varied according to public transport accessibility.
Destinations for each purpose were entered into the model as points. A slightly different
level of aggregation was used for each purpose. These are listed below in Table 15.2, together
with the source of the data.
The employment index used for the Leisure attraction measure was calculated using the
following method. National travel survey data was used to derive the mean number of trips
per year made per person to each leisure purpose type—to eat/drink, for entertainment, to
participate in sports, and for a day trip. Assuming that over 95% of trips in each category are
made to destinations within the study area and its buffer zone, then the trips per employee in
the study area can be derived. Each business was allocated to one of the four leisure purpose
categories. The number of employees of that business was then multiplied by the appropriate
trips per employee figure to give an attraction figure for each business comparable between
leisure types. The attraction figures were then aggregated by summation to town centres.
The 1991 census journey to work data for Gloucestershire shows that 91% of all work trips
are to destinations within the county (OPCS, 1994). Assuming similar proportions for other
journey purposes, then it is estimated that the model set up as described above includes 52%
of all trips made by the residents of Gloucestershire and 73% of motorised trips, based on data
from the National Travel Survey 97/99 (DETR, 2000b).
15.4.2
Calibration of the Model
Finding suitable data with which to calibrate the Gloucestershire model proved difficult. No
suitable local travel surveys had been carried out. The 1991 Census of Population included
questions on travel but only on commuting trips to the normal place of work. The National
Travel Survey includes data on the county of residence for each respondent, so trips by residents
of Gloucestershire could be retrieved. However, the sample sizes are too small to produce
statistically reliable information on a single county. Thus the solution was to calibrate the
Table 15.3 Mean Trip Lengths derived from National Travel Survey data 1988–95, for the
calibration sample and for all counties and regions in Great Britain.
* Excludes trips to visit friends at their home, trips to a holiday base and just walk trips.
model using mean journey distances and travel times based on counties with similar
characteristics to Gloucestershire.
Gloucestershire is predominantly a rural county. UWE (2000) found the following factors
distinguished rural and urban areas: Population Density, Employment Density, Settlement Size
and The percentage of the population in Agricultural Industries. For simplicity only two of
these determinants were used in selecting ‘like’ counties—population density and the
percentage of the population employed in agricultural industries.
Using these two determinants, counties were selected if both parameters were within plus
or minus a factor of two of the values for Gloucestershire. This gave a list of 24 counties including
Gloucestershire. National Travel Survey (NTS) data for several years either side of 1991 was
used to boost the number of trips included in the sample further. Trips and Journey Distance
were then aggregated by mode and purpose to give mean trip distance for each trip type, i.e.
commute journeys by car (Table 15.3).
The model was then run for each mode-purpose combination. In each case, the distance-
decay exponent was adjusted until the modelled mean trip distance and the NTS mean trip
distance converged. Because, the data used to calibrate the model, was not specifically for
Gloucestershire and contained no further spatial disaggregation, it was not possible to assesses
the statistical fit of the calibrated model at ward level, as done previously when calibrating
ESTEEM (see Titheridge et al., 1998).
Problems were encountered calibrating leisure trips by bus using the Gloucestershire
equivalent sample. The NTS trip data includes all trips ending in the UK. The destination data
included in the model however, is limited to locations within Gloucestershire. For most
purposes, the percentage of external trips is small, so this is not too much of a problem.
However, for rural areas like Gloucestershire with poor public transport provision, leisure
trips by bus will tend to include a large proportion of coach excursions, with the associated
long journey lengths. To get around this problem, leisure trips by bus were calibrated against
NTS data for all counties rather than the smaller sample representing Gloucestershire
equivalents.
15.5
CURRENT TRAVEL PATTERNS
After calibration the model input data was updated to represent the situation in
Gloucestershire in 1998 (the most recent year for which population estimates were available).
Figure 15.2 The Pattern of Mean Trip Lengths in Gloucestershire, 1998.

As more recent population estimates are not available at the enumeration district level or for
all the variables used within the model, a number of assumptions about the changing level of
car ownership and structure of the population had to be made (see Titheridge, 2000, for a full
description of these assumptions). The model was then run for the entire county to establish
current travel patterns.
It was found that those living in the Severn Vale area tend to travel on average shorter
distances than those living elsewhere in the county (Figure 15.2). This is not surprising given
that this region has the highest density of employment and housing within the county, contains
the towns of Gloucester and Cheltenham, and the majority of services. Short mean journey
distances are also found in the Forest of Dean. Service provision in this area is not huge but
connections to the rest of the county are poor, so possibly the distances to the full range of
services and facilities on offer are too great thus inhabitants make do with the limited range of
facilities available locally.
The majority of travel in Gloucestershire is by car2, accounting for 89% of all trips, 92% of
the total distance travelled by residents of Gloucestershire and 97% of the fuel consumed for
travel (Table 15.4). Nationally, car travel accounts for 93% of road vehicle mileage, using the
same base for journey purposes. The 6,000 km per annum travelled per capita for
Gloucestershire also compares favourably with the 6900 km travelled per person per year on
average within the UK for the same range of journey modes and purposes (DETR, 2000b).
Although it should be noted that the figure for Gloucestershire excludes journeys to
destinations beyond the county boundaries. As already discussed, the 1991 census showed that
9% of commuter trips by residents of Gloucestershire were to destinations outside the county.
Taking this into account, and the fact that these trips are likely to be longer than the average
for the area, then travel in Gloucestershire is very similar to the national average
2 Car travel includes journeys by motorcycle, taxi and minicab.

Table 15.4 Annual modelled travel and energy for trips within Gloucestershire, 1998.
15.6
MODELLING THE TRANSPORT IMPLICATIONS OF VARIOUS
DEVELOPMENT OPTIONS
A variety of development options were tested for Gloucestershire, representing different
methodologies for assigning housing, services and facilities. Option 1 was our interpretation of
the Gloucestershire structure plan. Option 2 tackled under provision of services in
Gloucestershire settlements. Option 3 added housing to those settlements that were over-
provided in terms of services. Option 4 combined options 2 and 3, in an attempt to balance
service provision and housing across the county. In option 5 housing was added to a small
number of settlements, bringing them above population thresholds for medium to high order
services. Lastly, a slightly different approach was taken for option 6 in which public transport
provision was increased substantially.
A limited range of services and facilities were considered in each option. These services and
facilities, namely primary schools, higher education colleges, theatres, leisure centres, pubs,
supermarkets, convenience stores, clinics, banks and garages, were chosen to represent the
range of activities a person undertakes and a range of high to lower order services. Williams
(2000) established a range of threshold populations need to support each of these facilities
using a variety of different techniques. A mid-range population threshold for each facility was
used to determine the number of each of these key services that each settlement within the
County could support. Settlements of less that a 1000 population were not considered. No
employment was added in any option except for employment directly related to the services
and facilities listed above. The characteristics of the population added through the new
developments considered in each option, reflected the population currently resident in that
area. Thus there is slight variation between the population totals for each option.
15.6.1
Option 1
This option tested the impact of the housing allocations given in the structure plan for
Gloucestershire (Gloucestershire County Council, 1999) on transport. Local plans for each of
Table 15.5 Annual modelled travel and energy for trips within Gloucestershire, Option 1—based
on structure plan housing allocations.
the districts were used in conjunction with the structure plan to determine more precise
locations for the housing allocations. The structure plan policy is to locate as much of the
housing allocation as possible within the Severn Vale area, covering Gloucester, Cheltenham
and Stroud. In addition, there is to be some development in Tewkesbury/Ashchurch area and
Cirencester. Within the Forest of Dean, housing would go to support the forest ring—a
cluster of settlements including Lydney (Forest of Dean, 1996). In addition, a number of
villages were identified in the local plans as appropriate for small infill developments. It should
be noted that at the time of analysis, many of the local plans were under review or predated
the structure plan, so the results of this analysis may not truly represent the travel implications
of the structure plan.
The structure plan strategy resulted in a small reduction in travel per capita and an increased
proportion of journeys made by bus (Table 15.5). This gives a reduction in energy
consumption per capita and per trip. These changes reflect the reduction in mean trip length
and an increase in bus use across work, shopping and leisure purposes. However, the structure
plan strategy did result in an increased use of the car for education trips. This is possibly due to
the nature of school bus provision; Gloucestershire education authority provide school buses in
areas where the distance to the nearest appropriate school exceeds a set distance—this distance
varies depending on the type of school. In areas where no school bus is provided, parents are
more likely to escort their children the entire length of the journey to school—often by car.
Thus where the distance is large to the nearest school, bus use prevails whilst those living
closer to the schools rely more heavily on the car.
The structure plan strategy had no impact on the spatial pattern of mean trip lengths. As no
attempt had been made to model the employment or transport policies of the structure plan,
the proportion of trips made by car as opposed to bus remained unchanged and the choice of
destinations and accessibility of them was also unchanged.
Table 15.6 Modelled annual travel resulting from Option 2—a development strategy to tackle
under provision in Gloucestershire.
15.6.2
Option 2
Option 2 aimed to test the implications for travel of boosting the number of services and facilities
in those settlements that our research showed were under provided, i.e. the number of each
facility within the urban area was lower than a settlement of that size could support. For each
settlement that was found to have underprovided in one or more of the nine services studied,
facilities were added to the shopping, leisure and education themes to bring the number of
each facility type up to the expected level. The attraction values used were based on the
average for facilities of that type within Gloucestershire. Jobs were added to the employment
theme based on the number of people expected to be employed by each facility. Again this was
based on the average employee size of similar facilities in Gloucestershire. Thus residents in
those settlements were provided with access to a better range of facilities.
The model was rerun and the following results produced (Table 15.6). In general, this
strategy resulted in a very slight reduction in average trip lengths. There was not much change
for work trips but whether or not a settlement was under provided in terms of employment
was not taken into account in the methodology for allocating services. Average trip lengths for
leisure purposes have changed the most—reducing from 18.2 km per trip to 17.9 km per trip
in the case of leisure trips by bus, a reduction of 1.75%. On the whole, bus trips seemed to be
affected to a greater extent than car trips. This resulted in a very slight increase in the
proportion of travel made by car—91.59% in 1998, 91, 63%. The effect on fuel usage was
even less noticeable with a decrease of 0.2% in fuel used per km travelled and an increase of 0.
2% in fuel used per trip. Spatially, this strategy resulted in decreased trip lengths around
Stroud, Cirencester and the Forest Cluster (Figure 15.3), which were the main areas where
facilities were added.
15.6.3
Option 3
In this strategy housing was allocated to those settlements which had been identified as having
excess capacity in their levels of facility provision, i.e. were over provided for. Population was
Figure 15.3 The Pattern of Mean Trip Lengths Resulting From Option 2.
added to each settlement so that none of the nine services included in the study were then over
provided. Characteristics of the population added, such as the number in employment,
number of school-aged children and car ownership levels, were based on the existing
population of that area. In all, an extra 28,000 population were added. Gloucestershire’s
forecast population for 2001 is 567,100 and for 2011—583,900, i.e. an increase of 26,900. So
this method covers any projections. No additional facilities, services or employment were
added as part of this allocation strategy.
In general this created longer trip lengths as those settlements over provided in terms of
facilities tended to be smaller, more isolated settlements (Table 15.7). Also by adding
population to balance the maximum difference, resulted in some facilities then being under
provided. Finally, the type of facility that tended to be over provided were pubs and garages,
these don’t necessarily serve the local community, i.e. depend on passing trade in respect of
pubs or are specialists, dealing with one type of car, in respect of garages. The percent of
journeys made by car increased slightly (by 0.1% over 1998 levels), as did the distance
travelled per capita (2%) and the fuel consumed per capita (2%). There were no discernable
effects on the spatial distribution of mean trip lengths.
15.6.4
Option 4
In this strategy it was assumed that planning gains from developments would be used to
provide additional services and facilities in locations that were under provided, whilst housing
Table 15.7 Modelled annual travel resulting from Option 3—a development strategy to tackle
over provision in Gloucestershire.
Table 15.8 Modelled annual travel resulting from Option 4—a development strategy to tackle both
under and over provision in Gloucestershire.
allocations would be used to support existing services and facilities in areas which are currently
over provided. The method for implementing this strategy within the model was to combine
the above two strategies, using origin data from the over provision strategy and facility data
from the under provision strategy.
This strategy produced slightly shorter mean trip distances than the strategy to tackle over
provision of services (option 3) but still resulted in a slight increase (less than 1%) in mean trip
lengths and a more substantial increase of 7% for both total distance travelled and fuel
consumed compared with the situation in 1998 (Table 15.8). Fuel consumed per capita
increased by 2% whilst there was no change in fuel consumed per km travelled or in the
percentage of trips made by car as opposed to bus. As for the strategy for tackling under
provision of services, this strategy resulted in shorter trip lengths for those living in Stroud,
Cirencester and the Forest of Dean.
Table 15.9 Modelled annual travel resulting from Option 5—a development strategy to aimed at
supporting an increased number of higher order services in Gloucestershire.
15.6.5
Option 5
The previous two strategies concentrated on placing additional population to support lower
order services and facilities by adding small pockets of housing to existing settlements. This
strategy takes a slightly different approach, concentrating instead on supporting higher order
services, particularly those that require a population in the region of 25,000 to survive.
In this scenario additional population was placed in such a way as to take a number of
settlements above the threshold required to support higher order services than they currently
provide. Research at the Bartlett showed that residents of settlements with a population of
over 25,000 tend to travel shorter distances and use the car less than those in smaller
settlements (Banister, 1999 and Williams, 1997). Williams (2000) also identified a number of
services and facilities that require a population of this size to support them. Thus, settlements
within Gloucestershire that could be expanded to 25,000 in size were identified. Of the
settlements under the threshold, two were closest to this level—Dursley and Cirencester.
Housing was allocated to each of these settlements to bring them up to the required size. Key
services were then added to the level that could be supported by the expanded settlements.
The remaining housing allocation was insufficient to bring a third settlement up to the 25,000
threshold, so was allocated to four smaller settlements to bring them above the 5,000
threshold, thus of a size that could support services such as banks and supermarkets.
This had considerable impact on the travel patterns of the county (Table 15.9). The total
distance travelled by the residents of the county decreased by 9% over 1998 levels despite a 5%
increase in the population of the county. Total fuel consumption dropped by 4%, whilst
average journey length decreased by 13%. These changes are very similar to those achieved
through the structure plan strategy. The most dramatic decreases in journey lengths are for
journeys by bus. This is partially due to the high densities assumed within the two major
settlement expansions of Dursley and Cirencester and the inadequacies of the model in coping
with assigning trips to different modes under these conditions. In reality it is likely that many of
the very short bus journeys would have been made by non-motorised modes, thus an even
more dramatic reduction in fuel consumption could be expected. Journey lengths in the areas
surrounding all five expanded settlements decreased substantially (Figure 15.4). This was
accompanied by a slight increase in mean journey lengths in the area just to the northwest of
Figure 15.4 The Pattern of Mean Trip Lengths Resulting From Option 5.
Gloucester, with residents in these areas being attracted to the new facilities in Dursley and the
smaller centres such as Micheldean which were expanded as part of this option.
15.6.6
Option 6
The final option tested takes a different approach. The basic premise was that housing could be
allocated in such a way as to support public transport services. As no good quality data was
available on the size of population needed to support particular types and frequencies of service
the effect of an improved county-wide bus service on travel patterns of the current residents of
Gloucestershire was modelled, rather than adding the population required for these changes to
take place. This option assumed that all existing bus routes were run with a service frequency
of a least once an hour, in line with findings that the proportion of trips made by bus increases
dramatically if service frequencies are above this level (Titheridge, 2000).
The results showed a small but noticeable decrease in trip lengths (Table 15.10), distance
travelled and energy used (1%, 2% and 1% respectively) compared with 1998 levels. The
number of trips also decreases slightly as bus users are more likely to combine trips, so tend to
make less journeys in total. The distance travelled per capita and fuel consumed per capita also
decreased by 1% compared with 1998. More interestingly, this was the only strategy that
resulted in a reduction in the proportion of trips made and distance travelled by car compared
to bus.
Table 15.10 Modelled annual travel resulting from Option 6—a transport strategy to tackle poor
public transport provision in Gloucestershire.
15.7
CONCLUSIONS
Only small changes in travel patterns were achieved through options 3 and 4, which involved a
housing allocation strategy that concentrated on lower order services and facilities such as
convenience stores and pubs (Table 15.11). Much greater reductions were achieved through
option 5—the strategy to expand as many towns as possible to above a threshold population
size of 25,000. The travel reduction resulting from expanding Dursley and Cirencester to this
threshold was comparable with the travel reduction achieved through the structure plan
(option 1). Thus, development strategies based around expanding smaller settlements to above
a 25,000 population threshold could have considerable impact on travel reduction. This
conclusion is supported by the findings of Banister (1999) and Williams (1997) from analysis
of national travel survey data for different settlement sizes that residents of settlements with a
population of over 25,000 tend to travel shorter distances and use the car less than those in
smaller settlements. Allocating housing in this way could not only reduce travel but also
significantly increase the level of access to key services and facilities for rural populations.
The travel reduction resulting from option 5 was comparable with the travel reduction
achieved through the structure plan (option 1). It should be remembered that the technique
was applied to only a very limited range of facilities and employment sites were not
considered. If a wider range of services, facilities and employment had been included in the
analysis it is likely that much greater reductions in travel could have been achieved.
The results of the modelling exercise also suggest that the key to moving travel onto more
sustainable modes is to accompany any new development with improved public transport
services. Consideration needs to be given to the size of population need to support frequent bus
services. The minimum population levels could be perhaps be obtained through development
along public transport corridors linking major towns or a ring of smaller settlements served by
a circular bus route.
It is recognised that expanding all existing settlements to a population of 25,000 is
impractical. Whilst the resulting changes from allocating additional housing to support lower
order services and facilities were small, it is felt there are additional benefits to be derived from
allocating housing to support services and facilities in smaller settlements. Also, the methods
used for identifying areas of under and over provision and then for balancing population with
Table 15.11 Summary of changes in travel resulting from different development strategies.
* Includes internal trips only.
facility provision were crude, possibly effecting the results that could be achieved in travel
reduction through assigning development based on current facility provision. Consequently, it
is felt that this technique shows potential for being a useful tool for prioritising developments.
Several ways in which the methodology for assigning housing locations suggested in this
paper could be improved have been identified. Clearer guidelines are needed when identifying
suitable sites for development. Some locations are clearly inappropriate for housing
development in terms of access to higher-order facilities, lack of public transport provision etc.
It was also found that no settlement was consistently over provided for in all facility types.
Priority needs to given when allocating housing to supporting certain facility types, more work
is needed to identify which facilities should be prioritised. Finally, the methodology needs to
be expanded to include wider range of facilities and some measure of facility quality.
15.8
REFERENCES
Action for Communities in Rural England, 1998, ACRE’S Detailed Comments on the Comprehensive
Spending Review for the Department of the Environment, Transport and the Regions. (Cirencester:
ACRE).
Banister, D. 1999, Planning more to travel less: land use and transport. Town Planning Review, 703,
pp. 331–338.
Council for the Protection of Rural England, 1998, Rural Transport Policy and Equity. (London:
CPRE, Royal Development Commission and The Countryside Commission).
Department of the Environment, Transport and the Regions, 1997, National Travel Survey 1994/96.
(London: The Stationary Office).
Department of the Environment, Transport and the Regions, 1999, Projections of Households in
England to 2021. (London: HMSO).
Department of the Environment, Transport and the Regions, 2000a, Planning Policy Guidance Note 3:
Housing Revised. (London: HMSO).
Department of the Environment, Transport and the Regions 2000b, National Travel Survey 1997/99
Update. (London: The Stationary Office).
Forest of Dean, 1996, Local Plan Adopted Version. (Coleford: Forest of Dean District Council).
Gloucestershire County Council, 1999, Gloucestershire Structure Plan Adopted Version November
1999.
OPCS, 1994, 1991 Census Workplace and Transport to Work: Great Britain. Part 1, Vol.1. (London:
HMSO).
RTPI, 1998, 1995 GIS Survey. (London: RTPI).
RTPI, 2000, IT in Local Planning Authorities 2000. (London: RTPI).
Titheridge, H. 2000, Balancing housing and facility provision: the transport implications. URBASSS
Working Paper 8, Sept. 2000. (London: The Bartlett School of Planning, University College
London).
Titheridge, H. and Rana, S. 2000, ESTEEM: a technical report, ESTEEM Working Paper 8, June 2000.
(London: The Bartlett School Of Planning, UCL).
Titheridge, H., Hall, S. and Gardner, R. 1998, Sustainable settlements—a model for the estimation
of transport, energy and emissions: model development and calibration. ESTEEM Working Paper 3,
Oct. (London: The Bartlett School of Planning, UCL).
Titheridge H, Hall S And Banister D 2000, Assessing The Sustainability Of Urban Development
Policies In Achieving Sustainable Urban Form edited by Williams, K. Burton, E. and Jenks, M.,
(London: E & FN Spon), pp. 149–159
UWE, 2000, The Interdependence between Urban and Rural Areas in the West of England. (Bristol:
University of West of England), June 2000.
Williams, J. 1997, A study of the relationship between settlement size and travel patterns in the UK.
URBASSS Working Paper 2. (London: The Bartlett School Of Planning, University College
London).
Williams, J. 2000, Tools for Achieving Sustainable Housing Strategies in Rural Gloucestershire.
Planning Practice and Research, 15 (3), pp. 155–174.
PART V
GIS in Socio-Economic Policy

16
Using GIS for sub-ward measures of urban
deprivation in Brent, England
Richard Harris and Martin Frost
16.1
INTRODUCTION
Recently there has been increased interest in defining and locating areas of poverty,
deprivation and social exclusion in the UK. Such terms are difficult to define in any precise and
apolitical sense. Nevertheless, new measures have been devised that aim to calculate
deprivation and poverty rates in consistent, robust and (pseudo-) scientific ways. These
measures include the Department of the Environment, Transport and the Regions’ (DETR)
Index of Multiple Deprivation—IMD 2000—and also the Poverty and Social Exclusion Survey
of Britain (DETR, 2000b, Gordon et al., 2000). The IMD 2000 statistics are easily accessed
from the National Statistics Service (NSS), a website developed by the Office of National
Statistics (ONS), in partnership with central and local government (see
www.statistics.gov.uk). The aim of the NSS is to make statistical information available for
small areas across the UK. Presently the service offers statistics at only the Ward (electoral
district) level. However, the intention is to introduce smaller geographical units based on the
2001 UK Census Output Areas. Average household income estimates will then be assigned to
those units. On the basis that better information begets better policy-making—a rationale
behind the NSS—then this is an important development. This chapter highlights the need for
geographically meaningful income estimates, based on flexible approaches to model building
and area classification.
According to ONS (2000), one of the benefits of the NSS will be to assist neighbourhood
renewal; “to facilitate a better understanding of local problems and effective targeting of
solutions” by allowing assessment of local need. It should therefore prove invaluable to the
various policy-makers and agencies that have responsibility to counter the effects of rising
proportions of deprived households within the UK (14% of households in 1983; 20% in 1990;
24% in 1999: Gordon et al., 2000). At a national scale, income inequality has also risen,
reaching its highest level since 1981 (when comparable data were first collected). The 1999/
2000 annual Family Expenditure Survey (ONS, 2001) shows that the poorest fifth of
households had 6% of national income after tax, while the share held by the top fifth has risen
under the Labour government, from 44% to 45% (Ward, 2001)—not a statistically significant
rise, but one that nevertheless reveals the difficulties of reversing the legacies of income inequality
in the UK.
In our view a greater sensitivity to geographical difference is required to really begin

unlocking the true picture of deprivation, inequality and social exclusion across both the UK
and within specific regions. A social paradox of current times is that processes of both ‘ghetto-
isation’ and fragmentation appear to exist, and in close proximity (Urban Task Force, 1999,
Hall and Pfeiffer, 2000). The consequence is that whereas some neighbourhoods will contain a
large proportion of deprived households, others will also contain ‘hidden’ deprivation,
localised at a more niche scale. The first urban White Paper to be published by a British
government for some two decades recognises that “even in those towns and cities with
significant deprivation there remains a sharp contrast between prosperous areas and those with
most deprivation. For example, Sheffield has two wards amongst the least deprived in the
country just across the city from some deeply deprived areas” (DETR, 2000a). Despite this
insight and its emphasis on wards, ward boundaries were actually designed for administrative
purposes other than monitoring urban deprivation or for targeting neighbourhood renewal per
se. There is little reason to suppose that geographies of deprivation are adequately described by
crisp dividing lines that happen to coincide with the boundaries of the UK’s electoral
geography.
Areal indicators such as IMD 2000 are suited to identifying neighbourhoods of deprivation
at a ward scale only but are insufficient for either identifying households living in poverty at a
sub-ward scale, or for identifying flows of deprivation across administrative boundaries. They
are designed to enable between ward comparisons but without knowledge of within-ward
heterogeneity. It is implicitly assumed that wards provide a suitable foundation on which to
build spatial comparison. That is an assumption which we contest, using two additional
datasets to highlight how deprivation can vary within wards and also cross ward boundaries (a
typical ward contains, on average, about 2000 households, although individual wards vary
considerably in both their population and physical size). The first data source is based on local
administrative records from the London Borough of Brent that allow identification of
households receiving key benefits. Using a simple, point-pattern analysis undertaken using
analytical tools common to many GIS packages we identify local concentrations of households
whose children receive free school meals. We then compare the distribution of those
households against the income deprivation domain of IMD 2000 (DETR, 2000a). The second
source of information is a commercial, lifestyles dataset collected from individual respondents
to a national survey but tabulated at a unit postcode level (for reasons of privacy and data
protection). In the UK, a unit postcode contains an average of 12 to 16 residential, delivery points
(letterboxes). Using income data taken from this dataset we again undertake a point pattern
analysis, identifying, within the Brent study region, local concentrations of low income
households. The geography of low income revealed by the lifestyles-based analysis is also
compared against the IMD 2000. Our analysis suggests that innovations in GIS, and in data
collection and data handling facilitate the use of ‘unconventional’ sources of data to at least
complement ward-based deprivation indicators. The emphasis is more on that contention than
on the detail of the rather UK specific datasets examined.
16.2
DEPRIVATION, THE INCOME DOMAIN AND BRENT
Deprivation is by no means an exclusively urban phenomenon, and, of course, not all urban
areas are deprived. Yet, based on the DETR (2000b) statistics, over 70% of the population
who live in one of the 10% most deprived wards (boundaries as at 1st April 1998), also live in
USING GIS FOR SUB-WARD MEASURES OF URBAN DEPRIVATION 229
one of the main conurbations: Greater London; Greater Manchester; Merseyside; South
Yorkshire; Tyne and Wear; West Midlands; West Yorkshire; or the former county of
Cleveland (DETR, 2000a). DETR (2000a) reports that people living in the English
conurbations have on average, and by comparison to the population-at-large: lower educational
results; lower employment rates; more children living in poverty; an higher mortality rate;
and an increased exposure to violent crime and theft.
The IMD 2000 income domain—which we now refer to as simply ‘the income domain’—is
used to monitor income deprivation by identifying low income families from Department of
Social Security benefits data. It is one of six domains that are combined, with weighting, to
form the composite IMD 2000. The full set of domains and their weightings are: income
(25%); employment (25%); health deprivation and disability (15%); education, skills and
training (15%); geographical access to services (10%); and housing (10%). Both the income
and employment domain scores are calculated as rates. Consequently, if ward X has a score of
40% on the income domain, then it is estimated to have twice as much income deprivation as
ward Y with a score of 20%. In our analysis we consider only the income domain as,
intuitively, this ought to be the most closely associated with the other income information we
have available to us. Further details about IMD 2000 are available in DETR (2000b) or at
http://www.regeneration.detr. gov.uk/research/id2000/index.htm.
We have chosen Brent as a study region because it is indicative of a relatively small urban
area containing a diverse mix of social, economic and demographic conditions. For instance,
the housing stock has fragmented into a complex mix of tenures and despite some
redevelopment (e.g. of 1960s/70s tower blocks) there remain problems of ‘suburban decay’
and a diminishing local economy. At the time of writing the largest disused building within the
Borough is probably Wembley Stadium! Figure 16.1 shows the income domain scores for each
of the 31 wards within Brent. The scores range from 53% of the population living in income
deprived families in Stonebridge, to 12% of the population in Keniton. For comparison, the
most income deprived ward in England has 74% of its population living in low income
families. That is a 40% increase over Stonebridge but is exceptional. Stonebridge ranks as the
111th most deprived ward in England (of 8,414), meaning it is measured to be within the top
two percent most income deprived wards nationally. In fact, four Brent wards are within the
top two percent nationally, and a further three are within the top ten percent. None of the
Brent wards are in the least deprived quartile nationally. DETR (2000b) reports that there is a
total of 72,381 people who are income deprived in Brent.
Figure 16.2 maps the data shown in Figure 16.1. It is a choropleth map with the income
domain scores organised into quartiles—four classes with an (approximately) equal number of
wards in each (but not necessarily equal population). As with any map based on the UK census
or electoral geography, it is not a realistic representation of the urban morphology insofar as
each area contains an undisclosed mixture of residential, commercial, derelict and open land
(plus some areas of water). Harris and Longley (2000) suggest using Ordnance Survey UK’s
Code-Point product as a basis for space-filling out from unit postcode centroids (which are
listed in the dataset with a one metre precision in the majority of cases), using surface
estimation and modelling procedures to identify areas of either residential or commercial land-
use. Here we are interested in locating residents (who are deprived) and so our map would
provide better contextual information if we excluded non-residential areas. To do so we have
used less sophisticated techniques than those adopted by Harris and Longley (2000), but ones
that are essentially standard to any GIS package: we first created a 100 metre buffer zone
around each of the properties defined as residential within Brent authority’s central property
database (the database was recently updated and ascribes a geo-reference of one metre
Figure 16.1 Income deprivation measured for Brent wards.

precision to every property listed within it—ideally, every property in Brent); we then dissolved
the buffer zones into one theme (or map layer); finally, we used the new theme as a template
to ‘cookie cut’ out the residential areas of Brent from the non-residential, census geography. The
resulting map layer is shown in Figure 16.3 (with the property locations also shown) and is, of
course, an artefact of the modelling procedure—specifically, the source data and the buffer
width. That buffer width, of 100 metres, is somewhat arbitrary but was selected to ensure that
buffers drawn around individual points overlapped, meaning they can be merged (dissolved)
together to produce a map that does not appear overly fragmented in terms of its residential
area. Although a generalization of the true residential limits, Figure 16.3 is still a more realistic
model of Brent’s residential geography than the census geography of Figure 16.2.
16.3
FREE SCHOOL MEAL ELIGIBILITY AS A MEASURE OF
DEPRIVATION
Two wards are highlighted in Figure 16.3, and labelled as A and B. Ward A has an income
domain score in the upper quartile for Brent; ward B a score in the lower quartile. Since it is
low income households who are most eligible to receive free school meals, so we would
expect to find a greater concentration of free meal households in ward A, than in ward B. A
cursory inspection of Figure 16.4—which maps the geographical distribution of free meal take-
up within the study region—suggests the proposition to be true: there appears to be a greater
number of free meal recipients in ward A than in ward B.
The centre of each circle plotted in Figure 16.4 is defined by a geo-reference assigned from
the Brent property database to an household receiving free school meals. Approximately 80%
of free school meal recipients listed in the benefits database have been matched, by address, to
the property database. This ascribes a precise location to each household but, given that this is
personal information, requires permission to be obtained from the UK Data Protection
Registrar for data storage and analysis. The size and shading of each circle indicates the local
concentration of free school meal households around and including each single household
point. In detail, a simple point-pattern analysis has been undertaken that operates by centring a
Figure 16.2 Income domain scores for Brent wards.

circular window, of radius 500 metres, on each of the 1,634 household points that receives
free school meals. The total number of recipient households within the 500 metre radius focal
region is then found and that value assigned back to the central point (the household record).
The outcome of this procedure is one of aggregation, drawing-out local trends in the data,
whilst, in principle, partially smoothing out random errors. The procedure is analogous to
operating a low pass filter on remotely sensed imagery. Again, interpretation of the results is
contingent on the modelling procedure; specifically, the somewhat arbitrary radius
specification, subsequent classification of the results (quartiles in Figure 16.4) and also the map
symbols (varying the size and the density of their shading can emphasise or de-emphasise
apparent geographical trends). Note that because the circular, focal windows overlap across
the study region, so the sum of the modified data values will be greater than the original sum
of 1,634 households. The effect is one of double counting (or more). If the circles did not overlap
then only the central point would ever be found in each focal window and, consequently, the
data values would not be changed. The 500 metre radius has been selected to ensure overlap
and because, from experience, it strikes a suitable balance that neither over-, nor under-
smooths the data. However, it should be noted that our implicit assumption that the 500 metre
value is ‘correct’ is a generally unproven assertion which has not, in the context of this
chapter, been subject to more rigorous testing.
In Figure 16.4, households that receive free school meals appear geographically
concentrated in wards with relatively low income domain scores. Such households are not
exclusively contained in the most income deprived wards, however. Table 16.1 shows the
distribution of free meal households across the quartile grouping of Brent wards by income
domain score. The table shows that whilst there is an higher incidence of free meals allocated
to households in the poorest quartile wards (516 households, a 32% share), over two-thirds of
households receiving free school meals are within other wards; indeed, 315 (19%) are in the
relatively most affluent wards.
Figure 16.3 Residential areas in Brent.

In the analysis we are not actually comparing like with like, but an absolute quantity
(number of free meal households) against a rate (income domain score). It could be that the
poorest quartile of wards contains the highest number of free meal households simply because
it also contains the largest share of the total population (there being no guarantee that 1991
Census electoral wards in any part of the UK will have near equal population sizes). In other
words, a larger number of people would, ceteris paribus, lead to a greater number of eligible
households. As it happens, the observed share of free meal households for the poorest quartile
is (at 32%) greater than the expected share from the total population alone (27%).
Conversely, the observed share in the most affluent quartile (19%) is less than the expected
share from the population (23%)—see Table 16.1. However, the observed and expected
values are also positively correlated: a 0.60 correlation using Spearman’s rank statistic; 0.61
with Perason’s correlation coefficient; and a 0.55 likelihood that the observed and expected
values are not independent, based on a chi-squared analysis. Caution should be applied when
interpreting these statistics since the magnitude of any correlation tends to increase with
aggregation and here we are considering a broad grouping of wards into only four classes.
Furthermore, free school meal up-take is a narrow and unreliable measure of income
deprivation that necessarily excludes families without children of school age and the more aged
members of society. It is also not perhaps the case that take-up of free school meals correlates
well with eligibility to that right. What we can say with certainty, however, is that there are at
least 315 households that are income deprived (insofar as they receive free meals) but who are
also living in relatively affluent wards.
Imagine the results reflect a general trend: that one-fifth of all income deprived households
are living in wards that are not classified as particularly deprived on the income domain scale.
This would raise obvious concerns about the use of broad scale, area measures for redistributing
remedial funds to the most needy. Of course, we cannot and do not substantiate such concerns
based our results alone. The one fifth value should rightly be considered speculative.
Figure 16.4 Simple point-pattern analysis of free school meal households.
Table 16.1 Distribution of free school meal households across quartile grouping of Brent wards by
income domain score.
* Quartile 1 has least income deprivation.

+ Quartile 4 has most income deprivation.
Nevertheless, what we do argue, and can show, is that pockets of deprivation exist at sub-ward
scales but these are ‘hidden’ or ‘averaged-out’ by ward scale indicators that have no
corresponding measures of internal variation or diversity (and to define a mean without also
defining the variance is to only tell half the story). That contention is a prelude to the next section
where we use simple GIS techniques to search for ‘niche’ pockets of deprivation on the basis of
information taken from a commercial dataset that was originally complied primarily for the
purposes of direct marketing.
Figure 16.5 Household incomes in two Brent wards.
16.4
TARGETING CLUSTERS OF DEPRIVATION WITH LIFESTYLES
DATA
Figure 16.5 shows estimates of the proportion of households in each of 11 income bands for
the two wards named previously as A and B. The income bands increase at £5,000 intervals
from zero, until reaching the eleventh band which is open-ended (total family income of £50,
000 or more, per year). The data source is a commercial, ‘lifestyles’ database formed by replies
to a national, consumer survey. That survey was postal, undertaken in England, Wales and
Scotland by a commercial data vendor during the spring of 1999, and sent out to a high
proportion of households for whom at least one member was listed on the Electoral Register
for the three countries. (In principle all adults should be listed on the Register since it is a legal
offence not to be so. However, that is somewhat of an ideal). The full lifestyles dataset
contains a wide range of socio-economic, behavioural and consumer information that describe
approximately one million households. From that dataset we have extracted 1,825 records
corresponding to households within Brent who responded to the survey question ‘which group
best describes your combined household income?’ (survey’s original emphasis). The Royal Mail’s
Postal Address File (PAF) for the same year lists 98,314 residential, mail delivery-points in the
Borough. On that basis the lifestyles sample is of just under 2% of households within Brent,
which is consistent with the national average. Within ward A the sample is of 2.4% and in
ward B, 2.3%. Strictly speaking these are slight over-estimates since a single mail delivery-
point is sometimes shared by two or more households (for example converted terraced
properties where a single letterbox serves two or more apartments).
Figure 16.5 reveals that in ward A, a minimum of 10 of the 11 income groups can be found.
We know this because the lifestyles survey asked real people about their actual income status.
Unless the respondents lied or were mistaken, then we know there is at least one household
present from each of ten income groups. In ward B, all 11 groups are present. The results point
to diversity at sub-ward scales, but not chaos. Figure 16.5 also suggests evidence of
geographical difference, with ward A containing higher proportions of lower income families
(notably those earning under £10,000 per year). Although that finding is entirely consistent
with ward A having a lower IMD 2000 income domain score, we are not claiming that use of
lifestyles data will necessarily yield precise and accurate estimation of income deprivation rates
at ward or sub-ward scales. To the contrary, Longley and Harris (1999) have previously
discussed the problems associated with using lifestyles data in socio-economic research. Although
some of their findings are specific to a different dataset, their general conclusions remain valid
here. In particular, the survey respondents are recognised as essentially self-selecting, deciding
whether or not to return a survey questionnaire, and, as such, they may not form a
representative sample of the population-at-large.
The self-selection of survey respondents should not be regarded as solely a random error.
For example, Harris (1999) gives limited evidence that it is the 18–24 age group, and also the
most and least affluent members of society who are least likely to respond to a consumer
survey of the type considered here. Yet, neither can the self-selection easily be treated as
giving rise to a systematic bias which could, in principle, be corrected by weighting the data
against a second (more accurate) dataset. The problem here is twofold. Firstly, there is no
obvious dataset that can be used to ‘ground truth’ the lifestyles data other, perhaps, than at
comparatively coarse scales of aggregation (where official social survey data might be used).
Secondly, the propensity to respond to the consumer survey is ultimately a matter of
individual choice and is not unambiguously related to any ‘obvious’ variables like age,
affluence, location or lifestage. The propensity to respond retains a random component.
Furthermore, though there is, arguably, a decreased response from the less affluent, that is not
to say no lower income households will respond (as Figure 16.5 proves). With all this in mind,
what we suggest is that the lifestyles data can be used to identify, at the unit postcode level, the
location of every survey respondent who indicated their annual family income to be less than
£10,000. If these locations can be shown to be clustered spatially, then we would have cause to
believe that they are likely areas of income deprivation, particularly if a reasonable number of
the clusters were shown to be within, or in proximity to wards with high income deprivation
according to the IMD domain scores.
In accordance with our argument, Figure 16.6 maps the location of households with annual
family income less than £10,000 who responded to the lifestyles survey. The point-pattern
methodology described in Section 16.3 has again been used to draw-out patterns in the data. A
difference is that the centre of each circle is now defined by the location of a unit postcode
centroid, ascribed to each household from the Postal Address File (PAF) with a precision of
100 metres or greater. In general there is a large degree of correspondence between the
geography of income deprivation revealed from the lifestyles analysis and that suggested by the
income domain scores. A visual inspection of Figure 16.6 shows that most clusters of low
income households are either fully or partly within wards that are in the upper quartile of
income domain scores for Brent (the most income deprived). Yet, not all are. Furthermore,
there are a number of instances where the ward boundaries are revealed to artificially partition
low income clusters into two or more different areas. In relation to cartography and its
application in policy analysis, Figure 16.6 is, perhaps, a more honest map than Figure 16.2,
say, because Figure 16.6 gives a more vivid impression of where income deprivation is to be
found in wards and also allows the reader to gain an impression, albeit partial, of social
heterogeneity within the region. The uniformity of populations that census-based mapping
Figure 16.6 Point-pattern analysis of lifestyles data to identify low income households.
implies is not assumed. In particular, Figure 16.6 makes clear a central message of this report:
income deprivation does not respect ward boundaries!
16.5
CONCLUSION: A FLEXIBLE APPROACH FOR AREA
CLASSIFICATION
Whilst we welcome the development of the National Statistical Service and the ready access it
offers to policy relevant data, our report has shown how too rapid (and possibly glib)
identification of ‘poor areas’ within cities can conceal a dispersion of income deprived households
in areas that are otherwise regarded as relatively affluent on average. The intention of the NSS
to provide smaller scale income estimates based on census output areas will be an important
step forward for neighbourhood analysis. However, average measures should be set alongside
measures of income variation, at the same scale. An average statistic is never truly meaningful
until a measure of diversity (variance) is also provided.
We have sought to develop the basis of a simple, ‘bottom-up’ approach to identifying income
deprived areas, using analytical tools available in ArcGIS (see also Harris, 2001). A refinement
to the methodology would be to replace the fixed radius in the point-pattern analysis with an
adaptive kernal that is more sensitive to the residential geography and urban morphology
surrounding each point. Measures of statistical significance could be introduced, identifying
whether a low income cluster really is ‘unusual’ or merely, perhaps, an artefact of residential
postcode geographies. The uncertainty of the lifestyles data could also be more explicitly
incorporated within the modelling procedure. Such methods exist, having been developed in
the fields of population surface modelling (Martin, 1998) and geocomputation (Longley,
Brooks and McDonnell, 1998, Caldwell, 2000: see also http://www.geog.leeds.ac.uk/
research/ccg.html). Here, however, we are more immediately concerned with applying tools
that are available to users of ‘standard GIS’.
Our report has suggested that it is possible to experiment with non-official or non-
governmental sources of data, within a GIS framework, at sub-ward levels of analysis. The
application of lifestyles data, that are regularly up-dated by annual survey, may provide a
useful component of the suggested bottom-up approach, if—and it is a ‘big if’—issues of
representation and survey/response bias are suitably resolved. We provisionally suggest that
the data can be aggregated in geographically sensitive ways to a level where use of the data is
statistically robust and where the scale of analysis is not overly coarse. There is no reason why
the aggregation should be bounded by geographically inappropriate administrative units and,
by the act of aggregation, some of the ethical and confidentiality concerns of using detailed,
household data are avoided. The possibility is opened-up of identifying, at small spatial scales,
subtle dimensions of poverty and exclusion to complement the accelerating development of
standard neighbourhood statistics. It is common practice in standard geodemographic or area
profiling to group areas according to their socio-economic ‘type’, then using a range of
descriptive statistics to identify the differences between the various classes. A similar approach
could be adopted here, using the lifestyles data to search for different components of
deprivation that distinguish the geographical clusters of, say, low income households. For
example, there is evidence of an age related dimension within Brent, with the smaller clusters
to the north west of the region (near ward B) having a greater incidence of widowers and
people wearing an hearing aid than the larger south-central cluster (near ward A).
These are avenues for future research, however. For the present, we note that, for the first
time, some lifestyles data have been made directly available to academics who register at
MIMAS (http://www.mimas.ac.uk/docs/experian/). Although the data are currently limited
to age, vehicle, property, household and population records at the postal sector level, the
information is up-to-date and will be of interest to a number of researchers.
16.6
ACKNOWLEDGEMENTS
We are grateful to Brent local authority and to Claritas Europe for the supply of the data
analysed in this report. The lifestyles data are copyright © Claritas Europe (http://
www.claritas.com) and are reproduced with permission. The responsibility for any errors or
omissions arising from the analysis are solely our own.
16.7
REFERENCES
Caldwell, D.R., 2000, Editorial: developments in geocomputation. Computers, Environment and
Urban Systems, 24, pp. 379–382.
DETR, 2000a, Our Towns and Cities: the Future: Delivering an Urban Renaissance. (London: The
Stationary Office).
DETR, 2000b, Indices of Deprivation 2000. http://www.regeneration.detr.gov.uk/rs/03100/
index.htm.
Gordon, D., Adelman, L., Ashworth, K., Bradshaw, J., Levitas, R., Middleton, S., Pantazis, C.,
Patsios, D., Payne, S., Townsend, P., Williams, J., 2000, Poverty and Social Exclusion in Britain.
(York: Joseph Rowntree Foundation).
Hall, P., Pfeiffer, U., 2000, Urban Future 21. (London: E & FN Spon).
Harris, R.J., 1999, Geodemographics and the Analysis of Urban Lifestyles. Unpublished Ph.D. Thesis
(University of Bristol: School of Geographical Sciences).
Harris, R., 2001, On the diversity of diversity; is there still a place for small area classification?
Area, 33, pp. 329–336.
Harris, R.J., Longley, P.A., 2000, New data and approaches for urban analysis: modelling
residential densities. Transactions in GIS, 4, pp. 217–234.
Longley, P.A., Brooks, S.M., McDonnell, R., (editors) 1998, Geocomputation: a Primer. (Chichester:
John Wiley & Sons).
Longley, P.A., Harris, R.J., 1999, Towards a new digital data infrastructure for urban analysis and
modelling. Environment and Planning B, 26, pp. 855–78.
Martin, D., 1998, Automatic neighbourhood identification from population surfaces, Computers,
Environment and Urban Systems, 22, pp. 107–120.
ONS, 2000, Internet document, http://www.statistics.gov.uk/neighbourhood/downloads/
nssrev5.pdf.
ONS, 2001, Briefing paper, http://www.statistics.gov.uk/pdfdir/etbrief0401.pdf.
Urban Task Force, 1999, Towards an Urban Renaissance. (London: E & FN Spon).
Ward, L., 2001, Earnings inequality widens under Labour. The Guardian, April 18th. http://
www.guardian.co.uk/Archive/Article/0,4273,4171558,00.html.
17
The spatial analysis of UK local electoral
behaviour: turnout in a Bristol ward
Scott Orford and Andrew Schuman
17.1
INTRODUCTION
There is growing interest in the factors that influence turnout in elections in the UK. This
concern has become paramount with the extremely low turnout in the 2001 General Election.
Although research in this area is well established, there has been very little concerned with the
geography of turnout, particularly at the local level. This research aims to address this omission
by examining the geographical factors that influence turnout in a local election in a ward in
Bristol, UK. A GIS of the ward was constructed using voting data taken from the marked-up
electoral register used in the local election. The results suggest that both contextual factors,
such as the size of the household in which the voter lives, and geographic factors, such as the
distance from the household to the polling station, are important in understanding the
propensity to vote.
17.2
BACKGROUND
Studying election turnout has long been a useful way of examining both democratic and societal
participation. Elections are frequent, occur at different spatial scales, and their results are
easily quantifiable. Recent trends in the UK show generally falling levels of turnout by voters
in elections. Although not universally acknowledged as problematic, most academics and
politicians see declining turnout as symptomatic of a general political and social malaise, which
produces a democratic deficit, apathy and indifference to societal issues. Both the EU and
British Government have not only expressed their concern, but also provided funds for
substantial research in this area, through for example the 5th Framework RTD programme and
the ESRC’s ‘Democracy and Participation’ programme. Similarly at the local level, many local
authorities in the UK (for example, Bristol) have established ‘Democracy Commissions’ tasked
with examining the problems of turnout and participation in localities.
17.3
ELECTORAL GEOGRAPHY
Geography remains important in all election studies for a number of reasons. First, elections
are organised geographically, through defined constituencies and wards. Second, election
results often show distinctive geographical variations in voting patterns, the most well known
in the UK being the North-South divide of the 1980s between a generally Conservative-voting
South and a Labour-voting North (see Johnston et al. 1988). Third, voting is always place-
specific: local factors and political attitudes will always affect voting decisions. Fourth,
electoral representation (seats rather than votes) is also distinctively geographical—for
example in the 1997 General Election the Conservative party gained 17.5 % of the votes in
Scotland but gained no seats. By contrast, the Liberal Democrats with only 13 % of the votes
gained 10 out of the available 72 seats—primarily because, unlike the Conservatives, they got
their votes in the right places. Lastly, and following on, there is a geography to the power and
policy that comes from such a political representation.
The reasons for low electoral turnout can be broadly categorised into social (social
exclusion, alienation), social/administrative (political institutions, structures and political
mobilisation) and administrative (voter facilitation). While evaluating ways of increasing
electoral participation has been much discussed (Miller, 1988; Rallings and Thrasher, 1990,
1994, 1997; Rallings et al. 1996; Rallings et al. 1994), little of this work has yet fully taken on
the embeddedness of all participatory process, social and administrative, posited by Agnew
(1987, 1996) and demonstrated by Schuman (1999). Generally overlooked in most voting
studies is the particular influence of local factors on political attitudes and voting decisions.
Since Cox’s (1969) seminal work on the importance of the local geographical context to
voting, much has been studied and developed both with specific regards to voting (Agnew,
1987, 1996) and broader regard to place and the contextuality of action (Johnston, 1991;
Thrift 1983). Whilst there remain sceptics (Rose and McAllister, 1990; McAllister and
Studlar, 1992) the relationship between voting activity and where the activity takes place has
been increasingly demonstrated in a number of areas. Strong links have been seen between
voting and spatial variations in economic prosperity (Pattie and Johnston, 1995; Pattie et al.
1997) as well as between voting and local conditions such as local campaigning (Denver and
Hands, 1997; Pattie et al. 1995; Schuman, 1999).
Less demonstrated however (with the exception of Zuckerman et al. 1994, 1998) is the
‘neighbourhood effect’ which suggests that an important factor in local context is the social
networks that the individual voter participates in. Only recently has this gap begun to be
remedied and Pattie and Johnston (1999, 2000) take some key steps forward. Despite an
extensive survey of existing work on the neighbourhood effect (see also Taylor and Johnston,
1979; Books and Prysby, 1991, 1999; Huckfeldt and Sprague, 1995; Miller, 1977) they
bemoan the fact that few studies have investigated the hypothesis that “people who talk
together vote together” directly rather than through inference. Most of the work to date, they
argue, fails both to uncover the mechanisms behind the neighbourhood effect, and fails to deal
with it at an appropriate scale, far too often relying on large scale constituency or regional data.
Using two hundred and eighteen polling districts as ‘sampling points’, Pattie and Johnston
(2000) conducted an investigation into party conversational effects using the 1992 British
Electoral Survey which had included questions about whom people talked to about politics.
While there were differences in the strengths of the observed conversational/political effects
between parties (usually because of other contextual effects like variations in party
campaigning intensity), they provide clear evidence that conversation and context influence
voting. Dividing the conversation effect between ‘family’ (spouses and other relatives) and
‘non-family’ (work, friends, neighbours etc.) a further distinction is noted: “…people who
spoke with their kin were 4.9 times more likely to switch their support to the Conservatives”
(pp. 59) whereas with non-family they were only 2.5 times more likely to switch. Similarly for
Labour, the figures were 5.6 times for family and only 2.2 times for non-family, a huge family
THE SPATIAL ANALYSIS OF UK LOCAL ELECTORAL BEHAVIOUR 241
bias. Their results strongly suggest that “people listen most attentively, and are converted by,
discussants from within their families. Non-family discussants are also influential, but not as
strongly: the main sway occurs as a result of within-family discussions.” (pp. 59).
In providing a challenge to the assumption of methodological individualism that underlies
many British voting studies Pattie and Johnston (2000) suggest a number of avenues. The first
concerns the neighbourhood effect itself. Pattie and Johnston (and others too) concentrate on
the party neighbourhood effect, looking at the way in which people socialise each other to vote
for a particular party. This will always have methodological problems in that voting is by secret
ballot. If, rather, we examine the voting neighbourhood effect (i.e. people socialising each other
to vote, not just for a party but rather to vote at all (and the two are obviously related)), then
there are a number of methodological opportunities, explored below, as well as many potential
policy applications.
17.4
CONTEXTUALISING VOTING BEHAVIOUR
In the tradition of place-based research (Holt and Turner, 1968; Milne and McKenzie, 1954,
1958; Sharpe, 1967; Denver and Hands, 1997; Bochel and Denver, 1971, 1972; Schuman,
1999) this paper brings together context and democratic renewal, attempting to confirm and
supplement Pattie and Johnston’s (2000) key findings in a place context. This is done by
examining individual and grouped voter turnout in a ward through the use of a marked-up
electoral roll. The only similar work of this type comes from Taylor (1973) who examined the
Victoria ward in Swansea at the 1972 local elections and Dyer and Jordan (1987) in Aberdeen.
Utilising returns from one of the party’s telling activity (in which party campaigners had asked
voters the candidate that they had voted for on leaving the polling station) Taylor was most
interested in studying if there was any distance decay effect from where voters lived to polling
stations. He found a weak relationship between actual distance and turnout but a much
stronger relationship between perceived distance and turnout.
With particular reference to Taylor’s and Dyer and Jordan’s work, there are a number of
ways in which this paper seeks to improve their findings. First, by using the marked-up
electoral register used by the polling station administrators we can get a definitive record of
who did and did not vote that is not dependent on party campaigners getting voting
information from electors (who are not legally obliged to give it to them). Second, a much
more accurate framework for analysing network distances, topography and other local aspects
can be achieved by constructing a GIS of the ward under investigation. Third, a more complete
view of the election can be gained by supplementing the turnout data with data relating to
other contextual effects like local party activity and advertising.
The paper presents results from an initial small scale study, examining turnout in a single
ward for a single election. The aim of the research is to move towards a more contextual
approach to understanding voter turnout behaviour; one that takes into account the different
scales at which voting behaviour takes place (the individual, the household, the polling
district). The paper uses simplified assumptions regarding spatial voting behaviour as a means
of identifying basic trends, although future adaptations are also discussed. Such an approach
allows a ‘complete’ electoral geography to be established, one that addresses all reasons for low
voter turnout; the social, the administrative, and the geographical.
17.5
METHODOLOGY
17.5.1
Study Area
The electoral ward of Westbury-on-Trym (or Westbury) in Bristol, UK, was chosen as a study
area. Westbury was chosen since historically it is one of the few wards in Bristol that
consistently experiences high electoral turnouts. Westbury is sub-divided into three polling
districts referred to as A, B and C. These are shown in Figure 17.1, which also shows the
location of the polling stations and the road network. The principal polling district boundary is
formed by the A4018, a main thoroughfare linking the centre of Bristol to the M5 to the west.
The boundary of A and C is formed by the main road linking the centre of Westbury with the
neighbouring ward of Stoke Bishop. With the exception of those living in north of polling
district A, no voter has to cross a main road to get to a polling station. Polling district B
contains the local shopping and administrative centre of Westbury. The social and
demographic characteristics of Westbury are presented in Table 17.1. This reveals that it has a
high percentage of old and retired people and also people of a high social class. These
demographics are reflected in the local election results with Westbury being a traditional
Conservative strong hold as presented in Table 17.2.
17.5.2
GIS and Voting Behaviour
To analyse the geography of voting behaviour at the micro-scale, a GIS was created to map and
model local voting patterns. Voting data relating to the 1998 Local Council Elections was
acquired for Westbury. The voting data comprised of a marked-up electoral register for
Westbury that recorded every person in the ward that had registered to vote, together with
information upon whether they had voted in person, voted by postal vote or had not voted at
all. These data are kept temporarily on record to avoid electoral fraud and as a means of
reference for legal challenges to election results and are in the public domain. Since this data is
held at the individual level, it can be used as a means of investigating the effects of location
upon voting. This individual data was aggregated into households based upon the addresses of
the voters and the percentage of votes per household was then calculated. Postcodes allowed
the households to be geo-referenced to Ordnance Survey grid co-ordinates relating to the
centroid of the unit postcode. These were then converted into a point coverage which formed
the basis of a GIS. Other coverages incorporated into the GIS included road network data,
footpath network data, topographical data, and data relating to the boundaries of the three
polling districts within the ward. The site of each polling station was also identified.
Socio-economic information relating to the voting population was modelled using Super
Profile data. This allocates a socio-economic category to every postcode in the ward.
Figure 17.2 shows the geography of Super Profiles in Westbury. The majority of voters are
classed as being affluent achievers or thriving greys, reflecting Westbury’s status as a wealthy
suburb with a high percentage of retired people. The only diversity is to be found to the east of
the ward in polling district B, which contains hard-pressed families and producers (blue-collar
workers).
Figure 17.1 The ward of Westbury-on-Trym in Bristol, UK showing polling districts and polling
stations.
Spatial analytical functions in the GIS were used to generate measures of local geographical
context. Distance decay measures were generated using two methods. First, a simple measure
of Euclidean distance was generated from each household to the corresponding polling station.
Second, a more sophisticated measure of distance was generated using the topology of the road
network and footpaths with the GIS used to calculate the minimum network distance from
each household to the polling station. The topographical information was used in the GIS to
ascertain whether a person has to travel up or down hill to the polling station. This was
estimated by calculating the differences in height between each household and the
corresponding polling station.
The GIS was used to construct summary statistics regarding the local context of each polling
district. These are shown in Table 17.3. Polling district A is the largest by area although
polling district B contains the largest number of registered voters. Polling district A also has
the voters who, on average, have to travel the farthest. Polling district C is by far the smallest
and compact with respect to both size and number of voters. With respect to travel behaviour,
polling district A has the least car parking provision whilst polling district B, located in the
village centre, has the most opportunities for multiple trip destinations.
Figure 17.3 shows an electoral density surface of registered voters in each polling district
constructed using the GIS. This allows a comparison to be made between the centre of the
electoral density in each polling district and the location of each polling station. In terms of
Table 17.1 The demographic and socio-economic profile of Westbury (from the 1991 Census of
Population).
Table 17.2 Local election results for Westbury 1994–1998.
optimal location, polling district C has the best-sited station with respect to electoral density
whilst A has the worst. It is also important to recognise the influence that the local campaign may
have on voting behaviour. Electorates in polling district A had an extra card to remind them to
vote a week before the election. This was a minor experiment undertaken by the Electoral
Services Unit in the city council to ascertain the affect that an extra reminder has on voting.
Finally, the weather on the day of the election was overcast but dry, commonly recognised as
encouraging turnout.
17.5.3
Postal Voting
When analysing voting behaviour it is important to distinguish between different methods of
voting. Fundamentally, there are three recognised methods: voting in person, voting by post
and voting by proxy in person (when a person registers to vote for someone else on their
behalf). In the context of this research, there is an important distinction between postal votes
and voting in person since the latter will be more significantly influenced by local contextual
effects, particularly issues of accessibility to the polling station. Postal votes tend to be utilised
by people who are absent on the day of the election and people with mobility problems (for
other reasons see Halfacree and Flowerdew, 1993).
The majority of postal votes within Westbury were for voters living in single households
(75%), reflecting the bias towards old people living on their own. Very few postal votes were
for households of more than two people. At least half the postal votes represented electorates
Figure 17.2 The geography of super profiles in Westbury.

resident in old people’s homes. Table 17.4 shows that polling district C has the greatest
proportion of postal votes, reflecting the large number of old people’s homes in this area.
Since the marked-up electoral register only indicates that a person has chosen to vote by post,
and not whether they have actually voted in the election, this may positively bias the turnout
figures if not taken into account. Hence postal votes were removed from the sample (see also
Dyer and Jordan (1987) for similar recommendations).
17.6
ANALYSING VOTING BEHVIOUR
17.6.1
Voting by Polling District
Table 17.4 summarises the election turnout for Westbury as a whole and the three polling
districts. Overall, 42.75% of the electorate voted compared to an average of 30% for Bristol
as a whole indicating that it was a good turnout. When the votes are proportioned by polling
district, a distinct geography emerges. Whereas polling districts A and C both had similar
Table 17.3 Polling district context.
Figure 17.3 A comparison of polling station location with electorate density.

above average turnout, polling district B had significantly less numbers of people voting. Broadly
this conforms to expectations. Polling district C is the smallest in terms of area, roughly
circular in morphology with the polling station almost coincidental with the centre of the
electorate density and has a resident socio-economic class that has a high propensity to vote.
Hence it is expected that this polling district should have the highest turnout. The result for
polling district A is a little more surprising given its morphology and distribution of voters.
However, its favourable socio-economic class with respect to the propensity to vote and the
fact that it received an extra reminder may account for its above average turnout. Polling
district B, with its combination of lower socio-economic classes and elongated morphology has
the lowest turnout.
Table 17.4 Voting turnout in the 1998 local elections.
17.6.2
Voting by Household
Pattie and Johnston (2000) have demonstrated that households are the basic unit of analysis
with respect to voting for a political party and hence this may also be the case for voting
turnout. In particular, we might expect a socialisation effect whereby people within a
household encourage or discourage each other to vote. Hence, voting behaviour was analysed
by households. Figure 17.4 shows the percentage of votes per one person household in each
polling district. It is clear that turnout in one person households is disproportionately higher in
C than in other polling districts, reflecting the importance of postal votes of single person
households in this polling district. Figure 17.5 shows the affect of removing these votes, with
the turnout in polling district C falling in-line with the rest of the polling districts in the ward.
Table 17.5 shows voting by household size (i.e. the number of registered voters resident in
the household) by polling district with postal votes removed. This reveals two aspects of voting
behaviour at the household level. First, there is a clear decline in voting in all three polling
districts as the size of the household gets larger. Second, within households there appears to be
a ‘dual-voter’ effect. This is the phenomenon that occurs in households of two or more
people, in which two people in the household have a greater propensity to vote than one
person. In other words, people are more likely to vote in pairs than vote on their own and so
there are less than the expected number of single voters in households of two, three and four
voters.
17.6.3
Contextual Effects on Voting Behaviour
To understand the effects of local context upon voting behaviour in more detail, the
geographical variation in the percentage of votes per household was analysed with respect to
differences in socio-economic class, accessibility to the polling station and topography.
Table 17.6 is a summary of the parameters of a regression model of percentage of votes per
household against socio-economic class. The omitted base variable is affluent achievers. It can
be seen that there is no significant difference in the propensity to vote between the omitted
variable of affluent achievers and thriving greys, but the propensity to vote becomes
significantly less (at the 5% level) in areas of lower socio-economic class. Hence the propensity
to vote is almost 10% less in areas of settled suburbans and hard-pressed families.
Interestingly, settled suburbans have a lower propensity to vote than producers do, but this
may be because they live farther from the polling stations and reflect a distance decay effect.
Table 17.7 summarises the parameters of a regression model of percentage of votes per
household against Euclidean distance controlling for differences in socio-economic class. The
Figure 17.4 Percentage of votes per one person household.
Figure 17.5 Percentage of votes per one person household with postal votes removed.
model reveals that, after controlling for distance, settled suburbans propensity to vote has
increased by almost 2.5%. In addition, distance has a significant influence on voting behaviour,
decreasing the propensity to vote by 5.4% per kilometre.
Table 17.8 summarises the parameters of a regression model of percentage of votes per
household against network distance controlling for differences in socio-economic class. The
model reveals that network distance has a much stronger effect on voting behaviour (T-statistic
is double that of Euclidean distance in Table 17.7) than simple Euclidean distance, reducing the
propensity to vote by 8% per kilometre. Moreover, the propensity of settled suburbans to
vote becomes similar to that of thriving greys and affluent achievers (T-statistic is not
significant at 5% level) implying that the initial reduction in the propensity to vote associated
with settled suburbans in the previous model was actually a bias introduced by the use
Euclidean distance as an inadequate measure of voting behaviour. Hence, it is only hard-
pressed families and producers that vary in their voting behaviour, being less likely to vote than
the other residents in the ward.
Table 17.9 summarises the effects of topography on the percentage of votes per household
after taking into account the differences caused by socio-economic class. It reveals that
topography has a very strong effect upon voting, reducing the turnout by 3.6% for every ten
Table 17.5 Percentage of people voting by household size.
Table 17.6 The effect of socio-economic characteristics upon voting.
Table 17.7 The effect of Euclidean distance on voting.
metres difference in height between the polling station and the household. This effect is
strongest in polling district A, which contains the highest point in the ward.
Table 17.10 shows the results of a regression model examining the relationship between
network distance and propensity to vote between the three polling districts. Within each
polling district, there tends to be very little variation in the socio-economic class of voters and
hence most of the Super Profile variables are insignificant in the models. In the larger of the two
polling districts, network distance has the greatest effect on the propensity to vote, reducing
the vote by over 7% per kilometre. This is only 3.6% per kilometre in the smaller and more
compact polling district C.
Table 17.8 The effect of network distance on voting.
Table 17.9 The effect of topography on voting.
17.7
DISCUSSION
The factors affecting the turnout in local elections are complex. The research has indicated that
turnout is influence by household size, polling district context and local geographical factors.
Household size is the most important factor influencing the propensity to vote, with the
number of votes decreasing with household size. The result was that twice as many voters did
not turnout in household of four voters than two voters. In addition, the dual-voter effect
suggests that households socialise each other to vote or not to vote, supporting Pattie and
Johnston (2000) research. The result is a significant decline in single voter turnout in
households of two voters or more.
Polling district context is important since it has been shown that the smallest and most
compact polling district (C) also has the largest turnout. Local campaigning also appears to
have an affect, indicated by the comparatively high turnout in polling district A given its
morphology. The fact that voters in this polling district had an extra card a few days before the
election reminding them to vote may have increased overall turnout in this district.
Perhaps the most interesting results are those associated with the local geography of voting.
It has been shown that factors such as social class, distance to the polling station and
topography are all influential in the propensity to vote. The effect of social class on voting is
well documented and the research confirmed the decline in turnout associated with areas of
lower social class. With respect to distance, the research supports Taylor’s (1973) findings of a
negative relationship between distance to the polling station and voting turnout but also
quantifies this reduction. A decrease of 8% per kilometre is a significant reduction in turnout,
one that may be crucial in election in a marginal ward. Although on average the distance
travelled by a voter was actually quite small (an average of half a kilometre in the ward as a
whole), the fact that network distance has a far stronger affect on turnout than Euclidean
distance suggests that voters are particularly sensitive in their travel behaviour. This is
exemplified by the fact that network distance has the least effect on voting in the most
Table 17.10 The effect of network distance on voting by polling district.
compact polling district (C). The sensitivity to local geography is also highlighted by the
significant influence of topography on voting turnout.
17.8
CONCLUSIONS
This research has shown that geography is important in the study of election turnout. Although
only a small scale study, it has highlighted some interesting features. The use of a GIS has allowed
the interaction between voting behaviour and local context to begin to be unravelled. It would
seem that the propensity to vote is partly a function of both household size and local
geographical context. Not only is there a dual-voter effect but the distance a voter has to travel
and the topography they encounter also influences their voting behaviour. The voter also
appears to be very sensitive to local context and even small variations in distance travelled would
seem to make a difference. However, the proportion of the variation in turnout explained by
these models was relatively small, indicating that more influential factors exist that were not
included. These factors are probably related to issues of voter apathy and a general
disillusionment with mainstream politics.
The research has been undertaken using simplified assumptions and measurements. The
geographical distribution of different sized households has not been taken into consideration
when measuring the affect of local geographical context on turnout. Since it has been shown that
the propensity for a person to vote is influenced by household size, the distribution of
households may become important. Geographical factors such as distance, topography and
social class may also interact, potentially confounding their effects on turnout. A possible
remedy would be to utilise the GIS to construct an aggregate measure of distance, topography
and social class that could also include other factors such as the number of turns that a voter
makes when travelling to the polling station. In addition, developing recent work in spatial
behaviourism (e.g. Golledge and Stimson, 1997) will provide a much more nuanced idea of
‘perceived’ distances to polling stations than solely using simple network distances. A final
limitation is that no account has been made to contextualise the voter at the individual level. In
this respect a voter’s age, mobility and other characteristics such as education may become
influential in determining propensity to vote. However, such information would be very
difficult to collect without recourse to a much larger survey
To conclude, electoral participation is an issue that is becoming an increasing concern to
both academics and political parties. Voter turnout is generally declining, particularly amongst
certain sections of society (for instance, the young and first time voters in particular). There
are numerous reasons for this, such as voter apathy and a general disillusionment with
mainstream politics. However, this research has demonstrated that administrative factors are
also influential and these have local policy implications. These include the siting of polling
stations, the demarcation of polling district boundaries and the importance of local
campaigning. Issues of accountablity and ‘Best Value’ will undoubtedly become more
important as voter turnout increases in its political significance. Hence it can be argued that
methods and techniques, such as GIS, that can be used to measure and monitor voter turnout
and facilitate the implementation of policy will become an important part in administrating
future elections.
17.9
REFERENCES
Agnew, J.A., 1987, Place and Politics: The Geographical Mediation of State and Society. (London: Unwin
Hyman).
Agnew, J.A., 1996, Mapping politics: how context counts in electoral geography. Political
Geography, 15, pp. 129–146.
Bochel, J.M. and Denver, D., 1971, Canvassing, turnout and party support: an experiment. British
Journal of Political Science, 1, July, pp. 257–269.
Bochel, J.M. and Denver, D., 1972, The impact of campaigning on the results of local government
elections. British Journal of Political Science, 2, pp. 239–244.
Books, J.W. and Prysby, C.L., 1991, Political Behavior and the Local Context. (New York: Praeger).
Books, J.W. and Prysby, C.L., 1999, Contextual effects on retrospective economic evaluation: the
impact of the state and local economy. Political Behavior, 21, pp. 1–16.
Cox, K.R., 1969, The voting decision in a spatial context. Progress in Human Geography, 1,
pp. 81–117.
Denver, D. and Hands, G., 1997, Modern Constituency Electioneering. (London: Lass).
Dyer, M.C. and Jordan, A.G., 1987, Who votes in Aberdeen? Marked Elector Registers as a Data Source.
Working paper, University of Aberdeen.
Golledge, R.G. and Stimson, R.J., 1997, Spatial Behavior: A Geographic Perspective. (London:
Guilford).
Halfacree, K. and Flowerdew, F., 1993, The relationship between inter-constituency migration and
postal voting. Electoral Studies, 12, pp. 247–252.
Holt, R.T. and Turner, J.E., 1968, Political Parties in Action: The Battle of Barons Court. (New York: The
Free Press).
Huckfeldt, R. and Sprague, J., 1995, Citizens, Politics, and Social Communication: Information and
Influence in an Election Campaign. (Cambridge University Press: New York).
Johnston, R.J., 1991, A Question of Place: Exploring the practice of Human Geography. (London:
Blackwell).
Johnston, R.J. Pattie, C.J. and Allsopp, J.G., 1988, A Nation Dividing? The Electoral Map of Great
Britain 1979–1987. (London: Longman).
McAllister, I. and Studlar, D., 1992, Region and voting in Britain: territorial polarization or
artefact? American Journal of Political Science, 36, pp. 168–199.
Miller, W.L., 1977, Electoral Dynamics in Britain since 1918. (London: Macmillan).
Miller, W.L., 1988, Irrelevant Elections. (Oxford: Oxford University Press).
Milne, R.S. and Mackenzie, H.C., 1954, Straight Fight. A study of voting behaviour in the constituency of
Bristol North-East at the General Election of 1951. (London: Hansard Society).
Milne, R.S. and Mackenzie, H.C., 1958, Marginal Seat. A study of voting behaviour in the constituency of
Bristol North-East at the General Election of 1955. (London: Hansard Society).
Pattie, C.J. and Johnston, R.J., 1995, ‘Its not like that round here’: region, economic evaluations
and voting at the 1992 British General Election. European Journal of Political Research, 28,
pp. 1–32.
Pattie, C.J. and Johnston, R.J., 1999, Context, conversation and conviction: social networks and
voting at the 1992 British general election. Political Studies, XLVII, pp. 877–889.
Pattie, C.J. and Johnston, R.J., 2000, “People who talk together vote together”: an exploration of
contextual effects in Great Britain. Annals of the Association of American Geographers , 19, pp. 41–66.
Pattie, C.J., Dorling, D. and Johnston, R.J., 1997, The electoral politics of recession: local economic
conditions, public perceptions, and the economic vote in the 1992 general election. Transactions of
the Institute of British Geographers NS, 22, pp. 147–161.
Pattie, C.J., Johnston, R.J. and Fieldhouse, E.A., 1995, Winning the local vote: the effectiveness
of constituency campaign spending in Great Britain, 1983–1992. American Political Science Review,
89, pp. 969–983.
Rallings, C. and Thrasher, M, 1990, Turnout in English local elections: an aggregate analysis with
electoral and contextual data. Electoral Studies, 9, pp. 79–91.
Rallings, C. and Thrasher, M., 1994, Explaining Election Turnout. (London: HMSO).
Rallings, C. and Thrasher, M., 1997, Local Elections in Britain. (London: Routledge).
Rallings, C. Temple, M. and Thrasher, M., 1994, Community Identity and Participation in Local
Democracy. (London: Commission for Local Democracy).
Rallings, C., Thrasher, M. and Downe, J., 1996, Enhancing Electoral Turnout-A Guide to Current
Practice and Future Reform. (York: Jospeh Rowntree Foundation).
Rose, R. and McAllister, I., 1990, The Loyalties of Voters. (London: Sage).
Schuman, A.W.E., 1999, Boundary Changes, Local Political Activism and the Importance of the Electoral
Ward: An Electoral Geography of Bristol 1996–1999. (Bristol: Unpublished PhD thesis, University of
Bristol).
Sharpe, L.J., 1967, Voting in Cities: The 1964 Borough Elections. (London: Macmillan).
Taylor, A., 1973, Journey time, perceived distance, and electoral turnout—Victoria Ward,
Swansea. Area, 5, pp. 59–62.
Taylor, P.J. and Johnston, R.J., 1979, Geography of Elections. (Harmondsworth: Penguin).
Thrift, N.J., 1983, On the determination of social action in space and time. Society and Space, 1,
pp. 23–57.
Zuckerman, A.S., Kotler-Berkowitz, L.A. and Swaine, L.A., 1998, Anchoring political
preferences: the structural bases of stable electoral decisions and political attitudes in Britain.
European Journal of Political Research, 33, pp. 285–321.
Zukerman, A.S., Valentino, N.A. and Zuckerman, E.W., 1994, A structural theory of vote choice:
social and political networks and electoral flows in Britain and the United States. Journal of
Politics, 56, pp. 1008–1033.
18
Towards a European peripherality index
Carsten Schürmann and Ahmed Talaat
18.1
INTRODUCTION
Article 2 of the Maastricht Treaty states as the goals of the European Union, inter alia, the
promotion of harmonious and balanced economic development, convergence of economic
performance, improvement of the quality of life and economic and social coherence between
the member states. The proposed Trans-European Transport Networks (TETN) will play a
prominent role in achieving these goals as they are to link landlocked and peripheral areas with
the central areas of the Community. The identification of those peripheral regions, whose
accessibility and transport infrastructure systems are to be improved, is becoming of great
political importance. This is underlined by the European Commission’s Cohesion Report (1997)
which emphasises that regions should measure policy success, regularly monitor results and
regularly inform the public and political authorities of progress.
This paper presents the results of a Study on Peripherality undertaken for DG XVI, Regional
Policy, of the European Commission. The purpose was to undertake, for the fifteen EU states
and twelve candidate countries, the calculation of an index of peripherality of the ‘potential’
type. The economic potential of a region is the total of destinations in all regions weighted by a
function of distance from the origin region. It is assumed that the potential for economic
activity at any location is a function both of its proximity or ‘travel time’ to other economic
centres and of its economic size or ‘mass’. The influence of each economic centre on any other
centre is assumed to be proportional to its volume of economic activity and inversely
proportional to a function of the distance between them. The economic potential of a given
location is found by summing the influence on it of all other centres.
The calculation of such peripherality indices involves the acquisition, integration, storage
and analysis of various spatial and statistical data sets. The use of GIS functions and techniques
was indispensable in conducting such calculations. GIS capabilities and techniques, such as
overlaying, network analysis, geo-database management, statistics and presentation were
comprehensively utilised demonstrating the benefits of the use of GIS in the field of measuring
locations peripherality. For this purpose, an integrated GIS-based European Peripherality Index
software system was developed to facilitate the calculation of peripherality indices, scenarios
comparison, data updating and results demonstration.
The paper presents selected results of the study, compares the different peripherality
indices, explains the software system developed and concludes with suggestions for further
refinements of the methodology.
18.2
PERIPHERALITY INDICATORS
Fundamentally, a peripherality indicator can be interpreted as an inverse function of
accessibility, i.e. the higher the accessibility, the less peripheral a region is located and vice
versa. Accessibility indicators can be used to analyse peripherality in several ways: regions can
be classified into central and peripheral regions, impacts of different policy measures such as
transport investments can be evaluated, or impacts of accessibility on regional development
can be analysed. The accessibility indicators used in this study are based on the assumption that
the attraction of a destination increases with size and declines with distance or travel time or
cost. Therefore both size and distance of destinations are taken into account. The size of the
destination is usually represented by regional population or some economic indicator such as
total regional gross domestic product (GDP) or total regional income. The activity function
may be linear or non-linear. Occasionally the attraction term Wj is weighted by an exponent α
greater than one to take account of agglomeration effects, i.e. the fact that larger facilities may
be disproportionally more attractive than smaller ones. One example is the attractiveness of
large shopping centres which attract more customers than several smaller ones that together
match the large centre in size. The impedance function is non-linear. This is the main idea of
the so called potential accessibility (Hansen, 1959; Keeble et al., 1982; 1988; Schürmann et
al., 1997; Schürmann and Talaat, 2000a; Wegener et al., 2000). Generally a negative
exponential function is used in which a large parameter β indicates that nearby destinations are
given greater weight than remote ones.
The mathematical formula that calculates the accessibility A for a region i over all regions j
can be expressed as follows:
(18.1)
where cij is the generalised cost of reaching region j from region i.

Potential indicators are frequently expressed in percent of average accessibility of all regions
or, if changes of accessibility are studied, in percent of average accessibility of all regions in the
base year of the comparison.
The model developed is capable of calculating a large number of different output indicators.
The range of indicators available will be briefly explained in the following paragraphs.
18.2.1
Spatial Aggregation
All calculations of peripherality indices are based on level 3 of the Nomenclature of Territorial
Units for Statistics (NUTS) defined by Eurostat and are then aggregated to levels 2, 1 and 0 of
the NUTS for the EU member states (Eurostat, 1999a) and equivalent geographical units as
identified by Eurostat for the candidate and EFTA countries (Eurostat, 1999b) by averaging
over NUTS-3 regions weighted by NUTS-3 region population. The smallest unit available for
this study, i.e. the NUTS 3 level, represents counties or local authority regions, whereas
NUTS 0 level represents countries; the two levels in between these extremes represent other
or standard regions, depending on the country considered.
TOWARDS A EUROPEAN PERIPHERALLY INDEX 257
18.2.2
Modes
Since speed limits for cars and trucks differ and statutory drivers’ resting periods affect freight
transport, all indicators were calculated separately for passenger and freight road transport.
Travel time matrices and peripherality indices for cars represent the perspective of service firms
and consumers, namely how many opportunities, such as clients, markets or tourist facilities
can be reached from location. Travel time matrices and peripherality indicators for lorries, i.e.
for goods transport, can be interpreted from the perspective of producers on (potential)
markets as the answer to the question which location has the highest market potential.
18.2.3
Mass Terms
Peripherality indices are calculated for each origin region by adding up the mass of each
destination region weighted by a function of distance from the origin region. Usually, the mass
is measured in terms of gross domestic product (GDP). In this study, also GDP in purchasing
power standards (PPS), employment and population are used as mass terms. Distance is
measured as the average travel time from one region to every other region in the form of a
matrix. The regions are represented by their ‘centroids’, i.e. their main urban centres.
Statistical data representing the four mass terms were compiled from Eurostat (1997) and
linked to the relevant regions in an integrated database.
18.2.4
Type of Indicator
All peripherality indices are derivatives of potential accessibility. Two different types of
peripherality indices are defined:
• Peripherality Index 1 (PI1): The region with the highest potential accessibility, i.e. the most
central region, is defined to have a peripherality index of zero. The region with the lowest
potential accessibility, i.e. the most remote region, is defined to have a peripherality index
of one hundred. The peripherality index of all other regions is a linear interpolation
between zero and one hundred proportional to their potential accessibility. The higher the
peripherality index, the higher the peripherality.
• Peripherality Index 2 (PI2): The average potential accessibility of all regions weighted by
regional population is defined to be one hundred. The peripherality index of all regions is
calculated as potential accessibility expressed in percent of average accessibility. The higher
the peripherality index, the lower the peripherality. Peripherality Index 2 is therefore in
fact a standardised accessibility indicator.
18.2.5
Spatial Scope of Standardisation
The standardisation was done for three different territories: EU member states, EU member
states plus five candidate countries (Estonia, Poland, Czech Republic, Hungary, Slovenia) and
EU member states plus twelve candidate countries (the five countries above plus Latvia,
Table 18.1 Available output.
Lithuania, Slovakia, Romania, Bulgaria, Cyprus, Malta). This implies that the values of the
regional peripherality indices differ depending on the territory covered.
Based on the above classification (4 NUTS levels, 2 modes, 4 mass terms, 2 types of
indicators, 3 territories), 4×2×4×2×3=192 possible output indicators can be calculated and
mapped. Table 18.1 summarises the 192 possible output indicators. The numbers in the table
are consecutive numbers used for identifying the resulting maps and output files (see
Schürmann and Talaat, 2000b).
18.3
IMPLEMENTATION IN ARCINFO
A software system was developed based on ESRI’s Arclnfo to calculate the European
Peripherality Index (E.P.I.) which consists of several macros written in Arc Macro Language
(AML) (Schürmann and Talaat, 2000b). The E.P.I, is intended to facilitate the updating of
input data, the definition of scenarios, the calculation of peripherality indices and the
presentation of results.
The system itself consists of three core components: the INITIAL macro, the CALCUL
macro and the PLOT macro. The INITIAL macro defines global variables and parameters and
initialises ASCII input files. The CALCUL macro calculates the indicators presented
in Table 18.1, stores them in output coverages and INFO tables and exports them into ASCII
files. The PLOT macro is used to produce output maps showing the resulting peripherality
indices. In addition, a number of add-ons were developed to support certain tasks for updating
the geodatabase or to perform error checking. As a principle, updates of the database or

changes of the parameters can be achieved by editing the input coverages and parameter files.
It is not necessary to edit the code of the macros itself to change default settings. The User
Manual (Schürmann and Talaat, 2000b) gives a comprehensive description of the model
system, explains how to run the system and how to edit the database and suggests how to
handle errors.
The CALCUL macro is the central core macro for the calculation of the peripherality
indices. It calculates all values for all indicators, modes, levels, territories and masses in one
run. Such calculation is conducted within the CALCUL macro according to the following
steps:
1. Link travel times for all links in the input network coverage are calculated or updated
based on national speed limits as initialised by the INITIAL macro. The basic road
network coverage used in this study was extracted from the European road network
database developed by IRPUD (1999). The national speed limits for cars and lorries for
urban roads, major roads, expressways and motorways were compiled from ADAC
(2000), IRU (2000) and UBA (1998). Moreover, link travel times take account of speed
constraints in urban and mountainous areas and sea journeys. Speed limits and congestion
in urban areas are estimated as a function of population density at NUTS-3 level. It is
assumed that the higher the population density, the slower will be the speeds. Road
gradients are estimated by overlaying the road network with a digital terrain model
(DTM) extracted from the U.S. Geological Survey (2000).
2. As a pre-requisite of calculating accessibility indicators, travel time matrices are
calculated based on average travel times between all NUTS-3 regions. These travel time
matrices are used to calculate regional accessibility indicators, which are then converted
to peripherality indices. Shortest path analysis models offered by Arclnfo were utilised to
derive such travel time matrices. Beyond link travel times, the travel time matrices take
account of border delays and ferry boarding times, and, in case of freight transport,
statutory drivers’ resting periods. The matrices are derived by applying the nodedistance
command between all regions’ centroids and with the link travel times as link impedances
and border waiting times and ferry boarding times as node impedances.
3. Based on the travel time matrices, accessibility indicators are calculated for NUTS-3
regions using formula 18.1.
4. Peripherality indices for the three different spatial scopes for standardisation are then
calculated and standardised for the NUTS-3 level.
5. The travel times between NUTS-3 regions are then aggregated to travel times between
NUTS-2, 1 and 0 regions as weighted averages over NUTS-3 regions.
6. Statistical spatial analyses offered by Arclnfo were run over the obtained accessibility
indicators for NUTS-3 region in order to aggregate values to levels 2, 1 and 0 of the
NUTS by averaging over NUTS-3 regions weighted by population.
7. The peripherality indices for all NUTS-0, 1 and 2 regions considered are then calculated
and standardised.
18.4
THE E.P.I. SYSTEM
The first two macros of the core components of the E.P.I software, i.e. the INITIAL and
CALCUL macros, run automatically, whereas the PLOT macro depends on continuous user
interaction. The INITIAL macro does not require user input after the macro is started; the
CALCUL macro invokes a selection menu after its start, but after all options are set no further
user input is requested; in case of the PLOT macro, a windows-based selection menu is
permanently invoked until the macro is stopped.
As the first step to calculate the required indicators, the INITIAL macro is executed. The
macro sets global variables, parameters and path names necessary to perform calculations,
reads and initialises ASCII input parameter files and establishes a set of INFO tables.
As a second step, the CALCUL macro is run to calculate all 192 peripherally indices. After
its execution a selection menu is called at which the input coverages and output options are
selected (Figure 18.1). Two input coverages must be specified, the roads network and the
regions coverage. Then, the reference year has to be chosen. The user can choose to assess the
peripherality at the base year 2000 or to apply a future scenario for 2016 (which is the target
year for full implementation of the TEN programme). A third selection refers to the output
options, which determine the way the results of the E.P.I, calculations are stored and
presented to the user. The standard output option is to present the indices as Arclnfo
coverage. The coverage will always be generated and this option cannot be unchecked.
Additionally, there are three output options available which can be checked or unchecked in
any combination. These are: (i) Travel time matrices (ASCII files) so that travel time matrices are
exported to ASCII files, (ii) Travel time matrices (INFO tables) so that travel time matrices are
maintained as INFO tables, or (iii) Peripherality indices (ASCII files), so that peripherality indices
are exported to ASCII files.
These options enable or disable output of the travel time matrices and peripherality indices
depending on which results are of interest and how they are going to be further processed. The
standard output coverage enables full Arclnfo capabilities with respect to further analysis or
way of presentation. However, if the results are to be integrated into a document or into
further analysis using other software, the internal Arclnfo formats might not be appropriate. In
that case, ASCII files may be generated which enable data exchange between different
hardware platforms and software systems.
When the index calculations finish, a report is displayed (Figure 18.2) which summarises
input and output coverages as well as the output ASCII files created.
Finally, the PLOT macro is run to allow the display and presentation of the final results of
peripherality indices produced by the CALCUL macro. The PLOT macro invokes an easy-to-use
user interface (Figure 18.3) which enables the user to set the combinations of peripherality
parameters to be displayed (map parameters) or to select a map by its number according to the
naming convention described earlier in Table 18.1. Map parameters are the parameters used to
calculate peripherality indices. There are five parameters which determine peripherality index
calculations: Index, Mode, Level, Territory and Mass. For each parameter several options exist.
Each combination of options produces a different peripherality index map. In total, it is
possible to produce 192 different maps based on 192 parameter combinations. After the map
parameters are set, the composed map can be displayed, printed and/or exported to a
different format. Available export format options are png, tiff or ai. The user can also choose
between A4 and A0 output size.
Figure 18.1 Menu interface of the CALCUL macro of the E.P.I, software.
Apart from the above three core macros, additional macros were developed for the E.P.L
software system to support updating of the geodatabase. These macros are applied if the road
network is edited, new centroids are chosen or new region data are introduced to the region
coverage.
18.5
RESULTS
The results showed that the general spatial patterns of peripherality are very similar across all
indicators calculated, reflecting the fact that distant geographical location cannot be fully
compensated by transport infrastructure (Schürmann and Talaat, 2000a). However, each
indicator emphasises certain aspects of peripherality. So, the choice of the type of peripherality
index to be used becomes a matter of concern. Depending on the purpose of the study, a
certain indicator type may be more appropriate than another type as certain subsets of regions
yield slightly better or worse results with respect to peripherality. As an example, Plate 8
shows the peripherality index with respect to GDP by lorry for NUTS 3 level regions. Regions
in the European core show the highest accessibility and so are most central. These regions are
located along the ‘Blue Banana’ in western Germany, Belgium, in the southern parts of the
Netherlands, in northern France and in southern England. A band of regions from the Po
estuary towards Milan and Lyon up to the Channel coast show above-average accessibility.
Regions in the Baltic countries, in Romania and Bulgaria show index values of less than 10, i.e.
the most peripheral regions, due to their—compared to EU member states—still relatively
poor economic performance.
Based on other studies (Fürst et al., 2000; Schürmann et al., 1997) and on theoretical
considerations, the peripherally index shown in Plate 8 together with peripherality with
Figure 18.2 CALCUL macro summary report menu.

respect to population by car were proposed as the standard peripherally indices. The
correlation of these two indicators confirms a high degree of similarity (Plate 7). Although the
overall correlation seems clear, there are some small, but nevertheless important differences
between both indices. In general, central regions in Benelux, Germany and France, but also in
the UK show comparably higher values for peripherality with respect to GDP than with
respect to population. On the other side, regions in the candidate countries have higher
accessibility to population than to GDP. This is because most of the candidate countries have
relatively poor economic performance but large populations which confirms the observation
that peripherality index with respect to population seems less polarised than peripherality with
respect to GDP. In other words, if peripherality with respect to GDP is used, central regions
appear less peripheral; if peripherality with respect to population is used the candidate
countries appear less peripheral.
However, a number of comparisons between the other peripherality indices showed that the
choice of indicator has great influence on the results. The overall spatial patterns of all
peripherality indices are very similar, so correlation between different indicators are rather
high. This reflects the fact that, irrespective of the kind of peripherality index used, the distant
geographical position of peripheral regions cannot be fully removed by transport infrastructure
improvements.
Peripherality with respect to population by car is less polarised than peripherality with
respect to GDP by lorry reflecting the fact that population is more evenly distributed across
Europe and that because of faster driving speeds for cars a greater number of opportunities can
be reached. On the other hand, peripherality with respect to lorry favours regions around the
Channel coast, since for lorries (i.e. freight transport) the ‘barrier effect’ of the Channel is
much less than for cars (i.e. passenger movements), because for private trips the Channel
Figure 18.3 User-interface of the plotting macro of the E.P.I. software.
seems to be a greater obstacle. Candidate countries benefit more if peripherality with respect
to population by car is used; conversely, central regions benefit more if peripherality with
respect to GDP by lorry is used.
The type of indicator has relatively little influence on the results. Standardisation between
the minimum and maximum shows slightly more differentiation among peripheral regions,
whereas standardisation on the European average shows slightly more polarisation between the
central regions. The greater the territory used for standardisation is (i.e. the more candidate
countries are taken into account), the lower is the European average, and the more will
regions in EU member states improve their relative position.
Due to the overall objective of using purchasing power standards, GDP in PPS has slight
balancing effects compared to GDP in Euro; nevertheless, peripherality with respect to both is
more polarised than peripherality with respect to population or employment.
Finally, the higher the NUTS level, the greater is the loss in spatial differentiation. Studies
based on the NUTS-3 level yield a great number of detail and differentiation between and within
peripheral and central regions. This is particularly true for the relatively small German, French
and Italian regions.
18.6
ROLE OF GIS
With its functionality, the software system can be seen as a GIS-based contribution to a wider
system for measuring and monitoring the success of EU policies with respect to peripherality
and cohesion. The use of GIS has enhanced the accessibility modelling process in various ways.
The integration between spatial and statistical data from different sources in one unified
database allowed easier data accessibility, retrieval, maintenance and analysis. The application
of the spatial statistics functions of GIS resulted in more accurate and faster calculation and
aggregation of the accessibility indicators. The execution of network analysis models offered by
Arclnfo made it feasible to calculate travel times among a very large number of centres. The
presentation capabilities of GIS allowed a clear and easy visualisation and interpretation of the
obtained results. Finally, the development of such integrated system allowed simple
construction of different scenarios and an easy comparison of scenarios.
For calculating distance measures, i.e. average road travel times of passengers and goods,
the developed software takes account of road types, speed limits for cars and lorries,
congestion in urban regions and delays due to mountainous areas, national borders and
maximum driving hours of lorry drivers. In this the system goes beyond the way usually travel
times are measured in accessibility studies. Moreover, peripherality indices are calculated for
NUTS-0, NUTS-1, NUTS-2 and NUTS-3 regions based on a unified and disaggregate
approach. That was achieved by incorporating GIS overlay capabilities along with network
analysis functions such as nodedistance, impedance and turntable.
Additional strengths of the system can be seen in its flexible modular structure which is
expandable, in the core macro that calculates all peripherality indices in one model run, in the
variety of output options available, in the minimised number of user interactions, in the
possibility to run the system under UNIX or Windows NT/Windows 2000, in the fact that all
input coverages and input files can be manually edited, adjusted or exchanged, in the
combination of windows-based menu operations designed for user-friendliness and command
line executions designed for efficiency, and in the capabilities provided for designing future
scenarios.
The integration of all types of data storage, analysis, updating, presentation and scenario
building in one integrated system, without the need to export/import data to/from another
software, is another very important advantage of the use of GIS in such application.
Compared to these strengths, the software has only little weaknesses. One is the relatively
long processing time of the core macro which is due to the fact that it calculates all 192
indicators in one run. Also the relatively large amount of disc space required for temporary
coverages and for storing results might limit the applicability of the model. In the current
version, the model considers only road traffic and neglects rail, air and inland waterways and
so is not able to calculate intermodal accessibilities. Moreover, only accessibility of the
potential type can be calculated, whereas daily accessibility or average travel costs are not
taken into account.
18.7
CONCLUSIONS
The present study showed how the developed system utilises GIS capabilities to serve and
support the assessment of EU policies with respect to peripherality and cohesion. In summary,
for all kind of indicators, regions in western Germany, northern France, Belgium, the
Netherlands, southern England and northern Italy show the highest accessibilities and can be
considered the most central regions. When NUTS-3 regions are considered, great differences
in peripherality can be found between peripheral regions, for example in Scandinavia, Greece
and on the Iberian Peninsula. This indicates that the model is able to capture relatively small,
but nevertheless important differences. When higher NUTS levels are considered, these details
partly disappear.
The system evaluated the peripherality of EU member states and candidate countries by a
number of different peripherality indices with respect to NUTS levels, modes, mass terms,
types of indicator and spatial scope of standardisation. The software system developed offers
the full range of combinations of these parameters, totalling 192 indicators. The general spatial
patterns of peripherality are very similar across all these indicators, reflecting the fact that
distant geographical location cannot be fully compensated by transport infrastructure. Each
indicator emphasises certain aspects of peripherality. So, the choice of the type of peripherality
index to be used becomes a matter of concern. Depending on the purpose of the study, a
certain indicator type may be more appropriate than another type.
From a theoretical point of view, it would be of great interest to incorporate also the other
modes, namely rail, air and inland waterways, into the system to enable calculations of
intermodal accessibilities and peripherality indices. Also of interest would be the possibility to
calculate daily accessibility or average travel costs. A more practical extension would be to
incorporate a ‘scenario manager’ which would allow generation, management and application
of different scenarios.
18.8
ACKNOWLEDGEMENTS
The theoretical concepts of accessibility and peripherality is based on previous work performed
at IRPUD in the EU project Socio-economic and Spatial Impacts of Transport Infrastructure
Investments and Transport System Improvements (SASI) commissioned by DG VII (Transport) of the
European Commission as part of the 4th RTD Framework Programme and in the Working
Group ‘Geographical Position’ of the Study Programme for European Spatial Planning
(SPESP) organised by DG XVI.
The definition of the centroids used for the accessibility calculations and the compilation of
the socio-economic database used in this study were contributed by Andrew Copus, Rural
Policy Group, Scottish Agricultural College (SAC), Aberdeen.
18.9
REFERENCES
Allgemeiner Deutscher Automobil Club (ADAC), 2000, Tempolimits. http://www. adac.de.
European Commission, 1997, Cohesion and the Development Challenge Facing the Lagging Regions.
Fourth Cohesion Report. Regional Development Studies. (Luxembourg: Office for Official
Publications of the European Communities).
Eurostat, 1997, New Chronos Database. Tables d3pop, xe_gdp, pvdOe. (Luxembourg: Office for Official
Publications of the European Communities).
Eurostat, 1999a, Regions. Nomenclature of Territorial Units for Statistics—NUTS. (Luxembourg: Office
for Official Publications of the European Communities).
Eurostat, 1999b, Statistical Regions in the EFTA Countries and the Central European Countries (CEC) .
(Luxembourg: Office for Official Publications of the European Communities).
Fürst, F., Schürmann, C, Spiekermann, K. and Wegener, M, 2000, The SASI model. Demonstration
Examples. SASI Deliverable D15. (Dortmund: Institute of Spatial Planning).
Hansen, W.G., 1959, How accessibility shapes land-use. Journal of the American Institute of Planners
25, pp. 73–76.
International Road Union (IRU) (editor), 2000, Speed Limits, Goods Transport. IRU Information
Center. http://www.iru.org/IRU_NEWS/IRU/ECOMMERCE/Main/Ecommerce.E.asp,
October 2000.
IRPUD, 1999, European Transport Networks. Dortmund: Institute for Spatial Planning, http://
irpud.raumplanung.uni-dortmund.de/irpud/pro/ten/ten_e htm.
Keeble, D., Offord, J. and Walker, S., 1988, Peripheral Regions in a Community of Twelve Member
States. Luxembourg: Office for Official Publications of the European Communities.
Keeble, D., Owens, P.L. and Thompson, C., 1982, Regional accessibility and economic potential
in the European Community. Regional Studies 16, pp. 419–432.
Schürmann, C. and Talaat, A., 2000a, Towards a European Peripherality Index: Final report. Report for
General Directorate XVI Regional Policy of the European Commission. (Dortmund: Institute of
Spatial Planning).
Schürmann, C. and Talaat, A., 2000b, Towards a European peripherality Index: User Manual. Report for
General Directorate XVI Regional Policy of the European Commission. (Dortmund: Institute of
Spatial Planning).
Schürmann, C., Spiekermann, K. and Wegener, M., 1997, Accessibility Indicators: Model and Report.
SASI Deliverable D5. (Dortmund: Institute of Spatial Planning).
Umweltbundesamt (UBA), 1998, Geschwindigeitsbegrenzungen. http://www. umweltbundesamt.de/
uba-info-daten/daten/geschwindigkeitsbegrenzung.htm. Last updated May 09 1998, (Berlin:
UBA).
U.S. Geological Survey, 2000, GTOTO30 Digital Terrain Model. EROS Data Center. http://
edcdaac.usgs.gov//gtopo30/gtopo30.html.
Wegener, M., Eskelinen, H., Fürst, F., Schürmann, C. and Spiekermann, K., 2000, Indicators of
Geographical Position. Final Report Part 1 of the Working Group ‘Geographical Position’ of the
Study Programme on European Spatial Planning (ESPON). (Bonn: Bundesamt für Bauwesen und
Raumordnung). Draft version availabe at:http://www.nordregio.se/spespn/Files/1.
1.final1.pdf.
19
Using a mixed-method approach to
investigate the use of GIS within the UK
National Health Service
Darren P.Smith, Gary Higgs and Myles I.Gould
19.1
INTRODUCTION
This chapter describes an ESRC-funded research project which is examining the current and
potential uses of Geographical Information Systems (GIS) within the UK National Health
Service (NHS). We contend that previous questionnaire-based studies in this area have
provided partial understandings of this issue but have not, to date, fully teased out the
underlying factors which facilitate and/or constrain GIS uptake and use within primary and
secondary care. In this paper, we outline the research design that is being used to address this
caveat which, in turn, serves to illustrate Gatrell’s (2001) call for the adoption of mixed-
method research frameworks within health-based research.
In this chapter we describe a mixed-methods research project which is examining the types
of factors influencing variations in the use of GIS within the UK NHS. Previous studies,
described below, have provided important insights into the current levels of take-up of GIS,
but have tended to involve quantitative analysis usually based principally on postal and
telephone questionnaire approaches. Our current research project has also used questionnaire
surveys, but merely as a context for exploring factors influencing both levels of adoption and
current utilisation of GIS, via qualitative techniques based around semi-structured interviews.
In addition, we are considering the contribution GIS can make to ‘new’ policy agendas in the
UK, particularly the objectives set out in The NHS Plan (Department of Health, 1998) and
other national policy documents.
The main aim of this chapter is to discuss the benefits of employing a ‘sequential’ mixed-
method approach in order to investigate the take-up, and application, of GIS in the UK NHS.
We define mixed-methods, in line with Philip (1998:264), as a research process which employs
two or more methods “to address a research question at the same stage in the research process,
in the same place, and with the same research subjects”. In the context of our specific research
concerns, we assert that a mixed-method approach will provide a more comprehensive analysis
of the use of GIS within the NHS, in order to fully tease out the interplay between
organisational, technical and project management barriers that underpin the (non)uptake of
GIS in different settings within the NHS.
The chapter begins by discussing arguments for using mixed-methods approaches (Section
2). It then reviews current knowledge and previous studies of the use of GIS within the NHS
(Sections 3 and 4). We then discuss our project which is currently underway (Section 5)
before outlining the mixed-methods research design that we are using (Section 6).
19.2
ARGUMENTS FOR A MIXED-METHOD APPROACH
In a recent review of the methods and techniques employed by researchers of health
geographies, Gatrell (2001) points to a chasm between work which adopts ‘measurementled’
and ‘interpretation-based’ approaches. Noting the key differences, Gatrell subsumes the
former approach under the broad heading of ‘mapping’, and this is equated with quantitative
spatial data analysis. For purposes of convenience, Gatrell claims that this strand of health
geography involves a three-fold distinction between the tasks of ‘visualisation’ (identifying spatial
patterns), ‘exploratory spatial data analysis’ (investigation of association between variables to
explain spatial patterns), and ‘modelling’ (formal statistical procedure to test hypothesis). The
interpretation-based approach, Gatrell notes, involves a range of clearly defined methods to
collect qualitative data and which usually seeks to “understand human beliefs, values and actions”;
incorporating the use of interviews, participant observation and focus groups.
These different approaches are illuminated further in a selective summary of previous health
studies, and this serves to show that: “those researching the geography of health have tended to
use one or the other set of tools” (Gatrell, 2001:82). Importantly, Gatrell concludes the
discussion by problematising the dichotomous relationship between the quantitative (mapping)
and qualitative schools of thought, proclaiming:
“The danger, however, is that the analytical methods give very little attention to what
the points or dots on the map really represent. The dots are not inanimate objects; they
are real people, and while the ways in which they are arranged spatially may shed some
light on disease causation the [quantitative] approaches…give no consideration at all to
the feelings, experiences, beliefs and attitudes that the individuals have. We need to
have these people speak to us rather than reducing them to a collection of dots on a
map” (Gatrell, 2001:78).
To overcome this dualism, and contrary to the above orthodoxy, Gatrell advocates a
‘triangulation’ approach; whereby both quantitative and qualitative methods and techniques
are mixed, and findings integrated to corroborate the findings of one another (see Brannen,
1992 for discussion of the merits of triangulation). To exemplify the benefits of such a research
design, Gatrell draws attention to a number of contemporary studies:
“which have taken a more pragmatic and eclectic stance, using whatever methods are
appropriate to the problem under investigation” (Gatrell, 2001:82).
This mixing of multiple methods, Gatrell suggests, has proved valuable; “with insights from in-
depth interviews adding colour and explanatory power to quantitative studies” (ibid).
This standpoint parallels similar calls for multi-method research in other areas of human
geography. For example, McKendrick (1999), Sporton (1999), Findlay and Li (1999) and
Graham (1999) provide insightful discussions of the potential of multi-methods research in
population geography.
USING A MIXED-METHOD APPROACH TO INVESTIGATE THE USE OF GIS 269
19.3
CURRENT UNDERSTANDINGS OF THE USE OF GIS WITHIN
THE NHS
Overall, the take-up of GIS within the NHS is unclear with disagreements about the levels and
types of GIS use within the NHS. Some commentators have demonstrated that there are good
examples of GIS uptake within the NHS (e.g. Taylor and Allgar, 1997), and claim that many of
the ‘obstacles of the past’ have been removed. For example, emphasis is often made to the
reduced financial and staff resources required to implement and maintain GIS, particularly the
lower ‘start-up’ costs and less technical problems associated with GIS, when compared to the
1980s and early 1990s (Smith, 1999). This, coupled with improved access to cheaper, higher
quality, digitally coded health data and a growing organisational recognition within the NHS of
the need for spatial analysis within health service planning and monitoring, would appear to
provide a conducive context for GIS to flourish within the NHS. This point, to a certain
extent, is reflected in recent conferences held in the UK and further afield, such as the USA
where there is a longer tradition of GIS in the health sector.
By contrast, some authors are less optimistic, and claim that the current implementation
and use of GIS within NHS organisations “is still very much in its infancy in the healthcare
sector” (Burns, 1996:37). It is argued that previous forecasts of the widespread uptake of GIS
within the NHS (e.g. Wrigley, 1991) have not been realised (Barlow, 2000). This contention
would appear to have some validity. For example, in a recent review of a leading professional
health journal, Smith et al. (2001) identified only one example of health-related research which
had utilised GIS. Similarly, Higgs and Gould’s (2000) analysis of recent NHS policy documents
found no mention of the potential for GIS within the ‘New NHS’. Previous surveys of GIS use
within the NHS also support this notion of low levels of GIS uptake within NHS organisations
(Cummins and Rathwell, 1991; Gould, 1992; Smith and Jarvis, 1998; Cooper, 2000).
Evidence from other European countries also suggests a ‘patchy’ and ‘uncoordinated’ level of
GIS adoption in health organisations (e.g. Ireland—Houghton, 2001).
But why is there a limited uptake of GIS within the NHS? What are the factors impeding the
current and potential use of GIS? How valid are optimistic forecasts of extensive GIS use
within the NHS, such as Burns (1996:39) view that “mapping applications will soon form part
of the core suite of office applications, sitting alongside word-processor and spreadsheet
packages, and will become an integral part of any manager’s desktop”.
To comprehensively address current and future uses of GIS, we operationalise (using mixed-
methods) the call for enhanced knowledge and understandings of the reasons why the NHS has
not fully realised the benefits of GIS-uptake in relation to new technical and organisational
opportunities (Higgs and Gould, 2001). The following section now reviews earlier surveys of
the use of GIS within the NHS to elucidate some of the factors that are perceived to have
hindered GIS uptake within the NHS in the past.
19.4
PREVIOUS STUDIES OF THE USE OF GIS WITHIN THE NHS
An early investigation of the use of GIS within the NHS was commissioned by the Association
for Geographic Information (AGI), and carried out by Cummins and Rathwell (1991). This
study, using a telephone survey of the 14 Regional Health Authorities (RHAs) and 190 District
Health Authorities (DHAs) in England, found significant geographical variations of GIS use
between, and within, Health Regions. Despite noting a ‘surge’ in interest in GIS within the
health sector during the late 1980s, a number of concerns regarding the future uptake of GIS
were highlighted. First, it was claimed that GIS uptake was being hindered by a low awareness
of the value of GIS in spatially representing population and health needs-based information
(itself a legislative requirement of DHAs in 1991). This was attributed to previous low levels
of spatial data handling by health professionals, managers and IT staff within the NHS. Indeed,
other cultural and organisational criteria were discussed, which were seen to partly explain the
under-utilisation of GIS. In particular, it was noted that health planners had not previously
allocated health-based resources on the needs of the local population, and therefore geographic
information was not seen as an important factor during policy making and health planning
activities. In addition, Cummins and Rathwell (1991) point to a lack of infrastructure to enable
staff, training, resource management and financial budgets to implement GIS, and assert that
the small size of many IT departments and high staff turnover hindered the effective
implementation of GIS. This was further influenced by the lack of understanding of GIS
suppliers for the needs of health authorities, and how organisations within the NHS could make
best use of GIS. Hence, Cummins and Rathwell (1991:14) conclude: “The potential in the
NHS for geographically derived information is enormous; it is unfortunate that so far many in
the service have yet to realise or appreciate this”.
A wider-geographic survey of directors of public health and information, and IT officers, in
all of the 197 District Health Authorities in England and Wales in 1991, was undertaken by
Gould (1992). In contrast to Cummins and Rathwell (1991), Gould points to a generally high
level of awareness of the value of GIS amongst health officials, yet found that their use was
largely confined to relatively low-level operational tasks, such as desktop mapping, in order to
identify local population health needs. GIS was not being used for strategic tasks within the
NHS, and Gould identified a number of factors influencing such trends. As a result, two key
recommendations are put forward to encourage the greater use of GIS within the NHS. First,
Gould (1992:399) stresses that: “something must be done to improve the GIS skills and
understanding in the NHS if the use of such systems is to become more widespread”. Second,
it is argued that “there is a need to widen the functional capabilities of GIS software for spatial
analysis and spatial modelling”. Given the organisational and technical changes that have taken
place since Gould’s survey, an important aim of the current project has been to update these
survey findings.
Similarly, when examining the impact of NHS reforms from the early 1990s on the use of GIS,
Smith and Jarvis (1998) found relatively low-level, uncoordinated and sporadic uses of GIS
within the NHS. It was argued that this was partly due to technical problems, such as data
quality, but that the lack of policies and guidance regarding the use of GIS technology across
the NHS was also a major contributory factor. Much of the case study material appears in so-
called ‘grey literature’ and the authors conclude by espousing the benefits of improved data
availability, detailed meta-data, appropriate organisational mechanisms, and greater levels of
awareness of the potential for GIS in order to promote the exchange of data between agencies
(see also, Martin, 1996). Smith and Jarvis also suggest that GIS uptake was being hindered by a
limited transfer of ideas between the GIS research community in the academic sector and those
based in health departments despite the existence of more informal networks where advice and
ideas were being exchanged with external agencies and organisations (see Reeve and Petch,
1999) for a comprehensive discussion of technical, human and organisational factors
influencing the effectiveness of GIS).
More recently, a questionnaire-based study of GIS use within the 13 Health Authorities of
the West Midlands was undertaken by Cooper (2000). In line with earlier studies, a series of
obstacles were identified which were hindering the use of GIS technology across the region. It
was noted that two-thirds of the Health Authorities in the West Midlands stated that GIS was
being under-utilised and not being fully exploited. The main hurdle was the high costs of
‘unaffordable’ digital geographical data, which restricted the potential uses of GIS. Other
constraints included the lack of resources for training and available work time for NHS
personnel, due to the low priority (as deemed by senior management) given to GIS within the
organisations. To overcome these problems, Cooper (2000:33) calls for the establishment of:
“Regional health GIS centres, integrated with the newly formed public health laboratories”,
which “would be able to take advantage of regionally collected health data, achieve economies
of scale with regards GI, reduce local duplication of effort and provide GIS support for the
local health services”. This is an important recommendation that requires further investigation,
and which will be fully examined in the course of our research project.
Without doubt, the above studies have provided valuable insights into understanding the
current, and projected, roles of GIS within the NHS. Significantly these studies have pointed to
sporadic levels of GIS uptake within primary and secondary health care organisations, and have
been relatively consistent in identifying the types of factors that are hindering the
implementation and use of GIS within the NHS. Despite this contribution, we would contend
that more in-depth research is required in order to examine the interplay between constraining
and enabling factors in specific contexts and their influence on the level and type of GIS
uptake, particularly in light of recent structural changes and new technical developments.
Questionnaire-based studies have arguably failed to fully tease out the processes that
underpin the use of GIS within the NHS, a fact that is generally acknowledged by the authors of
these studies who, in turn, have called for more in-depth studies to explore such issues. In
order to address these shortcomings, we would concur with Philip (1998: 273), who suggests
that:
“researchers should think beyond the myopic quantitative—qualitative divide when it

comes to designing a suitable methodology for their research, and select methods—
quantitative, qualitative or a combination of the two—that best satisfy the needs of the
specific research projects”.
In the following section we describe the phases of research design included in our current
project, which is adopting a mixed-method approach. The aim is to tease out the key research
questions and to justify the rationale for such an approach.
19.5
THE RESEARCH AGENDA
The on-going research project seeks to understand current levels of GIS use within primary
and secondary health care organisations, and to explore the hypothesis that GIS are being
under-utilised in the health sector. This ESRC-funded project has a number of components
including:
• Surveying levels and variations in GIS utilisation within the primary and secondary health
care sectors—building upon a previous cross sectional survey of GIS (e.g. Gould, 1992).
• Reviewing/survey types of GIS application being used in the light of recent changes in the
nature of structure of the primary and secondary health care sectors.
• Gauging the importance of organisational factors in the widespread adoption of such systems
—such factors have been deemed to be important in other branches of medical informatics
(e.g. Kaplan, 1997).
• Understanding reasons for variations in the use and wider implementation of GIS including
the types of technical and organisational barriers that influence wider application.
• Investigating the implications of recent NHS policy changes on the use of geographically
referenced data and GIS technology.
• Examining the nature and extent of intra-/inter-data exchanges within the NHS given the
perceived advantages of GIS as an integrating technology and the over arching aim of
encouraging inter-agency collaboration in ‘Our Healthier Nation’.
• Highlighting ‘best-practice’ case studies of health service organisations currently using GIS
for addressing key policy tasks.
• Exploring the potential use of other NHS C&IT (communications and information
technology) developments and initiatives for sharing health information (e.g. NHSNet,
NHS direct and Internet-based GIS).
This research agenda seeks to capture and assess the impact of the recent drive from central
government to exploit the rich source of geographically referenced data held within the NHS,
to enhance both the delivery and monitoring of health care services (de Lusignan et al., 2000;
Bell, 2000; Gibbs, 2000). This is manifest in the creation of enabling organisations to translate
and implement the goals of a new Information Strategy for the NHS, as outlined in The NHS
Plan (Department of Health, 2000) and more specifically, Information For Health (NHS
Executive, 1998). In addition, we would contend that the application of GIS in the health
sector can, in turn, provide a testbed for many of the issues surrounding ‘joined-up
government’ debates which are driving the use of GIS in other sectors such as local
government and crime.
Interestingly, there is no mention of the use of GIS within recent policy documents despite
well documented examples of the potential for GIS within the NHS (Wrigley, 1991; Birkin et
al., 1996; Todd and Forbes, 1998; Gatrell and Loytonen, 1998), and the numerous academic
projects that have shown the benefits of GIS for health-based application tasks (Gatrell and
Senior, 1999; Higgs and Gould, 2001). This can also be examined in the light of more
complex, integrated and pervasive information systems coming on stream within the NHS and
a relative proliferation in spatially referenced disaggregate data sources (Bryant, 1998).
Clearly, if NHS organisations are to fully benefit from these holistic information systems and
respond to recent government objectives, new technologies such as GIS, must be fully
harnessed to realise their potential.
Therefore, the current and potential use of GIS must be accurately measured within the
emerging ‘information-drive’ context of the NHS. This is essential if Cummins and Rathwell’s
(1991) vision for the integration, manipulation and presentation of locational and spatial data
within the NHS is to be realised. The new policy initiatives of The NHS Plan and Information for
Health, as well as the promotion of ‘joined-up’ government, collaboration, and joint
commissioning, have offered substantial encouragement for the enhanced use of GIS within the
NHS (Higgs and Gould, 2001). This project will examine if such influences are having a
significant impact on GIS usage in the health sector.
Table 19.1 Stages of Piloting and Consultations (Dates, Venues).
19.6
A MIXED-METHOD RESEARCH DESIGN
In a recent paper, Higgs and Gould (2001:19) comment: “people wanting to survey current
GIS usage within the NHS are faced with a difficult task as there is no single source of
information”. In the absence of such an accurate baseline of the current use of GIS within the
NHS, the starting point for our research process involved the preparation of an extensive
questionnaire-based survey, in order to provide a context for a wider exploration of the issues
influencing the effective implementation of GIS.
In order to ensure that the questionnaire schedule was comprehensive, it was deemed
essential to fulfil two key tasks prior to designing the questionnaire schedule and administering
the survey. First, we reviewed the questionnaire schedules employed for previous studies of
the use of GIS within the NHS (e.g. Gould, 1992), as well as questionnaire-based surveys of the
use of GIS in other public sector contexts (e.g. Campbell and Masser, 1993; Church et al.,
1993; Gill et al., 1999). This consolidated and informed our understanding of the breadth of
issues under investigation as well as helping us to estimate the realistic levels of response we
could expect from addressing relatively complex factors that may be influential in the effective
implementation of GIS. Secondly, the questionnaire schedule was sensitised further following
pilot discussions with GIS experts and with health-professionals, key policy makers and
institutional actors with a broader knowledge of the issues involved (Table 19.1).
The questionnaire schedules were tailored for Health Authorities and counterparts in
England, Wales, Scotland and Northern Ireland to take into account different policy directives
and organisational structures. Other important suggestions included the clarification and
modification of questions (e.g. adding definitions), the incorporation of additional questions
(e.g. questions relating to GIS support or advice for Primary Care Groups/Trusts), the
availability of a web-based option for answering the questionnaire and the degree of question
overlap with previous surveys (e.g. Gould, 1991), which in turn, permit some degree of time-
series analysis (Smith et al, in preparation).
The second phase of the research process involved the delivery of the questionnaire-based
survey. This was posted out, and also made available for completion on the WWW, during
June and July 2001 to appropriate Information Technology/Services personnel in all Health
Authorities of England (95) and Wales (4), Health Boards in Scotland (15), and Health and Social
Services Boards in Northern Ireland (4), and all Health Trusts (469), and to the General
Managers/Chief Executives from a 50% random sample of Primary Care Groups/Trusts
across the UK (193), stratified by Health Authority Region. The sampling frame was
constructed from Binley’s (27th Edition Spring 2001) databases for ‘NHS Management’ and
‘Primary Care Groups/Trusts’. In total 780 surveys were distributed.
The main purpose of the questionnaire was to provide standardised information, elicited
from both open and closed-ended questions, which would provide an accurate measure of
current and potential uses of GIS, and would facilitate comparisons between different
respondents. It was essential that the questionnaire also flagged up factors which respondents
perceived to hinder or enable use and uptake of GIS within their organisation. Issues such as
the importance of the commitment of key personnel in the effectiveness of GIS
implementation were sought at this stage—such factors having been seen as important in other
health contexts (Houghton, 2001). A further outcome of the survey was that ‘best practice’
examples of the use of GIS within the NHS could be identified. This latter element was
essential, in order to select appropriate case studies, which could then be explored more fully
via face to face qualitative interviews.
Based upon our pre-existing understanding of the factors which constrain or enable GIS use
within the NHS, the questionnaire-based survey was divided into six main sections. Section 1
sought to identify the diversity of institutional contexts within the NHS which may influence a
range of current GIS uptake and uses. The aim has been to: document the historical
background of GIS uptake and uses; identify the presence/absence of a GIS strategy; establish
whether the current GIS is fully operational, and; tease out how the GIS is administered and
maintained. The levels of intra- and inter-organisational usage of GIS have been examined in
order to identify to whom organisations turn for GIS advice and analysis, and to examine levels
of leadership and political support for such initiatives. In Section 2 we attempted to reveal any
future plans for GIS uptake and use, for example, by asking respondents if the organisation had
any plans to modify their current GIS or purchase a new GIS. The aim of Section 3 was to pin
down the tasks for which GIS is currently being used and to identify policy-related uses of GIS.
The projected use of GIS was also identified through subsidiary questions, as was the
importance of recent official NHS documents/policy guidance, and the implications of GIS on
the wider availability of health information. In Section 4 we specifically focused on the
technical constraints that are often viewed as key constraints hindering GIS within the NHS.
Thus, we asked questions about the type of datasets that are being used for GIS purposes and
the types of data that are being exchanged with other organisations, and the factors that are
limiting accessibility to datasets. Other potential constraining factors were then considered in
Section 5, such as the level of access to and types of training, the availability of advice,
guidance and support for GIS implementation and use. Crucial here was the need to obtain the
subjective and personal opinions of respondents regarding the types of organisation that they felt
should provide training and guidance, which could then be followed up in face to face
interviews in phase 2. The final section sought to corroborate earlier answers and illuminate the
significance of factors which respondents had previously cited in their responses to questions
posed about the types pf factors which enabled and constrained GIS uptake within their
respective organisations. As Kaplan (1997:97) suggests, with regard to wider studies
examining the impact of technology in medicine, “evaluators often could benefit from multiple
research methods”. The approach we have taken therefore aims to build on a questionnaire
approach to address these issues.
From a preliminary analysis of the survey results, ‘best-practice’ examples of GIS use within
NHS organisations can be identified. These case studies will form the basis for the next phase
Table 19.2 Examples of the themes under investigation in phase 2 of the project.
of the research involving semi-structured interviews with key informants in a variety of

organisational settings, building on findings of the initial survey. Qualitative data will be
gathered and analysed (using grounded theory techniques) to permit a deeper understanding of
some of the contingent social and cultural influences, and the dynamic processes underpinning
the uptake and uses of GIS within different organisational contexts of the NHS. Examples of
inter-related themes which will be investigated during the face to face interviews are outlined
in Table 19.2.
19.7
DISCUSSION
It is envisaged that the mixed-method approach adopted for this research project will enhance
previous studies of the use of GIS within the NHS, and provide understandings drawn from
different levels of explanation. Whilst the extensive questionnaire survey has contextualised
and provided evidence of trends and regularities in GIS uptake and use across NHS
organisations, it is anticipated that the intensive semi-structured interviews will provide a
fuller understanding of the links between causal factors, such as managerial commitment to
GIS and organisational stability on GIS uptake and levels of use. As Kaplan (1997:97) notes:
“combining qualitative with quantitative method, allows for a focus on the complex web of
technological, economic, organisational and behavioural issues”, and adds to the robustness of
research findings. This will be addressed in the second phase of the project.
Although we fully recognise that triangulating the different strands of the research process will
be problematic, we would argue that it is highly beneficial. The process of triangulation will
provide complimentary empirical evidence, and both support and/or contradict the findings
from the questionnaire survey and interviews. Moreover, this will offer the opportunity to
cross-check results and methods, and thus tease out any meaningful inconsistencies which
emerge, and reduce the risk of establishing erroneous conclusions. This is likely to offer new
insights into the levels of GIS usage in different NHS organisational contexts. Moreover, and in
contrast to previous studies of the use of GIS within the NHS, we would assert that the mixed
method approach will provide a rigorous exploration of the factors constraining GIS uptake
and use within NHS organisations, and is likely to facilitate greater confidence in the initial
research. This will be important if policy makers are to acknowledge the findings from the
research, and take into account the factors which are influencing GIS implementation within
the NHS.
In conclusion, this chapter has sought to draw attention to the benefits of utilising a mixed
methods approach to investigate the factors which enable or constrain the uptake, and use, of
GIS within NHS organisations. The main benefits of this approach are: complimentarity,
confirmation, reliability, increased confidence levels and the illumination of meaningful
contradictions and inconsistencies. This, in turn, may have benefits for researchers examining
the impacts of new technologies within other public and private sector organisations.
19.8
ACKNOWLEDGEMENTS
The project is being funded by an ESRC research grant (R000223473). We acknowledge the
support of the Department of Health GI Steering Group and the AGI Health SIG. Particular
thanks to Ralph Smith, Paul White, Duncan Cooper, lan Bullard and Christine Dunn for
comments on the design of the questionnaire. However, the opinions expressed in this paper
are those of the authors alone and may not necessarily be representative of the organisations
listed above.
19.9
REFERENCES
Barlow, T., 2000, GIS and mapping in healthcare. The British Journal of Healthcare Computing and
Information Management, 17, pp. 42–44.
Bell, N., 2000, Realising the vision—making a difference. The British Journal of Healthcare Computing
and Information Management, 17, pp. 17–19.
Birkin, M, Clarke, G., Clarke, M. and Wilson, A., 1996, Intelligent GIS: Location Decisions and
Strategic Planning. (New York: Wiley).
Brannen, J., 1992, Mixing-methods: Quantitative and Qualitative Research. (Aldershot: Avebury).
Bryant, J., 1998, The importance of human and organisational factors in the implementation of
computerised information systems: an emerging theme in European healthcare. The British Journal
of Healthcare Computing and Information Management, 15, pp. 27–30.
Burns, S., 1996, Mapping Healthcare: What’s the Point? The British Journal of Healthcare Computing
and Information Management, 13, pp. 37–39.
Cooper, D., 2000, Developing GIS within the NHS—obstacles and possible solutions. The British
Journal of Healthcare Computing and Information Management, 17, pp. 31–33.
Campbell, H. and Masser, I., 1993, The impact of GIS on local government in Great Britain. In
Geographical Information Handling—Research and Applications, edited by Mather, P.M. (London:
Wiley), pp. 273–286.
Church, A., John, S., Shepherd, W., Frost, D. and Macmillan, A., 1993, An information system
for the Eastern Thames Corridor. In Geographical Information Handling-Research and Applications,
edited by Mather, P.M., (London: Wiley), pp. 287–324.
Cummins, P. and Rathwell, T., 1991, Geographical Information Systems and the National Health Service
Education, Training and Research Publication. (London: Association of Geographic Information).
Department of Health., 2000, The NHS Plan. A Plan for Investment, a Plan for Reform. (London, HMSO,
White Paper).
Findlay, A.M. and Li, F.L.N., 1999, Methodological issues in researching migration. Professional
Geographer, 51, pp. 50–59
Gatrell, A.C., 2001, Geographies of Health: An Introduction. (Oxford: Blackwell).
Gatrell, A.C. and Loytonen, M., 1998, GIS and Health. (London: Taylor and Francis).
Gatrell, A.C. and Senior, M., 1999, Health and health care applications. In Geographical Information
Systems: Principles, Techniques, Applications and Management, edited by Longley, P., Maguire, D.,
Goodchild, M. and Rhind, D., (London: Wiley), pp. 925–938.
Gill, S., Higgs, G. and Nevitt, P., 1999, GIS in planning departments: preliminary results from a
survey of local authorities. Planning and Policy in Practice, 14, pp. 341–361.
Gould, M., 1992, The use of GIS and CAC by health authorities: results from a postal questionnaire.
Area, 24, pp. 391–401.
Graham, E., 1999, Breaking out: the opportunities and challenges of multi-method research in
population geography. Professional Geographer, 51, pp. 76–89.
Higgs, G. and Gould, M., 2000, Commentary: healthcare commissioning, the modern NHS, and
geographical information systems. Environment and Planning A, 32, pp. 1905–1908.
Higgs, G. and Gould, M., 2001, Is there a role for GIS in the ‘New NHS’. Health and Place, 72,
pp. 247–259.
Houghton, F., 2001, Health GIS in Ireland—lessons from New Zealand. Irish Geography, 34,
pp. 96–97.
Kaplan, B., 1997, Addressing organisational issues into the evaluation of medical systems. Journal of
the American Medical Informatics Association, 4, pp. 94–101.
Lusignan, S., Mimnagh, C, Kennedy, J. and Peel, V., 2000, Alignment of information for health
with the NHS plan—a case for substantial investment and reform. The British Journal of Healthcare
Computing and Information Management , 17, pp. 28–32.
Martin, D., 1996, Geographic Information Systems: Socioeconomic Applications, (London: Routledge).
McKendrick, J., 1999, Multi-method research: an introduction to its application in population
geography. Professional Geographer, 51, pp. 40–50.
NHS Executive., 1998, Information for Health: An Information Strategy for the Modern NHS, (London:
NHS Executive).
Philip, L.J., 1998, Combining quantitative and qualitative approaches to social research in human
geography—an impossible mixture? Environment and Planning A, 30, pp. 261–276.
Reeve, D. and Petch, J., 1999, GIS Organisations and People: A Socio-Technical Approach. (London: Taylor
and Francis).
Smith, R., 1999, Co-ordinating healthcare—taking a geographic perspective. The British Journal of
Healthcare Computing and Information Management, 16, pp. 40–41.
Smith, R. and Jarvis, C., 1998, Just the Medicine. Mapping Awareness, 12, pp. 30–33.
Smith, D.P., Higgs, G. and Gould, M., 2001, Clarifying the current and potential use of GIS within
NHS organisations: optimism or pessimism? The British Journal of Healthcare Computing and
Smith, D.P., Gould, M. and Higgs, G., in prep, Surveying the uses of GIS in the UK NHS: ten years on.
Sporton, D., 1999, Mixing methods in fertility research. Professional Geographer, 51, pp. 68–76.
Taylor, G. and Allgar, G., 1997, The creation of an enhanced dataset from disparate medical
records using a relational management system. The British Journal of Healthcare Computing and
Todd, P. and Forbes, H., 1998, Seeing statistics more meaningfully: a spatial analysis package for
epidemiology. The British Journal of Healthcare Computing and Information Management, 15,
pp. 32–34.
Wrigley, N., 1991, Market-based systems of health-care provision, the NHS Bill and Geographical
Information Systems. Environment and Planning A, 23, pp. 5–8.
Index
3D Studio Max 100, 104 buffer/buffering 42–47, 165, 187, 234

3D visualisation/modelling 4, 25, 74, 93–106, built environment 93–106
111, 136–138, 143 Burglary Reduction Initiative 38
4D 111 ByDesign 42
accessibility 119, 181–198, 200, 259–270
accuracy 21, 29, 54, 57, 94, 114–115, 128, CAAD 93
170, 205 CAD 93, 100
ActiveX 83 CADW 108
address matching 176 California Police Department 90
Address Point 40–42, 45, 122, 164, 170 Canoma 101–104
AGI 274 catchments 189
adjacency analysis 124–126, 130 Census 6, 41–42, 94, 122, 130, 151–152, 183
aerial photographs 115, 164 data 5, 22, 160–161, 183, 199–211, 216
aggregation 23–24, 29, 41–42, 111, 235, 237, output areas 41, 176, 216, 231, 236, 240
241, 247, 260, 268 population 51, 122, 163–164, 216
alleygating 37–50 tracts 24, 76–77, 176
ArcData Online 151 US 147, 151
ArcGIS 240 US Bureau 90
Arclnfo 172, 185–187, 201, 215, 262–268 CHEST 200
ArcView 42, 99–100, 150, 201, 215 Chi-square 127, 237
Area Master File 184 class entropy 58–59
artificial neural networks (ANNs) 3, 51–65 Classic NTF Translator 42
AutoCarto 9–11 classification 56–57, 109, 184–190, 193, 231,
autocorrelation 2, 5, 21, 23–24, 147–162, 235, 240, 262
165, 172–173 cluster/clustering 3, 5, 21–36, 51–65, 147–
Autodesk MapGuide 80 150, 153–156, 160–161, 172–174, 237–249
automated geography 12 Code-Point 234
Cohesion Report 259
Bartholomew 1:250000 scale digital map coincident point distance 29
database 184, 200–201 communications and information technology
best practice 276, 279 (C&IT) 276
blunders 15 complete spatial randomness (CSR) 23
boundary objects 14 Cohesion Report 259
Brent Authority 234–235 Common Gateway Interface (CGI) 81
British Crime Survey 39 connectivity 94–96
279
280 INDEX
continuous surface (smoothing) map 21–36 database 10, 16, 40–41, 51–52, 72, 74, 86–89,
contour data 111 103, 107, 111–116, 137, 151, 170, 176,
coordinate system 15, 170 186, 202, 210, 261, 263, 278
COPDAT 52–53, 63–64 applications 83
correlation 124, 129, 237 community 84
corridor 187 design 10, 137
Council Tax Register 122 digital map 181, 184, 200
crime 1, 3–4, 19–65, 121–122, 130 functions 164
analysis 2–4, 37–50 geospatial 140, 143
audits 2 GIS 100
British Crime Survey 37, 39 integration 268
Burglary Reduction Initiative 38 lifestyle 3, 7, 238–240
clusters 21–22 management 72, 259, 263
COPDAT 52–53, 63–64 Oracle 84, 89
Crime and Disorder Act 2 police 51
Crime Reduction Programme 38 property 234–235
CrimeStat 23 server 87
data sets 3, 22, 37, 40, 41 socio-economic 269
distribution maps 22 spatial 51, 165, 176
hotspots 3, 21–36 technology 165
incidence 2, 3 topological 11
mapping 22 updates 263, 266
patterns 1, 3, 21 utility 165
prediction (forcasting) 2–3, 51–65 decision support system 73–74, 164
prevention 2–3, 37 collaborative 89
records 21 Delaunay triangulation 187
volumes 22 demographic data 151, 247
cross-tabulation 194–195 Department of Transport 184, 201
deprivation 119, 123, 125–126, 130, 231–242
data deprivation index 125
accessibility 278 desktop mapping/GIS 51, 93, 112, 215, 273–
availability 274 274
capture 112–114, 117 digital elevation model (DEM) 164
census see Census digital terrain model (DIM) 108, 167, 263
contour 111 digitising 15, 40, 42, 113–114, 164, 176, 184
costs 275 disaggregated data 163, 176, 184
demographic 151, 247 distance decay measures 167, 248
exchange 275 distance proximity 165
O.S. see Ordnance Survey
population 163–164 e-commerce 97
processing 53–55 economic indicators 259
quality 41, 274 edge matching 112
socio-economic 151, 164 electoral geography 232, 243–244
stratification 60 electoral register 238
update 259, 263 elliptical clusters 22
English Heritage 107–108, 111
INDEX 281
enumeration district (ED) 41, 122, 124–127, GIS training 279

129, 150, 163–164, 183, 199–211, 216–217 GISRUK 9
environmental global positioning system (GPS) 2
applications 1 Gloucestershire Structure Plan 214, 219
infrastructure 122 Government agencies 181
studies 1 Greater London Authority 94
EPCS 202, 210 grid cell size 26–27, 29, 31
EPSRC 214 Gross Domestic Product (GDP) 259, 261, 267
error 15, 41, 59, 150, 165–166, 170, 176,
235, 239, 263 Health Authorities 274
error checking/detection 54, 263 health care sectors 275–276
ESRC 200 health care services 148, 181–198, 200
ESRI 150 health data 273
Avenue 201 health geography 272
Arc Data Online 151 Health Regions 274
ArcGIS 240 health service planning 273–274
Arclnfo 172, 185–187, 201, 215, 262–268 hedonic price modelling 165–166, 170
ArcInfo GRID 171 Historic Landscape Assessment 107–118
Arc Macro Language (AML) 262 Historic Scotland 110
ArcView 42, 99–100, 150, 201, 215 Home Office 2, 37–38, 52
Internet Map Server 70, 72, 76–77, 80–83, hotspot maps/prediction 3, 21–36, 51–65
103–105 house price modelling 122, 165
Java applets 104 housing 119–132
Map Objects 72 HTML 76–77, 81, 83, 98, 104
shape file 42 human geography 1
ESTEEM 214–217 hypermedia 72, 115
Euclidean distance 56–57, 63, 248, 253, 255 hypothesis testing 23, 272
European Union 259
expert systems 62, 70 IBM 14
exploratory spatial data analysis 272 imagery 164, 235
imageability 71–72
filter 235 incremental means 32
Flash Shockwave 78–80 incremental standard deviations 31–32
Food and Agriculture Organization (FAO) 15 Index of Multiple Deprivation (IMD) 231–242
format conversion 54 Information for Health 276–277
freehand sketching 71, 83, 85–87 Innovations in GIS 1
interactive mapping 69–92
Gamma test 58–63 interface design 69–92, 103
generalisation 53–54, 100 Intergraph 10
geocoding 21 Intergraph Geomedia Web Map 80
geocomputation 240 International Journal of GIS 10
geodemographic analysis 241 internet 4, 69–92, 93–98, 103–105, 135–140,
geographical units 184, 231 143–144, 148, 183, 276
geographical visual information systems 135– Internet Explorer 71
146 Internet Information Server 72
geo-referencing 23, 41, 69, 114–115, 143, 163– Internet Map Server 72, 103
164, 170, 176–177, 234–235, 247 interpolation 29, 163–176, 261
282 INDEX
inverse distance weighting (IDW) 29–30, 167, pin 21

172, 174 point 21
isochrones 187 raster 114
thematic boundary 21, 22
Java 76–83, 86, 98, 104 topographic 113
JISC 200 MasterMap 112, 113
Joined-up government 2, 276–277 Merseyside Address Referencing System 40–41
metadata 107–108, 111–115, 117, 274
Kent County Constabulary 52, 64 metropolitan districts 202
kernel 167–168, 172–174 Merseyside Police 39–41, 122
kernel density estimation 23–29, 32–34 Integrated Criminal Justice System 40–41
Kohonen Self-Organising Maps 51–65 Microsoft
Access 72
LandLine 98, 113–114, 164 MIMAS 184, 200
Land Registry 122 Area Master File 184
Landsat TM 164 mixed-method approach 271–273
landscape character assessment 107–118 modifiable area unit problem 22
landscape classification 109–110 Moran’s I 24, 153–161, 172–174
landscape history 109 Moran scatterplot 153, 155–158
landscape register 110 Multi-Layered Perceptron (MLP) 56–61
land use 163–164 multi-level modelling 165
lattice 187 multimedia 72–73, 115, 137, 143
layer based design 15 multivariate analysis 120, 148
learning systems 138–144
least squares regression 60, 165 National Grid reference 41–42, 164, 183
legend thresholds 31–33 National Health Service (NHS) 3, 181–198,
LiDAR 99–100 210, 271–282
lifestyles data 3, 7, 238–241 NHSNet 276
local authorities 2, 39, 109, 126, 130 NHS Plan, The 271, 276–277
Local Indicators of Spatial Association (LISA) National Statistics Survey (NSS) 231, 240
21–36 National Transfer Format (NTF) 42
Location Profiler 27–29 National Travel Survey 216–217
natural resource management 137
Maastricht Treaty 259 nearest neighbour 23–24, 31, 33
Maplnfo 25, 42, 100, 122, 124–130 neighbourhood analysis 23–24, 31, 33, 240–
MapXtreme 80 241
PAVAN 100 neighbourhood characteristics 165
Vertical Mapper 25 neighbourhood effects 244–245
Map Objects 72 network analysis 165, 186, 201, 248, 259, 263,
map merging 234 268
map overlay 77, 114, 165, 234, 259, 268 neural network 3, 51–65
map projection 170 NUTS (Nomenclature des Unités Territoriales
map registration 15, 170 Statistique) 261–269
map transformation 170
maps Office for National Statistics (ONS) 94, 183,
crime distribution 22 231
hotspot 3, 21–36 online planning 93–106
INDEX 283
Oracle database 84, 89 Merseyside Police 39–41, 122

ordinary least squares (OLS) 165 sector boundaries 53
Ordnance Survey (O.S.) 40, 98, 164 policing strategies 52
1950 1:25000 Series 113 population distribution 199
Address Point 40–42, 45, 122, 164, 170 population remoteness 199–211
Code-Point 234 population surface modelling 240
contour data 111 Postal Address File (PAF) 238–239
first edition data 113 postcode 122, 163–165, 170, 183, 186, 194,
LandLine 98–99, 113–114, 164 234, 239, 240, 247
MasterMap 112, 113 Postcode to Enumeration District Directory
National Grid 164, 247 183
National Transfer Format (NTF) 42 poverty 121m 123–125, 130
OSCAR (Road data) 201 Poverty and Social Exclusion Survey 231
raster maps (1:10000) 114 primary care services 181
topographic maps 113 property prices 165, 176
OSCAR 201 public participation GIS 69–92
overlay 77, 114, 165, 234, 259, 268
qualitative data 272
panoramas 97–98, 103–105, 143 quantitative analysis 272
pattern analysis 148–153 quantitative index 147–150
PAVAN 100
Perason’s correlation coefficient 237 R-trees 10
peripherality 7, 259–270 random error 15
index 259–270 raster maps 114
Personal Social Services (PSS) 202, 208, 210 raster-vector debate 10
Photogrammetric Engineering and Remote real estate valuation 164
Sensing 9 Regional Public Health Observatories 197
photogrammetry 15, 93, 101, 103, 164 registration 15
photorealism 93–104 regression 59–60, 165–166, 170, 199, 208
planning 2, 4, 5, 48, 52, 69–92, 93–106, 116, regression model 253, 254
122, 135–146, 147, 160–161, 213–228 remoteness 199–211
Planning Policy Guidance (PPG) 214 remote sensing 164, 235
point attribute table (PAT) 170–171 residential segregation 148–149, 160
point-in-polygon 183–184 residuals 166–167, 170–172, 174
point patterns 22 residual surface 163, 166, 174
point pattern analysis 21, 236, 240 risk assessment 53, 126
point retaining methods 23 risk index 125–130
police 2, 3, 26, 37, 40, 49, 52–53, 64, 199 road data 197, 201
authorities 2 Royal Mail PAF 238–239
beat areas 3, 41–43 Royal Commission on Ancient and Historical
California Police Department 90 Monuments of Scotland 108, 110
databases 51 RTPI 215
forces 52 Rural Development Council 214
Police Force Area (PFA) 43 Rural Transport Partnerships 182
Kent County Constabulary 52, 64 Rural White Paper 196
Metropolitan Police Crime Report Information
System 23 Safer Merseyside Partnership (SMP) 39
284 INDEX
satellite imagery 164 studies of technology and science (STS) 9–18

search radius 26–27 Super Profile data 247
seamless map 108, 115 surface smoothing methods 21
Shanghai 163 Sustainable Cities programme 214
Shortest Network Paths 201 systematic error 15
shortest path analysis 263
Sites and Monuments Records (SMR) 108 technological determinism 1
social cohesion 119, 121–123 telecommunications 137, 148
social exclusion 119, 121, 123–125, 130, 231– temporal autocorrelation 150
232, 244 thematic boundary maps 22
social issues 1, 9, 244 Tiger files 151–152
social networks 121, 244 TIGRIS 10
social services 2 TIN 187
socio-economic thematic contour map 53
applications 1–3 topographic maps 113
data 148, 151, 164 topography 111, 245, 254
database 269 topological data model 10–11
research 239 topology 150, 248
sociology of scientific knowledge (SSK) 13 training (GIS) 279
software vendors 1 Trans-European Transport Networks (TETN)
software development processes 1 259
SpaceStat 150, 153 transport 1, 259
sparsity 199–211 travel time 181, 186–188
spatial association 148, 150, 153, 159–160 matrices 261–265
spatial aggregation 23–24, 29, 235–241, 247, triangulated irregular network 187
260, 268 triangulation 187, 272
spatial analysis 39, 77, 112, 115, 117, 124,
137, 147–148, 160, 163–165, 240, 248, unit postcode level 239
259, 272–273 unitary authorities 202–203
spatial autocorrelation 2, 5, 21, 23–24, 147– University of Manchester 184, 200
162, 165, 172–173 urban planning 93–106, 135–146, 147
spatial clustering 21–36, 51–65 URBASS 214
spatial communication 70, 76–89 US Census Bureau 90
spatial data patterns 2, 266 US Federal Geographic Data Committee 10
spatial database technology 165 US Geological Survey 90
spatial disaggregation 163, 176, 184 user interface 137, 143, 265–266
spatial dispersion 21, 23
spatial ellipse methods 21–22 Valuation Office 94
spatial probability matrix 53 Vertical Mapper 25
spatial statistics 147–162, 172, 268 viewshed analysis 164–165
spatio-temporal models 107 virtual environments 93–106, 135–143
Spearman rank correlation coefficient 237 visibility analysis 165
SPSS 127, 171 visual impact 164
SQL 124–125 visualisation 2, 21–22, 27, 29, 34, 51–55, 72–
standard deviation 31 74, 77, 81, 93–98, 103–105, 135–146, 150,
Standard Spending Assessment 199 153, 170–171, 176, 268, 272
statistical analysis 40, 51, 148, 171, 240, 263 voting behaviour 243–257
Strategic Development Projects (SDPs) 38
INDEX 285
VR 93–106, 135–143
VRML 103, 138
web-based mapping 2, 69–92, 137

web interfaces 71, 77, 86
weighted travel times 200
ward boundaries 41, 200, 231–242
zooming 78

Socio Economic Applications in Geographical

Uploaded by

Copyright:

Available Formats

Socio Economic Applications in Geographical

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Socio Economic Applications in Geographical

Uploaded by

Copyright:

Available Formats

Socio Economic Applications in Geographical

Editors David Kidner, Gary Higgs and Sean White

ISBN 0-203-30107-2 Master e-book ISBN

ISBN 0-203-34562-2 (Adobe eReader Format)

PART I: GIS AND CRIME 21

PART II: GIS AND PLANNING 67

PART III: GIS AND URBAN APPLICATIONS 133

PART IV: GIS AND RURAL APPLICATIONS 178

PART V: GIS IN SOCIO-ECONOMIC POLICY 226

• to act as a focus for GIS Research in the UK;

GISRUK National Steering Committee (as of GISRUK 2001)

Steve Carver University of Leeds steve@geog.leeds.ac.uk

GISRUK 2001 Local Organising Committee

David Kidner (Co-Chair) Linus Mofor Nathan Thomas

GISRUK 2001 Sponsors

The Association for Geographic Information (AGI) Blackwell Publishers

Geo-Europe Ordnance Survey

The sociology of scientific knowledge developed a ‘strong program’ of researchers (Bloor,

GIS and Crime

• the method needed to be practical and simple to apply;

• Apply nearest neighbour analysis to test for clustering

Figure 4.1 Alleygate in-situ.

20 polygons—each polygon therefore representing an area protected from burglary by

2 A tile is simply a supply unit—for example a 1km×1km digital map sheet.

Figure 4.2 Police beat and Alleygated zone geography.

Table 4.2 Results in all ten buffer zones (109,654 properties).

• Short-term (tactical deployment);

Figure 5.1 Data preparation process in relation to ANN processing.

• Format conversion and integration;

• Data reduction, transformation and generalisation.

1. Transformation and scaling may include:

• Applying a mathematical function (e.g. logarithm or square) to an input;

2. Reduction of relevant data includes simple operations such as filtering combinations of

Figure 5.2 A Kohonen Self Organising Feature Map.

Figure 5.3 Using a KSOM to partition a dataset.

Figure 5.4 An Example Trained Kohonen Self Organising Feature Map.

Figure 5.5 An Example KSOM.

Figure 5.6 Example cluster mappings from input to output space.

• Multi-Layered Perceptron (MLP) Model (Chen 1997)

Figure 5.7 Interpreting the output from the Gamma Test.

Identify Cluster boundaries in KSOM

For each node in the KSOM

and Day is Friday

Figure 5.8 Proposed COPDAT Hybrid System.

GIS and Planning

Figure 6.8 Complicated client-server configuration required for ESRI ArcIMS.

Figure 6.11 Two-way spatial communication: compositional method.

1. public participation in final decision;

Present commercial Web-GIS software mainly enables one-way communication. The

Figure 7.2 Bottlenecks in access speed, adapted from Halabi, 1997.

Figure 7.3 Level of required interaction with application.

Figure 7.4 Panoramic image of Canary Wharf Square, London Docklands.

Figure 7.5 Landline derived building outlines in ArcView.

Figure 7.6 LIDAR-based City Models (http://www.globalgeodata.com)

Figure 7.7 Extrusion by average height derived from LIDAR in ArcView.

Figure 7.8 PAVAN output from MapInfo, illustrating Roof Morphology.

Figure 7.9 Canoma modelling stage 1.

Figure 7.10 Canoma modelling stage 2.

Figure 7.11 Canoma modelling stage 3.

Figure 7.12 Canoma modelling stage 4.